Drug repurposing is a promising supplement to de novo drug discovery and development due to time and cost savings (Parvathaneni et al., 2019; Pushpakom et al., 2019; Roy et al., 2021). Repurposed drugs have lower attrition rates across clinical trial and regulatory approval phases compared to their de novo counterparts (Parvathaneni et al., 2019; Pushpakom et al., 2019; Roy et al., 2021). This is particularly relevant to oncology, where on average only 3.4% of drugs successfully traverse from Phase I clinical trials to market approval (Wong et al., 2019). Real-world data, including administrative healthcare data, may be used to discover or validate drug repurposing candidates via pharmaco-epidemiological analyses (Pushpakom et al., 2019; Ozery-Flato et al., 2020; Xu et al., 2020; Chen et al., 2021). The benefits of administrative healthcare data are longitudinal individual-level records, large population-based sample sizes, and widespread availability (Katkade et al., 2018; Ilomaki et al., 2020; Franklin et al., 2021; Tan et al., 2023). These advantages are attributable to embedded data collection processes during routine provision of healthcare.

Beta-blockers have been identified as potential repurposing candidates for cancer due to emerging evidence from preclinical studies (Sloan et al., 2010; Cole and Sood, 2012; Magnon et al., 2013; Creed et al., 2015; Chang et al., 2016; Le et al., 2016), retrospective observational studies across various cancer types (e.g., melanoma (De Giorgi et al., 2013), ovarian cancer (Watkins et al., 2015; Couttenier et al., 2019), colorectal cancer (Jansen et al., 2014), prostate cancer (Lu et al., 2015; Malik et al., 2023), and breast cancer (Botteri et al., 2013; Lofling et al., 2022)), and recent prospective clinical trials (Shaashua et al., 2017; Hiller et al., 2020). Despite advances in breast cancer therapeutics, metastasis remains the key prognostic marker for poor survival and the main cause of death (Weigelt et al., 2005; Jung et al., 2011). Preclinical evidence suggests that beta-blockade reduces metastasis by limiting tumour cell invasion, inflammation in the tumor microenvironment, and lymphatic vasculature remodelling (Palm et al., 2006; Sloan et al., 2010; Creed et al., 2015; Chang et al., 2016; Le et al., 2016). Additionally, beta-blockers may interact with breast cancer treatments to exert their protective effects (Pasquier et al., 2011; Reeder et al., 2015; Chang et al., 2023). However, a meta-analysis of epidemiological and perioperative studies by Yap et al. (2018) raised the possibility that the effects of beta-blockers on cancer outcomes may be tumor type-specific, as significant benefits were observed only in melanoma and ovarian cancer (Yap et al., 2018). Several retrospective observational studies found that beta-blocker use is associated with improved cancer outcomes in breast cancer (Powe et al., 2010; Barron et al., 2011; Melhem-Bertrandt et al., 2011; Botteri et al., 2013; Sorensen et al., 2013; Choy et al., 2016; Gillis et al., 2021; Lofling et al., 2022), while other studies have reported no significant association (Cardwell et al., 2013; Ganz et al., 2011; Sakellakis et al., 2014; Modi et al., 2020-; Cardwell et al., 2016). Moreover, meta-analyses by Løfling et al. (2022), Caparica et al. (2021) and Spini et al. (2019) have demonstrated more consistent protective effects of beta-blockers amongst women with triple negative breast cancer (TNBC) (Spini et al., 2019; Caparica et al., 2021; Lofling et al., 2022). Previous observational studies predominantly used data from population-based cancer registries and hospital electronic medical records (Mohammadzadeh et al., 2017), which contain detailed histopathological and other clinical information about cancer. However, cancer registry data may be unavailable or unfit for research purposes, especially in some developing countries (Valsecchi and Steliarova-Foucher, 2008; Abdul-Sater et al., 2021).

There is currently a paucity of studies using administrative healthcare data alone to investigate oncology-related repurposing hypotheses. To address this, we aimed to evaluate a drug repurposing opportunity in cancer using administrative healthcare data by analysing the association between beta-blocker use and breast cancer survival. We used an active comparator study design that compared beta-blocker users to users of first-line anti-hypertensives to reduce confounding and healthy user bias (Lund et al., 2015). In the absence of detailed cancer-related clinical information in the administrative dataset, we used longitudinal hospital admission data and prescription medication claims to infer incident cancer cases, baseline medication use and comorbidities, as well as breast cancer molecular subtypes.

