Spotlight commentary: Using real‐world data for real‐world evidence generation to advance drug repurposing

Drug repurposing has been presented as an attractive alternative approach to de novo drug development due to its lower costs, shorter development time, and higher success rate across the developmental and regulatory pipeline. Analysis of real‐world data (RWD), collected from routine healthcare delivery, is one of the computational approaches to advance drug repurposing efforts. ¹ Several recent publications in the British Journal of Clinical Pharmacology (BJCP) have featured the use of RWD and/or drug repurposing.

Generation of real‐world evidence (RWE) from observational studies using RWD has been used to complement findings from randomized controlled trials (RCTs), ² , ³ given their merits in larger sample size, longer follow‐up and inclusion of population groups beyond those included in RCTs. The BJCP publication by Szigeti et al. highlighted several studies using Bayesian analyses of a small number of real‐world clinical observations from case series to support findings from larger RCTs, such as repurposing cannabis to treat childhood epilepsy. ³ Similar to RCTs, findings from studies using RWD can be pooled and meta‐analysed to provide more precise effect estimates or improve the generalisability of evidence. For example, an earlier systematic review and meta‐analysis of 32 observational studies in BJCP by Ganesh and Randall reported favourable effects of metformin on COVID‐19 outcomes. ⁴ Over the past decade, multiple collaborative research infrastructures have been formed to leverage the wealth of RWD, such as the National Patient‐Centered Clinical Research Network (PCORnet) in the United States. A recent large multi‐centered electronic health record (EHR)‐based cohort study using PCORnet data contributors further validated the association between metformin and COVID‐19 outcomes and continued to entertain the possibility of repurposing metformin for COVID‐19. ⁵ To date, major drug regulatory agencies, such as the US Food and Drug Administration (FDA) and EU European Medicines Agency, have issued frameworks and guidances to support the integration of RWE into regulatory decision‐making. ¹ There are also regulatory pathways, such as the US 505(b)(2) and EU Hybrid Marketing Authorisation, that allow the use of existing data, such as RWE from the literature, to support label extension applications including for new indications.

Most studies using RWD to inform drug repurposing efforts validate existing drug repurposing hypotheses, ¹ focusing on evaluating single drug exposure‐outcome relationships. These hypothesis‐validation studies help to confirm or refine preconceived hypotheses. For example, a cohort study utilizing Australian linked administrative healthcare data generated RWE to support the potential effects of beta blockers on breast cancer survival among women specifically with triple‐negative breast cancer. ⁶ Conversely, studies analysing RWD can also refute emerging drug repurposing hypotheses. These studies are important to deprioritise drug repurposing hypotheses for subsequent scrutiny. A BJCP publication by Shapiro et al. investigated the association between sodium‐glucose co‐transporter‐2 inhibitors (SGLT‐2i) use and lung cancer using UK primary care records and found no reduced short‐term risk of lung cancer. ⁷ RWD studies can also confirm short‐term benefits demonstrated in RCT for drug repurposing. For example, an observational study may be conducted to evaluate the potential benefits of metformin use on long‐term breast cancer survival of metformin to validate the findings from the METNEO RCT by Serageldin et al. published in BJCP, which demonstrated improved short‐term clinical and pathological response in breast cancer tumours following metformin use alongside chemotherapy. ⁸

Hypotheses for drug repurposing do not need to be restricted to existing marketed drugs for new indications. Instead, a different paradigm uses hypothesis‐validation studies to facilitate real‐world testing of a novel therapeutic concept. Following preclinical evidence of improved insulin resistance from cyclooxygenase‐2 (COX2) inhibition, findings from a cohort study using linked registry data capturing a cohort of people with type 2 diabetes partially supported the COX2 hypothesis. ⁹

With the increasing quantity, quality, and depth of clinical big data, RWD can be leveraged to generate novel drug repurposing hypotheses. In fact, over the past decade, there has been increasing research interest in generating drug repurposing signals using RWD. ¹ Pharmacovigilance signal detection methods such as disproportionality analyses have been “repurposed” to identify drug repurposing signals among adverse drug event (ADE) reports collected from spontaneous reporting systems. The BJCP publication by Böhm et al. conducted disproportionality analyses using the US FDA Adverse Event Reporting System (FAERS) database to detect inverse signals between drugs and viral respiratory infections. ¹⁰ The study identified putative drugs as potential antivirals, some supported by affirmative data in existing literature. Using Vigibase, the WHO global pharmacovigilance database, the BJCP publication by Viguer et al. identified drugs with similar adverse drug reaction (ADR) signatures to voltage‐gated calcium channel blockers to infer new indications for these drugs. ¹¹ The underlying concept is that these drugs with similar ADR signatures may have similar molecular targets and may be repurposed for similar indications as those calcium‐channel blockers. An outcome‐wide approach can also be leveraged to generate drug repurposing signals. For example, an outcome‐wide cohort study identified protective effects of SGLT‐2i, by comparing rates of hospitalizations across a wide range of clinical outcomes concurrently. ¹² This hypothesis‐free evaluation of multiple drug‐outcome associations identified reduced hospitalizations relating to infections, chronic obstructive pulmonary diseases, anaemia etc. with SGLT‐2i use, which motivates further investigations of these novel benefits of SGLT‐2i for drug repurposing. Recently, tree‐based scan statistics, a data mining tool predominantly used in active drug safety surveillance, has been implemented to generate drug repurposing hypotheses. ¹³ While there is increasing incorporation of artificial intelligence (AI) and machine learning (ML) into various aspects of clinical pharmacology and therapeutic development, as discussed by the BJCP publication by Ryan et al., ¹⁴ there is the prospect of applying AI/ML methodologies to efficiently mine RWD to generate drug repurposing hypotheses.

