A Systematic Review of Cochrane Anticoagulation Reviews

See the Introduction

See also Results section: Categories of Methodological Errors and Biases #4 -- RCTs are too small to evaluate risk for HITT and observational studies not evaluated for HITT.

My published commentary, "Evidence-based Medicine and the Cochrane Collaboration on Trial"^[68] comprehensively details the points of contention in this review:

The primary efficacy endpoint combines asymptomatic and symptomatic VTE. This means that it relies primarily on the more frequent surrogate endpoint asymptomatic DVT.

Only a purely clinical endpoint should be used to measure efficacy.

Even symptomatic VTE would be inappropriate as the sole primary outcome measure to determine recommendations, since it does not include adverse events (ie, bleeding).

Combining symptomatic and fatal VTE and fatal complications of the LMWH (major bleeding and HITT) would be the most clinically relevant primary endpoint.

Because of excluding observational studies from consideration in the safety analysis, the bleeding complications will very likely be understated.

Rebound hypercoagulability was not assessed.

The review was published 14 months after my critique of the protocol was received and acknowledged. My points were not addressed.

Authors' conclusions: Implications for Practice section, "Patients receiving venous femoropopliteal, infragenicular bypasses seem to benefit more from VKA than from ASA." Comment: No differences in clinical endpoints were seen in placebo-controlled RCTs while both drugs were associated with unacceptable rates of fatal bleeding (VKAs, 16 [1.2%], aspirin/dipyridamole, 12 [0.9%]).

Rebound hypercoagulability was not assessed.

See also Results section: Categories of Methodological Errors and Biases #11 -- A patented, expensive treatment (not the subject of the review) that is not evidence-based to work is endorsed in the Implications for Practice section.

See Results section: Categories of Methodological Errors and Biases #5 -- Rebound hypercoagulability was not assessed by collecting study data for at least 2 months after anticoagulant withdrawal in most patients

This is a meta-analysis of non-inferiority trials in which neither the control group (shorter-duration VKA treatment) nor the experimental group (longer-duration VKA treatment) is evidence-based to be safe and effective.

Implications for Practice: " ... (We) recommend that injectable anticoagulants should be reserved for patients at higher risk of thrombosis and those with contraindications to physical methods and/or aspirin (Scottish Intercollegiate Guidelines Network 2002)".

My comment: There is no evidence basis for this statement, although it is drawn from another guidelines-producing group (Scottish Intercollegiate Guidelines Network [SIGN]).

Authors reply: "....as we state in the review, the source for the recommendation you refer to is the Hip Fracture Guideline Development Group. Prevention and management of hip fracture in older people. A national clinical guideline (No. 56). Edinburgh: Scottish Intercollegiate Guidelines Network (www.sign.ac.uk), 2002."

My response: Cochrane systematic reviews are supposed to base conclusions and recommendations on RCTs and other published literature, not non-evidence-based opinions of other authorities.

See also Results section: Categories of Methodological Errors and Biases #9 -- Intracranial bleeding was not separately included in the primary or secondary endpoints.

See Results section: Categories of Methodological Errors and Biases #5 -- Rebound hypercoagulability was not assessed.

Because observational, population-based, and/or case-control studies were excluded from consideration in the safety analysis, the bleeding complications were very likely understated.

The primary endpoints (DVTs and PE demonstrated on venograms and noninvasive scans) are surrogates rather than clinical endpoints. Asymptomatic VTE does not correlate with fatal PE.

Standard prophylactic anticoagulation (typically 7-10 days: the control group) following total hip or knee replacement or hip fracture repair provides no proven survival benefit.

Anticoagulation does cause significant risks for bleeding and rebound hypercoagulation events.

Rebound hypercoagulability events were missed in the extended duration of anticoagulant group (about 30 days of low-dose heparins or other drugs) of this review, thus biasing the findings.

In a meta-analysis of RCTs involving VTE prophylaxis of total hip replacement patients, the crude risks for clinically important bleeding (usually wound hematoma) were 0% for compression stockings, 0.3% for controls, 0.4% for aspirin, 1.8% for LMWH, and 2.6% for unfractionated heparin.^[75]

Because of the seriousness of wound hematomas and the frequently associated infections in these patients, SIGN changed the hip fracture and hip and knee replacement surgical VTE prophylaxis guidelines from advising anticoagulants to advising aspirin.^[76]

The primary endpoints, DVTs on venograms and noninvasive scans, are surrogate rather than clinical endpoints. Asymptomatic DVTs do not correlate with fatal PE.

Rebound hypercoagulability events were not assessed.

Despite the absence of benefit of LMWH for any of the major clinical endpoints (death, vascular death, disability), the Implications for Practice appear to justify the use of LMWH and heparinoids in some cases of acute ischemic stroke on the basis of the nonstatistically significant superiority of LMWH and heparinoids over UH in DVT and PE.

The Implications for Practice section should read, "Do not use any anticoagulants in patients with acute ischemic stroke."

Likewise, the Implications for Research section should say, "It would be unethical to conduct further trials of anticoagulants in patients with acute ischemic stroke."

Authors' reply: "The first version of the review was prepared at a time when there was uncertainty about the overall effects of heparins in acute ischemic stroke. Since then, as the commenter states, evidence has emerged which shows that there is no net benefit from the immediate anticoagulation in this setting. However, as is often the case, this evidence did not alter the beliefs of some clinicians, and hence, heparin is still used in some countries, in certain types of patient with acute ischemic stroke, for specific reasons and especially for prevention of venous thromboembolism. This review is now based on the premise that if a clinician plans to treat a patient for some special reason with some form of heparin then the choice of agent should be evidence based."

My rebuttal: I question that premise as a basis for a Cochrane Review. Heparin is dangerous. If it is not evidence-based to improve clinical outcomes, don't prescribe it.

The authors correctly point out that there was no statistically significant increase in major extracranial or intracranial hemorrhage (OR, 1.82 [95% CI, 0.87-3.81] and OR, 1.46 [95% CI, 0.72-2.97], respectively). However, when these 2 categories of major hemorrhage are combined, there is a borderline statistically significant increase in major bleeding (OR, 1.63 [95% CI, 0.98-2.72]; P< .06), which represents 6.2 additional major bleeding episodes per 1000 patients (95% CI, - 0.1 to 16.8) by adding the low-dose heparin.

Risk for bleeding in clinical practice is probably greater than in closely monitored RCTs.

Since adding low-dose heparin to aspirin does not improve clinical outcomes and does add to the bleeding risk, the Implications for Research section should say that further trials of any anticoagulants in acute ischemic stroke would be unethical and any studies under way should be stopped.

Also see Results section: Categories of Methodological Errors and Biases example #9 -- Fatal bleeding was not separately included in the primary or secondary endpoints.

This review of anticoagulant trials, including over 22,000 patients with acute ischemic stroke, found no net benefit with the use of any anticoagulant.

The Implications for Practice section should say that anticoagulants should be contraindicated in patients with acute ischemic stroke.

The Implications for Research section should say that further trials of anticoagulants in acute ischemic stroke would be unethical.

This Cochrane Review should be updated to indicate that VKA treatment should be contraindicated in patients with acute ischemic stroke.

In 2005, the Warfarin and Aspirin for Symptomatic Intracranial Arterial Stenosis (WASID) trial was terminated early because of the high incidence of death in the warfarin group (9.7% in the warfarin group vs 4.3% in the aspirin group; hazard ratio for aspirin relative to warfarin, 0.46; 95% CI, 0.23 to 0.90; P=.02).^[53] The authors concluded, "Aspirin should be used in preference to warfarin for patients with intracranial arterial stenosis."

No further anticoagulant drug research is warranted in this patient group.

The review concluded that aspirin significantly decreases death and dependency in stroke patients.

The Implications for Research section states, ". . . There is also a case for further trials of low dose subcutaneous heparin (or low dose low molecular weight heparin) plus aspirin versus aspirin alone in the prevention of post-stroke deep vein thrombosis and pulmonary embolism, and in reducing neurological disability from the original or recurrent strokes. Such trials would need to include several tens of thousands of patients."

My comment: This review should be updated to reflect that, given the lack of efficacy of anticoagulants in multiple trials and high bleeding risk in stroke patients (see #16 above), further trials with low-dose heparin or low-dose LMWH would be unethical.

Authors' conclusions in the "Implications for Practice" section: "The thromboprophylactic benefits of aspirin in patients with heart failure who are in sinus rhythm are not convincing, but may be detrimental in view of the reduction of ACE Inhibitor benefit and other side effects. It should be emphasized that the evidence for a potential interaction between ACE inhibitors and antiplatelet agents is mainly based on nonrandomized studies, in light of the quality of the studies from which the data are derived and the post hoc nature of the study analyses, the possible interaction should be interpreted with caution."

My comment: The implications for practice are obfuscation -- do not prescribe antiplatelet agents!

The abstract of the WASH trial (the only trial reviewed in this review) concludes, "The benefits of warfarin for patients with heart failure in sinus rhythm have not been established. Antithrombotic therapy in patients with heart failure is not evidence based but commonly contributes to polypharmacy."

Whereas, VKA therapy for heart failure has been tested and is not evidence-based to work, this review gives the false impression that the jury is out and that such treatment may be reasonable.

Adverse events due to rebound hypercoagulability were not monitored.

The Implications for Practice section focuses on beneficial therapies for subgroups of patients: (1) patients with venous graphs and (2) patients with artificial graphs. These subgroups were not defined in the protocol, so post hoc statistical analyses of outcomes of these subgroups should not be emphasized in the Results or Conclusions.

In several places, the Cochrane Handbook warns against problems with post hoc analyses -- for example, page 63: "Post-hoc questions are more susceptible to bias than those asked a priori, and data-driven questions can generate false conclusions based on spurious results."

Adverse events due to rebound hypercoagulability were not monitored.

No benefit was shown in any clinical parameter for claudication patients.

The anticoagulants increased bleeding, including at least 1 case of fatal bleeding.

Rebound hypercoagulation adverse events were not assessed.

Further research of anticoagulants for claudication is inappropriate and unethical.

A major conclusion of the review: Oral anticoagulation prolongs life in people with extensive small-cell lung cancer.

My comment: The survival in small-cell lung cancer was based on a post hoc statistical analysis.

Author reply: It was an a priori determined subgroup, because the protocol states, "We will try to explain heterogeneity, if present, by conducting subgroup analyses. Our a priori factors to explain heterogeneity are the characteristics of participants... and we list 'type/location of cancer' as a patient characteristic."

My rebuttal: If all cell types of cancer were designated for subgroup analysis by this catch-all statement, then adjusting for multiple comparisons (eg, Bonferroni adjustment) would render the P value for the benefit of extensive small-cell lung cancer with oral anticoagulants insignificant.

The authors' reply to my rebuttal is pending.

The authors' concluded, "The evidence suggests that anticoagulants are beneficial, without serious adverse effects, for people with NRAF and recent cerebral ischemia."

Although neither of these small RCTs with careful selection of relatively young patients had intracerebral bleeding (ie, for the 500-plus subject-years of follow-up on anticoagulants), about 12 such bleeding events would be expected in a general practice setting with a similar number of patient-years on anticoagulants.^[43]

The relatively younger age of patients in these RCTs means that efficacy in preventing vascular events will also be lower than in general practice.

Adverse events due to rebound hypercoagulability were not monitored.

Authors' conclusion: "The evidence from two trials suggests that anticoagulant therapy is superior to antiplatelet therapy for the prevention of stroke in people with NRAF and recent non-disabling stroke or TIA. The risk of extracranial bleeding was higher with anticoagulant therapy than with antiplatelet therapy."

See Results section: Categories of Methodological Errors and Biases #6 -- Because of excluding observational, population-based, and/or case-control studies from consideration in the safety analysis, the bleeding complications were very likely understated.

Fatal and intracranial bleeding were not reported

Adverse events due to rebound hypercoagulability were not monitored.

Authors' plain language summary: "Home treatment of deep vein thrombosis (DVT) is cost effective and preferred by people with DVT."

Hospital treatment of DVT with anticoagulants (the control group) is not evidence-based to be safe or effective.^{[1,68,87,88,89]} Therefore, home anticoagulation treatment for DVT (the experimental group) is not proven to be safe and effective.

The relevant issue: As it stands, the guidelines for anticoagulation during pregnancy and the postpartum period by the American College of Chest Physicians^[91] and the Royal College of Obstetricians and Gynaecologists^[92] are arguably the standard of care in the United States and United Kingdom, respectively. Despite the lack of RCT evidence, these opinion-based guidelines recommend anticoagulants in many instances, and they can be referenced in medicolegal cases.

This review appropriately concludes that anticoagulant thromboprophylaxis during pregnancy is not evidence-based to be safe and effective. However, the implications for practice and research do not go far enough.

Since anticoagulation carries risks for bleeding, osteoporosis, and fetal deformity, the appropriate implication for practice would be that thromboprophylaxis with anticoagulants should not be used outside of an RCT.

The Implications for Research section should state that any RCTs of anticoagulation in pregnant women should be placebo controlled.

The issue: Despite the lack of RCT evidence, opinion-based treatment guidelines for pregnant women with thrombophilia by the Royal College of Obstetricians and Gynaecologists and the American College of Chest Physicians include thromboprophylaxis.^[91,92]

My suggestion: "Given the risks of heparin, the implications for practice for the review should say that heparin should not be used for pregnant women with acquired or inherited thrombophilias outside of a placebo or antiplatelet agent controlled RCT."

The lead author agreed to make the change with the next review update.

Authors' Implications for Practice: "Both unfractionated heparin and low molecular weight heparin can be used as effective prophylaxis against postoperative thromboembolic complications after colorectal surgery."

My comment: These conclusions are based on surrogate endpoints (eg, venograms which mostly show asymptomatic DVTs) rather than clinical endpoints (eg, death, fatal PE, symptomatic VTE).

Authors' reply: "We must not forget that the efficacy of heparin prophylaxis against fatal Postoperative PE was established in a methodologically correctly performed trial more than 30 years ago (International Multicentre Trial).^[93] Taken these facts in mind I still mean that the use of postoperative prophylaxis is evidence based, and for comparing different methods the use of venography for detecting subclinical DVT as endpoint is scientifically acceptable."

My rebuttal: While methodologically correct for its time, the International Multicentre Trial preceded the era of early mobilization and mechanical VTE prophylaxis. The follow-up period was only until discharge from hospital, so the deaths caused by rebound hypercoagulation after heparin was stopped were missed. Likewise, in this review adverse events due to rebound hypercoagulability were not monitored.

At my suggestion, the authors completed this review 6 years after they published the protocol. The efficacy and safety of LWMH for acute coronary syndromes depends on the evidence basis of heparin vs placebo for acute coronary syndromes. See correspondence for review #30 below

See Results section #5 -- Rebound hypercoagulability was not assessed by collecting study data for at least 24 hours after heparin withdrawal

The appropriateness of this review is predicated on the efficacy of heparins when compared with placebo where all patients are also treated with aspirin. The authors maintain that their recently published Cochrane Review of this topic establishes the benefit of heparin plus aspirin vs aspirin alone. I dispute that contention. (See my feedback letter for #29 above.)

If neither LMWH nor UH is evidence-based to be beneficial for people with acute coronary syndromes, comparing them is inappropriate.

With additional antiplatelet agents and invasive procedures in recent years, heparins are significantly more hazardous that when the trials in this review were done.

Heparins should not be used for acute coronary syndromes outside of placebo-controlled RCTs.

From the abstract: "The objective of this systematic review is to determine the effectiveness and safety of heparin compared to all other modalities in the treatment of patients with STEMI."

Rebound hypercoagulability-related adverse events should be monitored.

Fatal bleeding and/or intracranial bleeding should be separately included in the primary or secondary endpoints.

RCTs, like those included in this review, report low bleeding complication rates with VKA treatment (eg, major bleeding/yr = 2.2%; intracranial bleeding/yr =0.5%; fatal bleeding/yr = 0.4% [n = 1939]^[96]). In contrast, a recent series of consecutive patients started on warfarin for atrial fibrillation reported higher rates of bleeding complications (major bleeding/yr, 7.2%, intracranial bleeding/yr, 2.5%, fatal bleeding/yr, 0.8% [n = 472]^[43]).

Rebound hypercoagulability-related adverse events were not monitored.

See also Results section: Categories of Methodological Errors and Biases #10 -- Conclusions of review were based on data from RCTs with patients who are not representative of those in general clinical practice.

Plain-language summary: "Insufficient evidence to say if anticoagulants help women with recurrent pregnancy loss without antiphospholipid syndrome"

Authors' recommendation: Large, randomized, placebo-controlled trials

Given the risks of heparin and the potential for harm, if tens of thousands of women have heparin treatment during pregnancy, the main endpoint in the recommended RCT (anticoagulant vs placebo) should be a live, healthy baby in 2 or 3 pregnancies rather than in a single pregnancy.

Rebound hypercoagulability-related adverse events should be monitored.

The bleeding complications will very likely be understated unless observational and/or population-based studies are included in the safety analysis.

Fatal bleeding and intracranial bleeding should be separately reported and added to the primary or secondary endpoints. Patients and physicians would be particularly interested in these safety measures.

This protocol assumes that those oral anticoagulants are evidence-based to benefit patients. However, the critiques of multiple Cochrane reviews in this paper question this assumption.

The authors implied that heparin is effective for CSVT, although the 2 small trials showed no statistically significant reduction in mortality.

The nonsignificant trend depends on a single small trial that was subject to early termination bias.

The authors of this CSVT review called heparin safe on the basis of only 40 cases (about 800 days of heparinization).

CNS bleeding is a common cause of morbidity and mortality in CSVT, so heparin is not safe.

Further placebo-controlled RCTs were inappropriately discouraged ("....patients and doctors may be reluctant to embark upon a new trial that includes a placebo group.").

Authors' conclusions: "No strong evidence was found to conclude thromboprophylaxis is effective to prevent thromboembolic events and safe, in people with unknown risk factors for thrombosis, undergoing knee arthroscopy."

From plain-language summary: "Adverse events were most common in the intervention group (i.e., anticoagulant prophylaxis). The most common complication was minor bleeding with a RR of 2.23 (range 0.99 to 4.99). The number needed to harm was 20."

From Implications for Research section: "Surveillance studies will be important to assess the detection of both adverse events and thrombotic events, with and without the use of LMWH."

Rebound hypercoagulability-related adverse events were not monitored.

The implication for practice should be that anticoagulant prophylaxis should not be used.

The implication for research should be that further RCTs with knee arthroscopy patients exposed to anticoagulants would be unethical.

Anticoagulant prophylaxis statistically significantly benefits surrogate endpoints of DVTs as shown by venograms and other tests, but it has not been shown to improve clinical endpoints (eg, fatal pulmonary emboli, mortality, symptomatic VTE).

The control group for this review (VKAs and LMWHs) is not evidence-based to be effective.

This proposed meta-analysis of non-inferiority trials comparing VKAs and LMWHs with direct thrombin inhibitors is not appropriate to prove efficacy or safety of direct thrombin inhibitors.

The protocol does not call for monitoring for rebound hypercoagulability-related adverse events.

Authors' conclusions: "The use of heparin-bonded catheters is a promising therapy but warrants further studies."

See Results section: Categories of Methodological Errors and Biases #3 -- Placebo or no treatment control was used when saline flushes are widely used and are safe and considered effective.

The primary endpoint is "Days of catheter patency (duration of patency of first catheter, in days)." The most appropriate clinical endpoint would seem to be occlusion of CVCs.

Given the risk for bleeding and HITT, a safety measurement should also be included in a composite primary endpoint.

Fatal bleeding and intracranial bleeding should have been designated separately as primary or secondary endpoints because of the magnitude of these events.

Rebound hypercoagulability-related adverse events should have been monitored.

