Chronic diuretic use and increased mortality and hospitalization in heart failure

Dyspnea and fluid overload are hallmarks of chronic heart failure, and they are often treated with nonpotassium-sparing diuretics.¹ Despite the fact that diuretic therapy is associated with neurohormonal activation and disease progression in heart failure,^2-4 its use is often required to control fluid overload and relieve symptoms, and is recommended in the 2005 American College of Cardiology/American Heart Association chronic heart failure guidelines.¹ Little is known, however, about the long-term effects of diuretics in patients with chronic heart failure.⁵ The recommendation for the use of diuretics in heart failure is the only class I recommendation for stage C heart failure that is based on level C evidence (only consensus opinion of experts, case studies, or standard of care).¹ The purpose of our study was to determine the effect of diuretic use in a cohort of heart failure patients matched according to propensity score.

Subjects and methods

The current study was a post-hoc propensity score analysis⁶ of the Digitalis Investigation Group (DIG) trial.⁷ The DIG dataset was suitable for this study because 7788 subjects with chronic ambulatory systolic and diastolic heart failure who were predominantly in New York Heart Association (NYHA) class I and II were enrolled in the study. DIG participants were in sinus rhythm, and more than 90% of these individuals were receiving angiotensin-converting enzyme (ACE) inhibitors. Because the DIG trial was conducted in the early 1990s, before beta blockers were approved for use in heart failure, data on beta blocker use were not collected. This analysis focused on a subset of 2782 patients who were matched 1 to 1 based on their propensity to receive diuretics. At baseline, 1391 patients were receiving diuretics and 139 were not.

The primary outcome of this analysis was all-cause mortality; secondary outcomes included all-cause hospitalizations and mortality and hospitalizations due to heart failure during a median follow-up period of 40 months. Because subjects were not randomly assigned to receive diuretics, significant imbalance in most baseline covariates, resulting from selection or indication bias, existed between patients who received diuretics and those who did not (Table, pre-match panel). For example, patients receiving diuretics were older and sicker than those not receiving diuretics.

Click on table to view in larger window (PDF).

Table. Baseline subject characteristics by diuretic use before and after propensity score matching.

ACE indicates angiotensin-converting enzyme; NYHA, New York Heart Association; CT,

Click on table to view in larger window (PDF).

computed tomography; SD, standard deviation.

To reduce this imbalance in covariate distribution between patients receiving and not receiving diuretics, statistical probability of diuretic use was calculated for each subject based on his or her baseline characteristics.^7,8 This conditional probability for diuretic therapy, which is also known as a propensity score for diuretic therapy, was then used to match patients with or without diuretics. This is, in a sense, similar to randomization in that each patient within a matched pair has an equal probability of receiving therapy; however, 1 subject actually receives therapy and 1 does not. However, unlike a randomized clinical trial of diuretics, in which every patient would have a 50% probability of receiving diuretics, in a nonrandomized study, there may be a wide range of probabilities of diuretic receipt. The purpose of propensity score matching is to identify patients with similar propensity or probability who were receiving and not receiving a therapy and match them. For example, a subject with a 40% probability of receiving a diuretic would be matched with a subject not receiving a diuretic but with the same 40% probability of receiving one.

To determine the effectiveness of propensity score matching in reducing selection bias, absolute standardized differences for each covariate between patients receiving and not receiving diuretics were calculated for each covariate and were found to be less than 10%, suggesting satisfactory bias reduction.⁹ Kaplan-Meier and matched Cox proportional hazards analyses were used to estimate the effect of diuretics on outcomes. The key limitation of propensity score matching is that unlike randomization, it cannot ensure whether all unmeasured covariates were also balanced. Therefore, a formal sensitivity analysis was conducted to determine whether the effect of diuretics was sensitive to a potential hidden covariate. Subgroup analyses were performed to determine if the effect of diuretics on mortality was homogeneous across various groups.

Results

P =

P =

The mean age (± SD) of the subjects was 63 (± 11) years; 19% of subjects were women, and 11% were non-whites. After matching, all measured baseline covariates were balanced between subjects receiving and not receiving diuretics (Table, post-match panel). Overall, 25% of subjects died during a median follow-up period of 40 months, including 7% of subjects whose death was due to worsening heart failure. Kaplan-Meier plots for mortality and hospitalization due to all causes are shown in Figure 1. Death from all causes occurred in 21% of subjects not receiving diuretics and 29% of subjects receiving diuretics (hazard ratio [HR] = 1.31; 95% confidence interval [CI], 1.11-1.55; .002; Figure 1A). Death from heart failure occurred in 6% of subjects receiving a diuretic and 9% of subjects not receiving diuretic (HR = 1.36; 95% CI, 0.99-1.87; .056).

