Low-Energy Reporting in Women at Risk for Breast Cancer Recurrence¹

Estimates of EI⁴and the types of foods ingested are critical to studies relating nutrition to health outcomes. These estimates rely on self-reports of food intakes over varying periods of time. The most widely used methods are diet recalls, where subjects report on foods consumed over the previous 24 h; diet diaries, in which subjects record all foods consumed on a daily basis; and food frequency questionnaires, which require subjects to report food consumption patterns over longer periods of time, from 1 month to 1 year or more. Because all these methods have their limitations, there is no “gold standard” for assessing dietary intake, which makes it difficult to assess measurement error and its impact on study results.

Recently, researchers have begun to compare estimates of self-reported EI with estimates of total energy expenditure to provide insight into the validity of self-reported EIs. Using biological markers such as doubly labeled water, or other methods to estimate total energy expenditure, these studies have found that methods of self-reported dietary assessment tend to underestimate EI (1, 2, 3, 4, 5, 6). This phenomenon, termed “underreporting” or “low-energy reporting,”may result from difficulties in accurately reporting food composition and portion size; changing eating patterns to simplify reporting; not reporting on “unusual” days of large consumption (i.e.,weekends, parties); erroneous package labeling on locally produced foods; changing eating patterns or reported consumption to be more socially desirable; or not reporting complicated foods (mixed dishes)or small items (bites and tastes; Refs. 7 and 8).

Low-energy reporting is a concern in studies of diet-disease relationships. Nonsystematic low-energy reporting could bias results toward the null, whereas systematic low-energy reporting could bias results if participant characteristics are related both to the low-energy reporting and to the disease end point of interest. Prentice (9) has argued that the lack of relationship between dietary fat and breast cancer may be a result of nonsystematic underreporting of fat and EI.

Using techniques such as doubly labeled water to estimate potential dietary measurement biases is impractical in large-scale studies. As a surrogate, a number of large studies have estimated a measure of expected EI using estimates of a participant’s energy expenditure and basal metabolic rate (6, 10). These studies have demonstrated considerable variability across participants in the relationship between reported and expected EI. Low-energy reporting was frequently observed and was more likely to occur among women (4, 11, 12, 13), among those categorized as overweight (11, 14, 15, 16, 17, 18, 19), among African Americans compared with Caucasians (20), and among younger rather than older adults (19, 20). Other demographic differences in underreporting have also been observed (12, 16, 21). The differences observed in these comparisons include not only the discrepancy between self-reported intake and expected EI, but also differences in diet composition (22), the number of foods reported (23, 24), portion sizes (23), and intake of specific food groups (4).

The dietary assessment method chosen might influence underreporting: an analysis of diet records in one study revealed that the reported number of both foods and nutrients was considerably lower on the 4th day of record-keeping than on the 1st day (25), suggesting that participant burden contributed to underreporting. Several studies have suggested that underreporting could be a concern when food intake is assessed by 24-h recall (16, 26, 27), although Buzzard et al. (27) noted that skilled interviewers and probing techniques could reduce the amount of underreporting considerably.

To our knowledge, no studies have specifically examined underreporting in breast cancer survivors by comparing EI with energy expenditure,although some studies have estimated underreporting by comparing different dietary assessment methods. In one study in women with localized breast cancer (24), investigators found that EI,number of food items, add-on foods, and supplements were underreported more frequently in 24-h recalls than food records. Conversely, another study in breast cancer survivors suggested that underreporting was a greater problem with food records than with 24-h recalls (27). This study examined 290 postmenopausal women with localized breast cancer participating in a dietary intervention study. The authors compared unannounced 24-h recalls conducted by telephone to 4-day food records over the 1st year of the study. Compared with the 24-h recalls, the 4-day food records overestimated the extent of fat reduction in the low-fat diet intervention group by 41% at 6 months and by 25% at 12 months.

In this study, we examined the prevalence of low-energy reporting among a group of breast cancer survivors, participants of a large randomized controlled trial investigating the effect of diet on breast cancer recurrence. Low-energy reporting was a concern because baseline dietary assessments suggested that this group had a lower mean percent energy from fat (29%) than is reported in women in this age group in the general population (28). We report differences in the proportion of “LERs” across demographic and other health habits categories. Furthermore, we investigate whether this low-energy reporting is associated with lower reporting of a variety of specific nutrients and food groups.

