Applying a Neural Network to Prostate Cancer Survival Data

Prediction of treatment efficacy for prostate cancer therapies has proven difficult and requires modeling of survival-type data. One reason for the difficulty may be infrequent use of flexible modeling techniques, such as artificial neural networks (ANN). The purpose of this study is to illustrate the use of an ANN to model prostate cancer survival data and compare the ANN to the traditional statistical method, Cox proportional hazards regression. Clinical data and disease follow-up for 983 men were modeled by both an ANN and a Cox model. Repeated sampling of 200 training and testing subsets were supplied to each technique. The concordance index c was calculated for each testing dataset. As further validation, ANN and Cox models were applied to a totally separate dataset. The ANN outperformed the Cox model in internal validation datasets (ANN c = 0.76, Cox c = 0.74) and on the external validation dataset (ANN c = 0.77, Cox c = 0.74). ANNs were more discriminating than Cox models for predicting cancer recurrence. Calibration of the ANN remains a problem. Once solved, it is expected that an ANN will make the most accurate predictions of prostate cancer recurrence and improve treatment decision making.

[1]  J R Beck,et al.  Artificial neural networks for medical classification decisions. , 1995, Archives of pathology & laboratory medicine.

[2]  Galina Pizov,et al.  Correlation of pathologic findings with progression after radical retropubic prostatectomy , 1993, Cancer.

[3]  R. B. Smith,et al.  Prostate specific antigen levels after radical prostatectomy in patients with organ confined and locally extensive prostate cancer. , 1992, The Journal of urology.

[4]  P. Grambsch,et al.  Martingale-based residuals for survival models , 1990 .

[5]  M W Kattan,et al.  Evaluation of a Nomogram used to predict the pathologic stage of clinically localized prostate carcinoma , 1997, Cancer.

[6]  Brian J. Miles,et al.  A Decision Analysis for Treatment of Clinically Localized Prostate Cancer , 1997 .

[7]  Daniel B. Mark,et al.  TUTORIAL IN BIOSTATISTICS MULTIVARIABLE PROGNOSTIC MODELS: ISSUES IN DEVELOPING MODELS, EVALUATING ASSUMPTIONS AND ADEQUACY, AND MEASURING AND REDUCING ERRORS , 1996 .

[8]  P. Scardino,et al.  The new american joint committee on cancer and international union against cancer tnm classification of prostate cancer , 1994, Cancer.

[9]  F. Harrell,et al.  Evaluating the yield of medical tests. , 1982, JAMA.

[10]  D Faraggi,et al.  A neural network model for survival data. , 1995, Statistics in medicine.

[11]  Kurt Hornik,et al.  Multilayer feedforward networks are universal approximators , 1989, Neural Networks.

[12]  M. Cowen,et al.  Computer modeling in urology. , 1996, Urology.

[13]  M De Laurentiis,et al.  A technique for using neural network analysis to perform survival analysis of censored data. , 1994, Cancer letters.

[14]  James F. Watkins,et al.  Analysing Survival Data from Clinical Trials and Observational Studies. , 1995 .

[15]  A W Partin,et al.  Serum PSA after anatomic radical prostatectomy. The Johns Hopkins experience after 10 years. , 1993, The Urologic clinics of North America.