Please note that the text below is not a full transcript and has not been copyedited. For more insight and commentary on these stories, subscribe to the This Week in Cardiology podcast on Apple Podcasts, Spotify, or your preferred podcast provider. This podcast is intended for healthcare professionals only.
In This Week’s Podcast
For the week ending August 19, 2022, John Mandrola, MD comments on the following news and features stories.
Collider Bias and the Obesity Paradox
The first topic relates to an important bias in making conclusions about cause and effect or correlation. I want listeners of #TWICPodcast to be forever alert to the word “paradox” when it comes to observational studies. But before we talk specifically about a study of body mass index (BMI) in patients with atrial fibrillation (AF), ponder these questions:
What if you were interested in the association between athletic ability and IQ? Is there a relationship? Are jocks dumber? What group would you study to learn this association? In trial lingo, what group would you condition on?
Let’s say you decided to study individuals who attend an elite school, say the University of Virginia (UVA)? The trial lingo is that you condition on being a UVA student. The problem with this is that to get into an elite school like UVA you have to be really smart or really athletic.
Now you see the problem: if you look at athletes at elite schools, they will likely have lower IQs than the regular students. You will find a falsely negative association between IQ and athleticism. But if you looked at this in a general population, then there likely is no relationship.
The journal Open Heart published a paper looking at the influence of BMI on outcomes in patients with AF. This paper offers the #TWICPodcast an opportunity to speak about one of the most important biases in observational studies —selection bias or in this case collider bias.
More than 3000 articles have been published on the obesity paradox and all that I have seen are marred by systemic bias. Collider bias featured prominently in the COVID era. For example, papers showing that obesity in ICU patients with COVID seemed protective relative to normal weight — another paradox.
For this study, the authors used the GARFIELD-AF registry, a registry of more than 52,000 newly diagnosed patients with AF who had at least one stroke risk factor. In other words, the authors conditioned on having AF. Patients from 35 countries were enrolled and followed over time.
For this study, the authors divided the study population into five BMI groups:
|Underweight||BMI < 18|
|Normal weight||BMI 18.5-25|
|Extremely obese||BMI > 35|
The main clinical outcomes were death, stroke, bleeding, and heart failure (HF). Before we get into the results, stop and think. Here they are recruiting people with a problem — AF — and studying the effect of a cause of that problem — obesity. The study found:
A U-shaped relationship between BMI and hard outcomes; both low and high BMI were positively associated with events.
The lowest risk of all-cause mortality was observed at a BMI of about 30 kg/m2.
Below 30 kg/m2, there was a 32% higher risk of mortality per 5 kg/m2 lower BMI (95% confidence interval [CI] 25% to 40%).
Above 30 kg/m2, there was a 16% higher risk of mortality per 5 kg/m2 higher BMI (95% CI 9% to 23%).
Pause there; they are saying that it is worse to be normal weight than to have a BMI in the obese range. They report that a BMI of 30, thought to be obese, had the best outcomes. You had higher risk as your weight went down. A person with a BMI of 24 had a higher risk.
“Our findings support an inverse relationship between BMI and risk of death in AF but only for BMI < ~30 kg/m2.”
They speculate why this would be: because higher BMI has been associated with greater risks of hypertension (HTN), diabetes, and coronary artery disease, which in turn are risk factors for mortality and HF. But we never tell normal weight patients with AF to start eating Cheetos and pop to gain weight. My friends, whenever you see the word paradox always be on alert for collider bias, a form of selection bias. This is a classic case, as is every study I have ever seen with the words paradox and obesity. I will link to the University of Oxford Catalogue of Bias, where they describe the first case of collider bias from none other than David Sackett himself.
In this case, consider the two ways to get AF. Yes, obesity is a cause of AF, but then consider the other causes of why normal weight individuals might get AF — pulmonary embolism, alcohol abuse, cancer, etc. These others may be more sinister causes than obesity, and if you only look at AF patients, there is a false or paradoxical relationship between obesity and AF. Obesity looks protective relative to the other causes of AF. Another great example is from a paper from Steven Stovitz in the Journal of Epidemiological Community Health. Stovitz is an expert on collider bias. A lot of what follows is thanks to Steven.
He and co-authors Jay Kaufman and Hailey Banack looked at a cohort of kids who had HbA1C levels tested at the University of Minnesota. He conditioned on having an HbAIC test.
They then divided them into weight categories (underweight, normal, and obese) and found that when you restrict the sample to only those individuals who got HbA1C tests, obesity is protective against diabetes.
