Cancer Incidence and Mortality during the Intervention and Postintervention Periods of the Women's Health Initiative Dietary Modification Trial

During the 1990s, preclinical, early ecologic, and observational studies historically reported that dietary fat intake was positively associated with breast (1) and colorectal cancer (2) risk. A sizeable randomized, controlled trial with adequate statistical power was designed to test this hypothesis. The Women's Health Initiative (WHI) randomized low-fat dietary modification (DM) trial (3) was initiated with incidence of (invasive) breast cancer and colorectal cancer as primary outcomes, to test the hypothesis that an intervention involving reduced total fat (<20% energy) would reduce incidence of these outcomes over a 9-year period in 48,000 free-living postmenopausal women (4–6). Intervention dietary goals focused on lowering total dietary fat intake and increasing fruit, vegetable, and grain consumption (3).

The DM trial intervention period ended in 2005 with simultaneous reports on breast cancer (7), colorectal cancer (8), and cardiovascular disease (9). Subsequent reports were published for the DM trial on other cancers, including ovarian cancer (10). There was no overall intervention effect on colorectal or breast cancer, although there were fewer invasive breast cancers in the low-fat diet group than the comparison group with a hazard ratio (HR) of 0.91, 95% confidence interval (CI), 0.83–1.01, and a significant interaction (P = 0.04) between the baseline percentage of energy from fat as measured by 4-day food records. Women who reported relatively higher percentage of energy from dietary fat at baseline made larger reductions in fat intake and demonstrated a nominally significant reduction in breast cancer risk (HR, 0.78; 95% CI, 0.64–0.95; ref. 7).

This article updates the DM trial analyses by including an additional 5.2 years of postintervention follow-up, through September 2010. The specific objectives for this follow-up analysis were to evaluate the longer-term effects of 8.3 years on average of diet intervention on incident breast, colorectal, other, and total cancers as well as cancer-specific and overall mortality over the 12.3-year combined intervention and postintervention follow-up period.

Details of the WHI DM trial have been provided elsewhere (11–13). Briefly, postmenopausal women, ages 50 to 79 years, were recruited at 40 clinical centers across the United States between 1993 and 1998. Women entering the WHI DM trial also could participate in the WHI hormone therapy (HT) trial (14) or the calcium plus vitamin D supplementation trial (15). Exclusions for participation in the DM trial included prior breast cancer or colorectal cancer, other cancer within the past 10 years, predicted survival less than 3 years, adherence or retention concerns (e.g., alcoholism and dementia), or baseline fat intake <32% of total calories as estimated by a food frequency questionnaire (FFQ; ref. 16). No exclusions for body mass index (BMI) were applied. A mammogram, not suspicious for cancer, was required for study entry and subsequently at 2 year intervals, or annually for women concurrently enrolled in the hormone trial; colorectal cancer screening was not protocol-mandated but self-reported information regarding screening was collected. Women were randomized 40%/60% into the intervention and control/comparison arms of the trial; this analysis includes all DM participants.

Between 1993 and 1998, 48,835 women were randomized (40:60) to intervention (n = 19,541) or comparison (n = 29,294) groups. Demographic characteristics and medical history were collected by self-report using standardized questionnaires. Anthropometric measurements including height, weight, and waist circumference at the umbilicus were collected at baseline and annually throughout the trial by trained personnel in the local WHI clinics using standardized protocols (10); no postintervention period follow-up anthropometric measurements were taken. BMI was calculated as weight in kg/height in meters-squared.

Diet was assessed by a validated FFQ during the intervention period as previously described (16). Four-day food records were collected on all DM trial women at baseline. The stored records were analyzed early on in the trial in the 4.6% sample and later for cases for use in breast cancer case-only analysis. Diet also was assessed in the 4.6% subsample at year 1 by FFQ and for all DM women at baseline, year 1, and on a rotating basis for one third of DM women each year during the intervention period as previously described (11). Assessment of dietary intake after the intervention period was based on a sub-sample of single 24-hour dietary recalls for 1,311 DM trial participants who re-consented for continued assessment between the years of 2005–2010 as part of their post-intervention period participation. Recalls were collected by telephone using trained staff and applying the U.S. Department of Agriculture (USDA) multi-pass methods (17). No post-randomization 4-day food records were collected.