See Results section of this review: Categories of Methodological Errors and Biases #3 -- Placebo or no treatment control was used when saline flushes are commonly employed as an anticlotting measure clinically; saline flushes are safe and considered effective in preventing CVC-related thrombosis in children.

The primary outcome measure, "CVC-related thrombosis (along the length of, or at the tip of, the catheter) as determined by either color-coded Doppler ultrasonography or contrast venography," is a surrogate endpoint rather than a clinically relevant efficacy measurement.

Instead, the secondary endpoint, "occlusion of CVC (defined as inability to infuse fluids through the catheter due to blockage)," should be made the primary endpoint.

Because RCTs for this indication will probably be too small to evaluate risk for HITT and hemorrhage, observational studies should also be included in the safety analysis.

Fatal bleeding and intracranial bleeding should be designated separately in the primary or secondary endpoints because of the particular importance of these events.

Rebound hypercoagulability-related adverse events should be monitored.

The primary endpoint (symptomatic plus asymptomatic DVT or symptomatic PE) is inappropriate, because asymptomatic DVT (a surrogate endpoint) will dominate. There is no evidence that asymptomatic DVT correlates with clinically important outcomes.

See Results section: Categories of Methodological Errors and Biases #8 -- The primary endpoint should include major bleeding rather than just VTE.

Fatal and intracranial bleeding should be separately designated as primary or secondary endpoints because of the importance of these events.

To properly assess adverse events, including major bleeding and HITT, RCTs should be supplemented with large prospective or retrospective observational studies.

Rebound hypercoagulability-related adverse events should be monitored.

Authors' conclusions: Adjusted-dose warfarin and related oral anticoagulants reduce stroke, disabling stroke, and other major vascular events for those with nonvalvular atrial fibrillation by about one third when compared with antiplatelet therapy.

Given the risk for adverse effects with VKAs, major bleeding should be included in a composite primary outcome measure.

See Results section: Categories of Methodological Errors and Biases #10 -- Conclusions of the review were based on data from RCTs with younger, healthier, and more compliant patients than are typically seen in general clinical practice.

Representative safety data will not come from RCTs with anticoagulation researchers carefully monitoring the international normalized ratios of highly selected patients. Large observational studies need to be 1 way to assess the rates of major and fatal bleeding. For example, in this review VKA-related major bleeding (1.6%/yr) and intracranial bleeding (0.45%/yr) were considerably less frequent than in a recent observational study (7.2%/yr. and 2.5%/yr, respectively).^[43]

Rebound hypercoagulability-related adverse events were not monitored.

Authors' conclusion: "This review suggests a survival benefit of heparin in cancer patients in general, and in patients with small cell lung cancer in particular."

My comment: (See Results section: Categories of Methodological Errors and Biases #13) -- Biases exist in the selection of trials for inclusion in the review: 12 RCTs were eligible but only 5 were included, so the potentially biased selection of trials makes interpretation of the data "favorable to heparin" highly suspect.

Authors' reply: "We agree that it would have been ideal to include data from the trials published as abstracts. . . Interpretation of the findings of this review is limited by the moderate heterogeneity. . . The interpretation of findings is also limited by not including data from the 7 trials published as abstracts only."

These qualifications were not added to the abstract or plain-language summary where people without a subscription to the Cochrane Library could read it.

Authors' conclusion: "Combined unfractionated heparin and aspirin may reduce pregnancy loss by 54%."

Since all 3 RCTs of aspirin vs placebo showed no benefit, aspirin should be avoided and excluded from future RCTs.

Even if adding heparin to aspirin does increase the chances of live birth in a single pregnancy in this situation from 30/70 (43%) to 52/70 (74%), it is not significantly better that the chance of a live birth in 2 pregnancies with aspirin (or placebo) alone (ie, 1 - [0.57 x 0.57] = 68%). Given the risks of anticoagulants, the endpoint of any future RCT should be a healthy baby in 2 or 3 pregnancies rather than 1 pregnancy.

Safety (eg, osteopenia, bleeding, HITT, fetal abnormalities) cannot be determined with these small trials (n = 140).

The conclusion that heparin added to aspirin increases live births and should become the standard of care in women with recurrent miscarriages associated with antiphospholipid antibodies is premature.

In women with recurrent miscarriages, anticoagulants should not be given outside of RCTs.

Authors' conclusion: "Adding antiplatelet therapy, either dipyridamole or low-dose aspirin, to oral anticoagulation decreases the risk of systemic embolism or death among patients with prosthetic heart valves. The risk of major bleeding is increased with antiplatelet therapy."

The use of oral anticoagulation itself is not evidence-based to reduce morbidity and/or mortality in patients with prosthetic valves. It is based on anecdotal reports and historical precedent, not RCTs.

Even without the addition of aspirin, at least 1% per year of patients taking oral anticoagulants experience fatal bleeding each year.

Because adding antiplatelet drugs to oral anticoagulants substantially reduced the risk for thromboembolic events and mortality (61% and 45% respectively), we should have a non-inferiority RCT comparing aspirin alone with aspirin plus a VKA. As a prelude to such a definitive RCT, I am coauthoring a Cochrane protocol titled, "Anticoagulation for prevention of cardiovascular disease in people with prosthetic heart valves."

Authors' conclusion: "There were no randomized trials comparing either anticoagulants or antiplatelet drugs with control. There is, therefore, no evidence to support their routine use for the treatment of extracranial internal carotid artery dissection.... We suggest that a randomized trial including at least 1400 patients in each treatment group (oral anticoagulants and antiplatelet drugs) with this condition is clearly needed."

Antiplatelet agents -- the control group for the primary endpoint comparison in this review (antiplatelet agents vs anticoagulants) -- are not evidence-based to reduce death or dependence in people with stroke from carotid artery dissection.

However, aspirin and other antiplatelet agents are a potential cause of serious bleeding.

Consequently, the Implications for Research section should state that a large RCT of aspirin vs placebo is indicated.

Oral anticoagulants should be avoided in carotid artery dissection on the basis of the poor outcome in patient with acute ischemic stroke (Review #16 above).

Primary objective: To assess the effectiveness of heparin for prevention of catheter related thrombosis.

Main results: ...." There was no statistically significant difference in the risk of thrombosis (RR 0.93, 95% CI 0.58, 1.51)" . . . One of 4 secondary endpoints (catheter occlusion) statistically significantly favored heparin.

Authors' conclusion: "Evidence from this systematic review support the prophylactic use of heparin for PCVC in neonates at a dose of 0.5 IU/kg/hr."

For this review, the control was placebo or no treatment. A considerable amount of literature attests to the safety and efficacy of saline flushes.^[27,28,108] Consequently, saline flushes should have been the control.

Safety data from larger observational and case-control studies should have been included to evaluate these potential complications. For example, a case-control study involving 66 neonates with intraventricular bleeding found an increased risk for this devastating complication with heparin use (OR, 3.9 [95% CI, 1.4-11]), after adjusting for other risk factors.^[109]

Because there was no benefit with heparin for the primary endpoint and considerable risk for devastating intracranial or other bleeding, heparin should not be used.

Main results: "......Five studies reported data on the duration of use of the first catheter. Two of these studies found no statistically significant effect of heparin; two studies showed a statistically significant increase and one study showed a statistically significant decrease in the duration of PIV catheter use in the heparin group."

Authors' conclusion, Implications for Practice section: ".....Because of clinical heterogeneity and heterogeneity in treatment effect, recommendations for heparin use in neonates with PIV catheters cannot be made."

Authors' conclusions, Implications for Research section: "Further research on the effectiveness, the optimal dose, and the safety of heparin is required."

Premature infants are particularly prone to intracranial hemorrhage.

The small number of patients in the 2 RCTs reporting the incidence of intracranial hemorrhage (n = 322) gives no assurance that adding heparin to catheter infusions will not precipitate this dreaded complication.

Given that this meta-analysis found no benefit of heparin on first catheter occlusion (n = 485; OR, 1.00 [95% CI, 0.85-1.16]) or any clinical endpoint, further RCTs of heparin in neonates for this indication are not warranted.

Heparin should be avoided in premature infants.

Authors' conclusions. "Ticlopidine plus aspirin after coronary stenting is effective in reducing the risk of the revascularization, non fatal myocardial infarction and bleeding complications when compared with oral anticoagulants...."

My comment: In this meta-analysis of non-inferiority RCTs, the control group, aspirin plus a VKA, has never been shown to improve clinical outcomes for people with coronary stenting. Therefore, it cannot determine the effectiveness and safety of the use of ticlopidine plus aspirin after coronary stenting.

The lead author's reply: "The aim of the review was to compare ASA plus ticlopidine versus ASA plus vitamin K antagonists after coronary stenting and not the efficacy and safety of ASA plus ticlopidine per se."

Safety data concerning VKAs from previous RCTs, observational studies, and population-based studies should supplement the data from the 2 small RCTs in this review.

Given the lack of efficacy of warfarin in the RCTs of the review and the likely underestimate of the bleeding risk and unknown risk for rebound hypercoagulation, further RCTs of VKA treatment to reduce vascular events in people with hypertension would be unethical.

The authors rebutted my comments challenging the efficacy of platelet inhibitors in secondary prevention of patients with hypertension.

Regarding my point that further RCTs of VKA prophylaxis in hypertension, the authors said, "There is inadequate data on warfarin in hypertension per se."

Authors' general observation: "We are not sure what the author [me] refers to being 'evidence-based,' but from his comment we believe that s/he has a different understanding of evidence-based. Being 'evidence-based' means following the 'best evidence' whatever that may be. Evidence can be high quality or lower quality, it can be direct or indirect."

I disagree. By that definition, every test or treatment in medicine is evidence-based.

My comment: In this proposed meta-analysis (published as a review after my feedback letter) of RCTs comparing LMWHs with oral anticoagulants, the control group, VKAs, is not evidence-based to be safe and effective.

Authors' reply: None

My comment: Observational studies were not included in the review to evaluate for HITT and other safety endpoints.

Authors' reply: "We will add both HIT and HITT to the list of outcomes to assess." ....."We have designed the systematic review to include data from RCTs only. We will discuss in the final review the limitation of not including harm data from observational study."

My response: These endpoints were not systematically evaluated in the primary RCTs.

My comment: Rebound hypercoagulability-related adverse events were not monitored.

Authors' reply: "Dr. Cundiff raises the question, whether the incidence of thrombotic events increases within two months of discontinuing heparin therapy. We will investigate this hypothesis if the studies we include in our review report the relevant data."

My response: No RCTs included these data.

My comment: RCTs included in this meta-analysis were subject to selection bias because trials published only as abstracts were not included.

Authors' reply: "Dr. Cundiff describes publication bias, which we will investigate as recommended by the Cochrane Collaboration."

My comment: Besides major bleeding, the primary or secondary endpoints should include fatal bleeding and intracranial bleeding because of the particular importance to patients and clinicians of these events.

Authors' reply: The 2 proposed outcomes are accounted for by the outcomes listed in our protocol. Fatal bleeding events would be counted as deaths. Intracranial bleeding events would be counted as major bleeding events.

My response: Iatrogenic deaths should be counted separately from cancer deaths, and intracranial bleeding rates (fatal and nonfatal) are of special interest to patients and doctors.

None of the proposed primary outcome measures relate to safety concerns with anticoagulant drugs.

The primary endpoint, which determines the main conclusions of the review, should include efficacy and safety measures.

Concerning anticoagulants, given the relatively long natural history of the nephrotic syndrome, the slowed metabolism of anticoagulants with renal impairment, the significant bleeding risk, and the paucity of anecdotal evidence of efficacy, the ethics of conducting RCTs with heparins or VKAs should be questioned.

Bleeding (major or fatal bleeding)

Anticoagulant-associated intracranial hemorrhage

Heparin-induced thrombocytopenia with thrombosis

None of the proposed primary outcome measures relate to safety concerns with anticoagulant drugs.

The primary endpoint, which determines the main conclusions of the review, should include efficacy and safety measures.

Given the relatively long natural history of IgA nephropathy, the slowed metabolism of anticoagulants with renal impairment, the significant bleeding risk, and the paucity of anecdotal evidence of efficacy, the ethics of conducting RCTs with any anticoagulant for any clinical endpoint should be questioned.

See the Introduction

See also Results section: Categories of Methodological Errors and Biases #4 -- RCTs are too small to evaluate risk for HITT and observational studies not evaluated for HITT.

My published commentary, "Evidence-based Medicine and the Cochrane Collaboration on Trial"^[68] comprehensively details the points of contention in this review:

The primary efficacy endpoint combines asymptomatic and symptomatic VTE. This means that it relies primarily on the more frequent surrogate endpoint asymptomatic DVT.

Only a purely clinical endpoint should be used to measure efficacy.

Even symptomatic VTE would be inappropriate as the sole primary outcome measure to determine recommendations, since it does not include adverse events (ie, bleeding).

Combining symptomatic and fatal VTE and fatal complications of the LMWH (major bleeding and HITT) would be the most clinically relevant primary endpoint.

Because of excluding observational studies from consideration in the safety analysis, the bleeding complications will very likely be understated.

Rebound hypercoagulability was not assessed.

The review was published 14 months after my critique of the protocol was received and acknowledged. My points were not addressed.

Authors' conclusions: Implications for Practice section, "Patients receiving venous femoropopliteal, infragenicular bypasses seem to benefit more from VKA than from ASA." Comment: No differences in clinical endpoints were seen in placebo-controlled RCTs while both drugs were associated with unacceptable rates of fatal bleeding (VKAs, 16 [1.2%], aspirin/dipyridamole, 12 [0.9%]).

Rebound hypercoagulability was not assessed.

See also Results section: Categories of Methodological Errors and Biases #11 -- A patented, expensive treatment (not the subject of the review) that is not evidence-based to work is endorsed in the Implications for Practice section.

See Results section: Categories of Methodological Errors and Biases #5 -- Rebound hypercoagulability was not assessed by collecting study data for at least 2 months after anticoagulant withdrawal in most patients

This is a meta-analysis of non-inferiority trials in which neither the control group (shorter-duration VKA treatment) nor the experimental group (longer-duration VKA treatment) is evidence-based to be safe and effective.

Implications for Practice: " ... (We) recommend that injectable anticoagulants should be reserved for patients at higher risk of thrombosis and those with contraindications to physical methods and/or aspirin (Scottish Intercollegiate Guidelines Network 2002)".

My comment: There is no evidence basis for this statement, although it is drawn from another guidelines-producing group (Scottish Intercollegiate Guidelines Network [SIGN]).

Authors reply: "....as we state in the review, the source for the recommendation you refer to is the Hip Fracture Guideline Development Group. Prevention and management of hip fracture in older people. A national clinical guideline (No. 56). Edinburgh: Scottish Intercollegiate Guidelines Network (www.sign.ac.uk), 2002."

My response: Cochrane systematic reviews are supposed to base conclusions and recommendations on RCTs and other published literature, not non-evidence-based opinions of other authorities.

See also Results section: Categories of Methodological Errors and Biases #9 -- Intracranial bleeding was not separately included in the primary or secondary endpoints.

See Results section: Categories of Methodological Errors and Biases #5 -- Rebound hypercoagulability was not assessed.

Because observational, population-based, and/or case-control studies were excluded from consideration in the safety analysis, the bleeding complications were very likely understated.

The primary endpoints (DVTs and PE demonstrated on venograms and noninvasive scans) are surrogates rather than clinical endpoints. Asymptomatic VTE does not correlate with fatal PE.

Standard prophylactic anticoagulation (typically 7-10 days: the control group) following total hip or knee replacement or hip fracture repair provides no proven survival benefit.

Anticoagulation does cause significant risks for bleeding and rebound hypercoagulation events.

Rebound hypercoagulability events were missed in the extended duration of anticoagulant group (about 30 days of low-dose heparins or other drugs) of this review, thus biasing the findings.

In a meta-analysis of RCTs involving VTE prophylaxis of total hip replacement patients, the crude risks for clinically important bleeding (usually wound hematoma) were 0% for compression stockings, 0.3% for controls, 0.4% for aspirin, 1.8% for LMWH, and 2.6% for unfractionated heparin.^[75]

Because of the seriousness of wound hematomas and the frequently associated infections in these patients, SIGN changed the hip fracture and hip and knee replacement surgical VTE prophylaxis guidelines from advising anticoagulants to advising aspirin.^[76]

The primary endpoints, DVTs on venograms and noninvasive scans, are surrogate rather than clinical endpoints. Asymptomatic DVTs do not correlate with fatal PE.

Rebound hypercoagulability events were not assessed.

Despite the absence of benefit of LMWH for any of the major clinical endpoints (death, vascular death, disability), the Implications for Practice appear to justify the use of LMWH and heparinoids in some cases of acute ischemic stroke on the basis of the nonstatistically significant superiority of LMWH and heparinoids over UH in DVT and PE.

The Implications for Practice section should read, "Do not use any anticoagulants in patients with acute ischemic stroke."

Likewise, the Implications for Research section should say, "It would be unethical to conduct further trials of anticoagulants in patients with acute ischemic stroke."

Authors' reply: "The first version of the review was prepared at a time when there was uncertainty about the overall effects of heparins in acute ischemic stroke. Since then, as the commenter states, evidence has emerged which shows that there is no net benefit from the immediate anticoagulation in this setting. However, as is often the case, this evidence did not alter the beliefs of some clinicians, and hence, heparin is still used in some countries, in certain types of patient with acute ischemic stroke, for specific reasons and especially for prevention of venous thromboembolism. This review is now based on the premise that if a clinician plans to treat a patient for some special reason with some form of heparin then the choice of agent should be evidence based."

My rebuttal: I question that premise as a basis for a Cochrane Review. Heparin is dangerous. If it is not evidence-based to improve clinical outcomes, don't prescribe it.

The authors correctly point out that there was no statistically significant increase in major extracranial or intracranial hemorrhage (OR, 1.82 [95% CI, 0.87-3.81] and OR, 1.46 [95% CI, 0.72-2.97], respectively). However, when these 2 categories of major hemorrhage are combined, there is a borderline statistically significant increase in major bleeding (OR, 1.63 [95% CI, 0.98-2.72]; P< .06), which represents 6.2 additional major bleeding episodes per 1000 patients (95% CI, - 0.1 to 16.8) by adding the low-dose heparin.

Risk for bleeding in clinical practice is probably greater than in closely monitored RCTs.

Since adding low-dose heparin to aspirin does not improve clinical outcomes and does add to the bleeding risk, the Implications for Research section should say that further trials of any anticoagulants in acute ischemic stroke would be unethical and any studies under way should be stopped.

Also see Results section: Categories of Methodological Errors and Biases example #9 -- Fatal bleeding was not separately included in the primary or secondary endpoints.

This review of anticoagulant trials, including over 22,000 patients with acute ischemic stroke, found no net benefit with the use of any anticoagulant.

The Implications for Practice section should say that anticoagulants should be contraindicated in patients with acute ischemic stroke.

The Implications for Research section should say that further trials of anticoagulants in acute ischemic stroke would be unethical.

This Cochrane Review should be updated to indicate that VKA treatment should be contraindicated in patients with acute ischemic stroke.

In 2005, the Warfarin and Aspirin for Symptomatic Intracranial Arterial Stenosis (WASID) trial was terminated early because of the high incidence of death in the warfarin group (9.7% in the warfarin group vs 4.3% in the aspirin group; hazard ratio for aspirin relative to warfarin, 0.46; 95% CI, 0.23 to 0.90; P=.02).^[53] The authors concluded, "Aspirin should be used in preference to warfarin for patients with intracranial arterial stenosis."

No further anticoagulant drug research is warranted in this patient group.

The review concluded that aspirin significantly decreases death and dependency in stroke patients.

