Participant characteristics in the two randomly assigned groups were closely comparable in both the initial 16 608 randomly assigned participants[16,21] and the 12 788 participants reconsenting for extended follow-up with no statistically significant differences (Table 1). Reconsent rates were similar between randomly assigned groups, even when considered by baseline characteristics, with the exception that obese women in the placebo group had slightly lower reconsent rates.
In intent-to-treat analyses, data were available for a median follow-up of 5.6 years (interquartile range [IQR] = 4.8–6.5) from random assignment until termination of study medicine intervention and a median follow-up of 8.2 years (IQR = 6.6–8.2) postintervention and 13.2 years (IQR = 10.5–14.2) of cumulative median follow-up.
Over cumulative follow-up, continuous combined estrogen plus progestin use decreased endometrial cancer incidence (66 case patients, 0.06% per year) compared with placebo (95 case patients, 0.10% per year, HR = 0.65, 95% CI = 0.48 to 0.89, P = .007) (Figure 2). During intervention, fewer endometrial cancers were diagnosed in the estrogen plus progestin group, but the difference was not statistically significant (25 vs 30 case patients, HR = 0.77, 95% CI = 0.45 to 1.31, P = .33). Similar findings were apparent through the protocol defined completion date (March 31, 2005) (HR = 0.73, 95% CI = 0.49 to 1.09). A statistically significant difference in endometrial cancer incidence between randomly assigned groups emerged postintervention (41 case patients, 0.08% per year vs 65 case patients, 0.13% per year, respectively [HR = 0.59, 95% CI = 0.40 to 0.88, P = .008]), but intervention and postintervention phase hazard ratios were not different (P difference = .46).
Kaplan-Meier estimates of cumulative hazards of endometrial cancer in the Women's Health Initiative randomized trial of continuous combined estrogen plus progestin with the intent-to-treat principle. Hazard ratio (estrogen plus progestin vs placebo) with 95% confidence intervals and P values is from Cox regression models stratified by age and random assignment in the dietary modification trial. Estrogen indicates conjugated equine estrogens, and progesterone indicates medroxyprogesterone acetate. All statistical tests were two-sided. CI = confidence interval; HR = hazard ratio.
To address the potential imbalance of reconsent for obese participants between randomly assigned groups, analyses stratified by BMI group had little effect on postintervention incidence (HR = 0.59, 95% CI = 0.40 to 0.87, P = .008). Inclusion of three endometrial sarcoma cases (1 spindle cell sarcoma in the placebo group and 2 stromal sarcomas in the estrogen plus progestin group) yielded a cumulative hazard ratio of 0.67 (95% CI = 0.49 to 0.91, P = .01), as previously reported. Endometrial cancer incidence results were somewhat greater in analyses adjusted for adherence censoring events that occurred six months after consuming less than 80% of study pills or starting non-protocol hormone therapy (HR = 0.49, 95% CI = 0.30 to 0.80, P = .004).
In terms of tumor characteristics, there was no evidence of a differential estrogen plus progestin effect by grade or stage (Figure 3). In intent-to-treat analysis over the entire study period, there was a statistically nonsignificant reduction in deaths from endometrial cancer (5 vs 11 deaths, HR = 0.42, 95% CI = 0.15 to 1.22, P = .10), but the number of events was insufficient for definitive assessment. None of the 11 subgroup interactions examined were statistically significant (P > .20), suggesting a similar influence of estrogen plus progestin on endometrial cancer risk generally, including in women in the highest BMI categories and in cases with type II histology (Figure 4).
Cumulative number of events (annual percentages) and hazard ratios (95% confidence intervals) for incidence of endometrial cancer, tumor types, and deaths from endometrial cancer in the Women's Health Initiative randomized trial of continuous combined estrogen plus progestin by random assignment. Tumor size was estimated in accordance with the Surveillance, Epidemiology, and End Results 1988, 2nd edition rules for coding cancers with size measured as the length along the longest axis. Tumors were classified as type I or type II by methods previously described by Setiawan in the Endometrial Cancer Consortium.(12) Briefly, endometrioid carcinoma, adenocarcinoma NOS, and adenocarcinoma with squamous differentiation were classified as type I, serous/papillary serous and mixed cell adenocarcinoma as type II, and remaining tumors classified as other. Hazard ratios, 95% confidence intervals, and P values are from Cox proportional hazards models stratified according to age and random assignment in the dietary modification trial. Dotted line represents overall hazard ratio for endometrial cancer incidence attributed to estrogen plus progestin. P value corresponds to a score x2 test from a competing risks analysis that tested whether hazard ratios differed between tumor types. Estrogen indicates conjugated equine estrogens, and progesterone indicates medroxyprogesterone acetate. Number of events may not add up because of missing endometrial cancer type. All statistical tests were two-sided. CI = confidence interval; HR = hazard ratio.
Cumulative number of events (annualized percentages) and hazard ratios (HRs [95% confidence intervals] for endometrial cancer incidence) in the Women's Health Initiative randomized trial of continuous combined estrogen plus progestin by baseline characteristics and random assignment. Hazard ratios, 95% confidence intervals, and P values are from Cox proportional hazards models stratified according to age and random assignment in the dietary modification trial. Dotted line represents overall hazard ratio for endometrial cancer incidence attributed to estrogen plus progestin. P value is from a one degree-of-freedom score x2 test of the interaction between the given subgroup and random assignment. Estrogen indicates conjugated equine estrogens, and progestin indicates medroxyprogesterone acetate. Number of events may not add up because of missing baseline data. All statistical tests were two-sided. BMI = body mass index; CI = confidence interval; HR = hazard ratio.
There were four endometrial cancers diagnosed in the 331 women originally randomly assigned to estrogen alone, seven in the 573 women randomly assigned to combined hormones, and four in the 522 women randomly assigned to placebo during the same period. Removing these case patients did not change the results (55 vs 91 case patients, HR = 0.59, 95% CI = 0.43 to 0.83, P = .002), nor did only excluding women originally randomly assigned to estrogen alone (62 vs 95 case patients, HR = 0.64, 95% CI = 0.47 to 0.88, P = .006).
Overall, there were more hysterectomies in the combined hormone therapy group (510 [0.51% per year] vs 450 [0.48% per year], HR = 1.08, 95% CI = 0.95 to 1.23, P = .23), but the difference was not statistically significant. In addition, very few (2.2%) were performed for atypical hyperplasia (Table 2), and the sensitivity analysis that censored women at the time of hysterectomy did not change the endometrial cancer results (HR = 0.65, 95% CI = 0.48 to 0.90, P = .008).
The expected strong association between BMI and endometrial cancer risk was seen among all participants (HR = 1.0 [reference], HR = 1.71, 95% CI = 1.99 to 2.93, HR = 3.58, 95% CI = 2.11 to 6.07, HR = 6.66, 95% CI = 3.82 to 11.61, and HR = 7.40, 95% CI = 3.86 to 14.16 for BMI <25, 25<30, 30<35, 35<40, and ≥40kg/m2, respectively) in analyses that added covariates of age (linear) and race/ethnicity in the core model.
J Natl Cancer Inst. 2016;108(3) © 2016 Oxford University Press