This study used data collected for the WHEL Study, a multisite clinical trial testing whether a dietary pattern high in vegetables, fruits, and fiber and low in fat will affect the course of breast cancer. Women 18–70 years of age at the time of diagnosis, who presented with early breast cancer (stages I, II, or IIIA) within the previous 4 years were recruited from seven clinical sites in the western United States (four in California, one in Arizona, one in Texas, and one in Oregon). WHEL participants had completed any prescribed chemotherapy before enrollment; therefore, treatment-induced dietary differences (29) were eliminated as a confounder. Detailed eligibility criteria are reported elsewhere (30), and results from a feasibility study for this trial have been reported previously (31, 32). The WHEL Study will recruit 3000 women and randomly assign them to a dietary intervention group or to a control group. Women enrolled in the WHEL Study between August 19, 1995, and April 1, 1998 (n = 1137), are the focus of this report. The WHEL Study protocol was approved by the Human Subjects Committee of the University of California, San Diego, School of Medicine, and by the Institutional Review Board at each clinical site.

Information on smoking status and exercise level was obtained from the Personal Habits Questionnaire, developed for the Women’s Health Initiative clinical trial and observational study (30, 33). Adult weight history was also assessed with this instrument. Demographic data were collected by a telephone screening interview and study forms. All questionnaires were completed either before or at a baseline clinic visit. At this clinic visit, staff trained in the WHEL Study protocol weighed and measured women using standard procedures (34) and calculated BMI [weight(kg)/height (m²)]. The final stage of the baseline clinic visit was random assignment to the intervention or control group. Furthermore, 729 of the women had completed a 1-year clinic visit at the time of manuscript preparation, from which we could track weight status subsequent to enrollment. Stable weight was defined as a 1-year weight that was within 5% of the baseline weight.

A team of trained dietary assessors at the WHEL Study Coordinating Center collected dietary assessments by telephone under the direction of a dietary assessment supervisor, with consultation and oversight by the director of nutrition services (V. N.). Dietary intake was measured at baseline before randomization using four 24-h recalls randomly selected to include recall of 2 weekdays (Monday and Thursday) and 2 weekend days (Friday and Sunday) over a 3-week period. The Minnesota Nutrition Data System software (University of Minnesota, Minneapolis, MN) was used to collect dietary data, and the University of Minnesota Nutrition database (version 2.92, 1997;University of Minnesota) was used for nutrient analysis.

The study used several strategies to obtain accurate recalls of food intake. Before enrollment, a registered dietitian trained study participants to estimate serving sizes with food models and distributed measuring cups and spoons, along with two-dimensional food models for reference during the recalls. In addition, the Nutrition Data System software uses an interactive multiple-pass method (35)that improves dietary recall accuracy by providing several opportunities during the interview to review the participant’s daily diet at varying levels of detail. Computer-generated prompts ensure that all assessors obtain detailed data about the type and amount of food, as well as preparation methods. As an additional quality control measure, at the end of every recall, the assessor verified the dietary analysis for nutrient outliers and corrected any errors immediately. Also, the dietary assessment supervisor verified all completed sets of recalls and made corrections as needed. An estimate of inter-coder variability was obtained quarterly by verifying the mean (SEM) EI and intake of selected key nutrients by assessor for recalls completed during that period. No significant differences between assessors were noted.

All assessors successfully completed a 3-week training program emphasizing standardized data collection, proper interviewing technique, and efficient use of the dietary analysis software. This initial training was followed by a week of scheduled recalls supervised by an experienced assessor who was available to offer assistance and to provide guidance as needed. During the following month, two full shifts of recalls (at least eight) were taped and the dietary assessment supervisor randomly selected at least four of these recalls for review. Experienced assessors had their recalls taped for review quarterly. The dietary assessment supervisor reviewed any discrepancies with each assessor and recommended improvements as necessary. When problems with accuracy and completeness of dietary data were noted, tape reviews were scheduled weekly until the problem was resolved. All assessors were female, ranging in age from 20–53 years.

To classify LERs, we used the methodology described by Goldberg et al. (36), which derives cutoff limits for plausible EIs depending on the sample size and number of days of dietary data. Using this methodology, the EI:BMR ratio was calculated where BMR was estimated by the Schofield equation (37),using height and weight values measured at the baseline clinic visit. We then compared these ratios to cutoffs for a single individual across 4 days of dietary data (1.06 for the 95% CI and 0.88 for the 99.7%CI) to determine who was underreporting. These cutoffs determined whether each individual’s EI could be a valid estimate for a 4-day period “allowing for the known day to day and week to week variability, and without having to postulate any systematic reduction in intake which may have been caused by the measurement procedure” (36). The cutoff, therefore, accounts for other reasons respondents may have given for eating less on any day of report, such as traveling, celebrating a special occasion, or being bored, stressed,or not hungry (i.e., random, rather than systematic underreporting).

