The association between saxagliptin use and an increased risk of hospitalization for heart failure (HF) has generated considerable controversy.1,2 One reason is that the mechanisms of the potential deleterious effect are mainly unknown and speculative. Second, to make the scenario even more complex, we recently pointed out a high risk of type 1 error (chance finding) in the SAVOR-TIMI 53 trial due to an insufficient Bonferroni correction and an apparent deviation from the initial statistical analysis planned by the authors.3
Give the controversy surrounding the relationship between saxagliptin and HF and in an effort to help resolve it, the present article aims to provide a summarized review of the association using the Bradford Hill criteria of causation.
As described by Hill, the criterion strength of association attempts to determine whether there is a relationship between the putative causal factor and the effect under study. The more distant the relative risk is from 1, the larger is the strength of the association. With saxagliptin, the hazard ratio (HR) was 1.27 and the 95% confidence interval (95%CI), 1.07-1.51 (P = .007).1 A subsequent analysis including all HF hospitalizations (analysis of recurrent events or Andersen-Gill analysis) showed a slight attenuation of this relationship (HR, 1.26) as well as a reduction in the lower limit of the 95%CI to 1.02, that is, very close to the null hypothesis.1 This attenuation was produced because the greater risk of HF in saxagliptin-treated patients was only observed in the first 314 days, being virtually neutral thereafter (HR, 1.05; 95%CI, 0.81-1.35).1 An analysis “excluding the first hospitalization for HF” as part of a sensitivity analysis using another approach, known as the Prentice-Williams-Peterson model, did not find a higher associated risk (HR, 1.06; 95%CI, 0.75-1.50).1 Thus, different statistical models yielded distinct, even contradictory results. In addition, we should not overlook that HF hospitalization was a secondary endpoint, understood therefore, as exploratory, which would increase the probabilities of chance. Because of all these considerations, the strength of the association is likely to be weak.
Hill considered an association consistent if the relationship between the 2 variables were upheld in more than 1 study, in different populations and circumstances. For saxagliptin, the relationship shows little consistency, as it has not been confirmed by recent retrospective observational studies4 (although this is not true of all5) and there is no robust experimental evidence.
The criterion of specificity refers to an effect being attributable to a single cause. With regard to the excess HF risk in the saxagliptin group, one might speculate that it could be due to chance because of the multiplicity of secondary variables (up to 10),3 or to the fact that there were more deaths (nonsignificantly) in the saxagliptin group, which could lead to a smaller number of patients at risk and therefore, an upward bias in the HF incidence rate.3
Temporality, as related to an association, is essential to ensure that the risk factor appeared before the putative effect, which is verified in the study.1
The biological gradient or dose-response relationship encompasses the concept that increased exposure or dose increases the incidence of a disease. Currently, there is no evidence of a dose-response relationship for the association between saxagliptin and HF.
With regard to biological plausibility, the biological context should logically explain the etiology by which a cause produces an effect. This concept is closely related to reproducibility and experimental evidence. In this regard, it should be mentioned that the finding was completely unexpected, as previous studies did not show a greater risk of edema or water retention.1 Endothelial dysfunction and increased left ventricular volumes have been cited as potential mechanisms,1 although other studies indicate benefits.6
Coherence refers to an association being in line with previous knowledge regarding biological mechanisms. Again, the association was unexpected, in disagreement with later studies4 (although not all5), and did not have a constant mechanism relating cause and effect.
Lastly, analogy is based on established cause-effect relationships, whereby if one risk factor produces an effect, another with similar characteristics should have the same impact. In this line, previous data from the same family of drugs (alogliptin) indicate that there could be a nonsignificant numerical trend toward a higher risk of HF.
In conclusion, on application of the Bradford Hill criteria to evaluate the relationship between saxagliptin use and the risk of HF hospitalization, a robust association was not found. However, as safety is a priority, it is essential to carry out new, specific, prospective studies to confirm or rule out this association.