The Implications for Research section states, ". . . There is also a case for further trials of low dose subcutaneous heparin (or low dose low molecular weight heparin) plus aspirin versus aspirin alone in the prevention of post-stroke deep vein thrombosis and pulmonary embolism, and in reducing neurological disability from the original or recurrent strokes. Such trials would need to include several tens of thousands of patients."

My comment: This review should be updated to reflect that, given the lack of efficacy of anticoagulants in multiple trials and high bleeding risk in stroke patients (see #16 above), further trials with low-dose heparin or low-dose LMWH would be unethical.

Authors' conclusions in the "Implications for Practice" section: "The thromboprophylactic benefits of aspirin in patients with heart failure who are in sinus rhythm are not convincing, but may be detrimental in view of the reduction of ACE Inhibitor benefit and other side effects. It should be emphasized that the evidence for a potential interaction between ACE inhibitors and antiplatelet agents is mainly based on nonrandomized studies, in light of the quality of the studies from which the data are derived and the post hoc nature of the study analyses, the possible interaction should be interpreted with caution."

My comment: The implications for practice are obfuscation -- do not prescribe antiplatelet agents!

The abstract of the WASH trial (the only trial reviewed in this review) concludes, "The benefits of warfarin for patients with heart failure in sinus rhythm have not been established. Antithrombotic therapy in patients with heart failure is not evidence based but commonly contributes to polypharmacy."

Whereas, VKA therapy for heart failure has been tested and is not evidence-based to work, this review gives the false impression that the jury is out and that such treatment may be reasonable.

Adverse events due to rebound hypercoagulability were not monitored.

The Implications for Practice section focuses on beneficial therapies for subgroups of patients: (1) patients with venous graphs and (2) patients with artificial graphs. These subgroups were not defined in the protocol, so post hoc statistical analyses of outcomes of these subgroups should not be emphasized in the Results or Conclusions.

In several places, the Cochrane Handbook warns against problems with post hoc analyses -- for example, page 63: "Post-hoc questions are more susceptible to bias than those asked a priori, and data-driven questions can generate false conclusions based on spurious results."

Adverse events due to rebound hypercoagulability were not monitored.

No benefit was shown in any clinical parameter for claudication patients.

The anticoagulants increased bleeding, including at least 1 case of fatal bleeding.

Rebound hypercoagulation adverse events were not assessed.

Further research of anticoagulants for claudication is inappropriate and unethical.

A major conclusion of the review: Oral anticoagulation prolongs life in people with extensive small-cell lung cancer.

My comment: The survival in small-cell lung cancer was based on a post hoc statistical analysis.

Author reply: It was an a priori determined subgroup, because the protocol states, "We will try to explain heterogeneity, if present, by conducting subgroup analyses. Our a priori factors to explain heterogeneity are the characteristics of participants... and we list 'type/location of cancer' as a patient characteristic."

My rebuttal: If all cell types of cancer were designated for subgroup analysis by this catch-all statement, then adjusting for multiple comparisons (eg, Bonferroni adjustment) would render the P value for the benefit of extensive small-cell lung cancer with oral anticoagulants insignificant.

The authors' reply to my rebuttal is pending.

The authors' concluded, "The evidence suggests that anticoagulants are beneficial, without serious adverse effects, for people with NRAF and recent cerebral ischemia."

Although neither of these small RCTs with careful selection of relatively young patients had intracerebral bleeding (ie, for the 500-plus subject-years of follow-up on anticoagulants), about 12 such bleeding events would be expected in a general practice setting with a similar number of patient-years on anticoagulants.^[43]

The relatively younger age of patients in these RCTs means that efficacy in preventing vascular events will also be lower than in general practice.

Adverse events due to rebound hypercoagulability were not monitored.

Authors' conclusion: "The evidence from two trials suggests that anticoagulant therapy is superior to antiplatelet therapy for the prevention of stroke in people with NRAF and recent non-disabling stroke or TIA. The risk of extracranial bleeding was higher with anticoagulant therapy than with antiplatelet therapy."

See Results section: Categories of Methodological Errors and Biases #6 -- Because of excluding observational, population-based, and/or case-control studies from consideration in the safety analysis, the bleeding complications were very likely understated.

Fatal and intracranial bleeding were not reported

Adverse events due to rebound hypercoagulability were not monitored.

Authors' plain language summary: "Home treatment of deep vein thrombosis (DVT) is cost effective and preferred by people with DVT."

Hospital treatment of DVT with anticoagulants (the control group) is not evidence-based to be safe or effective.^{[1,68,87,88,89]} Therefore, home anticoagulation treatment for DVT (the experimental group) is not proven to be safe and effective.

The relevant issue: As it stands, the guidelines for anticoagulation during pregnancy and the postpartum period by the American College of Chest Physicians^[91] and the Royal College of Obstetricians and Gynaecologists^[92] are arguably the standard of care in the United States and United Kingdom, respectively. Despite the lack of RCT evidence, these opinion-based guidelines recommend anticoagulants in many instances, and they can be referenced in medicolegal cases.

This review appropriately concludes that anticoagulant thromboprophylaxis during pregnancy is not evidence-based to be safe and effective. However, the implications for practice and research do not go far enough.

Since anticoagulation carries risks for bleeding, osteoporosis, and fetal deformity, the appropriate implication for practice would be that thromboprophylaxis with anticoagulants should not be used outside of an RCT.

The Implications for Research section should state that any RCTs of anticoagulation in pregnant women should be placebo controlled.

The issue: Despite the lack of RCT evidence, opinion-based treatment guidelines for pregnant women with thrombophilia by the Royal College of Obstetricians and Gynaecologists and the American College of Chest Physicians include thromboprophylaxis.^[91,92]

My suggestion: "Given the risks of heparin, the implications for practice for the review should say that heparin should not be used for pregnant women with acquired or inherited thrombophilias outside of a placebo or antiplatelet agent controlled RCT."

The lead author agreed to make the change with the next review update.

Authors' Implications for Practice: "Both unfractionated heparin and low molecular weight heparin can be used as effective prophylaxis against postoperative thromboembolic complications after colorectal surgery."

My comment: These conclusions are based on surrogate endpoints (eg, venograms which mostly show asymptomatic DVTs) rather than clinical endpoints (eg, death, fatal PE, symptomatic VTE).

Authors' reply: "We must not forget that the efficacy of heparin prophylaxis against fatal Postoperative PE was established in a methodologically correctly performed trial more than 30 years ago (International Multicentre Trial).^[93] Taken these facts in mind I still mean that the use of postoperative prophylaxis is evidence based, and for comparing different methods the use of venography for detecting subclinical DVT as endpoint is scientifically acceptable."

My rebuttal: While methodologically correct for its time, the International Multicentre Trial preceded the era of early mobilization and mechanical VTE prophylaxis. The follow-up period was only until discharge from hospital, so the deaths caused by rebound hypercoagulation after heparin was stopped were missed. Likewise, in this review adverse events due to rebound hypercoagulability were not monitored.

At my suggestion, the authors completed this review 6 years after they published the protocol. The efficacy and safety of LWMH for acute coronary syndromes depends on the evidence basis of heparin vs placebo for acute coronary syndromes. See correspondence for review #30 below

See Results section #5 -- Rebound hypercoagulability was not assessed by collecting study data for at least 24 hours after heparin withdrawal

The appropriateness of this review is predicated on the efficacy of heparins when compared with placebo where all patients are also treated with aspirin. The authors maintain that their recently published Cochrane Review of this topic establishes the benefit of heparin plus aspirin vs aspirin alone. I dispute that contention. (See my feedback letter for #29 above.)

If neither LMWH nor UH is evidence-based to be beneficial for people with acute coronary syndromes, comparing them is inappropriate.

With additional antiplatelet agents and invasive procedures in recent years, heparins are significantly more hazardous that when the trials in this review were done.

Heparins should not be used for acute coronary syndromes outside of placebo-controlled RCTs.

From the abstract: "The objective of this systematic review is to determine the effectiveness and safety of heparin compared to all other modalities in the treatment of patients with STEMI."

Rebound hypercoagulability-related adverse events should be monitored.

Fatal bleeding and/or intracranial bleeding should be separately included in the primary or secondary endpoints.

RCTs, like those included in this review, report low bleeding complication rates with VKA treatment (eg, major bleeding/yr = 2.2%; intracranial bleeding/yr =0.5%; fatal bleeding/yr = 0.4% [n = 1939]^[96]). In contrast, a recent series of consecutive patients started on warfarin for atrial fibrillation reported higher rates of bleeding complications (major bleeding/yr, 7.2%, intracranial bleeding/yr, 2.5%, fatal bleeding/yr, 0.8% [n = 472]^[43]).

Rebound hypercoagulability-related adverse events were not monitored.

See also Results section: Categories of Methodological Errors and Biases #10 -- Conclusions of review were based on data from RCTs with patients who are not representative of those in general clinical practice.

Plain-language summary: "Insufficient evidence to say if anticoagulants help women with recurrent pregnancy loss without antiphospholipid syndrome"

Authors' recommendation: Large, randomized, placebo-controlled trials

Given the risks of heparin and the potential for harm, if tens of thousands of women have heparin treatment during pregnancy, the main endpoint in the recommended RCT (anticoagulant vs placebo) should be a live, healthy baby in 2 or 3 pregnancies rather than in a single pregnancy.

Rebound hypercoagulability-related adverse events should be monitored.

The bleeding complications will very likely be understated unless observational and/or population-based studies are included in the safety analysis.

Fatal bleeding and intracranial bleeding should be separately reported and added to the primary or secondary endpoints. Patients and physicians would be particularly interested in these safety measures.

This protocol assumes that those oral anticoagulants are evidence-based to benefit patients. However, the critiques of multiple Cochrane reviews in this paper question this assumption.

The authors implied that heparin is effective for CSVT, although the 2 small trials showed no statistically significant reduction in mortality.

The nonsignificant trend depends on a single small trial that was subject to early termination bias.

The authors of this CSVT review called heparin safe on the basis of only 40 cases (about 800 days of heparinization).

CNS bleeding is a common cause of morbidity and mortality in CSVT, so heparin is not safe.

Further placebo-controlled RCTs were inappropriately discouraged ("....patients and doctors may be reluctant to embark upon a new trial that includes a placebo group.").

Authors' conclusions: "No strong evidence was found to conclude thromboprophylaxis is effective to prevent thromboembolic events and safe, in people with unknown risk factors for thrombosis, undergoing knee arthroscopy."

From plain-language summary: "Adverse events were most common in the intervention group (i.e., anticoagulant prophylaxis). The most common complication was minor bleeding with a RR of 2.23 (range 0.99 to 4.99). The number needed to harm was 20."

From Implications for Research section: "Surveillance studies will be important to assess the detection of both adverse events and thrombotic events, with and without the use of LMWH."

Rebound hypercoagulability-related adverse events were not monitored.

The implication for practice should be that anticoagulant prophylaxis should not be used.

The implication for research should be that further RCTs with knee arthroscopy patients exposed to anticoagulants would be unethical.

Anticoagulant prophylaxis statistically significantly benefits surrogate endpoints of DVTs as shown by venograms and other tests, but it has not been shown to improve clinical endpoints (eg, fatal pulmonary emboli, mortality, symptomatic VTE).

The control group for this review (VKAs and LMWHs) is not evidence-based to be effective.

This proposed meta-analysis of non-inferiority trials comparing VKAs and LMWHs with direct thrombin inhibitors is not appropriate to prove efficacy or safety of direct thrombin inhibitors.

The protocol does not call for monitoring for rebound hypercoagulability-related adverse events.

Authors' conclusions: "The use of heparin-bonded catheters is a promising therapy but warrants further studies."

See Results section: Categories of Methodological Errors and Biases #3 -- Placebo or no treatment control was used when saline flushes are widely used and are safe and considered effective.

The primary endpoint is "Days of catheter patency (duration of patency of first catheter, in days)." The most appropriate clinical endpoint would seem to be occlusion of CVCs.

Given the risk for bleeding and HITT, a safety measurement should also be included in a composite primary endpoint.

Fatal bleeding and intracranial bleeding should have been designated separately as primary or secondary endpoints because of the magnitude of these events.

Rebound hypercoagulability-related adverse events should have been monitored.

See Results section of this review: Categories of Methodological Errors and Biases #3 -- Placebo or no treatment control was used when saline flushes are commonly employed as an anticlotting measure clinically; saline flushes are safe and considered effective in preventing CVC-related thrombosis in children.

The primary outcome measure, "CVC-related thrombosis (along the length of, or at the tip of, the catheter) as determined by either color-coded Doppler ultrasonography or contrast venography," is a surrogate endpoint rather than a clinically relevant efficacy measurement.

Instead, the secondary endpoint, "occlusion of CVC (defined as inability to infuse fluids through the catheter due to blockage)," should be made the primary endpoint.

Because RCTs for this indication will probably be too small to evaluate risk for HITT and hemorrhage, observational studies should also be included in the safety analysis.

Fatal bleeding and intracranial bleeding should be designated separately in the primary or secondary endpoints because of the particular importance of these events.

Rebound hypercoagulability-related adverse events should be monitored.

The primary endpoint (symptomatic plus asymptomatic DVT or symptomatic PE) is inappropriate, because asymptomatic DVT (a surrogate endpoint) will dominate. There is no evidence that asymptomatic DVT correlates with clinically important outcomes.

See Results section: Categories of Methodological Errors and Biases #8 -- The primary endpoint should include major bleeding rather than just VTE.

Fatal and intracranial bleeding should be separately designated as primary or secondary endpoints because of the importance of these events.

To properly assess adverse events, including major bleeding and HITT, RCTs should be supplemented with large prospective or retrospective observational studies.

Rebound hypercoagulability-related adverse events should be monitored.

Authors' conclusions: Adjusted-dose warfarin and related oral anticoagulants reduce stroke, disabling stroke, and other major vascular events for those with nonvalvular atrial fibrillation by about one third when compared with antiplatelet therapy.

Given the risk for adverse effects with VKAs, major bleeding should be included in a composite primary outcome measure.

See Results section: Categories of Methodological Errors and Biases #10 -- Conclusions of the review were based on data from RCTs with younger, healthier, and more compliant patients than are typically seen in general clinical practice.

Representative safety data will not come from RCTs with anticoagulation researchers carefully monitoring the international normalized ratios of highly selected patients. Large observational studies need to be 1 way to assess the rates of major and fatal bleeding. For example, in this review VKA-related major bleeding (1.6%/yr) and intracranial bleeding (0.45%/yr) were considerably less frequent than in a recent observational study (7.2%/yr. and 2.5%/yr, respectively).^[43]

Rebound hypercoagulability-related adverse events were not monitored.

Authors' conclusion: "This review suggests a survival benefit of heparin in cancer patients in general, and in patients with small cell lung cancer in particular."

My comment: (See Results section: Categories of Methodological Errors and Biases #13) -- Biases exist in the selection of trials for inclusion in the review: 12 RCTs were eligible but only 5 were included, so the potentially biased selection of trials makes interpretation of the data "favorable to heparin" highly suspect.

Authors' reply: "We agree that it would have been ideal to include data from the trials published as abstracts. . . Interpretation of the findings of this review is limited by the moderate heterogeneity. . . The interpretation of findings is also limited by not including data from the 7 trials published as abstracts only."

These qualifications were not added to the abstract or plain-language summary where people without a subscription to the Cochrane Library could read it.

Authors' conclusion: "Combined unfractionated heparin and aspirin may reduce pregnancy loss by 54%."

Since all 3 RCTs of aspirin vs placebo showed no benefit, aspirin should be avoided and excluded from future RCTs.

Even if adding heparin to aspirin does increase the chances of live birth in a single pregnancy in this situation from 30/70 (43%) to 52/70 (74%), it is not significantly better that the chance of a live birth in 2 pregnancies with aspirin (or placebo) alone (ie, 1 - [0.57 x 0.57] = 68%). Given the risks of anticoagulants, the endpoint of any future RCT should be a healthy baby in 2 or 3 pregnancies rather than 1 pregnancy.

Safety (eg, osteopenia, bleeding, HITT, fetal abnormalities) cannot be determined with these small trials (n = 140).

The conclusion that heparin added to aspirin increases live births and should become the standard of care in women with recurrent miscarriages associated with antiphospholipid antibodies is premature.

In women with recurrent miscarriages, anticoagulants should not be given outside of RCTs.

Authors' conclusion: "Adding antiplatelet therapy, either dipyridamole or low-dose aspirin, to oral anticoagulation decreases the risk of systemic embolism or death among patients with prosthetic heart valves. The risk of major bleeding is increased with antiplatelet therapy."

The use of oral anticoagulation itself is not evidence-based to reduce morbidity and/or mortality in patients with prosthetic valves. It is based on anecdotal reports and historical precedent, not RCTs.

Even without the addition of aspirin, at least 1% per year of patients taking oral anticoagulants experience fatal bleeding each year.

Because adding antiplatelet drugs to oral anticoagulants substantially reduced the risk for thromboembolic events and mortality (61% and 45% respectively), we should have a non-inferiority RCT comparing aspirin alone with aspirin plus a VKA. As a prelude to such a definitive RCT, I am coauthoring a Cochrane protocol titled, "Anticoagulation for prevention of cardiovascular disease in people with prosthetic heart valves."

Authors' conclusion: "There were no randomized trials comparing either anticoagulants or antiplatelet drugs with control. There is, therefore, no evidence to support their routine use for the treatment of extracranial internal carotid artery dissection.... We suggest that a randomized trial including at least 1400 patients in each treatment group (oral anticoagulants and antiplatelet drugs) with this condition is clearly needed."

Antiplatelet agents -- the control group for the primary endpoint comparison in this review (antiplatelet agents vs anticoagulants) -- are not evidence-based to reduce death or dependence in people with stroke from carotid artery dissection.

However, aspirin and other antiplatelet agents are a potential cause of serious bleeding.

Consequently, the Implications for Research section should state that a large RCT of aspirin vs placebo is indicated.

Oral anticoagulants should be avoided in carotid artery dissection on the basis of the poor outcome in patient with acute ischemic stroke (Review #16 above).

Primary objective: To assess the effectiveness of heparin for prevention of catheter related thrombosis.

Main results: ...." There was no statistically significant difference in the risk of thrombosis (RR 0.93, 95% CI 0.58, 1.51)" . . . One of 4 secondary endpoints (catheter occlusion) statistically significantly favored heparin.

Authors' conclusion: "Evidence from this systematic review support the prophylactic use of heparin for PCVC in neonates at a dose of 0.5 IU/kg/hr."

For this review, the control was placebo or no treatment. A considerable amount of literature attests to the safety and efficacy of saline flushes.^[27,28,108] Consequently, saline flushes should have been the control.

Safety data from larger observational and case-control studies should have been included to evaluate these potential complications. For example, a case-control study involving 66 neonates with intraventricular bleeding found an increased risk for this devastating complication with heparin use (OR, 3.9 [95% CI, 1.4-11]), after adjusting for other risk factors.^[109]

Because there was no benefit with heparin for the primary endpoint and considerable risk for devastating intracranial or other bleeding, heparin should not be used.

Main results: "......Five studies reported data on the duration of use of the first catheter. Two of these studies found no statistically significant effect of heparin; two studies showed a statistically significant increase and one study showed a statistically significant decrease in the duration of PIV catheter use in the heparin group."

Authors' conclusion, Implications for Practice section: ".....Because of clinical heterogeneity and heterogeneity in treatment effect, recommendations for heparin use in neonates with PIV catheters cannot be made."

Authors' conclusions, Implications for Research section: "Further research on the effectiveness, the optimal dose, and the safety of heparin is required."

Premature infants are particularly prone to intracranial hemorrhage.

The small number of patients in the 2 RCTs reporting the incidence of intracranial hemorrhage (n = 322) gives no assurance that adding heparin to catheter infusions will not precipitate this dreaded complication.