Normal weight kids had > 8 times the risk of having a positive AIC compared with obese individuals.
This is a brilliant paper because they did this on purpose to show the effects of collider bias. They tried to get it into the British Medical Journal Christmas issue but were rejected.
Their paper had all the appearances of a high-quality paper: conducted at an academic center, abided by STROBE checklist, diverse group of urban and rural kids, and it had large numbers and low P-values. But it was purposely biased to make a point about collider bias. The exposure in their paper was obesity, the outcome was an abnormal A1C or having diabetes. The collider was getting a A1C test. Think about it:
You only got in this study if clinicians determined that a blood test for A1C level should be obtained, meaning that they had some type of clinical indication to warrant such a test. Many clinicians draw an A1C level when evaluating obese children and adolescents because recommendations include the consideration of testing for diabetes.
Clinicians also draw an A1C level when children have symptoms of diabetes and appear ill, regardless of bodyweight status.
This is the key to understanding the selection bias: ordering the A1C is an effect of either being obese or appearing ill with symptoms and signs of diabetes.
That is what leads to the spurious association. Those who had the A1C because they appeared ill were much more likely to have diabetes than simply being obese.
Thus, obesity looked protective against diabetes, which is foolish.
In the case of the GARFIELD registry study, we have the exposure — obesity — which can cause the outcome — death, stroke, HF, and the third variable or collider, AF. In other words, the people with AF who did not have a high BMI as a possible cause, must have had something else as a possible cause or they wouldn't have had AF. That other AF-causing illness is what causes the lower-weight patients to have a higher risk for AF. The authors even mention this in the discussion, but it did not affect their conclusion. You need to know about this bias because it is a threat to evidence-based medicine. It is systematic bias that lies hidden in all these papers. In 2016, JAMA published a randomized controlled trial (RCT) looking at the effect of calorie restriction on VO2 max in obese patients with HF with preserved ejection fraction (HFpEF). They found that weight loss improved VO2 max. A positive finding in the primary endpoint.
But here is what they wrote in the discussion section: “Because of the reported HF obesity paradox (lower mortality observed in overweight or obese individuals), before diet can be recommended for obese patients with HFpEF, further studies likely are needed to determine whether these favorable changes are associated with reduced clinical events.”
This was in JAMA! HFpEF is tough to treat and here is a positive study. Of course overweight patients with HFpEF do better when they lose weight.
My friends, whenever you see a study that describes a paradox, look at the population studied. It is, in every instance I have come across, a select population. These studies are all marred by a systematic error of selection bias. Journalists, when you come across one of these purported paradox studies, look up Steven Stovitz at the University of Minnesota, or Jay Kaufman at McGIll, or Hailey Banack at University of Toronto. Get them to comment. They have all written extensively on collider bias.
JAMA-Internal Medicine published a report of an RCT led by researchers at University of California San Francisco looking at the efficacy of a self-measured blood pressure (SMBP) cuff that connects to a smartphone vs a regular self-measured BP cuff. The idea is thus: HTN is a big problem; guidelines recommend self-measured BP and, heck, so do I and probably you too. Yet simply measuring BP at home does nothing. You need feedback and action on these numbers.
Devices that enhance standard SMBP with additional digital support from a paired and connected smartphone application are commercially available. These apps interact with patients and do not require health system involvement. Translation: it makes sense that such an improvement would lead to better BP control.
Well, regular listeners of #TWICPodcas know that things that make sense don’t always pass muster when tested in the RCT — the best test.
This trial was conducted in 23 health systems.
About 1000 patients with HTN were randomly assigned to standard SMBP, and 1000 were assigned to enhanced SMBP.
The primary endpoint was reduction in systolic BP, defined as the difference between clinic BP at baseline and the most recent clinic BP extracted from electronic health records at 6 months.
The results – no difference. Both groups had about 10 mmHg drop in BP.
This was a pragmatic trial. It enrolled a diverse patient population. Pragmatic trials in spaces like this are the way to go because you want to test interventions as they will be used in the real world.
The authors discuss four trials of SMBP. Taken together, they found that the studies support the idea that SMBP produces reductions in systolic BP, but there is minimal augmentation from digitally mediated co-interventions without clinicians.
My point in highlighting this RCT is that we need more of these in the digital health space. There is a huge movement among entrepreneurs and corporations to try to make profits in the health space. Yet, true health is elusive, and before rendering a gadget as a norm, we ought to be teaching the public that digital devices should pass muster in RCTs, lest the benefit of these devices — like so much of the current screening industry — passes to the device makers rather than the device users.
