OBJECTIVES: Treatment duration is a time-to-event variable best characterized by its median, while financial models need mean values for building forecasts. This study investigates the effect of different definitions on treatment discontinuation in four agents and indications.

METHODS: Using data from the German AOK PLUS sickness fund (3.4mn patients), we investigated treatment duration of macitentan in pulmonary arterial hypertension (PAH), abiraterone acetate plus prednisone (AAP) in metastatic castration-resistant prostate cancer (mCRPC), darunavir in HIV, and paliperidone in schizophrenia (SCH). We measured treatment duration from first prescription between 1/1/2011-6/30/2020 until discontinuation, using various gap definitions (30, 60, 90, 120 and 180 days) between the runout date of the last prescription and the next prescription. We assumed that patients consumed all accumulated medications (stockpiling). Hospitalizations were considered part of the treatment periods. Median time to treatment discontinuation was described by Kaplan-Meier analysis, mean duration of treatment was estimated with parametric fitting (exponential and Weibull distributions).

RESULTS: 103 PAH, 1134 mCRPC, 147 HIV and 1538 SCH patients were identified in the database. Differences in median time to discontinuation across gap definitions were largest for darunavir in HIV (range: 5.8 months [30-day gap] – 26.2 months [90-day gap]) followed by paliperidone in SCH (46.6 months [90-day gap] – 57.9 months [180-day gap]), macitentan in PAH (26.7 months [30-day gap] – 35.8 months [180-day gap]), and AAP in mCRPC (7.4 months [30-day gap] – 9.5 months [90-day gap]). Mean estimates showed 3-fold differences across various gap-length assumptions (darunavir) and statistically significantly different estimates depending on the choice of distribution (paliperidone).

CONCLUSIONS: Estimates of treatment duration in real-world data vary substantially depending on the applied statistical methods and rules of gap filling and discontinuation. We recommend conducting sensitivity analyses of underlying assumptions and methods, and the consideration of scenario planning to ensure that medication consumption forecasts are robust and consistent.