In this example using administrative healthcare data to investigate a drug repurposing candidate for cancer, we found no association between either use of any beta-blocker, nonselective beta-blocker use, or selective beta-blocker use, and short-term breast cancer mortality (median 2.7 years of follow-up). The results are aligned with a meta-analysis of 13 studies by Caparica et al. (2021) who reported no significant difference between beta blocker use and breast cancer mortality (Caparica et al., 2021). However, notably recent studies using cancer registry data or hospital electronic medical records with longer follow-up times and larger sample sizes have demonstrated that beta-blockers prolong breast cancer survival in TNBC patients. A recent meta-analysis of four independent studies by Løfling et al. (2022) reported a 26% reduced risk of breast cancer death in women with TNBC with the use of beta blockers compared to no use (Lofling et al., 2022). For example, Botteri et al. (2013) observed a 48% reduced risk of breast cancer death in women with TNBC who used any beta-blockers (n = 74) over 5 years (Botteri et al., 2013). Løfling et al. (2022) demonstrated a 34% improved breast cancer-survival with beta blocker use in women with TNBC (n = 312) over 5 years (Lofling et al., 2022). Our analyses indicated a possible reduction in breast cancer mortality for women with TNBC using beta-blockers compared to comparators (sHR 0.16, 95% CI 0.02–1.51), although results were not statistically significant. This may be due to short follow-up, small sample size, fundamental differences in study cohorts, and residual confounding (see limitations below).

The key advantage of administrative healthcare data over cancer registry data or electronic medical records is the presence of longitudinal and comprehensive records of medication use. In this study, we ascertained baseline use of beta-blockers by using longitudinal prescription medication claims. This allowed us to accurately identify the type of beta-blocker use (selective vs. non-selective) dispensed prior to breast cancer diagnosis. Studies using cancer registry data without linkage to medication claims often rely on patient self-reports or hospital electronic medical records to identify baseline use of medications (Pop et al., 2019). Self-reporting in cancer registry data is subject to recall bias. Moreover, hospital electronic medical records may not have exact information when the medication was initiated and may include initiation of exposure medication use after a breast cancer diagnosis. Ascertaining exposure of medications after the index date may lead to an immortal time bias, where people by design need to survive event-free until the exposure starts at a time after the index date (Suissa, 2008). Immortal time bias could distort results towards favourable drug-outcome associations (Weberpals et al., 2016). This was demonstrated in a meta-analysis by Zhong et al. (2016), which found that post- but not pre-diagnostic use of beta-blockers was associated with improved cancer survival (Zhong et al., 2016).

Cancer registries remain the gold standard real-world data source for evaluating drug repurposing opportunities for cancer due to rich clinical information on cancer diagnosis and management. However, administrative healthcare data are considered as alternative data sources if cancer registry data are unavailable or unfit for research purposes, especially in developing countries (Valsecchi and Steliarova-Foucher, 2008; Abdul-Sater et al., 2021). Several previous studies have used administrative data to evaluate drug repurposing opportunities for cancers other than breast cancer (Chen et al., 2019; Chen et al., 2020; Ng et al., 2021; Park et al., 2021). Some of these studies even have enriched administrative healthcare data (Ng et al., 2021), or linkage to external registry to provide or confirm cancer-related clinical information (Chen et al., 2019; Chen et al., 2020).

Due to the lack of cancer-specific histopathological clinical data in our dataset, we acquired information on incident breast cancer cases from longitudinal hospitalization data and prescription medication claims. We identified the first hospitalization record where breast cancer was recorded as a primary diagnosis and used this as a proxy for the date of breast cancer diagnosis. Cancer registries usually record incident cancer cases and provide accurate dates of diagnosis (Yuen et al., 2011; Okuyama et al., 2018). However, Yuen et al. have demonstrated good sensitivity (84.8%) and specificity (99.9%) between administrative data and actual recorded dates of diagnosis in cancer registry (Yuen et al., 2011).