Given that this meta-analysis found no benefit of heparin on first catheter occlusion (n = 485; OR, 1.00 [95% CI, 0.85-1.16]) or any clinical endpoint, further RCTs of heparin in neonates for this indication are not warranted.

Heparin should be avoided in premature infants.

Authors' conclusions. "Ticlopidine plus aspirin after coronary stenting is effective in reducing the risk of the revascularization, non fatal myocardial infarction and bleeding complications when compared with oral anticoagulants...."

My comment: In this meta-analysis of non-inferiority RCTs, the control group, aspirin plus a VKA, has never been shown to improve clinical outcomes for people with coronary stenting. Therefore, it cannot determine the effectiveness and safety of the use of ticlopidine plus aspirin after coronary stenting.

The lead author's reply: "The aim of the review was to compare ASA plus ticlopidine versus ASA plus vitamin K antagonists after coronary stenting and not the efficacy and safety of ASA plus ticlopidine per se."

Safety data concerning VKAs from previous RCTs, observational studies, and population-based studies should supplement the data from the 2 small RCTs in this review.

Given the lack of efficacy of warfarin in the RCTs of the review and the likely underestimate of the bleeding risk and unknown risk for rebound hypercoagulation, further RCTs of VKA treatment to reduce vascular events in people with hypertension would be unethical.

The authors rebutted my comments challenging the efficacy of platelet inhibitors in secondary prevention of patients with hypertension.

Regarding my point that further RCTs of VKA prophylaxis in hypertension, the authors said, "There is inadequate data on warfarin in hypertension per se."

Authors' general observation: "We are not sure what the author [me] refers to being 'evidence-based,' but from his comment we believe that s/he has a different understanding of evidence-based. Being 'evidence-based' means following the 'best evidence' whatever that may be. Evidence can be high quality or lower quality, it can be direct or indirect."

I disagree. By that definition, every test or treatment in medicine is evidence-based.

My comment: In this proposed meta-analysis (published as a review after my feedback letter) of RCTs comparing LMWHs with oral anticoagulants, the control group, VKAs, is not evidence-based to be safe and effective.

Authors' reply: None

My comment: Observational studies were not included in the review to evaluate for HITT and other safety endpoints.

Authors' reply: "We will add both HIT and HITT to the list of outcomes to assess." ....."We have designed the systematic review to include data from RCTs only. We will discuss in the final review the limitation of not including harm data from observational study."

My response: These endpoints were not systematically evaluated in the primary RCTs.

My comment: Rebound hypercoagulability-related adverse events were not monitored.

Authors' reply: "Dr. Cundiff raises the question, whether the incidence of thrombotic events increases within two months of discontinuing heparin therapy. We will investigate this hypothesis if the studies we include in our review report the relevant data."

My response: No RCTs included these data.

My comment: RCTs included in this meta-analysis were subject to selection bias because trials published only as abstracts were not included.

Authors' reply: "Dr. Cundiff describes publication bias, which we will investigate as recommended by the Cochrane Collaboration."

My comment: Besides major bleeding, the primary or secondary endpoints should include fatal bleeding and intracranial bleeding because of the particular importance to patients and clinicians of these events.

Authors' reply: The 2 proposed outcomes are accounted for by the outcomes listed in our protocol. Fatal bleeding events would be counted as deaths. Intracranial bleeding events would be counted as major bleeding events.

My response: Iatrogenic deaths should be counted separately from cancer deaths, and intracranial bleeding rates (fatal and nonfatal) are of special interest to patients and doctors.

None of the proposed primary outcome measures relate to safety concerns with anticoagulant drugs.

The primary endpoint, which determines the main conclusions of the review, should include efficacy and safety measures.

Concerning anticoagulants, given the relatively long natural history of the nephrotic syndrome, the slowed metabolism of anticoagulants with renal impairment, the significant bleeding risk, and the paucity of anecdotal evidence of efficacy, the ethics of conducting RCTs with heparins or VKAs should be questioned.

Bleeding (major or fatal bleeding)

Anticoagulant-associated intracranial hemorrhage

Heparin-induced thrombocytopenia with thrombosis

None of the proposed primary outcome measures relate to safety concerns with anticoagulant drugs.

The primary endpoint, which determines the main conclusions of the review, should include efficacy and safety measures.

Given the relatively long natural history of IgA nephropathy, the slowed metabolism of anticoagulants with renal impairment, the significant bleeding risk, and the paucity of anecdotal evidence of efficacy, the ethics of conducting RCTs with any anticoagulant for any clinical endpoint should be questioned.

See the Introduction

See also Results section: Categories of Methodological Errors and Biases #4 -- RCTs are too small to evaluate risk for HITT and observational studies not evaluated for HITT.

My published commentary, "Evidence-based Medicine and the Cochrane Collaboration on Trial"^[68] comprehensively details the points of contention in this review:

The primary efficacy endpoint combines asymptomatic and symptomatic VTE. This means that it relies primarily on the more frequent surrogate endpoint asymptomatic DVT.

Only a purely clinical endpoint should be used to measure efficacy.

Even symptomatic VTE would be inappropriate as the sole primary outcome measure to determine recommendations, since it does not include adverse events (ie, bleeding).

Combining symptomatic and fatal VTE and fatal complications of the LMWH (major bleeding and HITT) would be the most clinically relevant primary endpoint.

Because of excluding observational studies from consideration in the safety analysis, the bleeding complications will very likely be understated.

Rebound hypercoagulability was not assessed.

The review was published 14 months after my critique of the protocol was received and acknowledged. My points were not addressed.

Authors' conclusions: Implications for Practice section, "Patients receiving venous femoropopliteal, infragenicular bypasses seem to benefit more from VKA than from ASA." Comment: No differences in clinical endpoints were seen in placebo-controlled RCTs while both drugs were associated with unacceptable rates of fatal bleeding (VKAs, 16 [1.2%], aspirin/dipyridamole, 12 [0.9%]).

Rebound hypercoagulability was not assessed.

See also Results section: Categories of Methodological Errors and Biases #11 -- A patented, expensive treatment (not the subject of the review) that is not evidence-based to work is endorsed in the Implications for Practice section.

See Results section: Categories of Methodological Errors and Biases #5 -- Rebound hypercoagulability was not assessed by collecting study data for at least 2 months after anticoagulant withdrawal in most patients

This is a meta-analysis of non-inferiority trials in which neither the control group (shorter-duration VKA treatment) nor the experimental group (longer-duration VKA treatment) is evidence-based to be safe and effective.

Implications for Practice: " ... (We) recommend that injectable anticoagulants should be reserved for patients at higher risk of thrombosis and those with contraindications to physical methods and/or aspirin (Scottish Intercollegiate Guidelines Network 2002)".

My comment: There is no evidence basis for this statement, although it is drawn from another guidelines-producing group (Scottish Intercollegiate Guidelines Network [SIGN]).

Authors reply: "....as we state in the review, the source for the recommendation you refer to is the Hip Fracture Guideline Development Group. Prevention and management of hip fracture in older people. A national clinical guideline (No. 56). Edinburgh: Scottish Intercollegiate Guidelines Network (www.sign.ac.uk), 2002."

My response: Cochrane systematic reviews are supposed to base conclusions and recommendations on RCTs and other published literature, not non-evidence-based opinions of other authorities.

See also Results section: Categories of Methodological Errors and Biases #9 -- Intracranial bleeding was not separately included in the primary or secondary endpoints.

See Results section: Categories of Methodological Errors and Biases #5 -- Rebound hypercoagulability was not assessed.

Because observational, population-based, and/or case-control studies were excluded from consideration in the safety analysis, the bleeding complications were very likely understated.

The primary endpoints (DVTs and PE demonstrated on venograms and noninvasive scans) are surrogates rather than clinical endpoints. Asymptomatic VTE does not correlate with fatal PE.

Standard prophylactic anticoagulation (typically 7-10 days: the control group) following total hip or knee replacement or hip fracture repair provides no proven survival benefit.

Anticoagulation does cause significant risks for bleeding and rebound hypercoagulation events.

Rebound hypercoagulability events were missed in the extended duration of anticoagulant group (about 30 days of low-dose heparins or other drugs) of this review, thus biasing the findings.

In a meta-analysis of RCTs involving VTE prophylaxis of total hip replacement patients, the crude risks for clinically important bleeding (usually wound hematoma) were 0% for compression stockings, 0.3% for controls, 0.4% for aspirin, 1.8% for LMWH, and 2.6% for unfractionated heparin.^[75]

Because of the seriousness of wound hematomas and the frequently associated infections in these patients, SIGN changed the hip fracture and hip and knee replacement surgical VTE prophylaxis guidelines from advising anticoagulants to advising aspirin.^[76]

The primary endpoints, DVTs on venograms and noninvasive scans, are surrogate rather than clinical endpoints. Asymptomatic DVTs do not correlate with fatal PE.

Rebound hypercoagulability events were not assessed.

Despite the absence of benefit of LMWH for any of the major clinical endpoints (death, vascular death, disability), the Implications for Practice appear to justify the use of LMWH and heparinoids in some cases of acute ischemic stroke on the basis of the nonstatistically significant superiority of LMWH and heparinoids over UH in DVT and PE.

The Implications for Practice section should read, "Do not use any anticoagulants in patients with acute ischemic stroke."

Likewise, the Implications for Research section should say, "It would be unethical to conduct further trials of anticoagulants in patients with acute ischemic stroke."

Authors' reply: "The first version of the review was prepared at a time when there was uncertainty about the overall effects of heparins in acute ischemic stroke. Since then, as the commenter states, evidence has emerged which shows that there is no net benefit from the immediate anticoagulation in this setting. However, as is often the case, this evidence did not alter the beliefs of some clinicians, and hence, heparin is still used in some countries, in certain types of patient with acute ischemic stroke, for specific reasons and especially for prevention of venous thromboembolism. This review is now based on the premise that if a clinician plans to treat a patient for some special reason with some form of heparin then the choice of agent should be evidence based."

My rebuttal: I question that premise as a basis for a Cochrane Review. Heparin is dangerous. If it is not evidence-based to improve clinical outcomes, don't prescribe it.

The authors correctly point out that there was no statistically significant increase in major extracranial or intracranial hemorrhage (OR, 1.82 [95% CI, 0.87-3.81] and OR, 1.46 [95% CI, 0.72-2.97], respectively). However, when these 2 categories of major hemorrhage are combined, there is a borderline statistically significant increase in major bleeding (OR, 1.63 [95% CI, 0.98-2.72]; P< .06), which represents 6.2 additional major bleeding episodes per 1000 patients (95% CI, - 0.1 to 16.8) by adding the low-dose heparin.

Risk for bleeding in clinical practice is probably greater than in closely monitored RCTs.

Since adding low-dose heparin to aspirin does not improve clinical outcomes and does add to the bleeding risk, the Implications for Research section should say that further trials of any anticoagulants in acute ischemic stroke would be unethical and any studies under way should be stopped.

Also see Results section: Categories of Methodological Errors and Biases example #9 -- Fatal bleeding was not separately included in the primary or secondary endpoints.

This review of anticoagulant trials, including over 22,000 patients with acute ischemic stroke, found no net benefit with the use of any anticoagulant.

The Implications for Practice section should say that anticoagulants should be contraindicated in patients with acute ischemic stroke.

The Implications for Research section should say that further trials of anticoagulants in acute ischemic stroke would be unethical.

This Cochrane Review should be updated to indicate that VKA treatment should be contraindicated in patients with acute ischemic stroke.

In 2005, the Warfarin and Aspirin for Symptomatic Intracranial Arterial Stenosis (WASID) trial was terminated early because of the high incidence of death in the warfarin group (9.7% in the warfarin group vs 4.3% in the aspirin group; hazard ratio for aspirin relative to warfarin, 0.46; 95% CI, 0.23 to 0.90; P=.02).^[53] The authors concluded, "Aspirin should be used in preference to warfarin for patients with intracranial arterial stenosis."

No further anticoagulant drug research is warranted in this patient group.

The review concluded that aspirin significantly decreases death and dependency in stroke patients.

The Implications for Research section states, ". . . There is also a case for further trials of low dose subcutaneous heparin (or low dose low molecular weight heparin) plus aspirin versus aspirin alone in the prevention of post-stroke deep vein thrombosis and pulmonary embolism, and in reducing neurological disability from the original or recurrent strokes. Such trials would need to include several tens of thousands of patients."

My comment: This review should be updated to reflect that, given the lack of efficacy of anticoagulants in multiple trials and high bleeding risk in stroke patients (see #16 above), further trials with low-dose heparin or low-dose LMWH would be unethical.

Authors' conclusions in the "Implications for Practice" section: "The thromboprophylactic benefits of aspirin in patients with heart failure who are in sinus rhythm are not convincing, but may be detrimental in view of the reduction of ACE Inhibitor benefit and other side effects. It should be emphasized that the evidence for a potential interaction between ACE inhibitors and antiplatelet agents is mainly based on nonrandomized studies, in light of the quality of the studies from which the data are derived and the post hoc nature of the study analyses, the possible interaction should be interpreted with caution."

My comment: The implications for practice are obfuscation -- do not prescribe antiplatelet agents!

The abstract of the WASH trial (the only trial reviewed in this review) concludes, "The benefits of warfarin for patients with heart failure in sinus rhythm have not been established. Antithrombotic therapy in patients with heart failure is not evidence based but commonly contributes to polypharmacy."

Whereas, VKA therapy for heart failure has been tested and is not evidence-based to work, this review gives the false impression that the jury is out and that such treatment may be reasonable.

Adverse events due to rebound hypercoagulability were not monitored.

The Implications for Practice section focuses on beneficial therapies for subgroups of patients: (1) patients with venous graphs and (2) patients with artificial graphs. These subgroups were not defined in the protocol, so post hoc statistical analyses of outcomes of these subgroups should not be emphasized in the Results or Conclusions.

In several places, the Cochrane Handbook warns against problems with post hoc analyses -- for example, page 63: "Post-hoc questions are more susceptible to bias than those asked a priori, and data-driven questions can generate false conclusions based on spurious results."

Adverse events due to rebound hypercoagulability were not monitored.

No benefit was shown in any clinical parameter for claudication patients.

The anticoagulants increased bleeding, including at least 1 case of fatal bleeding.

Rebound hypercoagulation adverse events were not assessed.

Further research of anticoagulants for claudication is inappropriate and unethical.

A major conclusion of the review: Oral anticoagulation prolongs life in people with extensive small-cell lung cancer.

My comment: The survival in small-cell lung cancer was based on a post hoc statistical analysis.

Author reply: It was an a priori determined subgroup, because the protocol states, "We will try to explain heterogeneity, if present, by conducting subgroup analyses. Our a priori factors to explain heterogeneity are the characteristics of participants... and we list 'type/location of cancer' as a patient characteristic."

My rebuttal: If all cell types of cancer were designated for subgroup analysis by this catch-all statement, then adjusting for multiple comparisons (eg, Bonferroni adjustment) would render the P value for the benefit of extensive small-cell lung cancer with oral anticoagulants insignificant.

The authors' reply to my rebuttal is pending.

The authors' concluded, "The evidence suggests that anticoagulants are beneficial, without serious adverse effects, for people with NRAF and recent cerebral ischemia."

Although neither of these small RCTs with careful selection of relatively young patients had intracerebral bleeding (ie, for the 500-plus subject-years of follow-up on anticoagulants), about 12 such bleeding events would be expected in a general practice setting with a similar number of patient-years on anticoagulants.^[43]

The relatively younger age of patients in these RCTs means that efficacy in preventing vascular events will also be lower than in general practice.

Adverse events due to rebound hypercoagulability were not monitored.

Authors' conclusion: "The evidence from two trials suggests that anticoagulant therapy is superior to antiplatelet therapy for the prevention of stroke in people with NRAF and recent non-disabling stroke or TIA. The risk of extracranial bleeding was higher with anticoagulant therapy than with antiplatelet therapy."

See Results section: Categories of Methodological Errors and Biases #6 -- Because of excluding observational, population-based, and/or case-control studies from consideration in the safety analysis, the bleeding complications were very likely understated.

Fatal and intracranial bleeding were not reported

Adverse events due to rebound hypercoagulability were not monitored.

Authors' plain language summary: "Home treatment of deep vein thrombosis (DVT) is cost effective and preferred by people with DVT."

Hospital treatment of DVT with anticoagulants (the control group) is not evidence-based to be safe or effective.^{[1,68,87,88,89]} Therefore, home anticoagulation treatment for DVT (the experimental group) is not proven to be safe and effective.

The relevant issue: As it stands, the guidelines for anticoagulation during pregnancy and the postpartum period by the American College of Chest Physicians^[91] and the Royal College of Obstetricians and Gynaecologists^[92] are arguably the standard of care in the United States and United Kingdom, respectively. Despite the lack of RCT evidence, these opinion-based guidelines recommend anticoagulants in many instances, and they can be referenced in medicolegal cases.

This review appropriately concludes that anticoagulant thromboprophylaxis during pregnancy is not evidence-based to be safe and effective. However, the implications for practice and research do not go far enough.

Since anticoagulation carries risks for bleeding, osteoporosis, and fetal deformity, the appropriate implication for practice would be that thromboprophylaxis with anticoagulants should not be used outside of an RCT.

The Implications for Research section should state that any RCTs of anticoagulation in pregnant women should be placebo controlled.

The issue: Despite the lack of RCT evidence, opinion-based treatment guidelines for pregnant women with thrombophilia by the Royal College of Obstetricians and Gynaecologists and the American College of Chest Physicians include thromboprophylaxis.^[91,92]

My suggestion: "Given the risks of heparin, the implications for practice for the review should say that heparin should not be used for pregnant women with acquired or inherited thrombophilias outside of a placebo or antiplatelet agent controlled RCT."

The lead author agreed to make the change with the next review update.

Authors' Implications for Practice: "Both unfractionated heparin and low molecular weight heparin can be used as effective prophylaxis against postoperative thromboembolic complications after colorectal surgery."

My comment: These conclusions are based on surrogate endpoints (eg, venograms which mostly show asymptomatic DVTs) rather than clinical endpoints (eg, death, fatal PE, symptomatic VTE).

Authors' reply: "We must not forget that the efficacy of heparin prophylaxis against fatal Postoperative PE was established in a methodologically correctly performed trial more than 30 years ago (International Multicentre Trial).^[93] Taken these facts in mind I still mean that the use of postoperative prophylaxis is evidence based, and for comparing different methods the use of venography for detecting subclinical DVT as endpoint is scientifically acceptable."

My rebuttal: While methodologically correct for its time, the International Multicentre Trial preceded the era of early mobilization and mechanical VTE prophylaxis. The follow-up period was only until discharge from hospital, so the deaths caused by rebound hypercoagulation after heparin was stopped were missed. Likewise, in this review adverse events due to rebound hypercoagulability were not monitored.

At my suggestion, the authors completed this review 6 years after they published the protocol. The efficacy and safety of LWMH for acute coronary syndromes depends on the evidence basis of heparin vs placebo for acute coronary syndromes. See correspondence for review #30 below

See Results section #5 -- Rebound hypercoagulability was not assessed by collecting study data for at least 24 hours after heparin withdrawal

The appropriateness of this review is predicated on the efficacy of heparins when compared with placebo where all patients are also treated with aspirin. The authors maintain that their recently published Cochrane Review of this topic establishes the benefit of heparin plus aspirin vs aspirin alone. I dispute that contention. (See my feedback letter for #29 above.)

If neither LMWH nor UH is evidence-based to be beneficial for people with acute coronary syndromes, comparing them is inappropriate.

With additional antiplatelet agents and invasive procedures in recent years, heparins are significantly more hazardous that when the trials in this review were done.

Heparins should not be used for acute coronary syndromes outside of placebo-controlled RCTs.

From the abstract: "The objective of this systematic review is to determine the effectiveness and safety of heparin compared to all other modalities in the treatment of patients with STEMI."

