In an expanded analysis of a previously published study of energy intake and risk of postmenopausal breast cancer in the PLCO cohort (24
), we again observed a modest, but statistically significant, positive association between energy intake and risk of subsequent breast cancer. With four additional years of follow-up time and nearly 75% more breast cancer cases, we found a 20% increase in breast cancer risk comparing extreme quartiles of energy intake. This increase was minimally confounded by BMI and physical activity and strengthened noticeably with time elapsed since energy intake assessment. Results among women in the most recent four years of follow-up appeared stronger than those in the earlier years of follow-up, which indicates that our overall findings were not driven by the results we have already published.
Energy intake estimated from the second FFQ administered in the study, the DHQ, which was given, on average, 3.3 years after the DQx, provided another opportunity to expand upon the previous study. While energy intake was positively and significantly associated with increased breast cancer risk when using DQx-based energy intake, there was no evidence of an overall association when using DHQ-based energy intake.
The difference in results by FFQ could be explained by the different analytic populations, the different follow-up times, or the different dietary assessment instruments. While 1,742 fewer women were included in the DHQ analytic cohort than in the DQx analytic cohort, the two analytic cohorts were very similar, with more than 24,000 women completing both FFQs and, therefore, in both cohorts. Analyses among the 24,263 women who completed both FFQs demonstrated a positive association between DQx-based energy intake and breast cancer risk comparable to that in the DQx analytic cohort; therefore, the difference in analytic populations is not likely the reason for the difference in results by FFQ.
Secondly, the difference in results by FFQ could be explained by the different follow-up times. Each FFQ quantified energy intake, but the DHQ measured energy intake at a later point in time for each woman: the DHQ was administered, on average, 3.3 years after the DQx. Mean and median follow-up times were 3.1 and 3.3 years longer, respectively, in the DQx analytic cohort than in the DHQ analytic cohort. It is possible that several years need to elapse before the full influence of an increased or decreased energy intake on breast cancer incidence becomes apparent. Given more follow-up time, energy intake assessed by the DHQ may also show positive associations with risk, as a statistically significant 38% increase in risk, comparing extreme quartiles, did appear when using DQx-based energy intake among the women completing the DHQ.
Evidence for the importance of elapsed time also came from time lag analyses in the DQx and DHQ analytic cohorts. In four-year periods of follow-up created from the DQx analytic cohort, the relative risk of breast cancer increased from 1.21 to 1.37 to 1.55 with time elapsed since energy intake assessment. A similar, but weaker, relationship, with relative risk increasing from 1.00 to 1.23 with elapsed time since energy intake assessment, was detected in four-year periods of follow-up created from the DHQ analytic cohort.
Lastly, differences in the dietary assessment instruments could explain the differential results by FFQ. The DQx3
is similar to earlier FFQs in its grid format and inclusion of typical foods in the U.S. diet (26
). However, it also incorporated the results of extensive cognitive research with volunteers on how to improve dietary assessment methodology (27
). Thus, the DQx includes an empirically derived comprehensive food list, rational categories for frequency of intake and portion size, and optimized wording and formatting. The DHQ,4
finalized later in time, incorporated the continuing cognitive research (26
). It is substantially longer and more comparable to a dietary history with a series of questions after specific food items eliciting additional detail. It does not rely on a conventional grid format and includes somewhat different food items, frequency of intake categories, and portion size definitions than the DQx. The DHQ has been compared to the Block and Willett FFQs in a calibration study using four 24-hour recalls, one in each season, conducted by telephone (29
). The DHQ performed best overall; with a standard measurement error model, the correlation for energy between estimated truth and the DHQ was 0.48 for women. The DQx has never been compared with 24-hour dietary recalls or food records so it is not possible to use calibration studies to compare the two FFQs. However, in the PLCO cohort, energy intake estimated by the DQx was, as would be expected, positively, but weakly, associated with both BMI (r
= 0.08) and physical activity (r
= 0.04). Energy intake estimated by the DHQ was barely associated with these two determinants (r
= 0.03 for BMI and 0.01 for physical activity), which suggests that the DQx may be modestly better at estimating energy intake.
In the PLCO cohort, the frequency distributions for energy intake estimated by the two FFQs only partially overlapped, the correlation between the energy estimates was only 0.56, and absolute energy intake estimated by the DHQ was lower than energy intake estimated by the DQx for approximately 70% of the women. It is conceivable, therefore, that the FFQs differ in their ability to assess true energy intake and the DQx is more accurate though the two instruments have not been compared side-by-side. When we compared sources of calories, the percent of calories from carbohydrate, fat, protein, and alcohol were similar for the two FFQs. Possible explanations for the different performance of the two FFQs include differences in design and respondent burden.
