(A-182) Validity of High-Throughput Pharmacoepidemiologic Methods Using Electronic Healthcare Databases for Drug Repurposing Research: Application to Systemic Lupus Erythematosus

Background: High-throughput Pharmacoepidemiologic (HTP-PE) methods applied to large electronic healthcare databases have been proposed to identify drugs for repurposing. However, their internal validity has not yet been formally assessed.

Objectives: To evaluate the sensitivity and specificity of an HTP-PE use-non-use design in blindly identifying drugs known to prevent flares of systemic lupus erythematosus (SLE) (“positive trackers”) and those highly unlikely to have such effects (“negative trackers”).

Methods: Patients with a diagnosis of SLE were identified from the French national healthcare database (SNDS), which covers the entire French population, for the period 2010-2020. All drugs used by at least 500 patients were included. Incident drug cohorts were stratified by the presence or absence of prior organ damage. Each drug exposure was defined as lasting 60 days from dispensation, with overlapping periods subtracted. For each drug, randomly selected comparator periods of non-use were matched on calendar date (± 90 days) and high-dimensional propensity scores (caliper≤0.2). SLE flares were defined as hospitalizations with ICD-10 codes for acute inflammatory organ damage (i.e., nephritis, myocarditis, pericarditis, endocarditis, pulmonary disorder), hematological abnormalities, or seizures. Positive trackers included hydroxychloroquine/chloroquine, methotrexate, belimumab, azathioprine, and mycophenolic acetate. Negative trackers were topical drugs. Sitagliptin and metformin, both previously proposed for repurposing in SLE (confirmed and unconfirmed, respectively in previous studies), were pre-specified as repurposing candidate trackers. Poisson regression adjusted for NSAID and corticosteroid use was applied to compare outcome rates between exposure status.

Results: A total of 52,146 SLE patients were identified, using 1,244 distinct drugs dispensed. Of these, 560 drugs were used by at least 500 patients, including 103 topical drugs. Hydroxychloroquine/chloroquine, methotrexate, and belimumab were blindly identified as significantly associated with the prevention of SLE flares (p < 0.05) in the overall population. Other immunosuppressants showed such associations only when patients with prior organ damage were excluded, indicating confounding by indication. Two topical drugs were falsely associated with flare prevention (false positive). Sitagliptin was blindly identified as protective against flares, while metformin was not.

Conclusion: The HTP-PE use-non-use design is a valid method for blindly detecting drug effects when applied to populations aligned with the drug’s actual indications. This method holds promise for future drug repurposing research using large-scale electronic healthcare data.