Rebound hypercoagulability-related adverse events should be monitored.

Fatal bleeding and/or intracranial bleeding should be separately included in the primary or secondary endpoints.

RCTs, like those included in this review, report low bleeding complication rates with VKA treatment (eg, major bleeding/yr = 2.2%; intracranial bleeding/yr =0.5%; fatal bleeding/yr = 0.4% [n = 1939]^[96]). In contrast, a recent series of consecutive patients started on warfarin for atrial fibrillation reported higher rates of bleeding complications (major bleeding/yr, 7.2%, intracranial bleeding/yr, 2.5%, fatal bleeding/yr, 0.8% [n = 472]^[43]).

Rebound hypercoagulability-related adverse events were not monitored.

See also Results section: Categories of Methodological Errors and Biases #10 -- Conclusions of review were based on data from RCTs with patients who are not representative of those in general clinical practice.

Plain-language summary: "Insufficient evidence to say if anticoagulants help women with recurrent pregnancy loss without antiphospholipid syndrome"

Authors' recommendation: Large, randomized, placebo-controlled trials

Given the risks of heparin and the potential for harm, if tens of thousands of women have heparin treatment during pregnancy, the main endpoint in the recommended RCT (anticoagulant vs placebo) should be a live, healthy baby in 2 or 3 pregnancies rather than in a single pregnancy.

Rebound hypercoagulability-related adverse events should be monitored.

The bleeding complications will very likely be understated unless observational and/or population-based studies are included in the safety analysis.

Fatal bleeding and intracranial bleeding should be separately reported and added to the primary or secondary endpoints. Patients and physicians would be particularly interested in these safety measures.

This protocol assumes that those oral anticoagulants are evidence-based to benefit patients. However, the critiques of multiple Cochrane reviews in this paper question this assumption.

The authors implied that heparin is effective for CSVT, although the 2 small trials showed no statistically significant reduction in mortality.

The nonsignificant trend depends on a single small trial that was subject to early termination bias.

The authors of this CSVT review called heparin safe on the basis of only 40 cases (about 800 days of heparinization).

CNS bleeding is a common cause of morbidity and mortality in CSVT, so heparin is not safe.

Further placebo-controlled RCTs were inappropriately discouraged ("....patients and doctors may be reluctant to embark upon a new trial that includes a placebo group.").

Authors' conclusions: "No strong evidence was found to conclude thromboprophylaxis is effective to prevent thromboembolic events and safe, in people with unknown risk factors for thrombosis, undergoing knee arthroscopy."

From plain-language summary: "Adverse events were most common in the intervention group (i.e., anticoagulant prophylaxis). The most common complication was minor bleeding with a RR of 2.23 (range 0.99 to 4.99). The number needed to harm was 20."

From Implications for Research section: "Surveillance studies will be important to assess the detection of both adverse events and thrombotic events, with and without the use of LMWH."

Rebound hypercoagulability-related adverse events were not monitored.

The implication for practice should be that anticoagulant prophylaxis should not be used.

The implication for research should be that further RCTs with knee arthroscopy patients exposed to anticoagulants would be unethical.

Anticoagulant prophylaxis statistically significantly benefits surrogate endpoints of DVTs as shown by venograms and other tests, but it has not been shown to improve clinical endpoints (eg, fatal pulmonary emboli, mortality, symptomatic VTE).

The control group for this review (VKAs and LMWHs) is not evidence-based to be effective.

This proposed meta-analysis of non-inferiority trials comparing VKAs and LMWHs with direct thrombin inhibitors is not appropriate to prove efficacy or safety of direct thrombin inhibitors.

The protocol does not call for monitoring for rebound hypercoagulability-related adverse events.

Authors' conclusions: "The use of heparin-bonded catheters is a promising therapy but warrants further studies."

See Results section: Categories of Methodological Errors and Biases #3 -- Placebo or no treatment control was used when saline flushes are widely used and are safe and considered effective.

The primary endpoint is "Days of catheter patency (duration of patency of first catheter, in days)." The most appropriate clinical endpoint would seem to be occlusion of CVCs.

Given the risk for bleeding and HITT, a safety measurement should also be included in a composite primary endpoint.

Fatal bleeding and intracranial bleeding should have been designated separately as primary or secondary endpoints because of the magnitude of these events.

Rebound hypercoagulability-related adverse events should have been monitored.

See Results section of this review: Categories of Methodological Errors and Biases #3 -- Placebo or no treatment control was used when saline flushes are commonly employed as an anticlotting measure clinically; saline flushes are safe and considered effective in preventing CVC-related thrombosis in children.

The primary outcome measure, "CVC-related thrombosis (along the length of, or at the tip of, the catheter) as determined by either color-coded Doppler ultrasonography or contrast venography," is a surrogate endpoint rather than a clinically relevant efficacy measurement.

Instead, the secondary endpoint, "occlusion of CVC (defined as inability to infuse fluids through the catheter due to blockage)," should be made the primary endpoint.

Because RCTs for this indication will probably be too small to evaluate risk for HITT and hemorrhage, observational studies should also be included in the safety analysis.

Fatal bleeding and intracranial bleeding should be designated separately in the primary or secondary endpoints because of the particular importance of these events.

Rebound hypercoagulability-related adverse events should be monitored.

The primary endpoint (symptomatic plus asymptomatic DVT or symptomatic PE) is inappropriate, because asymptomatic DVT (a surrogate endpoint) will dominate. There is no evidence that asymptomatic DVT correlates with clinically important outcomes.

See Results section: Categories of Methodological Errors and Biases #8 -- The primary endpoint should include major bleeding rather than just VTE.

Fatal and intracranial bleeding should be separately designated as primary or secondary endpoints because of the importance of these events.

To properly assess adverse events, including major bleeding and HITT, RCTs should be supplemented with large prospective or retrospective observational studies.

Rebound hypercoagulability-related adverse events should be monitored.

Authors' conclusions: Adjusted-dose warfarin and related oral anticoagulants reduce stroke, disabling stroke, and other major vascular events for those with nonvalvular atrial fibrillation by about one third when compared with antiplatelet therapy.

Given the risk for adverse effects with VKAs, major bleeding should be included in a composite primary outcome measure.

See Results section: Categories of Methodological Errors and Biases #10 -- Conclusions of the review were based on data from RCTs with younger, healthier, and more compliant patients than are typically seen in general clinical practice.

Representative safety data will not come from RCTs with anticoagulation researchers carefully monitoring the international normalized ratios of highly selected patients. Large observational studies need to be 1 way to assess the rates of major and fatal bleeding. For example, in this review VKA-related major bleeding (1.6%/yr) and intracranial bleeding (0.45%/yr) were considerably less frequent than in a recent observational study (7.2%/yr. and 2.5%/yr, respectively).^[43]

Rebound hypercoagulability-related adverse events were not monitored.

Authors' conclusion: "This review suggests a survival benefit of heparin in cancer patients in general, and in patients with small cell lung cancer in particular."

My comment: (See Results section: Categories of Methodological Errors and Biases #13) -- Biases exist in the selection of trials for inclusion in the review: 12 RCTs were eligible but only 5 were included, so the potentially biased selection of trials makes interpretation of the data "favorable to heparin" highly suspect.

Authors' reply: "We agree that it would have been ideal to include data from the trials published as abstracts. . . Interpretation of the findings of this review is limited by the moderate heterogeneity. . . The interpretation of findings is also limited by not including data from the 7 trials published as abstracts only."

These qualifications were not added to the abstract or plain-language summary where people without a subscription to the Cochrane Library could read it.

Authors' conclusion: "Combined unfractionated heparin and aspirin may reduce pregnancy loss by 54%."

Since all 3 RCTs of aspirin vs placebo showed no benefit, aspirin should be avoided and excluded from future RCTs.

Even if adding heparin to aspirin does increase the chances of live birth in a single pregnancy in this situation from 30/70 (43%) to 52/70 (74%), it is not significantly better that the chance of a live birth in 2 pregnancies with aspirin (or placebo) alone (ie, 1 - [0.57 x 0.57] = 68%). Given the risks of anticoagulants, the endpoint of any future RCT should be a healthy baby in 2 or 3 pregnancies rather than 1 pregnancy.

Safety (eg, osteopenia, bleeding, HITT, fetal abnormalities) cannot be determined with these small trials (n = 140).

The conclusion that heparin added to aspirin increases live births and should become the standard of care in women with recurrent miscarriages associated with antiphospholipid antibodies is premature.

In women with recurrent miscarriages, anticoagulants should not be given outside of RCTs.

Authors' conclusion: "Adding antiplatelet therapy, either dipyridamole or low-dose aspirin, to oral anticoagulation decreases the risk of systemic embolism or death among patients with prosthetic heart valves. The risk of major bleeding is increased with antiplatelet therapy."

The use of oral anticoagulation itself is not evidence-based to reduce morbidity and/or mortality in patients with prosthetic valves. It is based on anecdotal reports and historical precedent, not RCTs.

Even without the addition of aspirin, at least 1% per year of patients taking oral anticoagulants experience fatal bleeding each year.

Because adding antiplatelet drugs to oral anticoagulants substantially reduced the risk for thromboembolic events and mortality (61% and 45% respectively), we should have a non-inferiority RCT comparing aspirin alone with aspirin plus a VKA. As a prelude to such a definitive RCT, I am coauthoring a Cochrane protocol titled, "Anticoagulation for prevention of cardiovascular disease in people with prosthetic heart valves."

Authors' conclusion: "There were no randomized trials comparing either anticoagulants or antiplatelet drugs with control. There is, therefore, no evidence to support their routine use for the treatment of extracranial internal carotid artery dissection.... We suggest that a randomized trial including at least 1400 patients in each treatment group (oral anticoagulants and antiplatelet drugs) with this condition is clearly needed."

Antiplatelet agents -- the control group for the primary endpoint comparison in this review (antiplatelet agents vs anticoagulants) -- are not evidence-based to reduce death or dependence in people with stroke from carotid artery dissection.

However, aspirin and other antiplatelet agents are a potential cause of serious bleeding.

Consequently, the Implications for Research section should state that a large RCT of aspirin vs placebo is indicated.

Oral anticoagulants should be avoided in carotid artery dissection on the basis of the poor outcome in patient with acute ischemic stroke (Review #16 above).

Primary objective: To assess the effectiveness of heparin for prevention of catheter related thrombosis.

Main results: ...." There was no statistically significant difference in the risk of thrombosis (RR 0.93, 95% CI 0.58, 1.51)" . . . One of 4 secondary endpoints (catheter occlusion) statistically significantly favored heparin.

Authors' conclusion: "Evidence from this systematic review support the prophylactic use of heparin for PCVC in neonates at a dose of 0.5 IU/kg/hr."

For this review, the control was placebo or no treatment. A considerable amount of literature attests to the safety and efficacy of saline flushes.^[27,28,108] Consequently, saline flushes should have been the control.

Safety data from larger observational and case-control studies should have been included to evaluate these potential complications. For example, a case-control study involving 66 neonates with intraventricular bleeding found an increased risk for this devastating complication with heparin use (OR, 3.9 [95% CI, 1.4-11]), after adjusting for other risk factors.^[109]

Because there was no benefit with heparin for the primary endpoint and considerable risk for devastating intracranial or other bleeding, heparin should not be used.

Main results: "......Five studies reported data on the duration of use of the first catheter. Two of these studies found no statistically significant effect of heparin; two studies showed a statistically significant increase and one study showed a statistically significant decrease in the duration of PIV catheter use in the heparin group."

Authors' conclusion, Implications for Practice section: ".....Because of clinical heterogeneity and heterogeneity in treatment effect, recommendations for heparin use in neonates with PIV catheters cannot be made."

Authors' conclusions, Implications for Research section: "Further research on the effectiveness, the optimal dose, and the safety of heparin is required."

Premature infants are particularly prone to intracranial hemorrhage.

The small number of patients in the 2 RCTs reporting the incidence of intracranial hemorrhage (n = 322) gives no assurance that adding heparin to catheter infusions will not precipitate this dreaded complication.

Given that this meta-analysis found no benefit of heparin on first catheter occlusion (n = 485; OR, 1.00 [95% CI, 0.85-1.16]) or any clinical endpoint, further RCTs of heparin in neonates for this indication are not warranted.

Heparin should be avoided in premature infants.

Authors' conclusions. "Ticlopidine plus aspirin after coronary stenting is effective in reducing the risk of the revascularization, non fatal myocardial infarction and bleeding complications when compared with oral anticoagulants...."

My comment: In this meta-analysis of non-inferiority RCTs, the control group, aspirin plus a VKA, has never been shown to improve clinical outcomes for people with coronary stenting. Therefore, it cannot determine the effectiveness and safety of the use of ticlopidine plus aspirin after coronary stenting.

The lead author's reply: "The aim of the review was to compare ASA plus ticlopidine versus ASA plus vitamin K antagonists after coronary stenting and not the efficacy and safety of ASA plus ticlopidine per se."

Safety data concerning VKAs from previous RCTs, observational studies, and population-based studies should supplement the data from the 2 small RCTs in this review.

Given the lack of efficacy of warfarin in the RCTs of the review and the likely underestimate of the bleeding risk and unknown risk for rebound hypercoagulation, further RCTs of VKA treatment to reduce vascular events in people with hypertension would be unethical.

The authors rebutted my comments challenging the efficacy of platelet inhibitors in secondary prevention of patients with hypertension.

Regarding my point that further RCTs of VKA prophylaxis in hypertension, the authors said, "There is inadequate data on warfarin in hypertension per se."

Authors' general observation: "We are not sure what the author [me] refers to being 'evidence-based,' but from his comment we believe that s/he has a different understanding of evidence-based. Being 'evidence-based' means following the 'best evidence' whatever that may be. Evidence can be high quality or lower quality, it can be direct or indirect."

I disagree. By that definition, every test or treatment in medicine is evidence-based.

My comment: In this proposed meta-analysis (published as a review after my feedback letter) of RCTs comparing LMWHs with oral anticoagulants, the control group, VKAs, is not evidence-based to be safe and effective.

Authors' reply: None

My comment: Observational studies were not included in the review to evaluate for HITT and other safety endpoints.

Authors' reply: "We will add both HIT and HITT to the list of outcomes to assess." ....."We have designed the systematic review to include data from RCTs only. We will discuss in the final review the limitation of not including harm data from observational study."

My response: These endpoints were not systematically evaluated in the primary RCTs.

My comment: Rebound hypercoagulability-related adverse events were not monitored.

Authors' reply: "Dr. Cundiff raises the question, whether the incidence of thrombotic events increases within two months of discontinuing heparin therapy. We will investigate this hypothesis if the studies we include in our review report the relevant data."

My response: No RCTs included these data.

My comment: RCTs included in this meta-analysis were subject to selection bias because trials published only as abstracts were not included.

Authors' reply: "Dr. Cundiff describes publication bias, which we will investigate as recommended by the Cochrane Collaboration."

My comment: Besides major bleeding, the primary or secondary endpoints should include fatal bleeding and intracranial bleeding because of the particular importance to patients and clinicians of these events.

Authors' reply: The 2 proposed outcomes are accounted for by the outcomes listed in our protocol. Fatal bleeding events would be counted as deaths. Intracranial bleeding events would be counted as major bleeding events.

My response: Iatrogenic deaths should be counted separately from cancer deaths, and intracranial bleeding rates (fatal and nonfatal) are of special interest to patients and doctors.

None of the proposed primary outcome measures relate to safety concerns with anticoagulant drugs.

The primary endpoint, which determines the main conclusions of the review, should include efficacy and safety measures.

Concerning anticoagulants, given the relatively long natural history of the nephrotic syndrome, the slowed metabolism of anticoagulants with renal impairment, the significant bleeding risk, and the paucity of anecdotal evidence of efficacy, the ethics of conducting RCTs with heparins or VKAs should be questioned.

Bleeding (major or fatal bleeding)

Anticoagulant-associated intracranial hemorrhage

Heparin-induced thrombocytopenia with thrombosis

None of the proposed primary outcome measures relate to safety concerns with anticoagulant drugs.

The primary endpoint, which determines the main conclusions of the review, should include efficacy and safety measures.

Given the relatively long natural history of IgA nephropathy, the slowed metabolism of anticoagulants with renal impairment, the significant bleeding risk, and the paucity of anecdotal evidence of efficacy, the ethics of conducting RCTs with any anticoagulant for any clinical endpoint should be questioned.

See the Introduction

See also Results section: Categories of Methodological Errors and Biases #4 -- RCTs are too small to evaluate risk for HITT and observational studies not evaluated for HITT.

My published commentary, "Evidence-based Medicine and the Cochrane Collaboration on Trial"^[68] comprehensively details the points of contention in this review:

The primary efficacy endpoint combines asymptomatic and symptomatic VTE. This means that it relies primarily on the more frequent surrogate endpoint asymptomatic DVT.

Only a purely clinical endpoint should be used to measure efficacy.

Even symptomatic VTE would be inappropriate as the sole primary outcome measure to determine recommendations, since it does not include adverse events (ie, bleeding).

Combining symptomatic and fatal VTE and fatal complications of the LMWH (major bleeding and HITT) would be the most clinically relevant primary endpoint.

Because of excluding observational studies from consideration in the safety analysis, the bleeding complications will very likely be understated.

Rebound hypercoagulability was not assessed.

The review was published 14 months after my critique of the protocol was received and acknowledged. My points were not addressed.

Authors' conclusions: Implications for Practice section, "Patients receiving venous femoropopliteal, infragenicular bypasses seem to benefit more from VKA than from ASA." Comment: No differences in clinical endpoints were seen in placebo-controlled RCTs while both drugs were associated with unacceptable rates of fatal bleeding (VKAs, 16 [1.2%], aspirin/dipyridamole, 12 [0.9%]).

Rebound hypercoagulability was not assessed.

See also Results section: Categories of Methodological Errors and Biases #11 -- A patented, expensive treatment (not the subject of the review) that is not evidence-based to work is endorsed in the Implications for Practice section.

See Results section: Categories of Methodological Errors and Biases #5 -- Rebound hypercoagulability was not assessed by collecting study data for at least 2 months after anticoagulant withdrawal in most patients

This is a meta-analysis of non-inferiority trials in which neither the control group (shorter-duration VKA treatment) nor the experimental group (longer-duration VKA treatment) is evidence-based to be safe and effective.

Implications for Practice: " ... (We) recommend that injectable anticoagulants should be reserved for patients at higher risk of thrombosis and those with contraindications to physical methods and/or aspirin (Scottish Intercollegiate Guidelines Network 2002)".

My comment: There is no evidence basis for this statement, although it is drawn from another guidelines-producing group (Scottish Intercollegiate Guidelines Network [SIGN]).

Authors reply: "....as we state in the review, the source for the recommendation you refer to is the Hip Fracture Guideline Development Group. Prevention and management of hip fracture in older people. A national clinical guideline (No. 56). Edinburgh: Scottish Intercollegiate Guidelines Network (www.sign.ac.uk), 2002."

My response: Cochrane systematic reviews are supposed to base conclusions and recommendations on RCTs and other published literature, not non-evidence-based opinions of other authorities.

See also Results section: Categories of Methodological Errors and Biases #9 -- Intracranial bleeding was not separately included in the primary or secondary endpoints.

See Results section: Categories of Methodological Errors and Biases #5 -- Rebound hypercoagulability was not assessed.

Because observational, population-based, and/or case-control studies were excluded from consideration in the safety analysis, the bleeding complications were very likely understated.

The primary endpoints (DVTs and PE demonstrated on venograms and noninvasive scans) are surrogates rather than clinical endpoints. Asymptomatic VTE does not correlate with fatal PE.

Standard prophylactic anticoagulation (typically 7-10 days: the control group) following total hip or knee replacement or hip fracture repair provides no proven survival benefit.

