A data-driven pipeline to extract potential side effects through co-prescription analysis: application to a cohort study of 2,010 patients taking hydroxychloroquine with an 11-year follow-up

Abstract

Context: Real-life data consist of exhaustive and unbiased data to study drug-safety profiles but are underused because of their complex temporality (i.e., safety depends on the dose, timing, and duration of treatment) and the considerable number of potential side effects to study. We aimed to create a pipeline that manages the complex temporality of real-life data using a data-driven strategy (i.e., without any hypothesis on the potential side effects to search for) to highlight the safety profile of a given drug. We used hydroxychloroquine (HCQ) and its co-prescription in a real-life database to illustrate this pipeline. Methods: We incorporated a weighted cumulative exposure statistical model into a data-driven strategy. This pipeline makes it possible to highlight both long-term and short-term side effects, while avoiding false positives due to the natural course of the underlying disease. We applied the proposed pipeline to a cohort of 2,010 patients with a prescription of HCQ and used their drug prescription as the source of data to highlight the HCQ safety profile. Results: The proposed pipeline introduces a bootstrap strategy into weighted cumulative-exposure statistics estimates to highlight significant drug signals. As applied to HCQ, the proposed pipeline showed nine drugs to be significantly associated with HCQ exposure. Of note, one of them has therapeutic indications for known HCQ side effects. Other associations could be explained by therapeutic indications linked to conditions associated with HCQ indications in France. Conclusion: We propose a data-driven pipeline that makes it possible to provide a broad picture of the side effects of a given drug. It would be informative to pursue the development of this pipeline using other sources of data.