Case Control Studies
In a typical case control study of diet and breast cancer, the diet before diagnosis reported by women with breast cancer (cases) is compared with the diet reported by women who have not been diagnosed with breast cancer (controls). The largest such study is that of Graham et al., who used a food frequency questionnaire to compare the fat intake of 2,024 women with breast cancer to that of 1,463 female controls seen at the hospital with benign conditions. Animal fat and total fat intake were almost identical in the two groups. In a meta-analysis, Howe et al. summarized the results from 12 smaller case control studies comprising a total of 4,312 cases and 5,978 controls.
The overall pooled relative risk for a 100-g increase in daily total fat intake was 1.35; the risk was somewhat stronger for postmenopausal women (relative risk, 1.48). Because average total fat consumption is approximately 70 g per day for U.S. women, a reduction in fat intake as large as 100 g would be impossible for almost all women. The results of this pooled analysis suggest that, even if the reported positive association were correct, the relative risk for readily achievable changes in total fat intake would be relatively small; for example, the reduction in risk for a 20 g per day decrease among postmenopausal women (correspondingto a decrease from 40% to 29% of calories for a typical middle-aged woman), would be only approximately 0.9.
Furthermore, in case control studies, relative risks of this magnitude may easily be due to selection bias (the controls are drawn from a population with a different distribution of fat intake than the distribution in the population that gave rise to the cases) or recall bias (the cases, knowing their diagnosis, differentially misreport their prediagnosis diet).
In a cohort (prospective) study, the diets of a large group of women are measured, and the subsequent rates of breast cancer among those with different levels of dietary factors are compared. Selection bias should not be a problem, because the population that gave rise to the cases is known (the starting members of the cohort); recall bias should not occur, because dietary information is collected before knowledge of disease.
The results for postmenopausal breast cancer (for which fat intake has been hypothesized to be strongest, because the international differences are largest for this group) from prospective studies with at least 200 incident cases of breast cancer. The number of breast cancer cases in these studies is similar to the number in the pooled analysis of case control studies referred to above, and the size of the comparison series (i.e., noncases) is much larger. In not a single study was a significant association with total fat intake observed (when the highest and lowest categories of total fat intake were compared). A collaborative pooled analysis of 4,980 cases of breast cancer (among 337,819 women) has been conducted on all the prospective studies included in Table 15-3.
In addition to providing great statistical precision, the pooled analysis allowed the application of standard analytic approaches to all studies, an examination of a wider range of fat intake, and a detailed evaluation of interactions with other breast cancer risk factors. Overall, no association was observed between intake of total, saturated, monounsaturated, or polyunsaturated fat and risk of breast cancer. As noted in above, no reduction in risk was seen even for fat intakes as low as 20% of energy. When the relatively few women with fat intake lower than 15% of energy were examined, their risk of breast cancer was actually increased twofold; this could not be accounted for by other dietary or nondietary factors.
Substudies were available for each cohort in the pooled analysis in which the measurement errors of the dietary questionnaires were quantified, and these were used to adjust the overall relative risks and confidence intervals to take intoaccount errors in measuring diet. Without correction, the relative risk for an increment of 25 g per day in fat intake was 1.02 (95% confidence interval, 0.94 to 1.11). After accounting for measurement error, the relative risk was 1.07 (95% confidence interval, 0.86 to 1.34).
The upper bound of the adjusted 95% confidence interval excludes the relative risk of 1.4 to 1.5 predicted by the international correlations. In calculations based on a series of theoretical assumptions, Prentice has claimed that the pooled analysis of breast cancer failed to find a positive association because the measurement error correction did not account for underreporting of fat intake by more obese women. Actual studies do not support this assumption, however, and the other predictions based on this theoretical model are not supported by the data. Thus, these theoretical concerns appear groundless. In the Nurses’ Health Study, additional analyses have been conducted with 8 years (1,439 cases of breast cancer) and 14 years (2,956 cases) of follow-up. In the 14-year analysis, up to four assessments of fat intake were available, which substantially improved the measurement of long-term dietary intake.
The relative risk for a 5% increase in percentage of energy from total fat was 0.97 (95% confidence interval, 0.94 to 1.00); the overall weak inverse trend was actually statistically significant. No suggestion of any reduction in risk was seen for fat intakes even lower than 20% of energy. Thus, the prospective studies provide strong evidence that no major relation exists between total dietary fat intake over a wide range during midlife and breast cancer incidence. Although it remains possible that dietary fat intake during childhood or early adult life may affect breast cancer risk decades later, no evidence exists for this.