Using the receiver operating characteristic curve to select pretreatment and pathologic predictors for early and late...

Area of Science:

• Urology
• Oncology
• Medical Informatics

Background:

• Postprostatectomy prostate-specific antigen (PSA) levels are influenced by pretreatment PSA, Gleason score, margin status, and pathologic T stage.
• Predicting early and late PSA failure is crucial for patient management and treatment planning.

Purpose of the Study:

• To identify optimal predictors for early and late postprostatectomy PSA failure using receiver operating characteristic (ROC) curve analysis.
• To develop and validate a clinical outcome prediction expert system for individual patient risk assessment.

Main Methods:

• A clinical outcome prediction expert system was developed and validated on a database of 1022 patients.
• The database was split into 60% for training and 40% for validation.
• ROC areas were calculated for predictors over a 24- to 60-month cutoff period.

Main Results:

• The combination of pathologic T stage, prostatectomy Gleason score, and margin status yielded the highest ROC area (0.900).
• Patients with Stage T disease < T3, negative margins, and Gleason score ≤ 6 had a 90% probability of being PSA failure-free at 4 years.
• Pathologic T stage and margin status accurately predicted PSA failure within 24 months (ROC area = 0.800).

Conclusions:

• Actuarial analysis combined with ROC optimization effectively identifies patients at high risk for postprostatectomy PSA failure.
• This predictive model aids in stratifying patients for risk of early and late PSA recurrence.