Anticoagulation does cause significant risks for bleeding and rebound hypercoagulation events.

Rebound hypercoagulability events were missed in the extended duration of anticoagulant group (about 30 days of low-dose heparins or other drugs) of this review, thus biasing the findings.

In a meta-analysis of RCTs involving VTE prophylaxis of total hip replacement patients, the crude risks for clinically important bleeding (usually wound hematoma) were 0% for compression stockings, 0.3% for controls, 0.4% for aspirin, 1.8% for LMWH, and 2.6% for unfractionated heparin.^[75]

Because of the seriousness of wound hematomas and the frequently associated infections in these patients, SIGN changed the hip fracture and hip and knee replacement surgical VTE prophylaxis guidelines from advising anticoagulants to advising aspirin.^[76]

The primary endpoints, DVTs on venograms and noninvasive scans, are surrogate rather than clinical endpoints. Asymptomatic DVTs do not correlate with fatal PE.

Rebound hypercoagulability events were not assessed.

Despite the absence of benefit of LMWH for any of the major clinical endpoints (death, vascular death, disability), the Implications for Practice appear to justify the use of LMWH and heparinoids in some cases of acute ischemic stroke on the basis of the nonstatistically significant superiority of LMWH and heparinoids over UH in DVT and PE.

The Implications for Practice section should read, "Do not use any anticoagulants in patients with acute ischemic stroke."

Likewise, the Implications for Research section should say, "It would be unethical to conduct further trials of anticoagulants in patients with acute ischemic stroke."

Authors' reply: "The first version of the review was prepared at a time when there was uncertainty about the overall effects of heparins in acute ischemic stroke. Since then, as the commenter states, evidence has emerged which shows that there is no net benefit from the immediate anticoagulation in this setting. However, as is often the case, this evidence did not alter the beliefs of some clinicians, and hence, heparin is still used in some countries, in certain types of patient with acute ischemic stroke, for specific reasons and especially for prevention of venous thromboembolism. This review is now based on the premise that if a clinician plans to treat a patient for some special reason with some form of heparin then the choice of agent should be evidence based."

My rebuttal: I question that premise as a basis for a Cochrane Review. Heparin is dangerous. If it is not evidence-based to improve clinical outcomes, don't prescribe it.

The authors correctly point out that there was no statistically significant increase in major extracranial or intracranial hemorrhage (OR, 1.82 [95% CI, 0.87-3.81] and OR, 1.46 [95% CI, 0.72-2.97], respectively). However, when these 2 categories of major hemorrhage are combined, there is a borderline statistically significant increase in major bleeding (OR, 1.63 [95% CI, 0.98-2.72]; P< .06), which represents 6.2 additional major bleeding episodes per 1000 patients (95% CI, - 0.1 to 16.8) by adding the low-dose heparin.

Risk for bleeding in clinical practice is probably greater than in closely monitored RCTs.

Since adding low-dose heparin to aspirin does not improve clinical outcomes and does add to the bleeding risk, the Implications for Research section should say that further trials of any anticoagulants in acute ischemic stroke would be unethical and any studies under way should be stopped.

Also see Results section: Categories of Methodological Errors and Biases example #9 -- Fatal bleeding was not separately included in the primary or secondary endpoints.

This review of anticoagulant trials, including over 22,000 patients with acute ischemic stroke, found no net benefit with the use of any anticoagulant.

The Implications for Practice section should say that anticoagulants should be contraindicated in patients with acute ischemic stroke.

The Implications for Research section should say that further trials of anticoagulants in acute ischemic stroke would be unethical.

This Cochrane Review should be updated to indicate that VKA treatment should be contraindicated in patients with acute ischemic stroke.

In 2005, the Warfarin and Aspirin for Symptomatic Intracranial Arterial Stenosis (WASID) trial was terminated early because of the high incidence of death in the warfarin group (9.7% in the warfarin group vs 4.3% in the aspirin group; hazard ratio for aspirin relative to warfarin, 0.46; 95% CI, 0.23 to 0.90; P=.02).^[53] The authors concluded, "Aspirin should be used in preference to warfarin for patients with intracranial arterial stenosis."

No further anticoagulant drug research is warranted in this patient group.

The review concluded that aspirin significantly decreases death and dependency in stroke patients.

The Implications for Research section states, ". . . There is also a case for further trials of low dose subcutaneous heparin (or low dose low molecular weight heparin) plus aspirin versus aspirin alone in the prevention of post-stroke deep vein thrombosis and pulmonary embolism, and in reducing neurological disability from the original or recurrent strokes. Such trials would need to include several tens of thousands of patients."

My comment: This review should be updated to reflect that, given the lack of efficacy of anticoagulants in multiple trials and high bleeding risk in stroke patients (see #16 above), further trials with low-dose heparin or low-dose LMWH would be unethical.

Authors' conclusions in the "Implications for Practice" section: "The thromboprophylactic benefits of aspirin in patients with heart failure who are in sinus rhythm are not convincing, but may be detrimental in view of the reduction of ACE Inhibitor benefit and other side effects. It should be emphasized that the evidence for a potential interaction between ACE inhibitors and antiplatelet agents is mainly based on nonrandomized studies, in light of the quality of the studies from which the data are derived and the post hoc nature of the study analyses, the possible interaction should be interpreted with caution."

My comment: The implications for practice are obfuscation -- do not prescribe antiplatelet agents!

The abstract of the WASH trial (the only trial reviewed in this review) concludes, "The benefits of warfarin for patients with heart failure in sinus rhythm have not been established. Antithrombotic therapy in patients with heart failure is not evidence based but commonly contributes to polypharmacy."

Whereas, VKA therapy for heart failure has been tested and is not evidence-based to work, this review gives the false impression that the jury is out and that such treatment may be reasonable.

Adverse events due to rebound hypercoagulability were not monitored.

The Implications for Practice section focuses on beneficial therapies for subgroups of patients: (1) patients with venous graphs and (2) patients with artificial graphs. These subgroups were not defined in the protocol, so post hoc statistical analyses of outcomes of these subgroups should not be emphasized in the Results or Conclusions.

In several places, the Cochrane Handbook warns against problems with post hoc analyses -- for example, page 63: "Post-hoc questions are more susceptible to bias than those asked a priori, and data-driven questions can generate false conclusions based on spurious results."

Adverse events due to rebound hypercoagulability were not monitored.

No benefit was shown in any clinical parameter for claudication patients.

The anticoagulants increased bleeding, including at least 1 case of fatal bleeding.

Rebound hypercoagulation adverse events were not assessed.

Further research of anticoagulants for claudication is inappropriate and unethical.

A major conclusion of the review: Oral anticoagulation prolongs life in people with extensive small-cell lung cancer.

My comment: The survival in small-cell lung cancer was based on a post hoc statistical analysis.

Author reply: It was an a priori determined subgroup, because the protocol states, "We will try to explain heterogeneity, if present, by conducting subgroup analyses. Our a priori factors to explain heterogeneity are the characteristics of participants... and we list 'type/location of cancer' as a patient characteristic."

My rebuttal: If all cell types of cancer were designated for subgroup analysis by this catch-all statement, then adjusting for multiple comparisons (eg, Bonferroni adjustment) would render the P value for the benefit of extensive small-cell lung cancer with oral anticoagulants insignificant.

The authors' reply to my rebuttal is pending.

The authors' concluded, "The evidence suggests that anticoagulants are beneficial, without serious adverse effects, for people with NRAF and recent cerebral ischemia."

Although neither of these small RCTs with careful selection of relatively young patients had intracerebral bleeding (ie, for the 500-plus subject-years of follow-up on anticoagulants), about 12 such bleeding events would be expected in a general practice setting with a similar number of patient-years on anticoagulants.^[43]

The relatively younger age of patients in these RCTs means that efficacy in preventing vascular events will also be lower than in general practice.

Adverse events due to rebound hypercoagulability were not monitored.

Authors' conclusion: "The evidence from two trials suggests that anticoagulant therapy is superior to antiplatelet therapy for the prevention of stroke in people with NRAF and recent non-disabling stroke or TIA. The risk of extracranial bleeding was higher with anticoagulant therapy than with antiplatelet therapy."

See Results section: Categories of Methodological Errors and Biases #6 -- Because of excluding observational, population-based, and/or case-control studies from consideration in the safety analysis, the bleeding complications were very likely understated.

Fatal and intracranial bleeding were not reported

Adverse events due to rebound hypercoagulability were not monitored.

Authors' plain language summary: "Home treatment of deep vein thrombosis (DVT) is cost effective and preferred by people with DVT."

Hospital treatment of DVT with anticoagulants (the control group) is not evidence-based to be safe or effective.^{[1,68,87,88,89]} Therefore, home anticoagulation treatment for DVT (the experimental group) is not proven to be safe and effective.

The relevant issue: As it stands, the guidelines for anticoagulation during pregnancy and the postpartum period by the American College of Chest Physicians^[91] and the Royal College of Obstetricians and Gynaecologists^[92] are arguably the standard of care in the United States and United Kingdom, respectively. Despite the lack of RCT evidence, these opinion-based guidelines recommend anticoagulants in many instances, and they can be referenced in medicolegal cases.

This review appropriately concludes that anticoagulant thromboprophylaxis during pregnancy is not evidence-based to be safe and effective. However, the implications for practice and research do not go far enough.

Since anticoagulation carries risks for bleeding, osteoporosis, and fetal deformity, the appropriate implication for practice would be that thromboprophylaxis with anticoagulants should not be used outside of an RCT.

The Implications for Research section should state that any RCTs of anticoagulation in pregnant women should be placebo controlled.

The issue: Despite the lack of RCT evidence, opinion-based treatment guidelines for pregnant women with thrombophilia by the Royal College of Obstetricians and Gynaecologists and the American College of Chest Physicians include thromboprophylaxis.^[91,92]

My suggestion: "Given the risks of heparin, the implications for practice for the review should say that heparin should not be used for pregnant women with acquired or inherited thrombophilias outside of a placebo or antiplatelet agent controlled RCT."

The lead author agreed to make the change with the next review update.

Authors' Implications for Practice: "Both unfractionated heparin and low molecular weight heparin can be used as effective prophylaxis against postoperative thromboembolic complications after colorectal surgery."

My comment: These conclusions are based on surrogate endpoints (eg, venograms which mostly show asymptomatic DVTs) rather than clinical endpoints (eg, death, fatal PE, symptomatic VTE).

Authors' reply: "We must not forget that the efficacy of heparin prophylaxis against fatal Postoperative PE was established in a methodologically correctly performed trial more than 30 years ago (International Multicentre Trial).^[93] Taken these facts in mind I still mean that the use of postoperative prophylaxis is evidence based, and for comparing different methods the use of venography for detecting subclinical DVT as endpoint is scientifically acceptable."

My rebuttal: While methodologically correct for its time, the International Multicentre Trial preceded the era of early mobilization and mechanical VTE prophylaxis. The follow-up period was only until discharge from hospital, so the deaths caused by rebound hypercoagulation after heparin was stopped were missed. Likewise, in this review adverse events due to rebound hypercoagulability were not monitored.

At my suggestion, the authors completed this review 6 years after they published the protocol. The efficacy and safety of LWMH for acute coronary syndromes depends on the evidence basis of heparin vs placebo for acute coronary syndromes. See correspondence for review #30 below

See Results section #5 -- Rebound hypercoagulability was not assessed by collecting study data for at least 24 hours after heparin withdrawal

The appropriateness of this review is predicated on the efficacy of heparins when compared with placebo where all patients are also treated with aspirin. The authors maintain that their recently published Cochrane Review of this topic establishes the benefit of heparin plus aspirin vs aspirin alone. I dispute that contention. (See my feedback letter for #29 above.)

If neither LMWH nor UH is evidence-based to be beneficial for people with acute coronary syndromes, comparing them is inappropriate.

With additional antiplatelet agents and invasive procedures in recent years, heparins are significantly more hazardous that when the trials in this review were done.

Heparins should not be used for acute coronary syndromes outside of placebo-controlled RCTs.

From the abstract: "The objective of this systematic review is to determine the effectiveness and safety of heparin compared to all other modalities in the treatment of patients with STEMI."

Rebound hypercoagulability-related adverse events should be monitored.

Fatal bleeding and/or intracranial bleeding should be separately included in the primary or secondary endpoints.

RCTs, like those included in this review, report low bleeding complication rates with VKA treatment (eg, major bleeding/yr = 2.2%; intracranial bleeding/yr =0.5%; fatal bleeding/yr = 0.4% [n = 1939]^[96]). In contrast, a recent series of consecutive patients started on warfarin for atrial fibrillation reported higher rates of bleeding complications (major bleeding/yr, 7.2%, intracranial bleeding/yr, 2.5%, fatal bleeding/yr, 0.8% [n = 472]^[43]).

Rebound hypercoagulability-related adverse events were not monitored.

See also Results section: Categories of Methodological Errors and Biases #10 -- Conclusions of review were based on data from RCTs with patients who are not representative of those in general clinical practice.

Plain-language summary: "Insufficient evidence to say if anticoagulants help women with recurrent pregnancy loss without antiphospholipid syndrome"

Authors' recommendation: Large, randomized, placebo-controlled trials

Given the risks of heparin and the potential for harm, if tens of thousands of women have heparin treatment during pregnancy, the main endpoint in the recommended RCT (anticoagulant vs placebo) should be a live, healthy baby in 2 or 3 pregnancies rather than in a single pregnancy.

Rebound hypercoagulability-related adverse events should be monitored.

The bleeding complications will very likely be understated unless observational and/or population-based studies are included in the safety analysis.

Fatal bleeding and intracranial bleeding should be separately reported and added to the primary or secondary endpoints. Patients and physicians would be particularly interested in these safety measures.

This protocol assumes that those oral anticoagulants are evidence-based to benefit patients. However, the critiques of multiple Cochrane reviews in this paper question this assumption.

The authors implied that heparin is effective for CSVT, although the 2 small trials showed no statistically significant reduction in mortality.

The nonsignificant trend depends on a single small trial that was subject to early termination bias.

The authors of this CSVT review called heparin safe on the basis of only 40 cases (about 800 days of heparinization).

CNS bleeding is a common cause of morbidity and mortality in CSVT, so heparin is not safe.

Further placebo-controlled RCTs were inappropriately discouraged ("....patients and doctors may be reluctant to embark upon a new trial that includes a placebo group.").

Authors' conclusions: "No strong evidence was found to conclude thromboprophylaxis is effective to prevent thromboembolic events and safe, in people with unknown risk factors for thrombosis, undergoing knee arthroscopy."

From plain-language summary: "Adverse events were most common in the intervention group (i.e., anticoagulant prophylaxis). The most common complication was minor bleeding with a RR of 2.23 (range 0.99 to 4.99). The number needed to harm was 20."

From Implications for Research section: "Surveillance studies will be important to assess the detection of both adverse events and thrombotic events, with and without the use of LMWH."

Rebound hypercoagulability-related adverse events were not monitored.

The implication for practice should be that anticoagulant prophylaxis should not be used.

The implication for research should be that further RCTs with knee arthroscopy patients exposed to anticoagulants would be unethical.

Anticoagulant prophylaxis statistically significantly benefits surrogate endpoints of DVTs as shown by venograms and other tests, but it has not been shown to improve clinical endpoints (eg, fatal pulmonary emboli, mortality, symptomatic VTE).

The control group for this review (VKAs and LMWHs) is not evidence-based to be effective.

This proposed meta-analysis of non-inferiority trials comparing VKAs and LMWHs with direct thrombin inhibitors is not appropriate to prove efficacy or safety of direct thrombin inhibitors.

The protocol does not call for monitoring for rebound hypercoagulability-related adverse events.

Authors' conclusions: "The use of heparin-bonded catheters is a promising therapy but warrants further studies."

See Results section: Categories of Methodological Errors and Biases #3 -- Placebo or no treatment control was used when saline flushes are widely used and are safe and considered effective.

The primary endpoint is "Days of catheter patency (duration of patency of first catheter, in days)." The most appropriate clinical endpoint would seem to be occlusion of CVCs.

Given the risk for bleeding and HITT, a safety measurement should also be included in a composite primary endpoint.

Fatal bleeding and intracranial bleeding should have been designated separately as primary or secondary endpoints because of the magnitude of these events.

Rebound hypercoagulability-related adverse events should have been monitored.

See Results section of this review: Categories of Methodological Errors and Biases #3 -- Placebo or no treatment control was used when saline flushes are commonly employed as an anticlotting measure clinically; saline flushes are safe and considered effective in preventing CVC-related thrombosis in children.

The primary outcome measure, "CVC-related thrombosis (along the length of, or at the tip of, the catheter) as determined by either color-coded Doppler ultrasonography or contrast venography," is a surrogate endpoint rather than a clinically relevant efficacy measurement.

Instead, the secondary endpoint, "occlusion of CVC (defined as inability to infuse fluids through the catheter due to blockage)," should be made the primary endpoint.

Because RCTs for this indication will probably be too small to evaluate risk for HITT and hemorrhage, observational studies should also be included in the safety analysis.

Fatal bleeding and intracranial bleeding should be designated separately in the primary or secondary endpoints because of the particular importance of these events.

Rebound hypercoagulability-related adverse events should be monitored.

The primary endpoint (symptomatic plus asymptomatic DVT or symptomatic PE) is inappropriate, because asymptomatic DVT (a surrogate endpoint) will dominate. There is no evidence that asymptomatic DVT correlates with clinically important outcomes.

See Results section: Categories of Methodological Errors and Biases #8 -- The primary endpoint should include major bleeding rather than just VTE.

Fatal and intracranial bleeding should be separately designated as primary or secondary endpoints because of the importance of these events.

To properly assess adverse events, including major bleeding and HITT, RCTs should be supplemented with large prospective or retrospective observational studies.

Rebound hypercoagulability-related adverse events should be monitored.

Authors' conclusions: Adjusted-dose warfarin and related oral anticoagulants reduce stroke, disabling stroke, and other major vascular events for those with nonvalvular atrial fibrillation by about one third when compared with antiplatelet therapy.

Given the risk for adverse effects with VKAs, major bleeding should be included in a composite primary outcome measure.

See Results section: Categories of Methodological Errors and Biases #10 -- Conclusions of the review were based on data from RCTs with younger, healthier, and more compliant patients than are typically seen in general clinical practice.

Representative safety data will not come from RCTs with anticoagulation researchers carefully monitoring the international normalized ratios of highly selected patients. Large observational studies need to be 1 way to assess the rates of major and fatal bleeding. For example, in this review VKA-related major bleeding (1.6%/yr) and intracranial bleeding (0.45%/yr) were considerably less frequent than in a recent observational study (7.2%/yr. and 2.5%/yr, respectively).^[43]

Rebound hypercoagulability-related adverse events were not monitored.

Authors' conclusion: "This review suggests a survival benefit of heparin in cancer patients in general, and in patients with small cell lung cancer in particular."

My comment: (See Results section: Categories of Methodological Errors and Biases #13) -- Biases exist in the selection of trials for inclusion in the review: 12 RCTs were eligible but only 5 were included, so the potentially biased selection of trials makes interpretation of the data "favorable to heparin" highly suspect.

Authors' reply: "We agree that it would have been ideal to include data from the trials published as abstracts. . . Interpretation of the findings of this review is limited by the moderate heterogeneity. . . The interpretation of findings is also limited by not including data from the 7 trials published as abstracts only."

These qualifications were not added to the abstract or plain-language summary where people without a subscription to the Cochrane Library could read it.

Authors' conclusion: "Combined unfractionated heparin and aspirin may reduce pregnancy loss by 54%."

Since all 3 RCTs of aspirin vs placebo showed no benefit, aspirin should be avoided and excluded from future RCTs.