The principal goal of the DM low-fat diet intervention was to reduce fat intake to 20% of total energy (13). Women were instructed and supported through behavioral modification strategies to increase vegetable, fruit, and whole grain intake daily. Total energy intake was neither restricted nor was weight loss advocated. The intervention was largely implemented using 18 group meetings in the first year and quarterly thereafter. Comparison group women received printed health-related materials only.

Clinical outcomes were collected through annual clinic visits during the trial and semi-annual mailed questionnaires during the trial and follow-up period. Outcomes for cancer were verified, initially by trained physician adjudicators at the local clinical centers by medical record and pathology review, followed by final central blinded adjudication (18). Vital status of all participants were cross-checked against The National Death Index at 2- to 3-year intervals.

At the protocol-specified termination date of March 31, 2005 for the intervention period, representing a mean 8.1 years of trial participation, vital status was known for 96% of participants with 4.9% deceased. Subsequently, DM participants were contacted by mail for consent to participate in an additional postintervention period for collection of clinical outcomes (including cancers). The characteristics of the sample of WHI DM women who reconsented for the extended follow-up (81.1%, n = 14,769 intervention versus 84.4%, n = 23,089, comparison; P < 0.001) are provided in Supplementary Table S1.

The two primary study outcomes were invasive breast or colorectal cancers. Other cancer outcomes were assigned on the basis of immunohistochemistry tests described in the pathology reports obtained from medical records and adjudicated by WHI-trained physician adjudicators. The intervention period sample size calculation was based on an anticipated 13% lower invasive breast cancer incidence in the dietary intervention group that would be achieved approximately linearly during the intervention period. An external Data Safety and Monitoring Board (DSMB) monitored the trial.

The analytic plan for the combined intervention and postintervention follow-up periods was similar to the approach taken in prior reports of outcomes during postintervention follow-up in the WHI HT trials (19). All participants were included in analyses by randomized group assignment until they last provided follow-up information. Baseline characteristics for women who provided additional consent were compared by randomization group using χ² and t tests.

Annualized rates of clinical events were estimated for the intervention period, the postintervention period, and the entire follow-up period by dividing the number of events by the corresponding person-time in each period. Cumulative incidence curves were generated for the entire active study participation period (low-fat diet intervention and comparison control) and separately during the postintervention period. HRs, 95% CI, and P values were computed from Cox proportional hazards models that were stratified by age, prior disease (if appropriate), and randomization status in the WHI hormone trials. Models were constructed for each clinical endpoint where women contributed follow-up time until the end of the interval, the date of their first relevant clinical event, or the date of death or withdrawal from the study, whichever came first. For breast cancer subgroups, analysis by tumor subtype was also explored. Formal tests of the differences between HR in the intervention vs. postintervention period were calculated by inclusion of a binary term for period (intervention vs. postintervention) as a time-dependent variable. All statistical tests were two-sided. Nominal P values are reported without adjustment for multiple outcomes or sequential looks during the clinical trial follow-up period. Permutation tests were performed post hoc to help interpret the invasive breast cancer results by determining whether the cumulative data were more consistent with (i) an overall null hypothesis or (ii) with an alternative hypothesis in which there was a real, but not quite significant, risk reduction during the intervention period, followed by no effect during the postintervention period. In both scenarios, 10,000 permutations were generated.

To investigate potential imbalances due to differing consent rates, entry baseline characteristics for women who reconsented and reconsent rates were compared by randomization group and by age, race/ethnicity, and other pertinent demographics (Supplementary Table S1). Furthermore, a reconsent model was constructed and the inverse of the estimated probabilities of reconsent was used as weights in Cox regression models.