Means and SDs were calculated for EIs and EI:BMR ratios. For EI and EI:BMR ratio, differences between groups of demographic and behavioral risk factors were quantified using a linear regression model for each group. Using the Goldberg cutoffs, we classified anyone with an EI:BMR ratio below 1.06 as a LER and anyone with a ratio below 0.88 as a VLER. The VLER group is, thus, a subset of the LER group. Significant differences in low-energy reporting between groups of behavioral and demographic risk factors were tested using χ²contingency table analysis. Logistic regression was used to examine predictors of low-energy reporting, controlling for potential covariates. t tests were used to test for differences in grams or servings of specific nutrients or food groups between LER and non-LER. For each nutrient and food group, we computed a percentage difference for LER compared with non-LER.

Table 1 presents reported EIs, estimated EI:BMR ratios, and the percentage of LERs and VLERs at baseline. Mean EI was 1699 kcal/day. As expected, EIs decreased with age, and these differences were statistically significant. Significant differences were also observed across categories of BMI, with intakes generally increasing as BMI increased. As levels of alcohol increased, EIs also increased, but the differences disappeared when we subtracted alcohol calories from total energy. EIs decreased with increasing exercise. There were marginal associations of EI with supplement use (P = .10) and with weight change at 1-year follow-up (P = .07). No differences in EI were seen across ethnicity, adult weight history, or smoking status.

The mean EI:BMR ratio was 1.28. The mean EI:BMR ratio by subgroup ranged from a low of 1.07 in women with BMI ≥40 and 1.13 in African American women to 1.58 in the oldest age group (70–74 years) and 1.57 in women with the lowest BMI (<18.5). Significant differences in the EI:BMR ratio were observed for age, ethnicity, BMI, alcohol intake, and adult weight history. Smoking status, exercise level, supplement use,nonalcohol calories, and weight change at 1-year follow-up were not associated with the EI:BMR ratio.

LERs accounted for 25.6% of the total sample, and VLERs accounted for 10.8% of the total sample. The percentage of persons classified as LER reached substantial levels within certain subgroups, approaching 50%in women with a BMI ≥40. The percentage of VLER observed was much lower than the percentage of LER across all subgroups. The degree of LER varied significantly by age, ethnicity, BMI, alcohol intake, and adult weight history. The degree of VLER varied significantly by age,BMI, exercise level, supplement use, alcohol intake, adult weight history, and weight change at 1 year. In general, low-energy reporting was highest among African American women and among those with BMI >35.

Table 2 shows ORs of being classified as either LER or VLER, controlling for covariates. These results suggest that a woman’s age was associated with the risk of being both LER and VLER. Women, ages 35–59 years, had an OR of 6.84 for LER and 11.99 for VLER, whereas women <35 years of age had an OR of 19.44 for VLER. Alcohol intake was negatively associated with both LER and VLER. BMI was related to LER but not VLER;women with a BMI >30 had an OR of 1.95 for LER compared with those with BMI <25. Women who gained weight or whose weight fluctuated had a higher risk of LER (OR of 1.55) compared with women who lost weight or were weight stable. Women who exercised moderately were at increased risk of being a LER, and women who used five or more supplement formulations a day were at decreased risk of being a VLER.

Table 3 presents mean values for specific nutrients and food groups for LER and non-LER. Also presented in Table 3 are the percentage differences between LER and non-LER for each nutrient and food group. For all nutrients, except β-carotene, mean values were significantly higher for non-LER than LER. With reference to the percentage difference for EI, the percentage differences for fat, alcohol, and sucrose were higher, whereas the percentage differences for protein, fiber, vitamin C, and β-carotene were lower, with β-carotene showing the smallest percentage difference for nutrients between groups (−10.2). Mean values for all food groups were significantly higher for non-LER than LER, with percentage differences ranging from a low of −10.0 for vegetable servings to −23.5 for legume servings.

We report an average EI:BMR ratio of 1.28, indicating that a large proportion of women in this population are reporting lower than expected EIs. The Food and Agriculture Organization/World Health Organization/United Nations University guidelines (38) suggest that the average daily EI:BMR ratio for women engaged in light work is 1.56, and 1.35 is recommended as the lowest habitual value for the EI:BMR ratio compatible with a normal lifestyle (26).

Although this study estimated a low EI:BMR ratio, it is consistent with other published data. Black et al. (10), in their review of 37 published dietary studies on adults, found the average estimated EI:BMR ratio for women to be 1.37, and many of the studies reported an estimated EI:BMR ratio equal to or less than our finding of 1.28 (36). They also examined the EI:BMR ratio by dietary method and found that the average estimated EI:BMR ratio from 17 studies, all using 24-h dietary recalls similar to the WHEL Study methodology, was 1.31.

Furthermore, the average EIs that we found in this study are consistent with those reported in large nationwide surveys. EIs reported by women from 24-h dietary recalls in the Third National Health and Nutrition Examination Survey, Phase I (1988–1991; Ref. 28) were within 4% of those reported in our study across all age groups.

