Session Item

The diagnosis of tumor recurrence after external beam radiotherapy (EBRT) for prostate cancer (PCa) largely relies on the previous detection of a biochemical relapse. However, current criteria of biochemical relapse require PSA to exhibit a sustained increasing trend, which delays the detection of tumor recurrence and the delivery of a secondary treatment. To address this limitation, we propose to leverage patient-specific forecasts of PSA dynamics to early identify tumor recurrence. Our forecasts are based on a mathematical model describing the basic underlying mechanisms of PCa response to EBRT, which is fit to patient-specific PSA data collected during standard post-EBRT monitoring.

We performed a retrospective study in a cohort of 166 patients with localized PCa who only received EBRT as primary treatment with curative intent at San Raffaele Hospital (Milan, Italy). Ten patients had biochemical relapse. Inclusion criteria required more than 3 years of patient follow-up since EBRT onset and at least 5 post-EBRT PSA values. Our mathematical model relies on four key assumptions: PSA is proportional to the number of tumor cells, EBRT is delivered as an equivalent single dose, EBRT irreversibly damages a fraction of the tumor cells that ultimately undergoes programmed cell death, and the complementary fraction survives to EBRT and continues proliferating. For each patient, we fit the model to an increasing number of PSA values and then validate a model forecast against the remainder of the patient’s PSA data. We used ROC curve analysis to assess a panel of model-based metrics of tumor recurrence. We also assessed the days gained to tumor recurrence diagnosis using our forecasts versus standard clinical criteria. 

The median root mean squared error (RMSE) of our model fits ranged from 0.21 to 0.52 ng/mL for fitting scenarios using 5 to 21 PSA values including the baseline PSA. The median RMSE of the subsequent model forecasts ranged from 0.12 to 1.32 ng/mL. ROC curve analysis identified three metrics of tumor recurrence: the proliferation rate of tumor cells (AUC = 0.99), the ratio of tumor cell proliferation rate to EBRT-induced death (AUC = 0.97), and the PSA nadir (AUC = 0.81). The median and IQR of the days gained to standard tumor recurrence diagnosis using these metrics were 175 [90, 450], 222 [90, 450], and 972 [51, 1639], respectively. The three metrics detected tumor recurrence significantly earlier than standard practice (Wilcoxon sign-rank test p < 0.01).

Our model provided a promising performance in fitting and forecasting post-EBRT PSA values. We identified three model-based metrics enabling an early detection of tumor recurrence with respect to standard practice, and whose definition emanates from the mechanisms of PCa response to EBRT in our model. In the future, we plan to validate our model and metrics in larger cohorts with a higher proportion of tumor recurrences.