Subgroup analyses based on baseline percentage of energy from fat were conducted also. Instead of FFQ data, baseline 4-day diet records data were analyzed using NDS-R due to the fact that fat intake by FFQ was applied for eligibility screening, an approach that resulted in a truncated baseline FFQ fat intake and consequently upwardly biased assessment. The 4.6% sample was used only to provide estimates of baseline intake quartiles for the DM trial cohort as a whole, for use in the case-only HR analyses. Statistical software SAS, version 9.3 (SAS Institute Inc.) and R, version 2.15 (R Foundation for Statistical Computing; http://www.r-project.org/) were used for these analyses.

The DM participant flow including screening, consent, randomization, eligibility and follow up during intervention and postintervention periods is detailed in Fig. 1. At entry, baseline characteristics were similar in the two randomization groups (7–9). The percentages of DM women, who consented to follow-up, by treatment arm and participant characteristics, are described in Supplementary Table S1; comparison group participants were somewhat more likely to consent to follow-up. The characteristics of the women who consented to, and provided data during, extended follow-up are described in Table 1. The majority of the women were overweight or obese at baseline with mean (SD) BMI of 28.9 (5.8) kg/m² and, from FFQs, mean daily dietary intake of percentage calories from total fat, saturated and trans- fat, polyunsaturated fat of 37.6%, 12.7%, and 7.7%, respectively. Mean daily fiber intake was 15.5 grams. Mean servings per day of fruit and vegetables were 1.6 and 2.0, respectively.

Dietary fat intake by DM group assignment during the intervention period has been previously described (7). Longitudinal plots of mean (95% CI) percentage of energy as fat during follow-up and subsequently the extension period, stratified by quartiles of percentage energy as fat at baseline, suggest that this diet exposure is somewhat preserved by randomization arm over time (Supplementary Fig. S1). During the postintervention period, the intervention group reported a somewhat smaller percentage energy from fat than the comparison group with mean (95% CI) differences of −2.8 (−4.9, −0.7), −4.6 (−6.6, −2.5), −3.5 (−5.4, −1.6), and −3.6 (−5.5, −1.7) for increasing quartiles of baseline fat, respectively, based on 4-day diet records at baseline and year 1 followed by 24-hour dietary recalls during the remainder of follow-up (n = 1,311 with postintervention data; 2.7% of total sample).

The influence of the low-fat intervention on cancer and mortality outcomes over the combined intervention and postintervention periods (12.3 [3.4] years) as well as for each period separately (intervention period mean [SD] of 8.3 [1.7] years; postintervention period 5.2 [0.8] years) are summarized in Fig. 2 and Supplementary Table S2. There was no indication that the low-fat diet assignment during the combined periods was associated with lower risk for total cancer or total mortality (Figs. 2 and 3C and D). For total mortality, the cumulative HR associated with the low-fat diet was 0.97 (95% CI: 0.91–1.02, 0.80% [n = 1,904] vs. 0.83% [n = 3,030]).

The annual incidence of invasive breast cancer was not significantly lower in the low-fat diet intervention women compared with the comparison women (annual percentage 0.43 vs. 0.44%; HR, 0.97; 95% CI, 0.89–1.05) over the combined intervention and follow-up period. During the intervention period, mammogram frequency was closely comparable between randomization groups and remained similar during postintervention follow-up period (93% and 92%, in low fat vs. comparison, respectively). Log-rank tests suggested a possible benefit of a low-fat diet in relation to invasive breast cancer risk during the intervention period (HR, 0.92; 95% CI, 0.84–1.01), which was not sustained in the postintervention period (HR, 1.08; 95% CI, 0.94–1.24; P_diff = 0.07; Fig. 2; Supplementary Table S2; Fig. 3A). A permutation test confirmed the validity of P_diff under an overall null hypothesis (P value of permutation test = 0.07) and suggested the alternative hypothesis of a reduced HR in the intervention period, followed by an HR of one in the postintervention period was somewhat more plausible (P value of permutation test = 0.17).

