The expression of PKARIα
was determined in 313 (21%) of 1,521 assessable cases in RTOG 92-02. The patient characteristics are shown in . We then analyzed whether the PKA cohort was representative of the parent cohort by comparing the distribution of patients with respect to age (<70 versus ≥70 years), pretreatment PSA (≤30 versus >30 ng/mL), Gleason score (2–6 versus 7 versus 8–10), clinical T-category (T2 versus T3/T4), and assigned treatment (STAD + RT versus LTAD + RT). The only statistically significant difference was in Gleason score (P
= 0.03) because there were slightly more cases with Gleason score 8 to 10 (28% versus 23%) and fewer with Gleason score 2 to 6 (32% versus 40%) in the PKA cohort compared with those in whom PKA was not available. In terms of the various end points tested (Supplementary Table S1
), the only significant difference was that the PKA cohort had more LFs (P
= 0.03; HR, 1.35). Thus, some imbalances between those cases that had a PKA available and those that did not were apparent.
Patient characteristics by PKA status
displays a series of comparisons between manual and image analysis estimations of PKARIα
intensity. These comparisons were done with the goal of identifying a robust image analysis approach with the potential to be better replicated between laboratories compared with more subjective manual quantification. To limit comparisons, but to identify if image analysis cutpoints other than the median were best, we tested the median, first quartile (Q1, 25%), and third quartile (Q3, 75%) cutpoints as well as a cutpoint identified in the prior analysis of cases from RTOG 86–10. Thus, the image analysis–derived MIS cutpoints tested were at 101.7 (Q1), 111.8 (median), and 128.0 (Q3), as well as a previously tested median cutpoint of 135.5 (9
). The grouping of the manual PKARIα
determinations into negative (staining classified as level 0 or 1) versus positive (staining classified as level 2 or 3) and negative/low (level 0, 1, or 2 staining) versus high (level 3 staining) was modeled after our prior analysis. The comparisons in between the MISs by image analysis and the manual assessments revealed strong correlations between the two for all cutpoints tested.
Concordance of PKA scores obtained from manual and semiautomated image (ACIS) analysis
The relationships of PKARIα
expression to age, initial PSA, Gleason score, T-category, and assigned treatment, using the categorizations used in , were explored. As described above, manually determined PKARIα
expression was dichotomized as intensities of 0 to 1 versus 2 to 3 and 0 to 2 versus 3. Image analysis–determined PKARIα
MIS expression was dichotomized by the first quartile, median, third quartile, and the cutpoint from our prior study (9
). In none of the permutations tested was PKARIα
expression significantly associated with patient- or treatment-related factors.
Multivariate analyses were used to examine the relationship of the estimated PKARIα levels and the clinical outcomes. Multi variate analyses were adjusted for age (continuous), initial PSA (continuous), Gleason score, T-category, and assigned treatment and the results are presented in . With the exception of OM, in which Cox proportional hazards was used, Fine and Gray’s regression was applied to account for competing risks. Dichotomized PKARIα expression was most significantly associated with outcome using the third quartile cutpoint (MIS of 128.0) as well as the previously established cutpoint (MIS of 135.5) based on an analysis of RTOG 86-10. Using the 92-02–derived third quartile cutpoint or the 86-10–derived cutpoint, high PKARIα expression was significantly correlated with DM, LF, and BF. A borderline relationship (P = 0.07) was seen for CSM. As a continuous covariate, PKARIα expression was only significant for BF (P = 0.005; ).
Multivariate analyses of dichotomized and continuous PKARIa expression quantified by image analysis
Because the expression of PKARIα at the third quartile cutpoint (MIS of 128.0) was significantly associated with DM, but not CSM or OM, we examined the power to detect a significant relationship for the 313 analyzable patients in the cohort. For the BF, DM, and LF end points, at HRs of 1.8, 1.56, and 1.34, the statistical power would be at least 92%, 95%, and 72%. In contrast, for CSM and OM at HRs of 1.25 and 1.13, the statistical power would be at least 55% and 17%. These estimates were made using a one-sided log-rank test at the 0.05 significance level. Thus, the study was most adequately powered to detect associations with BF and DM, with relatively low statistical power to detect differences in CSM and OM based on dichotomized PKARIα at the third quartile.
displays adjusted cumulative incidence DM curves for patients subdivided by the third quartile MIS cutpoint of 128.0 () and the manual cutpoint of 0 to 1 versus 2 to 3 (). The 10-year risk of DM was 15.3% for those with a PKARIα of <128.0 and 29.1% for a PKARIα of ≥128.0. Similar results were observed using manual counts with a 10-year risk of DM of 14.8% for those with a manual intensity of 0 to 1 and 25.2% for a manual intensity of 2 to 3. shows the multivariate results of PKARIα expression quantified manually to the end points investigated. There was little difference between grouping the manual intensities as 0 to 1 versus 2 to 3 or 0 to 2 versus 3. The results were analogous to those using image analysis; however, the image analysis PKARIα results using the Q3 (75th percentile) cutpoint, and the cutpoint from the prior study, had more significant P values and higher HRs ().
Adjusted DM cumulative incidence curves by PKARIα expression quantified by image analysis (ACIS MIS cutpoint of 128.0; A) and manually (manual intensity of 0–1 versus 2–3; B).
Multivariate analyses of dichotomized PKARIα expression by manual scoring
We then asked whether PKARIα overexpression had a differential effect on outcome when the patients were subdivided by assigned treatment. displays this analysis and reveals that, in general, the gain from extending AD for an additional 2 years (LTAD + RT arm) was most pronounced when PKARIα expression was low. Manual PKARIα intensities of 0 to 1 were associated with HRs for CSM, DM, LF, and BF of 0.25 (P = 0.005), 0.23 (P = 0.003), 0.31 (P = 0.007), and 0.54 (P = 0.0003) when RT + LTAD was referenced to RT + STAD. In contrast, the only statistically significant treatment-related difference was for BF (HR, 0.45; P = 0.003) when the manual PKARIα intensities were 2 to 3. A very similar pattern was seen using the image analysis data (MIS cutpoint of 128.0). These results hold when the HRs were adjusted for other covariates.
Subgroup analyses for treatment effect [LTAD + RT versus STAD (RL) + RT]