Recursive partitioning for prognostic grouping of patients with clinically localized prostate carcinoma

Abstract

Patients treated with radical prostatectomy for clinically localized prostate carcinoma present considerable heterogeneity in terms of disease free survival outcome. Multiple studies have attempted to create prognostic groupings of these patients in the perioperative phase, using information available regarding several clinicopathologic variables. Such groupings allow physicians to make early yet prudent decisions regarding adjuvant combination therapies. The current study presents results from a statistical analysis that enables the natural identification of such prognostic groups. Examination of consecutive radical prostatectomy specimens was performed between January 1991 and December 1995 at Wayne State University, Harper Hospital, Detroit, Michigan. Disease free survival in a cohort of 485 of these men was analyzed using recursive partitioning and amalgamation technique. Clinicopathologic parameters evaluated included age, race, preoperative prostate specific antigen (PSA) level, clinical and pathologic stage, and Gleason grade of the fine-needle biopsy as well as the radical prostatectomy specimen. A binary decision tree representation was generated for classifying patients based on the clinicopathologic variables mentioned earlier. The worst prognosis was for patients with either advanced stage and a PSA level > 24.1 ng/mL or advanced stage, a PSA level </= 24.1 ng/mL, and age </= 65 years. This group had an estimated median disease free survival of only 10.3 months. Patients with lower pathologic stage, prostatectomy Gleason scores of </= 7, and a preoperative PSA level </= 22.7 ng/mL had the best prognosis. The recursive partitioning analyses allows easy characterization of a patient for prognosis shortly after radical prostatectomy. This will enable a physician to make more prudent decisions regarding whether to employ watchful waiting, proceed with accepted adjuvant therapy (radiotherapy or hormonal therapy), or refer patients to a research center that is able to administer experimental adjuvant therapy. [See editorial counterpoint on pages 232-3 and reply to counterpoint on pages 234-5, this issue.]