We retrospectively reviewed the records of 380 patients with AIPC who were treated at The University of Texas MD Anderson Cancer Center between 1988 and 1995. Androgen independence was defined as the clinical progression of prostate cancer despite castrate serum testosterone concentrations. At presentation, none of the patients had received any therapy for AIPC other than medical or surgical androgen ablation. All patients had at least one site of organ or bone metastasis. Before enrollment in chemotherapeutic protocols for AIPC, all patients underwent a thorough evaluation consisting of a history and physical examination, serum hemoglobin, alkaline phosphatase, prostate-specific antigen (PSA), chest radiography, computerized tomography of the abdomen and pelvis, and bone scan. Cross-sectional imaging studies were reviewed to assess the number of organs involved with prostate cancer. Bone scans were reviewed by a single examiner (CGW) to quantify the number of lesions of increased uptake. An extent-of-disease (EOD) scale was used to grade all bone scans as validated by Soloway and colleagues as level one (fewer than six metastatic sites), level 2 (between six and 20 sites), level 3 (more than 20 sites but not a superscan), and level 4 (more than 75% involvement or superscan). In addition, patients were classified by osseous stage to describe the extent of dissemination. As previously described, an osseous stage of 1 was assigned to patients with metastasis limited to the axial skeleton; patients with additional diaphyseal and distal extremity involvement were classified as stage 2.

Incomplete records were updated via chart review or phone contact with family members. Survival was defined as the time from AIPC diagnosis to death by disease. Kaplan-Meier analysis was used to estimate survival rates and median survival for African-American, Hispanic and Caucasian men. Patient race was determined by patient self-description. Hispanic ethnicity included patients with a Spanish surname who were self-described as Hispanic as defined by the Surveillance, Epidemiology, and End Results database of the National Cancer Institute. Univariate and multivariate survival analysis was performed using Cox pro portional hazards models to evaluate potential prognostic variables affecting survival, such as race, patient age, time to AIPC, serum hemoglobin, alkaline phosphatase, PSA, osseous stage, bone scan extent-of-dis-ease, number of organs involved and treatment group. The log rank test was used to determine the statistical significance of various cut points associated with survival subsequent to multivariate analysis. Log-transformed serum PSA and alkaline phosphatase values were used for statistical analysis, as their values were not normally distributed. The proportionality of hazards was confirmed by plotting log-minus-log survival plots. Parallel plots confirmed that the baseline functions were proportional (data not shown).