Development of a prognostic model for survival time prediction in castration-resistant prostate cancer patients

Background: To identify predictors of survival in patients treated with docetaxel chemotherapy for castration-resistant prostate cancer (CRPC). Methods: We retrospectively analyzed clinical data from 186 patients who underwent docetaxel chemotherapy for CRPC from 2005 to 2016 at a single center. Pretreatment baseline variables including demographic and clinicopathological data were reviewed. Disease progression was defined by imaging and/or consecutive prostate-specific antigen (PSA) elevation. The systemic immune-inflammation index (SII), the modified Glasgow Prognostic Score (mGPS), and the neutrophil-lymphocyte ratio (NLR) were calculated. Univariable and multivariable Cox proportional hazards regression analyses reporting hazard ratios assessed the risk for disease progression and overall survival (OS). A survival nomogram was constructed. Results: Most patients (n = 139, 74.7%) completed at least 6 cycles of docetaxel chemotherapy. 156 patients (82.9%) experienced disease progression during the studied period. Only mGPS was independently associated with disease progression in a multivariable model (P < 0.01). During the studied period, 98 patients (52.1%) died. The built survival nomogram included statistically significant variables for OS in univariable analysis: hemoglobin, PSA, alkaline phosphatase (AP), lactate dehydrogenase, SII, neutrophil-lymphocyte ratio, mGPS, and site of metastases; and had a concordance index of 0.703. At decision curve analysis, the nomogram led to superior outcomes for any decision associated with a threshold probability of above 40%. In multivariable analysis, only AP (P = 0.02), hemoglobin and PSA (P < 0.01, respectively) remained associated with OS. Conclusions: PSA, AP, and hemoglobin are independent prognosticators for OS. Although mGPS is a promising marker for tumor progression and SII is a plausible prognostic marker for OS, valid integration of inflammatory indices into a prognostic model requires validation studies. Predictive and prognostic biomarkers are desperately needed to guide physicians in treatment counseling given the heterogeneous nature of CRPC and the plethora of effective therapies.