This propensity-matched study, in which patients with and without diabetes were well balanced in all measured baseline characteristics, including traditional risk factors and comorbidities, found that diabetes was associated with increased mortality and hospitalization in ambulatory patients who had chronic, mild-to-moderate heart failure and were receiving angiotensin-converting enzyme inhibitors. These findings also highlight the sex- and age-related variations in the effect of diabetes in these patients.
Diabetes and heart failure often coexist, and the presence of diabetes in heart failure patients is associated with increased morbidity and mortality.1,2 The poor prognosis in these patients may be attributed, in part, to the risk factors and the comorbidity burden resulting from diabetes.1,2 Diabetes also appears to directly activate the renin-angiotensin-aldosterone and sympathetic nervous systems, which may cause a specific diabetic cardiomyopathy.3 Because heart failure patients with diabetes have many risk factors and comorbidities, it is often difficult to know to what extent the poor prognosis is intrinsic to diabetes and to what extent it results from the comorbidities associated with diabetes. We conducted a study to evaluate whether diabetes is associated with poor outcomes in heart failure independent of associated risk factors and comorbidities. Subjects with and without diabetes were well balanced in all measured baseline covariates.
We used observational data because it is not possible to randomize subjects with heart failure to have or not have diabetes. Although outcome-based multivariable risk adjustment models can account for potential confounding covariates (other risk factors and comorbidities), interpretation of these results is somewhat limited because of concern over possible bias.4 Propensity score matching, however, can be used to compile cohorts of subjects with and without an exposure (ie, diabetes) who would be equivalent in all other measured baseline factors.5-7 Even more important, propensity-matched studies are done without prior knowledge of the outcomes, just as investigators remain blinded during the design phase of a randomized clinical trial, and the magnitude of bias reduction may be assessed objectively with standardized differences.4,6,8,9 In the current study, propensity score matching was used to determine the degree to which the effects of diabetes on outcomes in heart failure are caused by shared risk factors and comorbidities. Furthermore, because data regarding the sex- and age-related variations in the effect of diabetes on outcomes in an ambulatory chronic mild-to-moderate heart failure patient population are sparse, we assessed sex- and age-related variations.
Subjects and methods
This study was a post-hoc propensity score-based analysis of the Digitalis Investigation Group (DIG) trial.10 From 1991 to 1993, 7788 subjects with chronic ambulatory systolic and diastolic heart failure were recruited from 302 centers in the United States and Canada. Most of these subjects had New York Heart Association functional class I or II heart failure and were receiving angiotensin-converting enzyme (ACE) inhibitors. The subjects were randomly assigned to receive either digoxin (Lanoxin) or placebo. DIG participants were suitable for this study because data on a history of diabetes as a comorbid condition were collected at baseline for all subjects in the study. A total of 2218 subjects of the 7788 DIG participants (29%) had diabetes.
The primary end points for this analysis were all-cause mortality and all-cause hospitalization. DIG participants were followed-up for a median of 38 months, and vital status data were complete for 99% of subjects.11 Heart failure subjects with diabetes tended to be sicker and to have a higher comorbidity burden, which could have increased their risk of poor outcomes. These differences in between-group baseline characteristics are shown in Table 1
. Baseline characteristics for subjects with and without diabetes before and after
propensity score matching.
The propensity score is the probability of having or the propensity to have an exposure, given a set of measured baseline characteristics.4-6,8,12-14 In a randomized clinical trial, patients generally have a 50% probability or propensity to receive a treatment, but in a nonrandomized setting, the probability of receiving a treatment may vary based on various patient and care characteristics. For example, a young heart failure patient with a low ejection fraction and normal serum creatinine and potassium levels being seen by a cardiologist in an academic institution will likely have a high probability of receiving an ACE inhibitor. An elderly heart failure patient with a normal ejection fraction and high serum creatinine and potassium levels being seen by a noncardiologist in a nonacademic setting, however, will likely have a low probability of receiving an ACE inhibitor. One can estimate these probabilities and use them to match subjects, ensuring that they will be balanced on all other measured baseline characteristics (Table 1). Unlike regression-based multivariable risk adjustment models, propensity score methods allow investigators to design nonrandomized studies without knowledge or access to outcomes data, a key feature of randomized clinical trials.
Although the propensity score technique was originally developed to reduce imbalances in baseline characteristics between subjects in 2 exposure or treatment groups, it also can be used to achieve balance in subjects in nondrug exposure groups (ie, based on diabetes status).7,9,15-18 Using a nonpar-simonious multivariable logistic regression model, the probability of having diabetes was measured for each subject based on baseline characteristics and then was used to match subjects with similar propensity scores. This resulted in a cohort of 2056 pairs of subjects with and without diabetes who were well balanced in all measured baseline covariates and who really only differed with regard to the presence of diabetes (Table 1). To determine whether the propensity score matching was effective in achieving balance, absolute standardized differences for each covariate between subjects with and without diabetes were estimated and found to be <10% for all and <5% for most covariates, suggesting a satisfactory baseline balance.6,8,14
Kaplan-Meier and matched Cox proportional hazards analyses were used to estimate the effect of diabetes on outcomes. The interaction effect of diabetes with sex was determined by using Mantel-Haenszel tests of homogeneity and a Cox proportional hazards model with sex, diabetes, and their multiplicative interaction term as covariates in the model, first in all subjects and then separately in those <65 years and ≥65 years.