This study used telephone interviewing to conduct the 24-h dietary recalls, which could influence comparisons with studies using face-to-face dietary assessment methods. However, a recent large-scale study using multiple-pass 24-h dietary recalls in 700 women, ages 20–49 (39), concluded that telephone interviewing was a valid alternative to face-to-face interviews for collecting 24-h dietary recall data.

One limitation to our methodology is that our measure of energy expenditure is not based on more precise methods, such as doubly labeled water measurements or indirect calorimetry. However,measurement error does occur with the doubly labeled water approach as well and is compounded by estimates of various physiological factors and calculation assumptions. The major component of total energy expenditure is BMR, which is most strongly determined by the fat-free mass (40). This study used body weight as a surrogate for fat-free mass to estimate BMR, which could lead to significant error.

One possible explanation for the apparent high level of low-energy reporting observed could be that the Schofield equation overestimates the BMR for this group of cancer survivors within 1–4 years of diagnosis. Demark-Wahnefried et al. (41)studied energy balance in breast cancer patients undergoing chemotherapy and reported that BMR decreased during treatment but returned to pretreatment levels at the end of treatment. However, lean body mass, which is the major determinant of BMR, also decreased during treatment but did not return to pretreatment levels in a small subgroup of women measured 1 year later. They hypothesized that decreases in lean body mass may contribute to the weight gain observed in women diagnosed with breast cancer. A lower lean body mass among women in our study may have contributed to having an estimated BMR that is higher than the true value.

It is important to note that the Goldberg methodology (36)used in this study provides a conservative estimate of low-energy reporting, because it does not identify those with high-energy needs(i.e., the very active) who might be low-energy reporting as well. It identifies only those persons who are at the extreme end of the distribution and report intakes that are not feasible to sustain a sedentary lifestyle.

The Goldberg methodology (36) also assumes that energy needs are stable and that EI reflects current needs. To examine whether weight change influenced our results, we divided a subgroup of women for whom we had weight measurements at 1 year after baseline into three groups: weight stable (within 5% of body weight), weight gainers(>5% of body weight), and weight losers (<5% body weight). We assumed that weight losers would be more likely to be labeled as LERs,because EI during weight loss would be less than their baseline energy needs. On the other hand, we assumed that weight gainers would be less likely to be labeled LERs, because energy needs to promote weight gain would be higher than baseline weight energy needs. However, our data showed no differences in low-energy reporting rates among these three groups of women, suggesting that the weight changes were not occurring during baseline data collection. Furthermore, 72% of the women in this subsample were weight stable; therefore, the majority satisfied the condition of intakes in balance with energy expenditure.

Our analysis showed that women who consumed >6 grams of alcohol a day were less likely to report low-EIs than those who consumed no alcohol(Table 1). This finding is intriguing and has been reported in one other study (21). One interpretation is that persons who report alcohol intake are more likely to also report other less socially desirable dietary constituents. However, when we examined the data stratified by alcohol intake for nonalcohol energy only, the differences seen with total energy disappeared, demonstrating that alcohol drinkers and nonalcohol drinkers reported similar nonalcohol EI. As expected, because nonalcohol EI did not differ significantly between categories of alcohol intake, neither did the rate of low-energy reporting differ between categories of alcohol intake. It seems that alcohol energy reported by heavier drinkers increased their reported EI above those who do not drink, and the discrepancy between their energy needs based on the Schofield equation and their total reported EI is less than for nondrinkers.

Table 3 shows that compared with energy the nutrients and food groups that are less socially desirable such as fat, sucrose, and alcohol are underreported to a greater extent than the socially desirable nutrients such as protein and β-carotene. Other investigators have reported similar findings (18, 23). Hietmann and Lissner (18)and Krebs-Smith et al. (23)have shown that participants who underreport seem to differentially underreport both sweet and savory snack foods.

If the desire to conform to social norms is related to underreporting,especially with regard to nutrients believed to affect breast health,then our power to detect associations between those nutrients and either breast cancer incidence or prognosis is severely diminished due to exposure misclassification. Results from epidemiological studies examining the association between dietary factors and risk for primary or recurring breast cancer must be interpreted with this awareness.

In summary, the results from this study support conclusions from the majority of other studies investigating underreporting, showing substantial underreporting of EIs among women and demonstrating that body weight is one of the major determinants of reporting low EIs. Having a previous diagnosis of breast cancer does not differentiate these women from the general population of women, either with regard to the prevalence of low-energy reporting or predictors of low-energy reporting. We also found that age, alcohol intake, and supplement use were related to reporting low EIs. This is the first study that has reported on a relationship between low-energy reporting and supplement use. Future studies examining dietary determinants of cancer recurrence must account for these apparent biases in energy reporting.