Adequate tissue was available for immunohistochemical staining of p53 in 780 (51.3%) patient cases, of Ki-67 in 637 (41.9%) patient cases, and of MDM2 in 589 (38.7%) patient cases, of the 1,521 total patient cases in the parent cohort. There were 478 patient cases (31.4%) with complete biomarker data for all three markers. The pretreatment characteristics of the groups with individual or complete biomarker data versus those without biomarker data were similar, except for Gleason Score, in which a higher proportion of the study cohort had Gleason scores of 8 to 10 (Ki-67, P = .04; MDM2, P = .007; all markers, P = .006) and had assigned treatment (MDM2, P = .0006; all markers, P = .02; ). Although these study cohorts may not represent a random sample of the parent cohort, there were no statistically significant differences between the groups with marker data (whether individual or complete) and those without for the end points DM, CSM, and OM (Appendix Table 1, online only).
Distribution of Patients by Missing v Determined Biomarkers
In terms of pretreatment characteristics, there were no significant differences in the distribution of patients by assigned treatment in each of these data sets. The median age of the patients with all three biomarkers was 70 years (range, 44 to 88 years), and the median iPSA was 21.5 ng/mL (range, 0.1 to 219.7 ng/mL). One hundred fifty-three patients (34%) and 134 patients (30%) had Gleason scores of 7 and of 8 to 10, respectively, and 262 (55%) patient cases had T3-4 disease. Two hundred sixty patients (54%) and 218 patients (46%) were assigned to the RT + LTAD and RT + STAD treatment arms, respectively. Median follow-up for all patients still alive with complete biomarker data was 9.3 years (0.04 to 13.9 years), and median follow-up was 11.4 years (range, 1.6 to 13.9 years).
We have observed that the median cut point is often not the most optimal for predicting outcome. Rather than undergo a random search for an optimal dichotomous cut point for each biomarker, we systematically investigated the relationships to OM, CSM, and DM for cut points at the 25th (Q1), 50th (median), and 75th percentiles (Q3). Also considered were the relative differences in these associations by using the manual and image-analysis PSP and image-analysis MIS parameters as dichotomous variables. Manual scoring of the MDM2 patient cases was not performed, as image analysis was found to be more reliable in our previous report.10
The optimal cut point for each marker was determined by the results of multivariate analyses that were adjusted for age, iPSA, Gleason score, clinical stage, and assigned treatment (). The manual PSP median (0%) for p53 (of note, any staining of ≤ 5% was counted as 0%), manual PSP Q3 (11.3%) for Ki-67, and image-analysis MIS Q3 (184 arbitrary units) for MDM2 were chosen on the basis of being associated with the highest HRs for all significant end points.
Multivariate Analyses of Markers as Dichotomous Variables: Individual Biomarker Models
lists the crude numbers of failures by each biomarker-dichotomous cut point for each end point. also displays the HR results of the multivariate analyses by each end point, after adjustment for age, iPSA, Gleason score, clinical stage, and assigned treatment and for all the covariates above plus the other biomarkers. After analysis was adjusted for the covariates, including the other markers, p53 lost its significance for DM (HR, 1.04; 95% CI, 0.68 to 1.61; P = .85), and MDM2 was no longer significant for CSM (HR, 1.55; 95% CI, 0.94 to 2.55; P = .08). The other previously significant relationships defined in individual biomarker multivariate analyses in remained. We also performed these analyses with iPSA as a continuous covariate (Appendix Table 2, online only), and the significant biomarker relationships to DM, CSM, and OM were not substantively different.
Summary of Failures and Adjusted MVA Results
The biomarker data were also considered as continuous variables for all end points (data not shown). In these multivariate analyses, only Ki-67 was significantly related to all the end points, after adjustment for all covariates, including the other markers (DM, P < .0001; CSM, P < .0001; OM, P = .001).
As dichotomous covariates, both Ki-67 and MDM2 were independent predictors of DM. As such, the combination of the Ki-67 and MDM2 data were considered together. Of note, when using either the Ki-67 manual or image-analysis data, the results were comparable with the four groups statistically different with respect to DM, CSM, and OM. The manual Ki-67 data were used in , because the results were stronger than when the image-analysis data were used; therefore, the manual Ki-67 results were combined with the image analysis results for MDM2. The group with high Ki-67 and MDM2 expression was more likely to experience failure events for all three end points than those groups with low Ki-67 and/or MDM2 expression (). A and B display the cumulative incidence curves (unadjusted) for DM and CSM, and C displays the Kaplan-Meier OM curves (unadjusted), subgrouped by the combination of Ki-67 and MDM2 expression data. The curves demonstrate that patients with high Ki-67 PSP and intensity of MDM2 had much greater risks of DM and death. These relationships remained after analysis was adjusted for age, iPSA, Gleason score, clinical stage, assigned treatment, and p53 (). As lists, high Ki-67 (manual PSP > 11.3%) and high MDM2 (image-analysis MIS > 184) were predictive of DM (P < .0001), CSM (P < .0001), and OM (P = .0002) independent of the other covariates. Of note, these findings were not changed substantively when iPSA was included as a continuous covariate (Appendix Table A3, online only).
Fig 1. Cumulative incidence failure curves of (A) distant metastasis and (B) cause-specific mortality. (C) Kaplan-Meier survival curves of overall mortality subdivided by the combination of Ki-67 percentage of tumor nuclei staining positive and MDM2 mean intensity (more ...)
Multivariate Analysis of Combined MDM2 and Ki-67 Marker Data
The outcome of the patients in RTOG protocol 92-02 was dependent on the length of androgen deprivation received. To study if this was associated with the combination of Ki-67 and MDM2 expression, tests for an interaction between the assigned treatment arm and each combined Ki-67/MDM2 group were performed (); there were no statistically significant differences for all end points tested. These results indicate that assigned treatment as a covariate was valid for interpretation, and they substantiated treatment association with DM and CSM (). We then questioned whether any of the combined Ki-67 and MDM2 subgroups (ie, ≤ 11.3% and ≤ 184 v ≤ 11.3% and > 184 v > 11.3% and ≤ 184 v > 11.3% and > 184) had reduced DM by RT + LTAD treatment. For example, if the high Ki-67 and high MDM2 (ie, > 11.3% and > 184) subgroup had reduced DM with RT + LTAD compared with RT + STAD, then we would have a better idea of which patients require RT + LTAD. Although a distinct trend was not observed, the small numbers of events in three of the four subgroups precluded meaningful statistical analyses (data not shown).
Test for an Interaction Between the Ki-67 and MDM2 Combination and Treatment Arm