Study population
The total cohort (N = 3460) consisted of patients treated with ADT plus AA plus prednisone (AAP cohort; N = 1930) and patients treated with ADT plus placebo with or without prednisone (ADT cohort; N = 1530). Baseline patient characteristics, accumulated for three cohorts based on PPI use, are summarized in Supplemental Tables 2 and 4. Overall, 475 patients (24.6%) used PPI in the AAP cohort, and 300 (19.6%) in the ADT cohort. Baseline characteristics were comparable before and after imputation. PPI use is listed in Supplemental Table 5. Omeprazole was the most widely used drug in the total cohort (N = 327 [42.2%]). The details of PPI use were similar between the AAP cohort and the ADT cohort.
After applying IPTW, patient characteristics were balanced between PPI users and non-users in all cohorts (Supplemental Table 2 to Supplemental Table 4). The distributions of estimated propensity scores were shown in Supplemental Fig. 1.
Association between PPI use and patient characteristics
Multivariable logistic regression analysis showed the following characteristics to be significantly associated with PPI use in the total cohort: performance status (OR: 2.05, 95% CI: 1.29 to 3.25), history of gastrointestinal disorder (OR: 4.05, 95% CI: 3.36 to 4.87), history of cardiovascular disorder (OR: 1.29, 95% CI: 1.05 to 1.58), statin use (OR: 1.29, 95% CI: 1.05 to 1.58), lymph node metastasis (OR: 0.79, 95% CI: 0.66 to 0.95), and trial (LATITUDE vs. COU-AA-301, OR: 2.51, 95% CI: 1.74 to 3.62; LATITUDE vs. COU-AA-302, OR: 1.78, 95% CI: 1.27 to 2.50) (Supplemental Table 6).
Treatment effectiveness of AA plus prednisone and ADT based on PPI use
Median follow-up was 49.3 months (interquartile range: 38.7 to 54.9) for the AAP cohort and 49.8 months (interquartile range: 39.4 to 54.0). During follow-up in the AAP cohort, radiographic progression developed in 1396 patients (72.3%) and all-cause mortality in 1274 (66.0%). In the ADT cohort, those numbers were 1186 (77.5%) and 1058 (69.2%), respectively.
The crude Kaplan-Meier curves showed that AA treatment was associated with improved survival of patients enrolled in LATITUDE, COU-AA-301, and COU-AA-302 compared with placebo, irrespective of PPI use (Supplemental Fig. 2 to 3). The crude Kaplan-Meier curves for the 2 cohorts, based on PPI use, are shown in Fig. 1.
Crude data from the AAP cohort showed that PPI users had a shorter RMOST (difference: -10.5 months, 95% CI: -12.6 to -8.4) and RMPFST (difference: -8.5 months, 95% CI: -10.7 to -6.3) than non-users (Table 1). After IPTW adjustment, PPI users continued to show shorter RMOST (difference: -4.2 months, 95% CI: -7.0 to -1.4) and RMPFST (difference: -3.5 months, 95% CI: -6.6 to -0.4) compared with non-users. Furthermore, the RMST difference curves showed that the differences in RMOST and RMPFST increased over time until truncation at 63.5 months (Fig. 2). PPI use was associated with higher HRs for all-cause mortality (HR: 1.25, 95% CI: 1.08 to 1.45) and radiographic progression (HR: 1.20, 95% CI: 1.03 to 1.39), respectively.
Similar associations were observed between PPI use and treatment outcomes from the crude data of the ADT cohort (RMOST, difference: -8.3 months, 95% CI: -10.6 to -5.9; RMPFST, difference: -4.7 months, 95% CI: -6.8 to -2.5). However, the differences in RMOST (difference: -2.6 months, 95% CI: -5.8 to 0.6) and RMPFST (difference: -1.7 months, 95% CI: -4.8 to 1.3) were no longer significant after applying IPTW (Table 1). The RMST difference curves in the ADT cohort also showed that the differences in RMOST and RMPFST remained non-significant until the truncation time (Fig. 3). The IPTW-adjusted Cox regression model did not show a significant association between PPI use and treatment outcomes in the ADT cohort (HR for all-cause mortality: 1.18, 95% CI: 0.99 to 1.41; HR for radiographic progression: 1.11, 95% CI: 0.92 to 1.33).
Interaction term analysis showed no evidence of significant interaction with treatment (AAP vs. ADT/ADTP, HR: 0.90, 95% CI: 0.72 to 1.13, P for interaction = 0.36). However, PPI notably affected mCSPC (mCRPC vs. mCSPC, HR: 1.48, 95% CI: 1.14 to 1.91, P for interaction = 0.003).
Sensitivity analysis
Sensitivity analysis based on the E-value was performed to assess the degree of possible unmeasured confounding within this study. In the ADT cohort and the AAP cohort, the E-values of HRs for all-cause mortality were 1.61 and 1.53, respectively, and for radiographic progression were 1.49 and 1.36, respectively.