Subjects had a mean (±SD) age of 64 (±11) years; 27% of subjects were women, and 17% were non-whites. After matching, essentially all measured baseline covariates were well balanced between subjects with and without diabetes (Table 1).
Overall, 36% of subjects died from all causes, and 69% were hospitalized for any cause during a median follow-up period of 38 months. Compared with a 40% rate of all-cause mortality in subjects with diabetes, only 33% of those without diabetes died (hazard ratio [HR] when diabetes was compared with no diabetes, 1.29; 95% confidence interval [CI], 1.16-1.42; <.001). Similarly, compared with a 73% rate of hospitalization among subjects with diabetes, only 65% of those without diabetes were hospitalized for any cause (HR when diabetes was compared with no diabetes, 1.28; 95% CI, 1.19-1.38; <.001).
Kaplan-Meier plots for cumulative risk of (A) all-
cause mortality and (B) all-cause hospitalization by diabetes
status and sex.
Kaplan-Meier plots for male and female heart failure subjects with and without diabetes are shown in Figure 1. The relative hazard for mortality associated with diabetes was significantly more pronounced in women (propensity score-adjusted HR, 1.73; 95% CI, 1.40-2.15; <.001) than in men (propensity score-adjusted HR, 1.17; 95% CI, 1.04-1.31; = .008; Table 2), with an adjusted for interaction of .002. Similarly, the hazard for hospitalization was significantly higher among women (propensity score-adjusted HR, 1.49; 95% CI, 1.29-1.72; <.001) than among men (propensity score-adjusted HR, 1.21; 95% CI, 1.11-1.31; <.001; Table 2), with an adjusted for interaction of .008.
Table 2. Effects of diabetes on all-cause mortality and all-cause hospitalization.
The sex-diabetes interactions on mortality in heart failure subjects <65 years and ≥65 years are shown in Figure 2. Diabetes-related death was similar between both sexes in younger subjects (6% and 8% for men and women, respectively), but was much higher in elderly women (19% and 4% for women and men, respectively). The increased mortality rate observed in elderly women compared with that of elderly men was significant (adjusted for interaction = .005; Figure 2).
Figure 2. Age-related variations in sex—diabetes interaction
Our study showed that in a propensity-matched sample of chronic heart failure subjects with and without diabetes who were well balanced in all measured baseline characteristics, the presence of diabetes was associated with a significant increase in the risk of death and hospitalization. These findings suggest an intrinsic effect of diabetes on outcomes in heart failure that was not mediated by coincidental traditional risk factors and comorbidities. These findings also show that the outcomes of heart failure with diabetes were generally worse in women, especially elderly women. This is an important finding because with the aging of the population, the number of elderly women with heart failure is projected to increase substantially in the coming decades. This finding also highlights the need for early diagnosis and better management of diabetes in patients with heart failure, prevention of diabetes in patients with heart failure, and prevention of heart failure in patients with diabetes.
The deleterious effects of diabetes in heart failure may well be through direct metabolic effects. Diabetes activates neurohormonal systems and reactive oxygen species that promote apoptosis and fibrosis in the heart.3,19 These changes at the myocardial level result in more severe left ventricular remodeling and lethal arrhythmias that may, in part, cause poor outcomes in diabetes. The underlying mechanism behind the differential effects of diabetes in women with heart failure, especially the elderly, is poorly understood. Diabetes was associated with a 19% increased risk of death in elderly women with heart failure, whereas it was only 4% in older men. Diabetes also seems to have negated the sex-related survival benefit of elderly women, so that elderly men and women with diabetes had a similar mortality rate (45% and 46%, respectively). The small magnitude of higher risk of death associated with diabetes in elderly men may be partly attributed to the high (41%) risk of death in elderly men with heart failure, even in the absence of diabetes (Figure 2).
Key limitations of our study include bias resulting from unmeasured covariates, lack of a central adjudication of the diagnosis of diabetes, and lack of data on the duration and control of diabetes. Subjects may also have developed diabetes during the follow-up period. Lack of routine use of beta blocking agents may somewhat limit generalizability to current practice. This study, however, provided an important opportunity to examine the effect of diabetes on the natural history of heart failure.
The findings of this study show that the presence of diabetes was associated with an increased risk of death and hospitalization in a wide spectrum of ambulatory subjects with chronic mild-to-moderate systolic and diastolic heart failure, who were well balanced in all other measured baseline covar-iates. The effect of diabetes was more pronounced in women, and was especially pronounced in elderly women. These findings highlight the need for aggressive evidence-based therapy for all heart failure patients with diabetes. Special attention should be paid to women with heart failure and diabetes, especially the elderly. Heart failure and diabetes share many risk factors, and clinicians should focus on the prevention of heart failure in diabetes and the prevention of diabetes in heart failure. New interventions need to be developed to reduce the adverse effect of diabetes in heart failure, particularly among elderly women.
DISCLOSURE This work was supported by the National Institutes of Health through grants from the National Heart, Lung, and Blood Institute (5-R01-HL085561-02 and P50-HL077100) and a generous gift from Ms Jean B. Morris of Birmingham, Alabama, to Dr Ahmed.