# Empirical Comparison of Exposure Set Definitions in the Prevalent New‐User Design

## TL;DR

This study compares different methods for defining exposure sets in prevalent new-user designs to assess treatment effects, using UK health data to evaluate how these choices impact results.

## Contribution

The study provides empirical evidence on the impact of exposure set definitions in PNU designs and emphasizes the importance of sensitivity analyses and calendar time.

## Key findings

- Prescription- and time-based exposure sets resulted in higher matching success compared to hybrid definitions.
- Variability in point estimates was observed across exposure set definitions, but overall conclusions remained consistent.
- Hybrid exposure sets provided the best balance for accounting for calendar time in the study.

## Abstract

Prevalent new‐user (PNU) designs aim to provide a wider assessment of treatment effects by incorporating users of a newer study drug who previously received the comparator. Similarity in terms of prior use of the comparator is accounted for via exposure sets based on time in study, prior prescription number, or a hybrid incorporating both plus calendar time. Given a current lack of consensus, we examine choice of exposure set definition using a study investigating upper gastrointestinal bleeding (UGIB) risk between users of non‐steroidal anti‐inflammatory drugs (NSAIDs) and cyclooxygenase‐2 inhibitors (COX‐2is).

We identified a cohort of individuals with osteoarthritis initiating NSAIDs or COX‐2is between 2000 and 2004 from the UK Clinical Practice Research Datalink. Considering prescription‐based, time‐based, and hybrid exposure set definitions, we estimated time‐conditional propensity scores (TCPS) using conditional logistic regression and matched COX‐2i users 1:1 to NSAID users. Analyses estimated the hazard ratio (HR) of UGIB bleed comparing COX‐2i and NSAID users overall, in the incident new‐user and prevalent new‐user subgroups.

We identified 100 185 individuals who received a prescription for either COX‐2is or NSAIDs; 25 742 patients were incident new‐users of COX‐2is and 17 952 were PNUs. Prescription‐ and time‐based exposure sets resulted in the highest proportion of COX‐2i users successfully matched (86% vs. 76% using hybrid definition). We observed variability in the point estimates obtained under the different exposure set definitions; however, conclusions remained consistent.

Given the potential for differences in the matched cohorts and substantive results obtained under different exposure set definitions, we encourage increased use of sensitivity analyses in PNU studies to explore the robustness of results to this decision.

This study investigates the impact of exposure set definition in the prevalent new‐user (PNU) design, considering prescription‐based, time‐based, and hybrid definitions in a study using UK electronic health record (EHR) data.In our case study, we observed some variability in the point estimates obtained under the different exposure set definitions, highlighting the need for increased adoption of sensitivity analyses to explore the robustness of results to this decision.Given the dynamic prescribing practice of settings where PNU designs are commonly applied, we highlight the importance of accounting for calendar time in exposure set definitions. In our case study, the best balance for calendar time was achieved under the hybrid exposure set definition.We provide R code for implementing the PNU design via GitHub, which addresses a key barrier to wider application and development of the design.

This study investigates the impact of exposure set definition in the prevalent new‐user (PNU) design, considering prescription‐based, time‐based, and hybrid definitions in a study using UK electronic health record (EHR) data.

In our case study, we observed some variability in the point estimates obtained under the different exposure set definitions, highlighting the need for increased adoption of sensitivity analyses to explore the robustness of results to this decision.

Given the dynamic prescribing practice of settings where PNU designs are commonly applied, we highlight the importance of accounting for calendar time in exposure set definitions. In our case study, the best balance for calendar time was achieved under the hybrid exposure set definition.

We provide R code for implementing the PNU design via GitHub, which addresses a key barrier to wider application and development of the design.

This study looks at how different ways of defining exposure sets in the prevalent new‐user (PNU) study design can affect study findings. Using UK electronic health record data, we compared three approaches: prescription‐based, time‐based, and a hybrid method that combines both. We found that the choice of method led to some differences in results, suggesting it is important to explore how sensitive findings are to this decision. Because prescribing patterns can change over time, we also emphasise the need to consider calendar time when defining exposure sets. To help other researchers use the PNU design more easily, we have provided software for implementation.