Machine-learning predicts time-series prognosis factors in metastatic prostate cancer patients treated with androgen deprivation therapy

Abstract

Machine learning technology is expected to support diagnosis and prognosis prediction in medicine. We used machine learning to construct a new prognostic prediction model for prostate cancer patients based on longitudinal data obtained from age at diagnosis, peripheral blood and urine tests of 340 prostate cancer patients. Random survival forest (RSF) and survival tree were used for machine learning. In the time-series prognostic prediction model for metastatic prostate cancer patients, the RSF model showed better prediction accuracy than the conventional Cox proportional hazards model for almost all time periods of progression-free survival (PFS), overall survival (OS) and cancer-specific survival (CSS). Based on the RSF model, we created a clinically applicable prognostic prediction model using survival trees for OS and CSS by combining the values of lactate dehydrogenase (LDH) before starting treatment and alkaline phosphatase (ALP) at 120 days after treatment. Machine learning provides useful information for predicting the prognosis of metastatic prostate cancer prior to treatment intervention by considering the nonlinear and combined impacts of multiple features. The addition of data after the start of treatment would allow for more precise prognostic risk assessment of patients and would be beneficial for subsequent treatment selection.

Figure 1

Comparison of accuracy of prognostic prediction models. (A) Comparison of C-index for each prognostic prediction model. Black, shaded, and horizontal bars indicate RSF, Cox proportional hazards, and PSA models, respectively. (B) Comparison of C-index for application of RSF model to patients with metastatic and non-metastatic prostate cancer. Black, striped, and dotted bars indicate all patients with prostate cancer, patients with metastatic prostate cancer, and patients with non-metastatic prostate cancer, respectively. (C to E) Permutation importance in prediction of progression (C), overall survival (D), and cancer specific survival (E) based on the RSF model.

Figure 2

Time-series of prognostic accuracy for patients with metastatic prostate cancer. (A to C) Accuracy of prediction of progression (A), overall survival (B), and cancer specific survival (C). The green, red, and blue lines indicate the RSF, Cox proportional hazards, and prognostic PSA-based models, respectively. The triangle mark indicates the time point at which prediction accuracy was the highest for RSF prediction. Error bars represent standard deviations of 10 independent RSF. Permutation importance in prediction of progression at 150 days after treatment initiation (D), overall survival at 120 days after treatment initiation (E), and cancer-specific survival at 120 days after treatment initiation (F). The number of factors was defined as the top 10 factors or those with positive importance.

Figure 3

Survival tree predicting overall survival (A) and cancer-specific survival (C). Kaplan–Meier curves of survival tree prognostic classification results for overall survival prediction (B) and cancer-specific survival prediction (D). P-values were calculated by the log-rank test.