Urinary Continence Recovery After Radical Prostatectomy Using Patient Reported Outcomes Data: Variability, Predictions, and Prediction Accuracy

Abstract

Despite advances in prostate cancer treatment, wide variability in post-operative oncologic and functional outcomes are observed. Effective management of prostate cancer requires accurate prediction of potential outcomes and setting realistic expectations of functional outcomes is an important challenge. Advances in computing are accelerating the development of prediction tools using routine clinical practice and patient reported outcomes data to support the treatment decision-making process; however, variability in post-operative functional outcomes remains despite the advancements. The development, deployment, and evaluation of prediction models in routine urology practices has the potential to enhance the treatment decision-making process and improve the overall quality of prostate cancer care. The objective of this dissertation is to assess the patterns of urinary continence recovery in a statewide registry. To achieve the objective of this dissertation, we conducted a retrospective analysis of urinary continence recovery after radical prostatectomy for individuals who completed the Michigan Urological Surgery Improvement Collaborative (MUSIC) - Patient Reported Outcomes (PRO) questionnaires. We assessed variability in urinary function outcomes at four post-operative time points and quantified the variability attributable to patients and surgeons (Aim 1). We trained pre- and post-operative prediction models that estimated long-term urinary continence recovery by predicting urinary domain scores and pad use at multiple time points (Aim 2). Lastly, we evaluated how temporal changes in practice patterns affect the robustness of models that estimate pad use by assessing various model building strategies (Aim 3). Using MUSIC registry and patient reported outcomes data, we found wide variability in continence recovery, with greater variability attributable to patients than surgeons (66% versus 7%) in mixed-effects models. Models that incorporated post-operative data and predicted outcomes at proximal time points performed better than pre-operative models or models that predicted outcomes at distal time points. Model discrimination remained stable and models while model calibration showed some indication of over-estimation in later years, but no evidence of calibration drift emerged. By conducting a retrospective study of statewide registry data to quantify variability in urinary continence, predicting long-term recovery, and assessing the robustness of prediction models to secular trends in registry data, our aim is to further the understanding of factors associated with urinary continence recovery and advance the science around prediction models using patient reported outcomes data. The results of these analyses fill gaps in our understanding of urinary continence recovery and further advance the goals of the Michigan Urological Surgery Improvement Collaborative to improve the quality of prostate cancer care in Michigan.