We thank Drs Qadir and Naeem for their insightful comments. We studied 1765 patients in the Americas cohort of TOPCAT, a randomized trial of spironolactone vs. placebo in patients with heart failure and preserved ejection fraction (HFpEF). We showed that use of and higher doses of loop diuretics were associated with higher risk of cardiovascular (CV) death and hospitalization for HF, and that thiazide diuretics alone were not associated with any endpoints, whereas thiazides added to loop diuretics were associated with additional risk for all endpoints
Drs. Qadir and Naeem point out that our primary analyses were not adjusted for natriuretic peptide levels or fluid status and congestion. This is correct. These variables were not available or only partly available in TOPCAT. This is a great example of confounding. Indeed, even though we studied a randomized controlled trial (RCT) cohort, the research question was observational and necessarily limited by confounding. Confounding becomes a problem when it affects both the independent variable (eg, more congestion would lead to more diuretic need), and the dependent variable (eg, more congestion would lead to worse outcomes, independently of diuretic need).
Some confounding variables are commonly measured, entered into multivariable analyses, and thus adjusted for. Examples could be age, estimated glomerular filtration rate (eGFR), or presence of atrial fibrillation. For example, observational studies of the adjusted association between ACE-inhibitors/ angiotensin receptor blockers (ARBs) and outcomes in heart failure with reduced ejection fraction (HFrEF) yield similar improvement in outcomes as do RCTs. , In contrast, other confounders are commonly not measured. Examples could be residual congestion, diuretic resistance, ascites, or hypokalemia. Accordingly, observational studies of the association between mineralocorticoid receptor antagonists (MRAs) and outcomes in HFrEF suggest no improved outcomes with MRAs (because MRA users had more severe HF), whereas RCTs have proven that MRA use improves outcomes in HFrEF.
Thus, treatment effect or causal risk factors cannot be assessed in observational studies. But for our purposes it may not matter. Our research question was whether loop and thiazide diuretic use could enrich clinical trials for events and our findings show that it can. It does not matter whether diuretics are causal risk factors or associated risk markers. “What you see is what you get”: If you require loop diuretic use for inclusion in a trial you will get higher event rates than if you do not. , Thus, loop diuretic use is a marker for greater risk (and thiazide on top of loop a marker of additional risk), just like natriuretic peptides are markers of risk, when used as inclusion criteria in trials It is not the natriuretic peptides that cause events, but natriuretic peptides are markers of HF severity, and HF severity causes events. Whether risk markers such as natriuretic peptide levels or loop diuretic or loop plus thiazide diuretic use also enrich for treatment effect is a completely different matter, and unknown.
Declaration of competing interest
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