Even if adding heparin to aspirin does increase the chances of live birth in a single pregnancy in this situation from 30/70 (43%) to 52/70 (74%), it is not significantly better that the chance of a live birth in 2 pregnancies with aspirin (or placebo) alone (ie, 1 - [0.57 x 0.57] = 68%). Given the risks of anticoagulants, the endpoint of any future RCT should be a healthy baby in 2 or 3 pregnancies rather than 1 pregnancy.

Safety (eg, osteopenia, bleeding, HITT, fetal abnormalities) cannot be determined with these small trials (n = 140).

The conclusion that heparin added to aspirin increases live births and should become the standard of care in women with recurrent miscarriages associated with antiphospholipid antibodies is premature.

In women with recurrent miscarriages, anticoagulants should not be given outside of RCTs.

Authors' conclusion: "Adding antiplatelet therapy, either dipyridamole or low-dose aspirin, to oral anticoagulation decreases the risk of systemic embolism or death among patients with prosthetic heart valves. The risk of major bleeding is increased with antiplatelet therapy."

The use of oral anticoagulation itself is not evidence-based to reduce morbidity and/or mortality in patients with prosthetic valves. It is based on anecdotal reports and historical precedent, not RCTs.

Even without the addition of aspirin, at least 1% per year of patients taking oral anticoagulants experience fatal bleeding each year.

Because adding antiplatelet drugs to oral anticoagulants substantially reduced the risk for thromboembolic events and mortality (61% and 45% respectively), we should have a non-inferiority RCT comparing aspirin alone with aspirin plus a VKA. As a prelude to such a definitive RCT, I am coauthoring a Cochrane protocol titled, "Anticoagulation for prevention of cardiovascular disease in people with prosthetic heart valves."

Authors' conclusion: "There were no randomized trials comparing either anticoagulants or antiplatelet drugs with control. There is, therefore, no evidence to support their routine use for the treatment of extracranial internal carotid artery dissection.... We suggest that a randomized trial including at least 1400 patients in each treatment group (oral anticoagulants and antiplatelet drugs) with this condition is clearly needed."

Antiplatelet agents -- the control group for the primary endpoint comparison in this review (antiplatelet agents vs anticoagulants) -- are not evidence-based to reduce death or dependence in people with stroke from carotid artery dissection.

However, aspirin and other antiplatelet agents are a potential cause of serious bleeding.

Consequently, the Implications for Research section should state that a large RCT of aspirin vs placebo is indicated.

Oral anticoagulants should be avoided in carotid artery dissection on the basis of the poor outcome in patient with acute ischemic stroke (Review #16 above).

Primary objective: To assess the effectiveness of heparin for prevention of catheter related thrombosis.

Main results: ...." There was no statistically significant difference in the risk of thrombosis (RR 0.93, 95% CI 0.58, 1.51)" . . . One of 4 secondary endpoints (catheter occlusion) statistically significantly favored heparin.

Authors' conclusion: "Evidence from this systematic review support the prophylactic use of heparin for PCVC in neonates at a dose of 0.5 IU/kg/hr."

For this review, the control was placebo or no treatment. A considerable amount of literature attests to the safety and efficacy of saline flushes.^[27,28,108] Consequently, saline flushes should have been the control.

Safety data from larger observational and case-control studies should have been included to evaluate these potential complications. For example, a case-control study involving 66 neonates with intraventricular bleeding found an increased risk for this devastating complication with heparin use (OR, 3.9 [95% CI, 1.4-11]), after adjusting for other risk factors.^[109]

Because there was no benefit with heparin for the primary endpoint and considerable risk for devastating intracranial or other bleeding, heparin should not be used.

Main results: "......Five studies reported data on the duration of use of the first catheter. Two of these studies found no statistically significant effect of heparin; two studies showed a statistically significant increase and one study showed a statistically significant decrease in the duration of PIV catheter use in the heparin group."

Authors' conclusion, Implications for Practice section: ".....Because of clinical heterogeneity and heterogeneity in treatment effect, recommendations for heparin use in neonates with PIV catheters cannot be made."

Authors' conclusions, Implications for Research section: "Further research on the effectiveness, the optimal dose, and the safety of heparin is required."

Premature infants are particularly prone to intracranial hemorrhage.

The small number of patients in the 2 RCTs reporting the incidence of intracranial hemorrhage (n = 322) gives no assurance that adding heparin to catheter infusions will not precipitate this dreaded complication.

Given that this meta-analysis found no benefit of heparin on first catheter occlusion (n = 485; OR, 1.00 [95% CI, 0.85-1.16]) or any clinical endpoint, further RCTs of heparin in neonates for this indication are not warranted.

Heparin should be avoided in premature infants.

Authors' conclusions. "Ticlopidine plus aspirin after coronary stenting is effective in reducing the risk of the revascularization, non fatal myocardial infarction and bleeding complications when compared with oral anticoagulants...."

My comment: In this meta-analysis of non-inferiority RCTs, the control group, aspirin plus a VKA, has never been shown to improve clinical outcomes for people with coronary stenting. Therefore, it cannot determine the effectiveness and safety of the use of ticlopidine plus aspirin after coronary stenting.

The lead author's reply: "The aim of the review was to compare ASA plus ticlopidine versus ASA plus vitamin K antagonists after coronary stenting and not the efficacy and safety of ASA plus ticlopidine per se."

Safety data concerning VKAs from previous RCTs, observational studies, and population-based studies should supplement the data from the 2 small RCTs in this review.

Given the lack of efficacy of warfarin in the RCTs of the review and the likely underestimate of the bleeding risk and unknown risk for rebound hypercoagulation, further RCTs of VKA treatment to reduce vascular events in people with hypertension would be unethical.

The authors rebutted my comments challenging the efficacy of platelet inhibitors in secondary prevention of patients with hypertension.

Regarding my point that further RCTs of VKA prophylaxis in hypertension, the authors said, "There is inadequate data on warfarin in hypertension per se."

Authors' general observation: "We are not sure what the author [me] refers to being 'evidence-based,' but from his comment we believe that s/he has a different understanding of evidence-based. Being 'evidence-based' means following the 'best evidence' whatever that may be. Evidence can be high quality or lower quality, it can be direct or indirect."

I disagree. By that definition, every test or treatment in medicine is evidence-based.

My comment: In this proposed meta-analysis (published as a review after my feedback letter) of RCTs comparing LMWHs with oral anticoagulants, the control group, VKAs, is not evidence-based to be safe and effective.

Authors' reply: None

My comment: Observational studies were not included in the review to evaluate for HITT and other safety endpoints.

Authors' reply: "We will add both HIT and HITT to the list of outcomes to assess." ....."We have designed the systematic review to include data from RCTs only. We will discuss in the final review the limitation of not including harm data from observational study."

My response: These endpoints were not systematically evaluated in the primary RCTs.

My comment: Rebound hypercoagulability-related adverse events were not monitored.

Authors' reply: "Dr. Cundiff raises the question, whether the incidence of thrombotic events increases within two months of discontinuing heparin therapy. We will investigate this hypothesis if the studies we include in our review report the relevant data."

My response: No RCTs included these data.

My comment: RCTs included in this meta-analysis were subject to selection bias because trials published only as abstracts were not included.

Authors' reply: "Dr. Cundiff describes publication bias, which we will investigate as recommended by the Cochrane Collaboration."

My comment: Besides major bleeding, the primary or secondary endpoints should include fatal bleeding and intracranial bleeding because of the particular importance to patients and clinicians of these events.

Authors' reply: The 2 proposed outcomes are accounted for by the outcomes listed in our protocol. Fatal bleeding events would be counted as deaths. Intracranial bleeding events would be counted as major bleeding events.

My response: Iatrogenic deaths should be counted separately from cancer deaths, and intracranial bleeding rates (fatal and nonfatal) are of special interest to patients and doctors.

None of the proposed primary outcome measures relate to safety concerns with anticoagulant drugs.

The primary endpoint, which determines the main conclusions of the review, should include efficacy and safety measures.

Concerning anticoagulants, given the relatively long natural history of the nephrotic syndrome, the slowed metabolism of anticoagulants with renal impairment, the significant bleeding risk, and the paucity of anecdotal evidence of efficacy, the ethics of conducting RCTs with heparins or VKAs should be questioned.

Bleeding (major or fatal bleeding)

Anticoagulant-associated intracranial hemorrhage

Heparin-induced thrombocytopenia with thrombosis

None of the proposed primary outcome measures relate to safety concerns with anticoagulant drugs.

The primary endpoint, which determines the main conclusions of the review, should include efficacy and safety measures.

Given the relatively long natural history of IgA nephropathy, the slowed metabolism of anticoagulants with renal impairment, the significant bleeding risk, and the paucity of anecdotal evidence of efficacy, the ethics of conducting RCTs with any anticoagulant for any clinical endpoint should be questioned.

See the Introduction

See also Results section: Categories of Methodological Errors and Biases #4 -- RCTs are too small to evaluate risk for HITT and observational studies not evaluated for HITT.

My published commentary, "Evidence-based Medicine and the Cochrane Collaboration on Trial"^[68] comprehensively details the points of contention in this review:

The primary efficacy endpoint combines asymptomatic and symptomatic VTE. This means that it relies primarily on the more frequent surrogate endpoint asymptomatic DVT.

Only a purely clinical endpoint should be used to measure efficacy.

Even symptomatic VTE would be inappropriate as the sole primary outcome measure to determine recommendations, since it does not include adverse events (ie, bleeding).

Combining symptomatic and fatal VTE and fatal complications of the LMWH (major bleeding and HITT) would be the most clinically relevant primary endpoint.

Because of excluding observational studies from consideration in the safety analysis, the bleeding complications will very likely be understated.

Rebound hypercoagulability was not assessed.

The review was published 14 months after my critique of the protocol was received and acknowledged. My points were not addressed.

Authors' conclusions: Implications for Practice section, "Patients receiving venous femoropopliteal, infragenicular bypasses seem to benefit more from VKA than from ASA." Comment: No differences in clinical endpoints were seen in placebo-controlled RCTs while both drugs were associated with unacceptable rates of fatal bleeding (VKAs, 16 [1.2%], aspirin/dipyridamole, 12 [0.9%]).

Rebound hypercoagulability was not assessed.

See also Results section: Categories of Methodological Errors and Biases #11 -- A patented, expensive treatment (not the subject of the review) that is not evidence-based to work is endorsed in the Implications for Practice section.

See Results section: Categories of Methodological Errors and Biases #5 -- Rebound hypercoagulability was not assessed by collecting study data for at least 2 months after anticoagulant withdrawal in most patients

This is a meta-analysis of non-inferiority trials in which neither the control group (shorter-duration VKA treatment) nor the experimental group (longer-duration VKA treatment) is evidence-based to be safe and effective.

Implications for Practice: " ... (We) recommend that injectable anticoagulants should be reserved for patients at higher risk of thrombosis and those with contraindications to physical methods and/or aspirin (Scottish Intercollegiate Guidelines Network 2002)".

My comment: There is no evidence basis for this statement, although it is drawn from another guidelines-producing group (Scottish Intercollegiate Guidelines Network [SIGN]).

Authors reply: "....as we state in the review, the source for the recommendation you refer to is the Hip Fracture Guideline Development Group. Prevention and management of hip fracture in older people. A national clinical guideline (No. 56). Edinburgh: Scottish Intercollegiate Guidelines Network (www.sign.ac.uk), 2002."

My response: Cochrane systematic reviews are supposed to base conclusions and recommendations on RCTs and other published literature, not non-evidence-based opinions of other authorities.

See also Results section: Categories of Methodological Errors and Biases #9 -- Intracranial bleeding was not separately included in the primary or secondary endpoints.

See Results section: Categories of Methodological Errors and Biases #5 -- Rebound hypercoagulability was not assessed.

Because observational, population-based, and/or case-control studies were excluded from consideration in the safety analysis, the bleeding complications were very likely understated.

The primary endpoints (DVTs and PE demonstrated on venograms and noninvasive scans) are surrogates rather than clinical endpoints. Asymptomatic VTE does not correlate with fatal PE.

Standard prophylactic anticoagulation (typically 7-10 days: the control group) following total hip or knee replacement or hip fracture repair provides no proven survival benefit.

Anticoagulation does cause significant risks for bleeding and rebound hypercoagulation events.

Rebound hypercoagulability events were missed in the extended duration of anticoagulant group (about 30 days of low-dose heparins or other drugs) of this review, thus biasing the findings.

In a meta-analysis of RCTs involving VTE prophylaxis of total hip replacement patients, the crude risks for clinically important bleeding (usually wound hematoma) were 0% for compression stockings, 0.3% for controls, 0.4% for aspirin, 1.8% for LMWH, and 2.6% for unfractionated heparin.^[75]

Because of the seriousness of wound hematomas and the frequently associated infections in these patients, SIGN changed the hip fracture and hip and knee replacement surgical VTE prophylaxis guidelines from advising anticoagulants to advising aspirin.^[76]

The primary endpoints, DVTs on venograms and noninvasive scans, are surrogate rather than clinical endpoints. Asymptomatic DVTs do not correlate with fatal PE.

Rebound hypercoagulability events were not assessed.

Despite the absence of benefit of LMWH for any of the major clinical endpoints (death, vascular death, disability), the Implications for Practice appear to justify the use of LMWH and heparinoids in some cases of acute ischemic stroke on the basis of the nonstatistically significant superiority of LMWH and heparinoids over UH in DVT and PE.

The Implications for Practice section should read, "Do not use any anticoagulants in patients with acute ischemic stroke."

Likewise, the Implications for Research section should say, "It would be unethical to conduct further trials of anticoagulants in patients with acute ischemic stroke."

Authors' reply: "The first version of the review was prepared at a time when there was uncertainty about the overall effects of heparins in acute ischemic stroke. Since then, as the commenter states, evidence has emerged which shows that there is no net benefit from the immediate anticoagulation in this setting. However, as is often the case, this evidence did not alter the beliefs of some clinicians, and hence, heparin is still used in some countries, in certain types of patient with acute ischemic stroke, for specific reasons and especially for prevention of venous thromboembolism. This review is now based on the premise that if a clinician plans to treat a patient for some special reason with some form of heparin then the choice of agent should be evidence based."

My rebuttal: I question that premise as a basis for a Cochrane Review. Heparin is dangerous. If it is not evidence-based to improve clinical outcomes, don't prescribe it.

The authors correctly point out that there was no statistically significant increase in major extracranial or intracranial hemorrhage (OR, 1.82 [95% CI, 0.87-3.81] and OR, 1.46 [95% CI, 0.72-2.97], respectively). However, when these 2 categories of major hemorrhage are combined, there is a borderline statistically significant increase in major bleeding (OR, 1.63 [95% CI, 0.98-2.72]; P< .06), which represents 6.2 additional major bleeding episodes per 1000 patients (95% CI, - 0.1 to 16.8) by adding the low-dose heparin.

Risk for bleeding in clinical practice is probably greater than in closely monitored RCTs.

Since adding low-dose heparin to aspirin does not improve clinical outcomes and does add to the bleeding risk, the Implications for Research section should say that further trials of any anticoagulants in acute ischemic stroke would be unethical and any studies under way should be stopped.

Also see Results section: Categories of Methodological Errors and Biases example #9 -- Fatal bleeding was not separately included in the primary or secondary endpoints.

This review of anticoagulant trials, including over 22,000 patients with acute ischemic stroke, found no net benefit with the use of any anticoagulant.

The Implications for Practice section should say that anticoagulants should be contraindicated in patients with acute ischemic stroke.

The Implications for Research section should say that further trials of anticoagulants in acute ischemic stroke would be unethical.

This Cochrane Review should be updated to indicate that VKA treatment should be contraindicated in patients with acute ischemic stroke.

In 2005, the Warfarin and Aspirin for Symptomatic Intracranial Arterial Stenosis (WASID) trial was terminated early because of the high incidence of death in the warfarin group (9.7% in the warfarin group vs 4.3% in the aspirin group; hazard ratio for aspirin relative to warfarin, 0.46; 95% CI, 0.23 to 0.90; P=.02).^[53] The authors concluded, "Aspirin should be used in preference to warfarin for patients with intracranial arterial stenosis."

No further anticoagulant drug research is warranted in this patient group.

The review concluded that aspirin significantly decreases death and dependency in stroke patients.

The Implications for Research section states, ". . . There is also a case for further trials of low dose subcutaneous heparin (or low dose low molecular weight heparin) plus aspirin versus aspirin alone in the prevention of post-stroke deep vein thrombosis and pulmonary embolism, and in reducing neurological disability from the original or recurrent strokes. Such trials would need to include several tens of thousands of patients."

My comment: This review should be updated to reflect that, given the lack of efficacy of anticoagulants in multiple trials and high bleeding risk in stroke patients (see #16 above), further trials with low-dose heparin or low-dose LMWH would be unethical.

Authors' conclusions in the "Implications for Practice" section: "The thromboprophylactic benefits of aspirin in patients with heart failure who are in sinus rhythm are not convincing, but may be detrimental in view of the reduction of ACE Inhibitor benefit and other side effects. It should be emphasized that the evidence for a potential interaction between ACE inhibitors and antiplatelet agents is mainly based on nonrandomized studies, in light of the quality of the studies from which the data are derived and the post hoc nature of the study analyses, the possible interaction should be interpreted with caution."

My comment: The implications for practice are obfuscation -- do not prescribe antiplatelet agents!

The abstract of the WASH trial (the only trial reviewed in this review) concludes, "The benefits of warfarin for patients with heart failure in sinus rhythm have not been established. Antithrombotic therapy in patients with heart failure is not evidence based but commonly contributes to polypharmacy."

Whereas, VKA therapy for heart failure has been tested and is not evidence-based to work, this review gives the false impression that the jury is out and that such treatment may be reasonable.

Adverse events due to rebound hypercoagulability were not monitored.

The Implications for Practice section focuses on beneficial therapies for subgroups of patients: (1) patients with venous graphs and (2) patients with artificial graphs. These subgroups were not defined in the protocol, so post hoc statistical analyses of outcomes of these subgroups should not be emphasized in the Results or Conclusions.

In several places, the Cochrane Handbook warns against problems with post hoc analyses -- for example, page 63: "Post-hoc questions are more susceptible to bias than those asked a priori, and data-driven questions can generate false conclusions based on spurious results."

Adverse events due to rebound hypercoagulability were not monitored.

No benefit was shown in any clinical parameter for claudication patients.

The anticoagulants increased bleeding, including at least 1 case of fatal bleeding.

Rebound hypercoagulation adverse events were not assessed.

Further research of anticoagulants for claudication is inappropriate and unethical.

A major conclusion of the review: Oral anticoagulation prolongs life in people with extensive small-cell lung cancer.

My comment: The survival in small-cell lung cancer was based on a post hoc statistical analysis.

Author reply: It was an a priori determined subgroup, because the protocol states, "We will try to explain heterogeneity, if present, by conducting subgroup analyses. Our a priori factors to explain heterogeneity are the characteristics of participants... and we list 'type/location of cancer' as a patient characteristic."

My rebuttal: If all cell types of cancer were designated for subgroup analysis by this catch-all statement, then adjusting for multiple comparisons (eg, Bonferroni adjustment) would render the P value for the benefit of extensive small-cell lung cancer with oral anticoagulants insignificant.

The authors' reply to my rebuttal is pending.

The authors' concluded, "The evidence suggests that anticoagulants are beneficial, without serious adverse effects, for people with NRAF and recent cerebral ischemia."

Although neither of these small RCTs with careful selection of relatively young patients had intracerebral bleeding (ie, for the 500-plus subject-years of follow-up on anticoagulants), about 12 such bleeding events would be expected in a general practice setting with a similar number of patient-years on anticoagulants.^[43]

The relatively younger age of patients in these RCTs means that efficacy in preventing vascular events will also be lower than in general practice.

Adverse events due to rebound hypercoagulability were not monitored.