Three other prospective studies have reported positive associations between energy intake and breast cancer risk (17
). In a large Canadian study of pre- and postmenopausal women (with 327,994 and 244,616 person-years of follow-up, respectively), multivariate HRs for highest to lowest quartiles of energy intake, in models including physical activity and BMI, were similar in magnitude to those found in our study and a significant trend was seen (19
). After stratification by menopausal status at baseline, the association with energy intake was relatively strong among premenopausal women (multivariate HR, 1.45; 95% CI, 1.13–1.85; Ptrend
= 0.001) but null among postmenopausal women (multivariate HR, 0.94; 95% CI, 0.72–1.23; Ptrend
= 0.86) with a marginally significant p-value for interaction by menopausal status (p = 0.06) (19
). Additionally, a small U.S. study of 590 postmenopausal women found increased breast cancer risk with each 500-kilocalorie increase in total energy intake (RR, 2.72; 95% CI, 1.51–4.89) (17
) and a Norwegian study of pre- and postmenopausal women found evidence of an energy intake-breast cancer association (RR, 1.50; 95% CI, 1.05–2.15) though the association lost statistical significance when BMI was added to the model (18
). Case-control studies in Argentina, China, Italy, and Switzerland have provided further evidence supporting a positive association between energy intake and risk of breast cancer (12
), as have international correlation studies of total calories and breast cancer incidence (r = 0.70) and breast cancer mortality (r = 0.60) (30
Surprisingly few epidemiologic studies have assessed the relationship between total energy consumption and breast cancer risk. The vast majority of studies of diet and breast cancer adjust nutrient estimates for total caloric intake by one of several accepted methods. Justifications include correcting for systematic underestimation and overestimation of consumption, reducing measurement error by controlling for variation in energy intake, controlling for confounding by total energy intake, and focusing on dietary composition rather than absolute nutrient intake (31
). Therefore, most studies, by adjusting for energy, have not been able to evaluate its independent role.
However, our findings do differ from those of three recent cohort studies and one recent case-control study (20
). Energy intake was not associated with risk among postmenopausal women in the Breast Cancer Detection Demonstration Project Follow-up Cohort Study (20
) nor among pre-and postmenopausal women in the Nurses’ Health Study cohort (21
); both studies presented results controlled for BMI but not physical activity. There was also no association among pre- and postmenopausal women in the California Teachers Study cohort (22
) nor among pre- and postmenopausal women in the case-control Shanghai Breast Cancer Study, which did not adjust for BMI or physical activity (23
The evidence from studies of populations exposed to severe energy restriction has been mixed (7
). A retrospective cohort study in Sweden found that in a group of women hospitalized for anorexia nervosa before age 40, calorie restriction was associated with a statistically significant 53% decreased risk of breast cancer, compared to the general female Swedish population (7
). In a Norwegian study, those women who had undergone puberty during World War II experienced lower breast cancer incidence than those who were older or younger during the same time period. As daily energy intake was 22% lower during the war, caloric restriction may have contributed to the lower incidence (8
). In contrast, a case-cohort study of women exposed to the briefer, but more severe famine conditions in the Netherlands at the end of World War II reported that caloric restriction early in life increased later breast cancer risk (10
). HRs, adjusted for adult BMI, were 1.13 (95% CI, 0.92–1.38) and 1.48 (95% CI, 1.09–2.01) for women moderately and severely exposed to the 1944–1945 Dutch famine, respectively. However, another study reported no association between place of residence in the Netherlands, which determined exposure to World War II famine, and risk of breast cancer (9
Strengths of our study include the prospective design, which minimized selection and participation bias. The large number of women in the cohort provided stable estimates of risk. Cancer ascertainment from the annual study update questionnaires was excellent, with 96% of the questionnaires returned, on average, and 96% of self-reports of breast cancer subsequently confirmed by hospital reports. In addition, prospective ascertainment of diet and breast cancer risk factors prevented recall bias. Diet was assessed with detailed, comprehensive, cognitively-designed FFQs. The FFQs were carefully completed with nearly all missing three food items or fewer. Weaknesses of our study include the difficulties inherent in measuring diet using FFQs, with energy intake being particularly challenging to assess, which leads to measurement error and, frequently, attenuation of RR estimates (32
). In addition, diet was assessed at only one point in time though each FFQ did ask the respondent to recall and integrate diet over the past year. At the time of our analyses, hormone receptor status information was being collected and available for only a portion of the confirmed breast cancer cases in our study population, which precluded additional analyses by breast cancer subtype.
In conclusion, findings from this extended analysis continue to suggest a modest positive association between total energy intake and risk of postmenopausal breast cancer, an association which was largely independent of BMI and physical activity. In this large cohort, breast cancer risk was increased 21% (95% CI, 3–42%) among women in the highest quartile of energy intake compared to those in the lowest. When follow-up time was divided into four-year periods, the RRs for high versus low energy intake increased from 1.21 to 1.37 to 1.55, with increasing time since dietary assessment. Our findings suggest a modest positive association between energy intake and postmenopausal breast cancer that strengthens with time since dietary assessment.