Prediction models for adverse events from prostate cancer curative radiotherapy: a systematic review of methodological quality

BACKGROUND: Predicting risks of urinary, bowel, sexual, and other adverse effects following prostate cancer curative radiotherapy (PCa-RT) is essential for treatment personalization and patient counseling. Several clinical prediction models (CPMs) have been published; however, their methodological quality remains unclear.

METHODS: We systematically reviewed studies developing or validating CPMs for adverse events after PCa-RT. Embase and Medline were searched for CPM studies for patient- or clinician-reported outcomes (PROs/ClinROs), published between January 1, 2004, and August 6, 2024. To focus the appraisal on methodologically more robust models, models were pre-selected based on events per variable ≥10 or the reporting of optimism-corrected performance metrics or effect estimates. We appraised models based on performance and ROB, using a six-item short form of the Prediction model Risk Of Bias ASsessment Tool (SF-PROBAST).

RESULTS: Of 3606 records screened, 136 CPM studies were identified, and 35 were included, yielding 107 CPMs. Most models (n = 87) were developed in external beam radiation therapy populations. 22 models were externally validated. Only two models (AUC= 0.59 and 0.80) - developed in two different studies - were classified as low risk of bias (fulfilled all the SF-PROBAST criteria). 32 models, from 13 studies, met at least four SF-PROBAST criteria and showed at least moderate discrimination (AUC ≥ 0.70) at internal (n = 25) and/or external (n = 11) validation.

CONCLUSION: Ready-to-use CPMs in PCa-RT remain scarce due to methodological limitations, miscalibration, and lack of external validation. Future efforts should prioritize validation and refinement of existing models rather than development of new ones.