Background: Insulin-sensitizing drugs of the thiazolidinedione class and metformin are commonly prescribed to treat diabetes in patients with heart failure despite strong warnings from the Food and Drug Administration against this practice. Whether this results in adverse outcomes is unknown.
Methods and results: We conducted a retrospective cohort study of 16,417 Medicare beneficiaries with diabetes discharged after hospitalization with the principal discharge diagnosis of heart failure. The association between antidiabetic drug prescriptions and outcomes was assessed in multivariable hierarchical Cox proportional hazards models, with adjustment for patient, physician, and hospital variables and accounting for the clustering of patients within hospitals. The primary outcome of the study was time to death due to all causes. Secondary outcomes included time to readmission for all causes or for heart failure. Crude 1-year mortality rates were lower among the 2226 patients treated with a thiazolidinedione (30.1%) or the 1861 treated with metformin (24.7%) compared with that among the 12,069 treated with neither insulin-sensitizing drug (36.0%, P= or <0.0001 for both comparisons). In multivariable models, treatment with the thiazolidinediones (hazard ratio [HR] 0.87, 95% CI 0.80 to 0.94) or metformin (HR=0.87, 95% CI 0.78 to 0.97) was associated with significantly lower risks of death. There was no association with treatment with sulfonylureas (HR=0.99, 95% CI 0.91 to 1.08) or insulin (HR=0.96, 95% CI 0.88 to 1.05) and mortality. Admissions for all causes did not differ with either insulin sensitizer. There was a higher risk of readmission for heart failure with thiazolidinedione treatment (HR 1.06, 95% CI 1.00 to 1.09) and a lower risk with metformin treatment (HR 0.92, 95% CI 0.92 to 0.99).
Conclusions: This observational study suggests that thiazolidinediones and metformin are not associated with increased mortality and may improve outcomes in older patients with diabetes and heart failure. Randomized trials are warranted to corroborate these findings.