Studies using RWD can also evaluate the safety of repurposed drugs. While this is often not necessary as repurposed drugs have well‐characterized safety profiles, further safety evaluation is crucial when these repurposed drugs are used beyond their original approved dose range or patient population groups, or when they are reformulated. Additionally, safety evidence for repurposed drugs may be limited if repurposed drugs have received accelerated approval based on preliminary safety data or are prescribed off‐label in practice. Observational studies using RWD can help to generate RWE to further support their ongoing expanded use in practice. The COVI‐PREG registry database described by the BJCP publication by Favre et al. collects data prospectively from EHR across multiple centres worldwide, focusing on medication use to manage COVID‐19 and treat its complications among pregnant women. ¹⁵ The primary intention of this registry was to generate RWE on the safety of commonly repurposed drugs for COVID‐19 during pregnancy, such as antivirals, anti‐interleukin‐6, hydroxychloroquine and corticosteroids.

Nonetheless, it is important to mention that observational studies using RWD to generate RWE are often susceptible to a wide range of confounding and biases. ¹ First, this can be mainly attributed to the lack of randomization of exposure to ensure exchangeability between exposure groups. In real‐world clinical practice, there is nonrandom prescribing of treatment to patients after considering a broad range of factors; this could lead to confounding if one or some of these factors also influence clinical outcomes. Drug–drug interactions may also lead to confounding: In the BJCP publication by Szigeti et al., ³ cannabidiol is a strong CYP inhibitor and half of the paediatric epilepsy patients use clobazam, the active metabolite of which can be increased from fivefold to sevenfold by cannabidiol, potentially influencing epilepsy outcomes. Second, there is also a wide range of time‐related biases, such as immortal time bias and protopathic bias, which are often the consequences of flawed study designs. A common approach when designing hypothesis‐validation studies is target trial emulation, as discussed in the BJCP publication by Suissa. ² Essentially, target trial emulation involves carefully designing an observational study to evaluate a causal association by emulating a theoretical RCT and focusing on time‐zero alignment to minimize time‐related biases. Time‐zero alignment refers to aligning the time points when eligibility criteria of study population are met, when drug exposure is ascertained and when follow‐up starts to assess outcomes. ² Misalignment, such as starting follow‐up before determining drug exposure, may lead to time‐related biases, which in this case immortal time bias. Overall, study confounding and biases can be addressed or minimized using careful real‐world study designs alongside rigorous statistical modelling for confounding control. ²

When analysing RWD, it is essential to consider and acknowledge its fitness for purpose and inherent limitations. ¹ First, data coverage affects the generalisability and external validity of study findings. RWD sources with broad, population‐level coverage enhance the applicability of results to the general population (if that is the target population). The population coverage also impacts the transportability of findings across population groups or jurisdictions. Second, data provenance and quality must be critically evaluated, as RWD is typically collected to support routine healthcare delivery or administrative functions rather than research. For example, incomplete data capture, such as for over‐the‐counter medication use, or missing data, such as unrecorded laboratory test results, can introduce misclassification bias and affect internal validity. Diagnostic coding in administrative healthcare data may also vary in completeness and granularity. Finally, the availability and quality of RWD sources may vary substantially across settings. High‐quality, population‐level and comprehensive RWD are more commonly available in high‐income countries with mature healthcare systems, whereas such RWD infrastructure may be limited or inconsistent in lower‐resourced settings.

Looking forward, future efforts should integrate data and knowledge beyond RWD to prioritize drug repurposing generated for further validation. These may include pharmacological data of existing drugs and pathophysiological knowledge of diseases. For example, the BJCP publication by Ryan et al. described the use of ML models to analyse knowledge graphs on ‐omics data, which describe relationships between data relating to gene, protein, signalling pathway, drug and other entities, to identify novel drug targets. ¹⁴ Integrating knowledge generated from other non‐RWD approaches allows early evaluation of the biological and mechanistic plausibility of the drug repurposing hypotheses before subsequent validation using real‐world observational studies and RCTs. Conclusively, RWD analysis should not be viewed as an isolated approach for drug repurposing but rather as a crucial component of a network of experimental and computational approaches to generate evidence for drug repurposing (Figure 1). Alongside traditional trial‐based evidence, RWE can be used to support future regulatory approval and real‐world clinical use of the repurposed drugs. To fully realize this potential, stakeholders across clinical pharmacology, regulatory science, and drug development must work collaboratively to integrate RWD systematically into the RWE generation ecosystem.

FIGURE 1.

Real‐world data analysis to generate real‐world evidence for drug repurposing, in the context of other approaches and trial‐based evidence generation.

AUTHOR CONTRIBUTIONS

Both authors made substantial contributions to the conception of the work. G.S.Q.T. drafted the work. Both authors revised it critically for intellectual content. Both authors gave final approval of the version to be published.

CONFLICT OF INTEREST STATEMENT

There are no competing interests to declare.

Tan GSQ, Ilomäki J. Spotlight commentary: Using real‐world data for real‐world evidence generation to advance drug repurposing. Br J Clin Pharmacol. 2025;91(9):2490‐2493. doi: 10.1002/bcp.70181