We read the article of Inampudi et al regarding the clinical benefit of spironolactone in the setting of older patients with heart failure (HF) with reduced glomerular filtration rate and reduced ejection fraction (left ventricle ejection fraction< 45%) and we wish to raise concern that should temper the conclusions of this observational study.

The study reports on patients who were prescribed on discharge after an HF hospitalization. However, the risk and/or benefit ratio of mineralocorticoid receptor antagonists (MRAs) has never been tested in the setting of patients with worsening HF. There is a sound pathophysiological rationale that would suggest a beneficial impact of MRAs in this setting. Aldosterone levels increase during hospitalization for worsening HF and higher aldosterone blood levels in patients with worsening HF are independently associated with a higher hazard of mid-term outcomes. In contrast, within this setting, patients are exposed to frequent and rapid changes in renal function that may increase the risk of worsening renal function and of hyperkalemia that may warrant more frequent monitoring and more careful dose optimization, than in stable outpatients. A clinical trial is needed to assess the clinical utility of MRAs in patients with worsening HF. Such a trial is being conducted, comparing 2 MRAs (A Randomized, Double-blind, Double-dummy, Multi-center Study to Assess Safety and Efficacy of Different Oral Doses of BAY94-8862 in Subjects With Emergency Presentation at the Hospital Because of Worsening Chronic Heart Failure With Left Ventricular Systolic Dysfunction and Either Type 2 Diabetes Mellitus With or Without Chronic Kidney Disease or Chronic Kidney Disease Alone Versus Eplerenone [ARTS-HF] study, NCT01807221).

Second, 38% of the patients included had also a prescription of potassium supplementation, although patients taking these drugs were usually excluded from MRAs trials. It is quite surprising to find such important proportion of potassium supplementation prescription in such frail population without any data on the biologic follow-up of serum potassium or creatinine.

The Randomized Aldactone Evaluation Study (RALES), Eplerenone Post-Acute Myocardial Infarction Heart Failure Effi cacy and Survival Study (EPHESUS), and Eplerenone in Mild Patients Hospitalization And SurvIval Study in Heart Failure (EMPHASIS-HF) trials, as mentioned by the investigators, did not include patients with serum creatinine >2.5 mg/dl (221 μm/l) for the first 2 and estimated glomerular filtration rate (eGFR) <30 ml/min/1.73 m ² for the latter. However, many patients included in these trials had an eGFR <60 ml/min/1.73 m ² (e.g., 32.2% in EMPHASIS-HF). In addition, most patients included in EMPHASIS-HF had a hospitalization for a CV event—including worsening HF—within 6 months before inclusion. We recently showed that in patients with eGFR <60 ml/min/1.73 m ² treated by eplerenone (mean age 71.1 ± 7.5 years old and mean eGFR in this group 48.59 ± 7.9 ml/min/1.73 m ² ) there was no significant excess of severe hyperkalemia with eplerenone, whereas eplerenone was effective to reduce the primary end point (hospitalization for HF or death for cardiovascular reason; 0.69 [95% confidence interval 0.56 to 0.86; p = 0.0008]). The same was true in patients aged >75. The planned post hoc analysis of a large trial has a lot more scientific value than the assessment of MRAs efficacy on observational data.

Finally, the investigators used propensity score–adjusted survival analysis. We believe that these analyses are suboptimal. Propensity score matching is much more efficient in reducing attribution bias as demonstrated by Austin. Actually 1/1 matching may expose to loss of power and they could have used 4/1 matching. This N/1 matching in conjunction with a Cox model analysis stratified on propensity score matching strata would have probably resulted in a minimal loss of power and a better assessment of treatment effect.

In an observational study, observations are collected without assignment, and no adjustment however elaborate it may be can protect against the “unavoidable risk of selection bias and of systematic differences in outcomes that are not due to the treatment itself”. Consequently, analyses assessing treatment effect using observational data set are at best hypothesis generating. In the setting of high attribution bias—as what is probably observed in patients with low eGFR discharged for acutely decompensated HF in whom MRA is currently nonindicated–results of an observational study should only inform about safety. The neutral effect of spironolactone on CV outcomes reported by Inampudi et al does not hamper the need for a controlled clinical trial assessing the risk-to-benefit ratio of MRAs and more generally of HF drugs specifically in these frail elderly patients with frequent co-morbidities and chronic kidney diseases, insufficiently represented in HF clinical trials so far.