International listeners may find this section unusual because so few countries that allow industry to advertise directly to consumers. Direct-to-consumer ads (DTCA) are common in the United States. You see them everywhere — TV, airports, internet, and sadly, in doctor’s offices. About a week ago, I Tweeted a framed poster that the makers of the WATCHMAN device assumed we would put in areas where patients would be.
You might wonder whether these ads actually work. There are two ways to know this. One is to assume that if a profit-driven company is using them, of course they work because profit-driven companies don’t routinely engage in low-value efforts. The second way to know if these things work, is to study them empirically.
JAMA-Health Forum has published an RCT of sorts in which the authors recruited about 2800 individuals from a US nationally representative sample of people at high risk for cardiovascular disease.
Participants were randomly assigned to one of three interventions: (1) DTCA for heart disease medications; (2) DTCA for heart disease medications with price disclosure; or (3) nonpharmaceutical advertising (control).
Each group watched five 1-minute videos for a total of 5 minutes of advertising exposure. The primary endpoint was an ordinal measure of medical- and lifestyle related intentions, beliefs, and brand perceptions.
The results were positive but in some surprising ways.
There was a positive association between DTCA and medication-related behavioral intentions, including intention to switch medication.
There was a positive association between DTCA and engagement in information-seeking behaviors.
They found no evidence for differential associations of price disclosures in the ads.
There was a positive association with consumers’ favorable perceptions of drug makers.
There was no evidence that DTCA discouraged use of diet and exercise.
Comments. The authors highlighted in their discussion and comments to media, that the DTCA likely does not affect lifestyle behaviors. But I would put emphasis on the other findings. Overall, ads work.
Specifically, the drug ads induced people to learn more information;
Created favorable views of the companies;
Showed that price notification had no effects, which has to be solely explained by many Americans are shielded from price, that is, they have no skin in that game.
I feel tension here. One side of me sees these as awful. I see the huge effect DTC ads have had in helping ensconce a dubious procedure, left atrial appendage closure, as a therapeutic fashion. Patients routinely ask whether they should have a WATCHMAN; 99/100 times these are patients who are doing fine either on or off oral anticoagulants. The ads alone aren’t the only cause of creating a therapeutic fashion. Overly generous guideline recommendations, and an embarrassingly rapturous audience of clinicians, many of whom are young enough to be removed from the debate over the seminal trials, as well as the fear of missing out are all on the causal path for a new therapeutic fashion.
The ads bolster the change in norms because they change public perception. This study finds that ad-watchers feel better about the companies; the ads create curiosity so that when that person enters a hospital lobby and sees more ads for WATCHMAN, they feel like this seems normal, and a big-name place is doing it.
So, part of my brain thinks DTCA should be banned.
But the part of my brain influenced by Hayek’s The Constitution of Liberty, a highly recommended read, thinks that banning such ads is wrong. The part of my brain that sees the benefit of liberty thinks these should be allowed because people aren’t so easily bamboozled. While politics and policies are above the level of this podcast, one of my goals here is to inoculate listeners with protection from being bamboozled.
Just because the hospital across the street is doing a dubious procedure for dubious reasons, does not mean you have to do it.
Just because patients hear about something, does not mean you should do it.
If you have the knowledge from the seminal trials, you can stand above the fray and say, ahem, y’all, this highly popular therapy X or Y doesn’t really seem that great.
Take sacubitril/valsartan, which is being heavily promoted in ads. Listeners here would know that its only positive trial came against mid-dose enalapril. All other sacubitril/valsartan outcome trials have failed to show statistically significant results.
I still use sacubitril/valsartan some but I recognize that it could simply be an expensive afterload reducer.
European Society of Cardiology
Next week, I leave for the ESC Congress in Barcelona. It starts Friday and #TWICPodcas tapes late Thursday or early Friday so I will do some previews. Look for my ESC preview column on theHeart,org | Medscape Cardiology early next week. If you are in Barcelona, say hello.
The week following is our Labor Day holiday weekend. Normally I would take that as a holiday, but since ESC is so important, I will do a recap and take a week off later in the month when I travel for a grand rounds talk.
© 2022 WebMD, LLC
Any views expressed above are the author's own and do not necessarily reflect the views of WebMD or Medscape.
Cite this: Aug 19, 2022 This Week in Cardiology Podcast - Medscape - Aug 19, 2022.