We read with great interest results provided by Barreto et al.
1 regarding the associations of statins and genetics with the progression of age-related macular degeneration (AMD). These findings are fundamental as they can deepen our understanding of the AMD pathophysiology. However, we believe that the analytical approach suffers from some limitations that might have induced important bias in the results.
First, as acknowledged by the authors in the Discussion section, the exact date of AMD progression between the two visits remained unknown.
1 The time to progression was thus interval censored between the two visits.
2 However, this was not accounted for in the analysis. Instead, the time to progression was defined as the time to the follow-up visit when the AMD diagnosis was made. Patients who had no AMD diagnosis at that follow-up visit were right-censored at that time and classified as non-progressors (table 1 in the original paper). The major issue with this analytical strategy is that the statin use, which was assessed between the two visits, may have occurred after the actual AMD progression for some patients. The association found between statin use and AMD progression, with the log rank test or with the time-dependent Cox model (tables 2–4 in the original paper), is therefore prone to important exposure misclassification and reverse causation bias.
3 Accordingly, an overestimation of statins effect may have occurred. Second, the Kaplan-Meier curves comparing statin users and non-users (fig. 2 in the original paper) indicates that individuals who died or dropped-out from the study for other reasons between the two visits, were excluded from the analysis. This exclusion may have induced a selection (attrition) bias.
3 Ultimately, if excluded individuals were less likely to develop AMD, their exclusion would likely overestimate the benefit of statins. In the opposite scenario, this benefit would be underestimated. Last, the authors adjusted for age, sex, smoking, diabetes, and genetic score, but this is unclear whether this is sufficient to control for the indication confounding pertaining to statins prescription. For example, a further adjustment for cardiovascular diseases and hypercholesterolemia would be warranted. In fact, statins are primarily prescribed to individuals at high cardiovascular risk, which is also associated with AMD.
4,5 Such a residual confounding bias has likely induced an underestimation of the protective effect of statins, given users are already at high risk of AMD. Propensity scores are therefore a relevant method to further decrease residual confounding due to the indication bias.
6
In summary, Barreto et al.
1 have provided insightful findings that would be worth confirming with a more appropriate modeling strategy, in particular, accounting for the actual timing between statin use and AMD progression, and including patients who dropped out between the two visits.
Disclosure: B. Kayembe Mulumba, None; K. Leffondré, None; P. Joly, None; M.-N. Delyfer, AbbVie (C), Bayer (C), Horus Pharma (C), Novartis (C), Roche (C); C. Delcourt, Allergan (C), Apellis (C), Chauvin-Bausch+Lomb (C), Théa Pharma (C), Novartis (C)