Discussion
In this retrospective analysis of 127 555 unmatched patients with T2D, extracted from a population of 18 million individuals, we show that the risk of HHF in patients with T2D taking DPP-4i over a period of 2.6 years was significantly lower than that in patients taking sulphonylureas. The finding was confirmed when the analysis was restricted to 39 465 propensity-matched patients with T2D. This resulted from a decrease of non-fatal HHF in DPP-4i-treated patients, as the rate of fatal HHF was not different among treatment groups.
Almost all oral glucose-lowering medications may increase the risk of HF. The excess cardiovascular risk of glitazones prompted regulatory agencies to require post-approval safety studies for all new products approved for the treatment of T2D. Given that this guideline has been implemented only recently, observational studies can bridge this gap of information, while awaiting results of cardiovascular safety studies.
Incretin-based therapies have pleiotropic effects that may reduce cardiovascular risk. However, there are several reasons whereby pre-clinical findings and clinical studies on surrogate endpoints may not translate into a protection from hard endpoints. DPP-4i can affect vascular function via GLP-1-dependent and -independent actions. Although GLP-1-dependent effects are common to GLP-1 receptor agonists, the DPP-4 enzyme inactivates several substrates different from incretin hormones (including cytokines, chemokines, and neurohormones), many of which can exert favourable, but unpredictable, vascular effects in the clinical setting and in the long run.
Unexpectedly, in the SAVOR trial, conducted on 16 492 patients with T2D and a history of, or at risk for, cardiovascular events, who were randomized to saxagliptin or placebo, an excess 27% HHF was detected in the saxagliptin group over a mean follow-up of 2.1 years. A subsequent re-analysis of this study reported that excess HHF risk decreased after the first 6 months of randomization and was highest among patients with elevated levels of natriuretic peptides, previous HF, or chronic kidney disease.
Primed by these data, several observational studies reported conflicting results on the association between DPP-4i and HF. In a population-based study in Taiwan, including 8288 matched pairs of patients, treatment with sitagliptin was associated with a significant 21% increase in HHF, with no change in all-cause mortality. In a nested case–control study conducted on 10 073 Danish patients with T2D, DPP-4i had a neutral effect on all cardiovascular outcomes, including HF. In a population-based retrospective cohort study including 7620 patients with T2D and incident HF in Canada, sitagliptin use was associated with an increased risk of HHF only among patients with pre-existing HF. Finally, Velez et al. reported that, among 4224 patients with T2D treated with an incretin- or non-incretin-based regimen (1 : 2 ratio), the use of DPP-4i was associated with a significantly lower risk of HHF, all-cause hospitalization, and mortality.
In addition to these observational data, three meta-analyses of phase III–IV randomized-controlled trials, including 55 141–85 224 patients, reported an increased risk of HHF from 16 to 19% associated with the use of DPP-4i. However, the increased risk of HHF was mainly driven by the SAVOR trial, which was the largest and longest study included in the aforementioned meta-analyses.
Uncertainty remains on the concerns raised by the SAVOR study regarding the risk of HF associated with DPP-4i therapy vs. placebo, but the question raises on the comparative effect of DPP-4i vs. other second-line oral agents for T2D. In this very large observational study, we specifically focused on the comparison between DPP-4i and sulphonylureas, which are still the most used oral medications for the treatment of T2D in Italy. In order to limit selection bias, we purportedly excluded patients on diet alone or on metformin monotherapy, as they would be predicted to be healthier, with shorter disease duration and lower HbA1c than patients who received treatment intensification with a second-line oral agent (sulphonylurea, TZD, or DPP-4i). The 28–30% lower risk of HHF detected in the DPP-4i group may thus derive from a beneficial protective effect of DPP-4i or from a detrimental effect of sulphonylureas on HF. In fact, sulphonylureas have been associated with an increased risk of HF compared with metformin. In any case, our findings may have implications for the care of T2D in routine clinical practice and may favour a shift in prescription trends towards oral medications, with a more favourable cardiovascular risk profile. We also show that use of glitazones was not associated with excess HHF, likely because clinicians aware of the potential HF-precipitating effect of glitazones do not prescribe such drugs to patients deemed at risk.
Our study has limitations inherent to its observational and retrospective nature. The typical bias is that differences in the outcomes according to ongoing therapies may not be attributable to specific effects of the drugs, but rather to the reasons whereby different patients receive different drugs. This is clearly demonstrated by the statistically significant and clinically meaningful differences in baseline characteristics, according to the glucose-lowering medication regimen (Table 1). Despite multivariable adjusting, residual bias is typically generated from unmeasured confounders. For instance, we have no data on body mass index, glucose control, disease duration, microvascular complications, and asymptomatic left ventricular dysfunction, all of which may affect the risk of HF. In addition, prescription of DPP-4i is subjected to a registry-based appropriateness evaluation and monitoring in Italy, thereby increasing the likelihood that patients on DPP-4i were prescribed the drugs more appropriately and followed-up more regularly in a specialty settings than those prescribed other medications. To improve reliability and cope with selection and prescription biases, the propensity score matching provides a means of reducing the differences among patient groups by accounting for the covariates that predict receiving the treatment. After matching, the study cohort was reduced to 39 465 patients, and use of DPP-4i was still associated with a significantly lower risk of HHF. These results strongly reinforce what was shown in the total population using multivariate adjustment. In addition, use of DPP-4i was associated with a lower HR risk than use of sulphonylureas, even when including hospitalization with both primary and secondary diagnoses of HF.
In the sensitivity analysis, the use of DPP-4i was associated with an increased risk of HHF compared with sulphonylureas in patients with less than 6 months of follow-up. This analysis does not allow proper adjustment for adherence to medications, which is critical when exposure is determined by prescription records, thereby generating uncertainties on whether the patients actually took the drugs. The results may be also explained by reverse causality, i.e. a higher likelihood of being prescribed a DPP-4i than a sulphonylurea in patients deemed at risk for HHF in the short term. In fact, when the analysis included patients with a previous history of HF, those in which the risk of subsequent HF is highest, the use of DPP-4i still showed trend protection against HHF.
In conclusion, this large observational study shows lower HHF risk in DPP-4i- vs. sulphonylurea-treated patients. These data do not confirm, nor contrast with what was shown by the SAVOR trial, which only compared saxagliptin with placebo. While waiting for the forthcoming phase IV randomized comparator-controlled trials on cardiovascular outcomes of DPP-4i, this reassuring finding may provide a basis for guiding the clinical care of patients with T2D.