Identifying Drugs Associated With Parkinson's Disease Risk Using Machine Learning

Abstract

Machine learning (ML)–based methods have been proposed as a potential approach for identifying candidate drugs to be repurposed as disease-modifying treatments for Parkinson's disease (PD). We applied an ML-based signal detection method to identify drugs associated with PD and evaluated the method's generalizability. An algorithm combining subsampling and lasso logistic regression was implemented in a case–control study of 12 257 PD cases and 81 103 matched controls identified in Finnish registers. Drug exposure was defined at the subgroup and drug substance levels of the Anatomical Therapeutic Chemical (ATC) classification, considering the frequency of dispensation over 2 years, starting 10 years before the index date (8-year lag). Three subgroups and two individual drugs were associated with reduced PD risk. Inhalant anticholinergics, in particular tiotropium bromide, showed the most robust signal. Other signals included antimalarial drugs (aminoquinolines) and the antibiotic subgroup lincosamides. Several drugs were associated with increased PD risk, as expected. In addition to direct pharmacological effects, observed associations could be due to treatment of prodromal symptoms of PD, increased comorbidity in individuals later diagnosed with PD or a combination of these factors. These results support the feasibility of the approach. Associations of decreased PD risk observed should be further investigated in view of drug repurposing.