P <

P <

When the analysis was repeated using traditional multivariable risk adjustment for all 7788 subjects in the full cohort, 34% of subjects died. Death due to all causes occurred in 21% and 37% of subjects receiving and not receiving a diuretic, respectively (adjusted HR = 1.28; 95% CI, 1.13-1.45; .001). When the analysis was repeated comparing subjects who always received diuretics during the first 24 months (n = 2984) and who never received diuretics during that period (n = 781), death from all causes occurred in 8% of never users and 19% of ever users of diuretics (adjusted HR = 1.81; 95% CI, 1.38-2.38; .001). The results of the sensitivity analysis showed that for the lower survival time to be attributed to an unmeasured covariate (that was not associated with any measured covariates and not a predictor of mortality) rather than to the effect of receiving diuretics, the odds of a subject receiving a diuretic would need to be increased by about 12%.

Figure 1. Kaplan-Meier plots for cumulative risk of (A) death and (B) hospitalizations due to all

causes. HR indicates hazard ratio; CI, confidence interval.

P =

P =

Hospitalizations due to all causes occurred in 61% of subjects, of which 20% were due to worsening heart failure. All-cause hospitalizations occurred in 59% of subjects not receiving diuretics and 64% of subjects receiving diuretics (HR = 1.15; 95% CI = 1.02-1.29; .023; Figure 1B). Hospitalizations due to worsening heart failure occurred in 18% of subjects receiving diuretics and 23% of subjects not receiving them (HR = 1.37; 95% CI, 1.13-1.65; .001). The effect of diuretics on mortality was observed in all subgroups of subjects, including those with NYHA class I and II symptoms and those receiving ACE inhibitors (Figure 2).

Hazard ratio (HR) and 95% confidence interval (CI) for all-cause mortality in subgroups of

patients with heart failure. NYHA indicates New York Heart Association; ACE, angiotensin-converting

enzyme; HR, hazard ratio; CI, confidence interval.

Figure 2.

Discussion

In a propensity score-matched cohort of heart failure subjects in which subjects receiving and not receiving a diuretic were well balanced in all measured baseline covariates, long-term use of diuretics was associated with a significant increase in long-term mortality and hospitalization. These effects were seen in a wide spectrum of ambulatory chronic systolic and diastolic heart failure subjects, 80% of whom had NYHA class I and II symptoms and 90% of whom were receiving ACE inhibitors.

This is the first report based on a nonrandomized study that used a propensity score technique to successfully reduce imbalance in all measured baseline covariates and yet show a significant negative effect on major natural history end points of heart failure. These findings are important because diuretics are commonly used, and a combination of diuretics and ACE inhibitors is the most commonly used drug regimen for heart failure therapy.

Diuretic use is often associated with worsening renal function and electrolyte imbalance, both of which may potentially increase the risk of death and hospitalization, especially in the short term. However, the long-term effect of diuretics in heart failure is most likely due to the activation of renin-angiotensin and sympathetic neurohormones. These neurohormones are already elevated in heart failure, and most evidence-based heart failure therapy tries to antagonize these neurohormones. Thus, by further elevating these neurohormones, diuretics probably increase disease progression and poor outcomes.

These findings raise concerns about chronic use of diuretics for ambulatory chronic mild-to-moderate heart failure (NYHA class I-II) and suggest that very tight control of symptoms in these patients might come at a cost of increased mortality and morbidity. Clinicians should emphasize tighter salt restriction; however, following a salt-restricted diet may not eliminate the need for diuretics. There is evidence that diuretics may increase one's desire for salt and that aldosterone-induced myocardial damage is worse in a high-salt environment, thus engendering a vicious cycle of diuretic-salt-diuretic.^10,11 Before increasing the dose of the diuretic, symptomatic heart failure patients may also be given low-dose digoxin.¹² Recent evidence suggests that torsemide (Demadex) may be superior to furosemide (Lasix).^4,13,14 The lack of data on the long-term effect on outcomes and cost of torsemide, however, temper enthusiasm for a wider role for this drug.

The key limitation of this study is its nonrandomized design and the potential effect of a hidden bias. The effect of diuretics on mortality and morbidity observed in this study could be potentially explained by a hidden covariate. For this to happen, the covariate would need to be associated with both diuretic use and mortality, and not be associated with any of the patient characteristics presented in the Table. The presence of such a covariate seems clinically implausible. Even though a sensitivity analysis can determine the effect of such a putative hidden confounder, sensitivity analysis cannot determine if such a confounder exists or not.

Conclusions

The results of this study show that the use of diuretics may be associated with poor outcomes in chronic mild-to-moderate heart failure. Although these results are based on a nonrandomized design, use of the propensity score is one of the best available techniques for bias reduction. Evidence derived from rigorous methodology such as this can provide temporary evidence and impetus for randomized clinical trials.¹⁵ Clinicians should be cognizant of the deleterious effects of diuretics when using these drugs in heart failure patients who are asymptomatic or minimally symptomatic.

Acknowledgment

Dr Ahmed is supported by the National Institutes of Health through grants from the National Institute on Aging (1-K23-AG19211-04) and the National Heart, Lung, and Blood Institute (1-R01-HL085561-01 and P50-HL077100).