Pérez García, Victor M.. 

The detection of prostate cancer recurrence after external beam radiotherapy relies on the measurement of a sustained rise of serum prostate-specific antigen (PSA). However, this biochemical relapse may take years to occur, thereby delaying the delivery of a secondary treatment to patients with recurring tumors. To address this issue, we propose to use patient-specific forecasts of PSA dynamics to early predict biochemical relapse. Our forecasts are based on a mechanistic model of prostate cancer response to external beam radiotherapy, which is fit to patient-specific PSA data collected during standard post-treatment monitoring. Our results show a remarkable performance of our model in recapitulating the observed changes in PSA and yielding short-term predictions over approximately one year (cohort median RMSE of 0.10 to 0.47 ng/mL and 0.13 to 1.39 ng/mL, respectively). Additionally, we identify three model-based biomarkers that enable an accurate identification of biochemical relapse (AUC>0.80) significantly earlier than standard practice (p<0.01).