We used diagnosis and procedural codes from hospitalization data and prescription medication claims data to infer breast cancer molecular subtypes and staging. We developed a novel algorithm to estimate breast cancer molecular subtypes from commonly prescribed antineoplastics used to treat each molecular subtype of breast cancer. Molecular subtype was used to stratify and adjust our analyses. Moreover, we used the presence of breast cancer surgery at index hospital admission and diagnosis codes for metastasis to infer early breast cancer cases. The meta-analysis by Caparica et al. (2021) evaluated the association between beta-blocker use and breast cancer survival in early breast cancer cases given their better prognoses (Caparica et al., 2021). Given that early breast cancer cases were likely to have localized operable tumors or no metastasis, we included both the presence of breast cancer surgery during index hospitalization and metastasis as independent covariates for adjustment in our analyses.

Strengths of our study include the use of active comparators and competing risk analyses. The use of active comparators minimizes confounding by indication (Lund et al., 2015), and other differences in baseline characteristics between the exposure groups. The active comparators also minimize healthy user bias as all women in the study were active users of preventive cardiovascular medications and the healthcare system (Lund et al., 2015). Despite this, a higher proportion of beta-blocker users had congestive heart failure and higher weighted Charlson comorbidity scores than the comparators, therefore, our model adjusted for both of these. We also considered competing risks in our time-to-event survival analyses using Fine and Gray’s subdistribution hazard model. Competing risk events can lead to overestimation of risk when using the conventional Cox proportional hazards regression model (Tan et al., 2018). This is especially true when analyzing a cohort with a substantial proportion of older people, as the incidence of diseases that are attributable to aging and frailty increases (Tan et al., 2018).

There are important limitations to this study. First, the limited sample size and short follow-up time (median 2.7 years) led to underpowered analyses. Second, the study cohort included people admitted to hospitals with diabetes, acute coronary syndrome, stroke, or hip fracture only. This limited the ability to generalize the results to a general population of women with breast cancer. To address these two limitations, future studies would benefit from using larger scale administrative data from a nationwide or general population-based cohort to both increase the power of study and generalizability of results. Third, there is the potential of confounding by severity as beta-blockers are not first-line anti-hypertensives and are prescribed for people with additional cardiovascular comorbidities, notably ischaemic heart disease and congestive heart failure (Blood Pressure Reduction, 2021). We considered first-line anti-hypertensives as the best possible active comparator medications as they have previously been used in other studies to reduce confounding (Gillis et al., 2021; Lofling et al., 2022), and included myocardial infarction and congestive heart failure as covariates in the models. Furthermore, there could be residual confounding due to undetermined histopathological characteristics of breast cancer tumors, for example, tumor size and Ki-67 expression, which are conventionally available from cancer registry data but not in our dataset. Fourth, we used dispensing records during follow-up to estimate breast cancer molecular subtype, which may lead to selection bias in the stratified analyses by molecular subtypes. This is because people need to survive during follow-up until receiving the breast cancer treatments (Acton et al., 2023). Furthermore, we were not able to validate the sensitivity and specificity of the algorithm used in inferring breast cancer molecular subtypes as we did not have information on the subtypes available. Future validation studies with data linkage to cancer registries are required to confirm the validity of the algorithm. Moreover, we were unable to classify breast cancer molecular subtype for 14% of women as they did not have dispensing records of either breast cancer antineoplastics or conventional chemotherapy. Therefore, the algorithm could not infer the breast cancer molecular subtypes for women who have undergone surgery or radiotherapy only. Lastly, given that this study only includes data from patients treated until mid-2018, our results do not include the effects of emerging immunotherapies used since 2019 (Barzaman et al., 2021). Preclinical studies have reported that beta-blockade enhances the response to immunotherapy, for example, anti-CTLA4 therapy (Fjaestad et al., 2022), raising the possibility that beta blocker use might enhance new immunotherapy regimens.

Use of administrative data as an alternative to registry data increases opportunities to identify novel drug repurposing candidates. In addition to being more widespread and accessible than registry data, administrative data often has longitudinal and comprehensive records of medication use that can be used to ascertain medication exposure. The current study emphasizes the need for use of large administrative datasets to provide sufficient power to identify associations. The current study describes the development of methodologies to infer cancer-related clinical information from variables available in administrative dataset. Future studies with linkage to clinical data sources are required to validate these methodologies.