(L) Identifying Protective Drugs for Parkinson's Disease: Large Screening of The French National Healthcare Database with Signal Detection Machine Learning Algorithms

Publication Number: 670

Background: Parkinson's disease (PD) is the neurological disease whose prevalence has increased most, and currently available treatments are only partially or transiently effective. Neuroprotective drugs are needed and a promising strategy is to identify existing molecules that may present a therapeutic or even preventive advantage for PD (drug repositioning).

Objectives: We aimed at detecting drugs that are potentially protective for PD, using a large-scale automated screening strategy.

Methods: We implemented a machine learning (ML) algorithm combining subsampling and lasso logistic regression in a case-control study nested in the French national healthcare database. Our study population consisted of 40 760 incident cases identified by a validated algorithm in 2016-2018, and 176 395 controls of similar age and sex, both followed since 2006. Drug exposure was defined by the penultimate level of the Anatomical Therapeutic Chemical (ATC) classification considering the frequency of prescriptions over a two-year period starting ten years before the date of PD to account for the potential for reverse causation. Sensitivity analyses were conducted considering an exposure period of five years and the finest substance level of the ATC classification. Models were adjusted for a wide range of covariates available in the database (chronic diseases, reasons for hospitalisations, medical consultations, biological procedures, deprivation and rurality indices).

Results: Six drug subgroups were detected by our algorithm among the 374 screened. Sulfonamides (ATC C03CA), in particular furosemide (C03CA01), showed the most robust signal. Other signals included adrenergics in combination with anticholinergics for obstructive airways diseases (R03AL), and insulins and analogues (A10AD). Drugs for nicotine dependence (N07BA) also stood out, likely as a proxy for smoking status which is known to be inversely associated with PD.

Conclusions: We identified several potentially interesting signals that deserve to be replicated in large studies with an appropriate consideration of the potential for reverse causation. Our results illustrate the value of ML-based signal detection algorithms for identifying drugs inversely associated with PD risk in medico-administrative databases.
This study was funded by the Michael J. Fox Foundation for Parkinson’s Research.