A Study Protocol for an Instrumental Variables Analysis of the Comparative Effectiveness of two Prostate Cancer Drugs

TL;DR

This paper outlines a protocol for using instrumental variables to compare the effectiveness of two prostate cancer drugs, leveraging regional prescription differences to estimate impacts on mortality and morbidity.

Contribution

It introduces a novel IV analysis design based on regional prescription variation, with validation and sensitivity analysis for its assumptions.

Findings

County prescription practices influence drug choice

Pre-treatment morbidity outcomes are unaffected by regional factors

The design's assumptions are supported by data

Abstract

This paper presents a protocol, or design, for the analysis of a comparative effectiveness evaluation of abiraterone acetate against enzalutamide, two drugs given to prostate cancer patients. The design explicitly make use of differences in prescription practices across 21 Swedish county councils for the estimation of the two drugs comparative effectiveness on overall mortality, pain and skeleton related events. The design requires that the county factor: (1) affects the probability to be treated (i.e. being prescribed abiraterone acetate instead of enzalutamide) but (2) is not otherwise correlated with the outcome. The fist assumption is validated in the data. The latter assumption may be untenable and also not possible to formally test. However, the validity of this assumption is evaluated in a sensitivity analysis, where data on the two morbidity outcomes (i.e. pain and skeleton related events) observed before prescription date are used. We find that the county factor does \emph{not} explain these two pre-measured outcomes. The implication is that we cannot reject the validity of the design.