Colorectal cancer risk did not differ by group assignment overall (HR, 1.04; 95% CI, 0.90–1.20) either during the intervention period or during the postintervention period (Figs. 2 and 3B). For ovarian cancer, the point estimate remained <1.0 in the intervention period and postintervention periods, but the overall HR was not significant (HR, 0.87; 95% CI, 0.68–1.12; Fig. 2). Similarly, a nonsignificant HR was observed for all other cancer types combined (Fig. 3C).

Figure 4 provides the HR for invasive breast cancer for the intervention and postintervention periods by baseline percentage energy from fat by quartiles as assessed by 4-day food records. A change in HR from the intervention to the postintervention period was evident for women who reported baseline percentage of energy from fat within the highest quartile [(HR, 0.76; 95% CI, 0.62–0.92) during the intervention period versus (HR, 1.11; 95% CI, 0.84–1.46) during the postintervention period (P_diff = 0.03)].

Given observational evidence that the relationship between diet and breast cancer may vary by tumor subtypes (20, 21), although not consistently (22), the association between the low-fat diet and incidence of breast tumors by subtype also was evaluated. The results supported a possible association between diet assignment and breast cancer for the ER⁺/PR⁻ tumor subtype over the entire period (HR, 0.70; 95% CI, 0.56–0.88; Table 2), with the suggested protective associations during the intervention period (HR, 0.69; 95% CI, 0.53–0.89) slightly attenuated during the postintervention period (HR, 0.76; 95% CI, 0.48–1.20). No association was shown among the remaining ER/PR subtypes (P_het = 0.03). Similarly, reclassification of the ER⁺/PR⁻ tumor subtype as luminal B tumors showed a nonsignificant reduced point estimate during both periods. A test of heterogeneity during the entire follow-up suggested a differential effect of the dietary intervention for PR⁻ tumors overall (P_het = 0.03; HR, 0.83; 95% CI, 0.71–0.97). In analyses by baseline BMI, we did not find effect modification of DM randomization by BMI overall or BMI × HT for any of the tumor subtypes (data not shown).

In general, estimates of HRs obtained by accounting for censoring due to lack of consent for postintervention follow-up via inverse probability weighting were similar to those presented above. For example, the postintervention HR for invasive breast cancer changed from 1.08 to 1.07.

The WHI DM trial was the first long-term, randomized controlled low-fat dietary intervention study to test the hypothesis that adoption of a low-fat eating pattern after menopause would yield a reduction in invasive breast cancer and/or colorectal cancer incidence (11). Here, we present the longer-term cancer outcomes for the WHI DM trial participants. These analyses show that risk reduction for invasive breast was not evident during the additional 5.2 years of postintervention follow-up, despite some suggestion of a risk reduction during the intervention period for a subgroup of women who entered the trial with relatively high self-reported dietary fat intake.

At the time, the WHI DM trial was designed, efforts to protect the statistical power included setting an exclusionary criterion for women who reported dietary intake below 32% of total energy as fat with the expectation that the greatest benefit would occur among women with habitual diets comparatively higher in fat and for whom a greater overall reduction in fat intake would be required to achieve study dietary goals. At the end of the intervention period, women who reported the highest dietary fat intake at baseline (≥36.8% of energy from fat) and were assigned to the low-fat diet arm demonstrated reduced risk for invasive breast cancer that was nominally significant compared to women randomized to the comparison group (HR, 0.78; 95% CI, 0.64–0.95), but this was not evident during the postintervention period.

The nonsignificant postintervention breast cancer HR of 1.08 suggests no influence of assignment to the low-fat diet on breast cancer risk during the postintervention period. However, the cumulative data are evidently less consistent with an overall null hypothesis and more consistent with an alternative hypothesis in which there is a real (but not quite significant) reduction during the intervention period, followed by no effect during the postintervention period. Speculatively, this may suggest that any protective effect of a low-fat intervention does not carry forward after stopping the intervention, when dietary fat intake tended to increase. The data may be most consistent with the possibility that the intervention induced a delayed onset of clinically detectable breast cancer, with catch-up occurring during the postintervention period.