Authors' conclusion: "The evidence from two trials suggests that anticoagulant therapy is superior to antiplatelet therapy for the prevention of stroke in people with NRAF and recent non-disabling stroke or TIA. The risk of extracranial bleeding was higher with anticoagulant therapy than with antiplatelet therapy."

See Results section: Categories of Methodological Errors and Biases #6 -- Because of excluding observational, population-based, and/or case-control studies from consideration in the safety analysis, the bleeding complications were very likely understated.

Fatal and intracranial bleeding were not reported

Adverse events due to rebound hypercoagulability were not monitored.

Authors' plain language summary: "Home treatment of deep vein thrombosis (DVT) is cost effective and preferred by people with DVT."

Hospital treatment of DVT with anticoagulants (the control group) is not evidence-based to be safe or effective.^{[1,68,87,88,89]} Therefore, home anticoagulation treatment for DVT (the experimental group) is not proven to be safe and effective.

The relevant issue: As it stands, the guidelines for anticoagulation during pregnancy and the postpartum period by the American College of Chest Physicians^[91] and the Royal College of Obstetricians and Gynaecologists^[92] are arguably the standard of care in the United States and United Kingdom, respectively. Despite the lack of RCT evidence, these opinion-based guidelines recommend anticoagulants in many instances, and they can be referenced in medicolegal cases.

This review appropriately concludes that anticoagulant thromboprophylaxis during pregnancy is not evidence-based to be safe and effective. However, the implications for practice and research do not go far enough.

Since anticoagulation carries risks for bleeding, osteoporosis, and fetal deformity, the appropriate implication for practice would be that thromboprophylaxis with anticoagulants should not be used outside of an RCT.

The Implications for Research section should state that any RCTs of anticoagulation in pregnant women should be placebo controlled.

The issue: Despite the lack of RCT evidence, opinion-based treatment guidelines for pregnant women with thrombophilia by the Royal College of Obstetricians and Gynaecologists and the American College of Chest Physicians include thromboprophylaxis.^[91,92]

My suggestion: "Given the risks of heparin, the implications for practice for the review should say that heparin should not be used for pregnant women with acquired or inherited thrombophilias outside of a placebo or antiplatelet agent controlled RCT."

The lead author agreed to make the change with the next review update.

Authors' Implications for Practice: "Both unfractionated heparin and low molecular weight heparin can be used as effective prophylaxis against postoperative thromboembolic complications after colorectal surgery."

My comment: These conclusions are based on surrogate endpoints (eg, venograms which mostly show asymptomatic DVTs) rather than clinical endpoints (eg, death, fatal PE, symptomatic VTE).

Authors' reply: "We must not forget that the efficacy of heparin prophylaxis against fatal Postoperative PE was established in a methodologically correctly performed trial more than 30 years ago (International Multicentre Trial).^[93] Taken these facts in mind I still mean that the use of postoperative prophylaxis is evidence based, and for comparing different methods the use of venography for detecting subclinical DVT as endpoint is scientifically acceptable."

My rebuttal: While methodologically correct for its time, the International Multicentre Trial preceded the era of early mobilization and mechanical VTE prophylaxis. The follow-up period was only until discharge from hospital, so the deaths caused by rebound hypercoagulation after heparin was stopped were missed. Likewise, in this review adverse events due to rebound hypercoagulability were not monitored.

At my suggestion, the authors completed this review 6 years after they published the protocol. The efficacy and safety of LWMH for acute coronary syndromes depends on the evidence basis of heparin vs placebo for acute coronary syndromes. See correspondence for review #30 below

See Results section #5 -- Rebound hypercoagulability was not assessed by collecting study data for at least 24 hours after heparin withdrawal

The appropriateness of this review is predicated on the efficacy of heparins when compared with placebo where all patients are also treated with aspirin. The authors maintain that their recently published Cochrane Review of this topic establishes the benefit of heparin plus aspirin vs aspirin alone. I dispute that contention. (See my feedback letter for #29 above.)

If neither LMWH nor UH is evidence-based to be beneficial for people with acute coronary syndromes, comparing them is inappropriate.

With additional antiplatelet agents and invasive procedures in recent years, heparins are significantly more hazardous that when the trials in this review were done.

Heparins should not be used for acute coronary syndromes outside of placebo-controlled RCTs.

From the abstract: "The objective of this systematic review is to determine the effectiveness and safety of heparin compared to all other modalities in the treatment of patients with STEMI."

Rebound hypercoagulability-related adverse events should be monitored.

Fatal bleeding and/or intracranial bleeding should be separately included in the primary or secondary endpoints.

RCTs, like those included in this review, report low bleeding complication rates with VKA treatment (eg, major bleeding/yr = 2.2%; intracranial bleeding/yr =0.5%; fatal bleeding/yr = 0.4% [n = 1939]^[96]). In contrast, a recent series of consecutive patients started on warfarin for atrial fibrillation reported higher rates of bleeding complications (major bleeding/yr, 7.2%, intracranial bleeding/yr, 2.5%, fatal bleeding/yr, 0.8% [n = 472]^[43]).

Rebound hypercoagulability-related adverse events were not monitored.

See also Results section: Categories of Methodological Errors and Biases #10 -- Conclusions of review were based on data from RCTs with patients who are not representative of those in general clinical practice.

Plain-language summary: "Insufficient evidence to say if anticoagulants help women with recurrent pregnancy loss without antiphospholipid syndrome"

Authors' recommendation: Large, randomized, placebo-controlled trials

Given the risks of heparin and the potential for harm, if tens of thousands of women have heparin treatment during pregnancy, the main endpoint in the recommended RCT (anticoagulant vs placebo) should be a live, healthy baby in 2 or 3 pregnancies rather than in a single pregnancy.

Rebound hypercoagulability-related adverse events should be monitored.

The bleeding complications will very likely be understated unless observational and/or population-based studies are included in the safety analysis.

Fatal bleeding and intracranial bleeding should be separately reported and added to the primary or secondary endpoints. Patients and physicians would be particularly interested in these safety measures.

This protocol assumes that those oral anticoagulants are evidence-based to benefit patients. However, the critiques of multiple Cochrane reviews in this paper question this assumption.

The authors implied that heparin is effective for CSVT, although the 2 small trials showed no statistically significant reduction in mortality.

The nonsignificant trend depends on a single small trial that was subject to early termination bias.

The authors of this CSVT review called heparin safe on the basis of only 40 cases (about 800 days of heparinization).

CNS bleeding is a common cause of morbidity and mortality in CSVT, so heparin is not safe.

Further placebo-controlled RCTs were inappropriately discouraged ("....patients and doctors may be reluctant to embark upon a new trial that includes a placebo group.").

Authors' conclusions: "No strong evidence was found to conclude thromboprophylaxis is effective to prevent thromboembolic events and safe, in people with unknown risk factors for thrombosis, undergoing knee arthroscopy."

From plain-language summary: "Adverse events were most common in the intervention group (i.e., anticoagulant prophylaxis). The most common complication was minor bleeding with a RR of 2.23 (range 0.99 to 4.99). The number needed to harm was 20."

From Implications for Research section: "Surveillance studies will be important to assess the detection of both adverse events and thrombotic events, with and without the use of LMWH."

Rebound hypercoagulability-related adverse events were not monitored.

The implication for practice should be that anticoagulant prophylaxis should not be used.

The implication for research should be that further RCTs with knee arthroscopy patients exposed to anticoagulants would be unethical.

Anticoagulant prophylaxis statistically significantly benefits surrogate endpoints of DVTs as shown by venograms and other tests, but it has not been shown to improve clinical endpoints (eg, fatal pulmonary emboli, mortality, symptomatic VTE).

The control group for this review (VKAs and LMWHs) is not evidence-based to be effective.

This proposed meta-analysis of non-inferiority trials comparing VKAs and LMWHs with direct thrombin inhibitors is not appropriate to prove efficacy or safety of direct thrombin inhibitors.

The protocol does not call for monitoring for rebound hypercoagulability-related adverse events.

Authors' conclusions: "The use of heparin-bonded catheters is a promising therapy but warrants further studies."

See Results section: Categories of Methodological Errors and Biases #3 -- Placebo or no treatment control was used when saline flushes are widely used and are safe and considered effective.

The primary endpoint is "Days of catheter patency (duration of patency of first catheter, in days)." The most appropriate clinical endpoint would seem to be occlusion of CVCs.

Given the risk for bleeding and HITT, a safety measurement should also be included in a composite primary endpoint.

Fatal bleeding and intracranial bleeding should have been designated separately as primary or secondary endpoints because of the magnitude of these events.

Rebound hypercoagulability-related adverse events should have been monitored.

See Results section of this review: Categories of Methodological Errors and Biases #3 -- Placebo or no treatment control was used when saline flushes are commonly employed as an anticlotting measure clinically; saline flushes are safe and considered effective in preventing CVC-related thrombosis in children.

The primary outcome measure, "CVC-related thrombosis (along the length of, or at the tip of, the catheter) as determined by either color-coded Doppler ultrasonography or contrast venography," is a surrogate endpoint rather than a clinically relevant efficacy measurement.

Instead, the secondary endpoint, "occlusion of CVC (defined as inability to infuse fluids through the catheter due to blockage)," should be made the primary endpoint.

Because RCTs for this indication will probably be too small to evaluate risk for HITT and hemorrhage, observational studies should also be included in the safety analysis.

Fatal bleeding and intracranial bleeding should be designated separately in the primary or secondary endpoints because of the particular importance of these events.

Rebound hypercoagulability-related adverse events should be monitored.

The primary endpoint (symptomatic plus asymptomatic DVT or symptomatic PE) is inappropriate, because asymptomatic DVT (a surrogate endpoint) will dominate. There is no evidence that asymptomatic DVT correlates with clinically important outcomes.

See Results section: Categories of Methodological Errors and Biases #8 -- The primary endpoint should include major bleeding rather than just VTE.

Fatal and intracranial bleeding should be separately designated as primary or secondary endpoints because of the importance of these events.

To properly assess adverse events, including major bleeding and HITT, RCTs should be supplemented with large prospective or retrospective observational studies.

Rebound hypercoagulability-related adverse events should be monitored.

Authors' conclusions: Adjusted-dose warfarin and related oral anticoagulants reduce stroke, disabling stroke, and other major vascular events for those with nonvalvular atrial fibrillation by about one third when compared with antiplatelet therapy.

Given the risk for adverse effects with VKAs, major bleeding should be included in a composite primary outcome measure.

See Results section: Categories of Methodological Errors and Biases #10 -- Conclusions of the review were based on data from RCTs with younger, healthier, and more compliant patients than are typically seen in general clinical practice.

Representative safety data will not come from RCTs with anticoagulation researchers carefully monitoring the international normalized ratios of highly selected patients. Large observational studies need to be 1 way to assess the rates of major and fatal bleeding. For example, in this review VKA-related major bleeding (1.6%/yr) and intracranial bleeding (0.45%/yr) were considerably less frequent than in a recent observational study (7.2%/yr. and 2.5%/yr, respectively).^[43]

Rebound hypercoagulability-related adverse events were not monitored.

Authors' conclusion: "This review suggests a survival benefit of heparin in cancer patients in general, and in patients with small cell lung cancer in particular."

My comment: (See Results section: Categories of Methodological Errors and Biases #13) -- Biases exist in the selection of trials for inclusion in the review: 12 RCTs were eligible but only 5 were included, so the potentially biased selection of trials makes interpretation of the data "favorable to heparin" highly suspect.

Authors' reply: "We agree that it would have been ideal to include data from the trials published as abstracts. . . Interpretation of the findings of this review is limited by the moderate heterogeneity. . . The interpretation of findings is also limited by not including data from the 7 trials published as abstracts only."

These qualifications were not added to the abstract or plain-language summary where people without a subscription to the Cochrane Library could read it.

Authors' conclusion: "Combined unfractionated heparin and aspirin may reduce pregnancy loss by 54%."

Since all 3 RCTs of aspirin vs placebo showed no benefit, aspirin should be avoided and excluded from future RCTs.

Even if adding heparin to aspirin does increase the chances of live birth in a single pregnancy in this situation from 30/70 (43%) to 52/70 (74%), it is not significantly better that the chance of a live birth in 2 pregnancies with aspirin (or placebo) alone (ie, 1 - [0.57 x 0.57] = 68%). Given the risks of anticoagulants, the endpoint of any future RCT should be a healthy baby in 2 or 3 pregnancies rather than 1 pregnancy.

Safety (eg, osteopenia, bleeding, HITT, fetal abnormalities) cannot be determined with these small trials (n = 140).

The conclusion that heparin added to aspirin increases live births and should become the standard of care in women with recurrent miscarriages associated with antiphospholipid antibodies is premature.

In women with recurrent miscarriages, anticoagulants should not be given outside of RCTs.

Authors' conclusion: "Adding antiplatelet therapy, either dipyridamole or low-dose aspirin, to oral anticoagulation decreases the risk of systemic embolism or death among patients with prosthetic heart valves. The risk of major bleeding is increased with antiplatelet therapy."

The use of oral anticoagulation itself is not evidence-based to reduce morbidity and/or mortality in patients with prosthetic valves. It is based on anecdotal reports and historical precedent, not RCTs.

Even without the addition of aspirin, at least 1% per year of patients taking oral anticoagulants experience fatal bleeding each year.

Because adding antiplatelet drugs to oral anticoagulants substantially reduced the risk for thromboembolic events and mortality (61% and 45% respectively), we should have a non-inferiority RCT comparing aspirin alone with aspirin plus a VKA. As a prelude to such a definitive RCT, I am coauthoring a Cochrane protocol titled, "Anticoagulation for prevention of cardiovascular disease in people with prosthetic heart valves."

Authors' conclusion: "There were no randomized trials comparing either anticoagulants or antiplatelet drugs with control. There is, therefore, no evidence to support their routine use for the treatment of extracranial internal carotid artery dissection.... We suggest that a randomized trial including at least 1400 patients in each treatment group (oral anticoagulants and antiplatelet drugs) with this condition is clearly needed."

Antiplatelet agents -- the control group for the primary endpoint comparison in this review (antiplatelet agents vs anticoagulants) -- are not evidence-based to reduce death or dependence in people with stroke from carotid artery dissection.

However, aspirin and other antiplatelet agents are a potential cause of serious bleeding.

Consequently, the Implications for Research section should state that a large RCT of aspirin vs placebo is indicated.

Oral anticoagulants should be avoided in carotid artery dissection on the basis of the poor outcome in patient with acute ischemic stroke (Review #16 above).

Primary objective: To assess the effectiveness of heparin for prevention of catheter related thrombosis.

Main results: ...." There was no statistically significant difference in the risk of thrombosis (RR 0.93, 95% CI 0.58, 1.51)" . . . One of 4 secondary endpoints (catheter occlusion) statistically significantly favored heparin.

Authors' conclusion: "Evidence from this systematic review support the prophylactic use of heparin for PCVC in neonates at a dose of 0.5 IU/kg/hr."

For this review, the control was placebo or no treatment. A considerable amount of literature attests to the safety and efficacy of saline flushes.^[27,28,108] Consequently, saline flushes should have been the control.

Safety data from larger observational and case-control studies should have been included to evaluate these potential complications. For example, a case-control study involving 66 neonates with intraventricular bleeding found an increased risk for this devastating complication with heparin use (OR, 3.9 [95% CI, 1.4-11]), after adjusting for other risk factors.^[109]

Because there was no benefit with heparin for the primary endpoint and considerable risk for devastating intracranial or other bleeding, heparin should not be used.

Main results: "......Five studies reported data on the duration of use of the first catheter. Two of these studies found no statistically significant effect of heparin; two studies showed a statistically significant increase and one study showed a statistically significant decrease in the duration of PIV catheter use in the heparin group."

Authors' conclusion, Implications for Practice section: ".....Because of clinical heterogeneity and heterogeneity in treatment effect, recommendations for heparin use in neonates with PIV catheters cannot be made."

Authors' conclusions, Implications for Research section: "Further research on the effectiveness, the optimal dose, and the safety of heparin is required."

Premature infants are particularly prone to intracranial hemorrhage.

The small number of patients in the 2 RCTs reporting the incidence of intracranial hemorrhage (n = 322) gives no assurance that adding heparin to catheter infusions will not precipitate this dreaded complication.

Given that this meta-analysis found no benefit of heparin on first catheter occlusion (n = 485; OR, 1.00 [95% CI, 0.85-1.16]) or any clinical endpoint, further RCTs of heparin in neonates for this indication are not warranted.

Heparin should be avoided in premature infants.

Authors' conclusions. "Ticlopidine plus aspirin after coronary stenting is effective in reducing the risk of the revascularization, non fatal myocardial infarction and bleeding complications when compared with oral anticoagulants...."

My comment: In this meta-analysis of non-inferiority RCTs, the control group, aspirin plus a VKA, has never been shown to improve clinical outcomes for people with coronary stenting. Therefore, it cannot determine the effectiveness and safety of the use of ticlopidine plus aspirin after coronary stenting.

The lead author's reply: "The aim of the review was to compare ASA plus ticlopidine versus ASA plus vitamin K antagonists after coronary stenting and not the efficacy and safety of ASA plus ticlopidine per se."

Safety data concerning VKAs from previous RCTs, observational studies, and population-based studies should supplement the data from the 2 small RCTs in this review.

Given the lack of efficacy of warfarin in the RCTs of the review and the likely underestimate of the bleeding risk and unknown risk for rebound hypercoagulation, further RCTs of VKA treatment to reduce vascular events in people with hypertension would be unethical.

The authors rebutted my comments challenging the efficacy of platelet inhibitors in secondary prevention of patients with hypertension.

Regarding my point that further RCTs of VKA prophylaxis in hypertension, the authors said, "There is inadequate data on warfarin in hypertension per se."

Authors' general observation: "We are not sure what the author [me] refers to being 'evidence-based,' but from his comment we believe that s/he has a different understanding of evidence-based. Being 'evidence-based' means following the 'best evidence' whatever that may be. Evidence can be high quality or lower quality, it can be direct or indirect."

I disagree. By that definition, every test or treatment in medicine is evidence-based.

My comment: In this proposed meta-analysis (published as a review after my feedback letter) of RCTs comparing LMWHs with oral anticoagulants, the control group, VKAs, is not evidence-based to be safe and effective.

Authors' reply: None

My comment: Observational studies were not included in the review to evaluate for HITT and other safety endpoints.

Authors' reply: "We will add both HIT and HITT to the list of outcomes to assess." ....."We have designed the systematic review to include data from RCTs only. We will discuss in the final review the limitation of not including harm data from observational study."

My response: These endpoints were not systematically evaluated in the primary RCTs.

My comment: Rebound hypercoagulability-related adverse events were not monitored.

Authors' reply: "Dr. Cundiff raises the question, whether the incidence of thrombotic events increases within two months of discontinuing heparin therapy. We will investigate this hypothesis if the studies we include in our review report the relevant data."

My response: No RCTs included these data.

My comment: RCTs included in this meta-analysis were subject to selection bias because trials published only as abstracts were not included.

Authors' reply: "Dr. Cundiff describes publication bias, which we will investigate as recommended by the Cochrane Collaboration."

My comment: Besides major bleeding, the primary or secondary endpoints should include fatal bleeding and intracranial bleeding because of the particular importance to patients and clinicians of these events.

Authors' reply: The 2 proposed outcomes are accounted for by the outcomes listed in our protocol. Fatal bleeding events would be counted as deaths. Intracranial bleeding events would be counted as major bleeding events.

My response: Iatrogenic deaths should be counted separately from cancer deaths, and intracranial bleeding rates (fatal and nonfatal) are of special interest to patients and doctors.

None of the proposed primary outcome measures relate to safety concerns with anticoagulant drugs.

The primary endpoint, which determines the main conclusions of the review, should include efficacy and safety measures.