PURPOSE:

RWD is pivotal in generating evidence for drug discovery, regulatory, and post-authorisation purposes. However, inherent biases can compromise the validity of RWD. While instrumental variable methods are widely utilised in economics and epidemiology research in academia, their adoption in the pharmaceutical industry has been limited until recently. In this discussion, we present use-cases of instrumental variables within the UK Biobank and CPRD, illustrating their role in generating evidence pertaining to drug discovery, post-authorisation effectiveness, safety evaluations, and facilitating market access.

DESCRIPTION:

Case 1: IVs in the Clinical Practice Research Database (CPRD) were used to examine real-world safety and effectiveness in specific populations, unexplored outcomes, and long-term follow-ups (1–5 years) for smoking cessation medicines. Notably, our studies provided precise estimates overcoming underpowered analyses in the largest comparable trials to-date.

Case 2: Assessed the safety and effectiveness of two type-2 diabetes treatments in older adults (+70 years) using IVs and hospital-linked UK primary care data. Current guidelines recommend individualized treatment decisions for this demographic, but limited evidence exists due to their frequent exclusion from trials. Our approach addressed this gap, offering insights based on larger sample sizes and overcoming issues related to residual confounding in previous observational studies.

Case 3: In large real world cohort studies like the UK Biobank, participants undergo a comprehensive baseline assessment to capture clinical, genetic and omics measurements. This has made it an indispensable tool in contemporary drug development using a genetic IV analyses termed 'Mendelian randomization' (MR). We describe how MR provides key lines of evidence to inform early drug target discovery as well as post-licensing expansion of an existing therapy to new indications.

In conclusion, we argue that IVs provide an attractive alternative to other adjustment methods in the evolving landscape of evidence generation, particularly compared to propensity scores and multivariable regression analysis.