Overall dietary fat, although not reduced on average to the goal of 20% of total energy intake, was significantly lower for women in the intervention versus comparison diet group during the trial. Diet measures in a subgroup of DM women who continued to be observed during the extension period (n = 1,311) suggest that dietary fat intake remained somewhat lower in the women randomized to the low-fat diet group, although fat intake appeared to increase in the intervention group with time. Intervention women who reported fat intake being in the highest quartile of energy intake as fat at baseline may have experienced a risk reduction that reflected reduced energy exposure given the higher caloric value of dietary fat versus other macronutrients (23). Earlier analysis of calibrated energy intake in DM women supports an increased breast cancer risk in relation to energy intake (24).

In the subgroup of women wherein risk differed during the intervention period versus postintervention period, one might argue that any liberalization of dietary fat (and potentially energy) intake after the intervention period may have contributed to an increase in body weight and as a consequence, mediating factors such as circulating cytokines, insulin, or estrogen levels could have promoted breast cancer events. In fact, there was an overall reduction in body weight and body fat in women randomized to the low-fat diet versus comparison arms of WHI DM during the early (12 months) period of intervention (25, 26) partly supporting this hypothesis. But the mean difference in BMI between the two randomization groups was not significant after 7 years of follow-up. What is unknown is if the weight loss was associated with a lowering of inflammatory, metabolic, and/or estrogen levels such that modification of risk for additional cancer events could be realized in such a short follow-up period. Evidence from WHI did show a lower estradiol level in women randomized to the low-fat intervention versus comparison arms, but while significant, the difference between the randomization groups may not have been pronounced enough to lead to a differential risk for breast cancer. Importantly, adjustment for change in body weight during the trial period did not appreciably change the HRs for the remaining cancer outcomes (data not shown). The potential modulating effects of body weight, and related biologic exposures, could be evaluated in future analysis of WHI data.

In WHI, any evidence for breast cancer risk reduction with the low-fat diet intervention was restricted to ER⁺/PR⁻ tumor subtypes (HR, 0.69; 95% CI, 0.53–0.89 during intervention period; HR, 0.76; 95% CI, 0.48–1.20 during the postintervention period). Importantly, these tumors express higher proliferation markers and thus could be fueled by insulin and inflammation as well as estrogen, metabolic exposures also associated with obesity. Only one other intervention trial evaluated dietary fat in relation to breast cancer risk by tumor subtypes. In a trial of 4,690 conducted in Canada, women with high breast density, many of whom were premenopausal, were randomized to a diet of 15% total energy as fat. A total of 118 invasive breast cancer occurred over an average follow-up period of 10 years. No overall intervention effect was demonstrated. Elevated risk for ER⁺ breast cancer was described for women who reported lower carbohydrate intake at baseline and during active trial participation (27). Furthermore, results of observational studies have reported that dietary fat, and in particular animal fat, may be associated with greater risk across all common tumor subtypes (22).

Earlier analysis suggested a potential reduction in ovarian cancer risk related to the low-fat diet assignment, but only after 4 years of intervention (10). Results from this analysis are generally equivalent when evaluating ovarian cancers occurring after 4 years on intervention through the end of this extended observation period for events (HR, 0.79; 95% CI, 0.59–1.06). Although not significant, these results suggest that the role of low-fat diet in ovarian cancer risk warrants further study. A lack of significant effect of the low-fat eating plan on colorectal and total cancers and on cancer-specific and overall mortality was observed throughout the intervention and postintervention periods.

