Phosphodiesterase-5 Inhibitors and Dementia Risk: Confounding by Indication in Real-World Studies

In their recent report, Gohel et al. suggested a 54% lower risk of dementia among users of sildenafil, a phosphodiesterase-5 (PDE-5) inhibitor. 1 While the result may seem to identify a compelling repurposing opportunity, we caution that the observed association could be explained by biases introduced by the study design. Most importantly, the study was susceptible to confounding by indication due to the active comparators chosen.

In their study, Gohel et al. 1 compared sildenafil to 3 diuretics (bumetanide, furosemide, spironolactone), and a calcium channel blocker (nifedipine). Although like sildenafil these can be prescribed to people with pulmonary hypertension, they are also used for other conditions such as chronic kidney disease, heart failure, hypertension, and angina. Supplementary Tables 1 and 2 in Gohel et al. 1 show that users of these active comparators were much more likely than users of sildenafil to have renal and heart disease, implying many of the users of the active comparators might have initiated them for other indications. Many of these indications are associated with dementia risk. In contrast, sildenafil is mainly used for erectile dysfunction, but the study was unable to distinguish between users for this indication versus pulmonary hypertension, suggesting that healthier people might have been among the sildenafil users. Under these circumstances, covariate adjustment was unlikely to address substantial unmeasured confounding (e.g., due to frailty, health-seeking behaviors, etc.). 2

The first sildenafil study to use electronic health records was conducted by Fang et al., which was also susceptible to confounding by indication. Fang et al. compared sildenafil users to non-users, or to users of diltiazem, losartan, glimepiride, or of metformin. 3 Those comparisons were even less appropriate than those made by Gohel et al. 1 , and the results are likely to be explained by similar mechanisms of confounding, especially healthy user bias. 2

A recent cohort study by Adesuyan et al. found that PDE-5 inhibitors were associated with a 16% lower risk of dementia. 4 To mitigate confounding by indication, that study restricted people entering the cohort to those newly diagnosed with erectile dysfunction. The study used a time-varying exposure definition to address immortal time bias; 5 however, it appears that inverse probability treatment weights were derived from a baseline logistic regression model predicting the probability of treatment initiation during follow-up, rather than from a model (e.g., marginal structural model) predicting the probability of treatment initiation in a time-varying manner. 6 Selection bias due to the use of future exposure data for propensity score calculation, and residual confounding due to failure to properly adjust for time-varying confounders, could account for the results of that study.

An active-comparator new-user cohort design with an appropriate comparator can effectively reduce confounding by indication without introducing immortal time or selection bias.7,8, 7,8 Also, an active-comparator new-user cohort emulates an active-controlled randomized trial to provide a clear clinical question and to strengthen causal inference.7,8, 7,8 However, an appropriate active comparator should have the same indication and a similar clinical position (e.g., two first-line treatments for the same condition). 9 As PDE-5 inhibitors have multiple indications, it would be essential to first restrict the cohort to people with pulmonary hypertension or erectile dysfunction, as in Adesuyan et al.. As a difficulty inherent in the context of erectile dysfunction, an active comparator may not be feasible. According to the guidelines for pulmonary hypertension treatment, endothelin-1 receptor antagonists and PDE-5 inhibitors have similar clinical positions. 10 A study by Desai et al. found no association between new users of PDE-5 inhibitors and dementia risk when comparing them to new users of endothelin-1 receptor antagonists in people with pulmonary hypertension. 11 In contrast to Gohel et al. and several previous reports,1,3,12, 1,3,12 Table 1 from Desai et al. showed small differences in most characteristics before propensity score matching, 11 indicating minimal likelihood of confounding by indication.

As a secondary consideration, an active comparator might ideally have no effect on the outcome. If the active comparator were to be harmful, then the drug of interest could appear to have benefit. On the other hand, if the active comparator were to have benefit, then the drug of interest could appear to have no benefit. Pertinent to Desai et al. 11 , evidence of effects of endothelin-1 receptor antagonists on dementia risk remains lacking in humans, although theoretical benefits have been suggested in preclinical studies. 13 Considering this, it is possible that both the PDE-5 inhibitor and the active comparator in Desai et al. were neuroprotective, or that neither was neuroprotective. Nonetheless, that study offers internal validity, and an inference relevant to a valid clinical question—“which of these two competing therapeutic options might be best for the brain?”. 11 Despite these strengths, that study was limited by a short duration of follow-up, an important consideration in weighing the evidence that weakened biological plausibility, and which might be addressed in further studies.

In summary, several real-world studies suggesting profound benefits of PDE-5 inhibitors for dementia prevention have lacked internal validity due to confounding by indication introduced by inappropriate comparators,1,3,12, 1,3,12 while a dissenting study was limited by a short duration of follow-up. 11 Further observational comparative effectiveness studies using appropriate methodology may be helpful to address the important hypothesis that PDE-5 inhibitors might prevent dementia and to suggest whether randomized controlled trials should be pursued.