If It Seems Too Good to Be True...
Since the publication of the CVD-REAL trial last year—and its sequel, CVD-REAL 2, in March—a lot of confusion has ensued among clinical practitioners about whether to believe the "noise" around the purported primary prevention cardiovascular benefits of sodium-glucose cotransporter 2 (SGLT2) inhibitors in patients with diabetes.
The original study included 309,056 subjects with type 2 diabetes from the United States and five European countries. A 51% relative risk reduction (RRR) in all-cause mortality was observed in favor of SGLT2 inhibitors—a benefit miraculously accrued in less than 9 months of average follow up.
The 2.0 version of the study now "corroborates" these findings in a study sample of 470,528 participants from six different countries in the Asia Pacific, Israel, and Canada, with a 49% RRR for death over a 224-day mean duration of treatment. Of note, both CVD-REAL studies suggest a similar mortality reduction in both the primary and secondary prevention subgroups.
So, here are some obvious next questions that need to be asked about this too-good-to-be-true mortality benefit identified in the primary prevention subgroup analyses:
Why are these observational data inconsistent with randomized, controlled trials of SGLT2 inhibitors?
Could this observational result be influenced by residual or unmeasured confounding and/or bias? What is the utility of the propensity score matching technique that was employed in the CVD-REAL studies?
Most importantly, what are the clinical and guideline implications of the CVD-REAL studies?
Inconsistencies With Randomized, Controlled Trials
To date, two large cardiovascular outcome trials with SGLT2 inhibitors have shown superiority for major adverse cardiovascular events (MACEs): EMPA-REG OUTCOME and CANVAS.
However, MACE benefit in both trials was only observed in secondary prevention (ie, those with established cardiovascular disease [CVD]).
The EMPA-REG trial only included subjects with CVD history. About one third of the CANVAS subjects had no previous CVD, but in a stratified analysis for MACE, the hazard ratio was statistically significant in the secondary prevention subgroup but not in the primary prevention subgroup.
Hence, these results are contrary to the large mortality benefit suggested by CVD-REAL in the primary prevention subgroup.
How Would It Work?
Mechanistically, also, it appears that the cardioprotection induced by SGLT2 inhibitors may be limited to individuals with an already "diseased" heart. Both of the leading hypotheses (hemodynamics and the ketone bodies as "thrifty" fuel) explain the CVD benefits of this drug class only when a malfunction of cardiomyocytes is present at baseline.
In a recent pathophysiology-based review of cardiovascular outcomes trials in diabetes, Dr Bernard Zinman and I contended that cardioprotection from SGLT2 inhibitors in early diabetes remains uncertain. We believe that the upcoming DECLARE trial results offer the best opportunity to answer this question reliably.
DECLARE—a large, randomized trial with dapagliflozin—has followed about 10,000 primary prevention subjects for over 4 years. This nearly completed trial is expected to report results within the next 6-12 months.
CVD-REAL: Confounded and Biased?
The correct conclusion from both of the CVD-REAL studies should be that subjects who were prescribed an SGLT2 inhibitor were destined to die less—not that the death reduction was due to their prescription for an SGLT2 inhibitor. That inference of causality, to me, is the basic difference between observational studies and randomized, controlled trials. Only the latter can test a hypothesis and infer causality.
The CVD-REAL studies offer a clear example of how confounding and biases invalidate study results.
Propensity score matching is just another fancy statistical technique that can adjust only the known and measured confounders and hence should be considered similar in value (not superior) to the standard multivariate analysis employed in most observational studies.
The important point is that neither of the above statistical techniques (multivariate or propensity matched analysis) can account for the unknown or unmeasured confounding in observational studies.
It is well known that randomization in large cardiovascular outcomes trials matches both the known as well as the unknown confounders among trial arms and hence represents a much higher level of evidence.
Baseline characteristics from both CVD-REAL studies suggest major confounding problems due to unmeasured patient- and physician-level factors.
Here are some clues to the unmeasured confounding issues in CVD-REAL: The SGLT2 inhibitor cohort in both studies received more cardioprotective medications (statins, angiotensin-converting enzyme [ACE] inhibitors, angiotensin II receptor blockers [ARBs] , metformin, glucagon-like peptide-1 [GLP-1] receptor agonists, etc), despite being younger and having fewer comorbidities. This suggests the presence of unmeasured confounding from patients' socioeconomic factors, health education status, or their physician-perceived adherence to medications, which likely influenced the decision to prescribe an SGLT2 inhibitor medication versus another glucose-lowering drug.
Additionally, a number of biases seem to be at play in the twin CVD-REAL studies.
One of the obvious ones is the selection bias. Could physician assessment of a patient's limited life expectancy have influenced their medication choice (ie, those subjects who were more likely to die within 9 months may not have been prescribed an SGLT2 inhibitor)?
Another potential bias is the immortal time bias, which means that the study neglected to account for the time duration when other medications were being used prior to initiation of the SGLT2 inhibitor within the SGLT2 study cohort (eg, if a subject was on a sulfonylurea before an SGLT2 inhibitor was added, the time that he/she lived [ie, did not die] while on the sulfonylurea was not counted).
Clinical and Guideline Implications
In my opinion, the CVD-REAL studies have no practical clinical implications whatsoever.
Indeed, CVD-REAL results failed to influence the recently updated clinical practice guidelines in the United States and Canada.[12,13] These guidelines continue to reserve their preferential recommendation for SGLT2 inhibitors only within the secondary prevention population.
Over the past two decades, preventive diabetology has adopted a cardiology-like reliance on high-level randomized, controlled trial evidence to guide clinical decisions. This paradigm shift in the evidence base, which started with the publication of the landmark DCCT (type 1 diabetes) and UKPDS (type 2 diabetes) trials in the mid to late 1990s, has been accelerated by the US Food and Drug Administration-mandated cardiovascular outcomes trials over the past 10 years.
My suggestion: Let's "keep calm and carry on" until randomized, controlled trial evidence provides us a clear answer to the important question of primary prevention of heart disease with SGLT2 inhibitors.
Medscape Diabetes © 2018 WebMD, LLC
Any views expressed above are the author's own and do not necessarily reflect the views of WebMD or Medscape.
Cite this: SGLT2 Inhibitors: Unreal Primary Prevention? - Medscape - May 15, 2018.