The current report providing data on the longer term, postintervention period of cancer outcomes in the WHI DM trial has limitations. First, while the assessment of clinical outcomes over time followed the same protocol used during the intervention period, information on dietary intake is limited to a small subset of women providing serial 24-hour dietary recalls and may not reflect dietary changes in the overall postintervention sample. Furthermore, it is unknown whether other lifestyle changes, such as changes in body weight, alcohol consumption, and physical activity, or even health screening behavior could have influenced clinical outcomes during the postintervention period as reassessment of these variables was not conducted during the postintervention follow-up period. Finally, measurement error in self-reported dietary intake, including dietary fat intake, a factor used to select women for trial participation, and energy intake and/or expenditure continue to be of concern. This is particularly important given the lack of a well-validated biomarker of dietary fat intake, with related uncertainty in the magnitude of dietary difference between randomization groups and with possibly reduced overall statistical power for the trial (28, 29). Furthermore, dietary differences between the intervention and comparison groups are uncertain, as there may be differential reporting by randomization assignment. In the case of energy consumption, for which a reliable biomarker (DLW) exists, there appeared to be greater underreporting in the intervention versus comparison group, a difference of about 100 kcal/d when evaluated near the end of the intervention period (30). Future studies are intended that will apply blood and urine metabolomics for objective assessment of this difference during the intervention period.

The WHI DM low-fat diet intervention did not result in a significantly lower risk for breast, colorectal, ovarian, or other cancers or cancer-specific or total mortality over the combined intervention and postintervention period. Whether other diet interventions, integrated within a comprehensive lifestyle approach that considers physical activity and weight control will show efficacy in primary prevention of cancer is yet to be determined. Our observational analysis from WHI supports this hypothesis (31).

No potential conflicts of interest were disclosed.

Conception and design: C.A. Thomson, A.K. Aragaki, J.E. Manson, L. Hou, K.C. Johnson, M.Z. Vitolins, R.L. Prentice

Development of methodology: C.A. Thomson, L. Van Horn, R.L. Prentice

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Van Horn, R.T. Chlebowski, J.E. Manson, L.F. Tinker, D.S. Lane, K.C. Johnson, M.Z. Vitolins, R.L. Prentice

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Van Horn, A.K. Aragaki, R.T. Chlebowski, L.H. Kuller, L. Hou, R.L. Prentice

Writing, review, and/or revision of the manuscript: C.A. Thomson, L. Van Horn, B.J. Caan, A.K. Aragaki, R.T. Chlebowski, J.E. Manson, T.E. Rohan, L.F. Tinker, L.H. Kuller, L. Hou, D.S. Lane, K.C. Johnson, M.Z. Vitolins, R.L. Prentice

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.E. Manson

Study supervision: L. Van Horn, D.S. Lane, K.C. Johnson, R.L. Prentice

Program Office: (National Heart, Lung, and Blood Institute, Bethesda, MD) Jacques Rossouw, Shari Ludlam, Dale Burwen, Joan McGowan, Leslie Ford, and Nancy Geller.

Clinical Coordinating Center: (Fred Hutchinson Cancer Research Center, Seattle, WA) Garnet Anderson, Ross Prentice, Andrea LaCroix, and Charles Kooperberg.

Investigators and Academic Centers: (Harvard School of Public Health, Boston, MA) JoAnn E. Manson; (MedStar Health Research Institute/Howard University, Washington, DC) Barbara V. Howard; (Stanford Prevention Research Center, Stanford, CA) Marcia L. Stefanick; (The Ohio State University, Columbus, OH) Rebecca Jackson; (University of Arizona, Tucson/Phoenix, AZ) Cynthia A. Thomson; (University at Buffalo, Buffalo, NY) Jean Wactawski-Wende; (University of Florida, Gainesville/Jacksonville, FL) Marian Limacher; (University of Iowa, Iowa City/Davenport, IA) Robert Wallace; (University of Pittsburgh, Pittsburgh, PA) Lewis H. Kuller; (Wake Forest University School of Medicine, Winston-Salem, NC) Sally Shumaker.

The authors also acknowledge Nicole Bergier (University of Arizona), who supported the formatting and citation reporting for this work.

The WHI program is funded by the National Heart, Lung, and Blood Institute, NIH, U.S. Department of Health and Human Services through contracts HHSN268201100046C, HHSN268201100001C, HHSN268201100002C, HHSN268201100003C, HHSN268201100004C, and HHSN271201100004C as well as Arizona Comprehensive Cancer Center Support grant funded by NIH-NCI grant CCSG-CA023074.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.