Empirical assessment of constraints and credibility when emulating a placebo-controlled trial with the clone-censor-weight approach

To demonstrate nuances of emulating placebo-controlled trials in the target trial emulation framework using claims, we aimed to expand evidence from the TOPCAT trial on spironolactone effectiveness in heart failure with preserved ejection fraction (HFpEF) to the U.S. HFpEF population. We compared spironolactone initiation and continued use versus non-initiation in 2012-2020 Medicare claims with the clone-censor-weight approach for five effectiveness endpoints. Anticipating threats to validity, we pre-specified 1) benchmarking against TOPCAT Americas results, and 2) a negative control outcome (non-cardiovascular mortality). To demonstrate frequent investigator-induced biases outside the target trial emulation framework, we additionally implemented a ‘naïve’ ever- vs never-user comparison. Here we show results based on 320,881 patients with heart failure and preserved ejection fraction (mean age 80.6 years (standard deviation 8.37); female 62%) from the overall Medicare cohort, of which 49,729 qualify for benchmarking against TOPCAT. In the benchmarking cohort, relative risks with spironolactone use compared to non-use for effectiveness outcomes range from 0.97 (95% confidence interval: [0.94; 1.01]) for the composite of heart failure hospitalization, cardiac arrest and cardiovascular death to 1.14 (95% confidence interval: [1.11; 1.18]) for all-cause mortality. Non-cardiovascular mortality results suggest presence of residual confounding. Bias correction improves agreement with the TOPCAT estimates, while the naïve analysis produces substantially biased results. In emulations of placebo-controlled trials, residual confounding remains a persistent threat. It is critical to pre-specify built-in guardrails to detect and address this bias. We used U.S. Medicare insurance data from over 320,000 patients to understand if the drug spironolactone benefits patients with heart failure caused by a stiffened heart muscle, after a previous clinical trial (TOPCAT) had shown mixed results. We compared heart failure hospitalization and death outcomes between real-world spironolactone users and similar non-users. We designed the study to closely mimic a randomized clinical trial, rather than a face-value comparison, to reduce distortions of results by hidden differences between treated and untreated patients. We used built-in checks to detect these distortions (biases). After adjustments, spironolactone reduced heart failure hospitalizations, but not deaths. We also showed that less careful analyses produced misleading results, highlighting that real-world studies need safeguards against bias. Stephan et al. emulate a placebo-controlled target trial of spironolactone for heart failure with preserved ejection fraction in U.S. Medicare claims. They demonstrate that clone-censor-weighting helps avoid immortal time bias, while benchmarking against a reference trial and use of a negative control outcome help assess residual confounding.