# Estimation of the Duration of Antihypertensive Prescriptions: Validation of a Data‐Driven Approach Using Rotterdam Study Data

## TL;DR

This study validates a method for estimating how long patients are prescribed antihypertensive medicines using real-world data.

## Contribution

The study shows that adjusting for dispensed medicine quantity improves the accuracy of duration estimation in variable data.

## Key findings

- rWTD models without quantity adjustment performed poorly in variable data.
- Adjusting for dispensed quantity reduced relative differences to ≤20%.
- Stratification by medicine subclass improved estimation accuracy.

## Abstract

Administrative medicine dispensing data often omit prescribed duration, which is important for research on adherence or other pharmacoepidemiological topics. While the reverse waiting time distribution (rWTD) method has been widely used to estimate prescribed durations, its accuracy in real‐world dispensing data is unknown. We assessed the performance of the rWTD method against the actual prescribed duration recorded in the Rotterdam Study.

100 725 antihypertensive (AHT) prescriptions from 2018 to 2019 were extracted from the Rotterdam Study's medicine data. Data were constructed into five scenarios with increasing variability in the number of medicines included and variations in prescribed duration. The rWTD with 10 random index dates with or without adjustment for the quantity of dispensed medicine was conducted in all scenarios. Relative differences and limit of agreement ratio based on Bland–Altman analysis were used to examine agreement between estimated and actual prescribed durations.

rWTD models without adjustment for the quantity of dispensed medicine performed well only in the most homogenous scenario. In scenarios with greater data variability, performance improved significantly when adjusted for the quantity of dispensed medicine. Relative difference decreased from ≥ 65% in models without covariates to ≤ 20% with covariates, and the limit of agreement ratio decreased from ≥ 36.8 in models without covariates to ≤ 5.3 with covariates. Stratification analysis by subclass of the AHT medicines provided similar results.

The study demonstrated that as data variability increased, the accuracy of the rWTD estimations decreased. However, the rWTD can produce good estimates (relative difference from 0% to 28%) of prescribed duration for AHT medicines, with the highest accuracy in the model adjusting for the quantity of dispensed medicine or stratification of the data with a relative difference less than 20% and the limit of agreement ratio less than 5.3 for the estimation at the 90th percentile of inter‐arrival density. Since this validation was limited to antihypertensive medicines, generalizing the finding to other chronic‐use medicines should be undertaken with caution.