Dynamic predictive joint models to characterise localised prostate cancer prognosis after radiotherapy

Abstract

This thesis aims to develop and validate dynamic predictive joint models (JMs) to characterise the prognosis of patients with localised prostate cancer who are treated with moderately hypofractionated radiotherapy with neoadjuvant and concurrent hormone therapy. Current clinical prediction models rely on baseline features, e.g. tumour severity and age at diagnosis, which do not adequately predict cancer recurrence. This thesis proposes using routinely collected longitudinal prostate-specific antigen (PSA) measurements, in addition to baseline prognostic factors, to obtain more accurate and dynamically updated predictions. This thesis uses CHHiP, the largest known moderately hypofractionated phase-III trial for localised prostate cancer, to develop a mixed-effects submodel for longitudinal PSAs and a relative risk submodel for time-to-recurrence. The dynamics of PSA trajectories, including concentration and rate-of-change, are considered. Predictions are compared across patient subgroups with contrasting prognostic factors, and PSA thresholds are explored to correlate with prognosis. The performance of the JM is validated using bias-corrected bootstraps and on external cohorts to assess its utility and generalisability. The model is extended to account for the competing risk of deaths unrelated to prostate cancer. This study finds that patients who developed recurrence generally had higher baseline and overall PSA values during follow-up and experienced an exponentially rising PSA in the two years before recurrence. Most baseline risk factors were significant in both submodels, and PSA value and rate-of-change were predictive of future recurrence. PSA thresholds 0.23ng/mL after treatment correlated with good prognosis. The model’s predictive performance was good across differing external cohorts and prediction times. Overall, this thesis demonstrates that dynamically updated PSA information can improve prognostication, which can be used to guide follow-up and treatment management options. It provides evidence for the potential use of JMs in clinical practice, for instance, instigating PSA-driven imaging in high-risk patients and recommending fewer PSA collections for low-risk patients.