Estimating treatment effects on mortality and competing risks using real-world data, with application to prostate cancer

Randomised controlled trials (RCTs) are the gold standard for evaluating medical treatments, but the role of "real-world data" (RWD), such as electronic health records for assessing treatment effects is increasingly recognised. Using RWD requires careful analysis to control for potential sources of bias, including selection biases, immortal time bias, and confounding. Target trial emulation applies RCT design principles to RWD, to facilitate estimation of causal effects of treatments. Benchmarking emulated trials against existing trials can help to assess the accuracy of treatment effect estimates obtained using RWD, enhancing confidence in the use of such data to address questions that have not been addressed in RCTs.

This thesis investigates prostate cancer treatments for high-risk patients, by emulating the PR07 trial using RWD from UK national linked datasets. The PR07 trial compared the effects of radiotherapy plus hormone therapy within an 8-week grace period from randomisation versus hormone therapy alone on all-cause and disease-specific mortality. This work addresses key challenges of emulating the PR07 trial using the available data, including defining time zero, handling treatment grace periods, and confounding. It also tackles the complexity of accurately defining treatment effects when there are competing risks, as deaths from other causes are common in prostate cancer patients.

The cloning-censoring-and-weighting (CCW) approach is used to estimate the effects of treatments on overall survival. An extension combining CCW approach and landmarking is also described, and further analyses use the sequential stratification approach. To consider competing events, causal estimands are specified, going beyond cause-specific hazard ratios, which are commonly used but have been criticised for lacking a clear causal interpretation. The CCW approach and the landmarking extension are used in the competing risks setting.

The trial emulation findings are broadly consistent with those from the PR07 trial, though differences emerged in the risk estimates for competing events. A simulation study explored possible explanations for this discrepancy. This research highlights the considerations needed when using RWD to estimate effects of treatments in prostate cancer, and demonstrates the application of trial emulation and of different causal inference methods for survival and competing risks outcomes.