Does an Apple a Day Keep Breast Cancer Away?

Can an apple a day keep breast cancer away? The latest answer to this question, based on an analysis of data reported by Smith-Warner and colleagues¹ in this issue of THE JOURNAL, is "no." However, anyone closely monitoring this question can find conflicting conclusions in the literature. Examining data from numerous epidemiological studies, the World Cancer Research Fund² concluded that the evidence that fruits and vegetables may reduce the risk of oral, pharyngeal, esophageal, lung, stomach, and colon cancer was convincing and that the evidence for an association between intake of fruits and vegetables with breast, pancreatic, and bladder cancer was probable. Since release of the World Cancer Research Fund report, several studies,^{3 - 5} including the one by Smith-Warner and colleagues,¹ have challenged the findings that were the basis for the 1997 report.²

In contrast to a meta-analytic approach in which risk estimates are derived from published data, the data used by Smith-Warner and colleagues in the Pooling Project of Prospective Studies of Diet and Cancer were obtained directly from participating cohort studies. Although these data were not collected in a standardized manner, the analyses performed in the Pooling Project were standardized. In those analyses, Smith-Warner and colleagues observed that for fruit intake, the risk estimates for breast cancer ranged from a 3% (95% confidence interval [CI], 0.91-1.04) reduction to a 1% (95% CI, 0.94-1.08) increase in risk per 100-g/d intake (slightly <1 apple); for vegetables, the risk estimates ranged from a 10% reduction (95% CI, 0.81-1.00) to a 10% increase in risk per 100-g/d intake (slightly >½ cup of broccoli).

To evaluate the significance of this and other studies of diet and cancer risk, it is important to remember the premises underlying epidemiological research. From a theoretical perspective, a hierarchy of study designs has been established that allows confidence to be placed in observations of disease associations. Clinical trials are at the top of the pyramid because exposure can be randomly assigned and change in outcome can be evaluated. Cohort studies, such as those included in the report by Smith-Warner and colleagues, assess exposure prior to disease and are therefore given a higher place on the research pyramid than are case-control studies, which by design must evaluate prior exposure after the outcome is known and case and control status have been assigned.

In a recently reported clinical trial, Schatzkin et al³ found that a low-fat diet, with high intake of fruit, vegetables, and fiber, did not influence the risk of recurrence of colorectal polyps. A positive outcome would have suggested that a higher intake of fruits and vegetables is protective against polyp initiation; a negative outcome reveals little about the role of fruits and vegetables as they relate to cancer. Knowledge of the underlying biological mechanisms of the association between diet and cancer is needed to design informative clinical trials. This information is critical to avoid the testing of interventions at inappropriate time(s) in the disease process. In addition, researchers must remain alert to the possibility that associations with intake of fruits and vegetables can be modified by other factors, some of which may change as a result of enrollment in a clinical trial. Without adequate knowledge of disease mechanisms, the likelihood of detecting associations in a clinical trial is diminished. On a theoretical basis, the clinical trial is the criterion standard, but on a practical basis this standard can be tarnished.

To evaluate the study by Smith-Warner and colleagues, it is important to understand strengths and limitations inherent in cohort studies. One advantage of cohort studies is that they allow assessment of factors that may change as a result of disease, such as hormone levels. However, a limitation is that the data are only as good as the component parts. Misclassification of exposure, in this case consumption of fruits and vegetables, can result if collection of data about fruit and vegetable intake is incomplete. The number of food items reported as being eaten is directly related to the number of questions included on the questionnaire.^{1 ,6 - 7} In the Pooling Project study, the number of questions about fruit and vegetable intake ranged from 9 items in the Sweden Mammography Cohort to 54 items collected as part of the Nurses' Health Study (b). These items included potatoes, legumes, and juices that were not included in the main analyses. Since several studies included in the Pooling Project analyses did not collect information on intake of apples, bananas, oranges, peaches, apricots, broccoli, brussels sprouts, cabbage, carrots, lettuce, spinach, string beans, or tomatoes (some of the most commonly consumed fruits and vegetables in the United States), interpretation of the reported results for total fruit and vegetable consumption is difficult.

This absence of information for certain fruits and vegetables may be especially problematic if only certain types of vegetables or fruits confer protection. Smith-Warner and colleagues evaluated specific foods and observed a nonsignificant 40% reduction in risk (95% CI, 0.33-1.15) per ½ cup serving of spinach; a 33% reduction in risk (95% CI, 0.35-1.27) per 1½ cup serving of brussels sprouts; a 14% reduction in risk per ½ cup of broccoli (95% CI, 0.72-1.02); and a 15% reduction in risk (95% CI, 0.66-1.09) per 1 cup serving of string beans per day. It is not unreasonable to theorize that certain vegetables may be associated with reduced risk of breast cancer. Since several of the study cohorts did not report intakes of common vegetables, stronger associations may have been observed if the diversity of vegetable consumption had been measured consistently. Furthermore, when study participants failed to supply information on specific foods, it was assumed that they were nonconsumed items. This treatment of missing information could influence fruit and vegetable associations as well as energy intake, one of the adjustment variables used in analyses.

Cohort studies, such as the ones included in the Pooling Project, involve select populations that may have underlying characteristics that differ from the population at large. Individuals who enroll in cohort studies are often in better health than those who do not. This self-selection of study participants can influence the study results through homogeneous exposures, thereby precluding the opportunity to detect a threshold effect. In the Pooling Project, the Netherlands cohort (the only cohort in which a significant protective effect from vegetables was observed) was derived from the general population. Underlying population characteristics could further influence the ability to detect associations if these characteristics influence a disease risk factor. For example, if vegetable intake has an important protective effect in sedentary populations, but the cohort consisted of active people, a meaningful association with vegetable intake would not be detected.

Much of the knowledge about diet and disease associations has come from case-control studies. These studies carry their own set of potential biases, including selection bias with enrolled cases and controls thought to be healthier than nonparticipants and recall bias in that information on disease-free diet is obtained postdiagnosis. Attempts to evaluate the impact of these potential biases vary.^{8 - 10} Since both cases and controls tend to be healthier, one would expect a nondifferential bias toward the null. One advantage of case-control studies is that the interviewer-administered questionnaires, which frequently are used in case-control studies, elicit more detail than self-administered questionnaires used in cohort studies, and have been shown to be more accurate in assessing exposure.⁷ Population-based case-control studies, by virtue of their wide sampling of the population, tend to ensure a wide range of exposures to the factors of interest. A further advantage of case-control studies is that large samples can be assembled in relatively short periods and uniform methods of data collection as well as analyses can be applied.

A difficulty shared by all studies assessing diet and disease is that diet is a moving target. A clinical trial usually follows the study participants for a relatively short time; a cohort study can evaluate exposures at different times; and case-control studies have a predetermined point of reference. In the Pooling Project, dietary intake was evaluated at baseline, usually in the early to mid-1980s. The period between the assessments of dietary intake and disease ascertainment varied from just a few to more than 20 years. If women made dietary changes, perhaps in response to mass media messages urging an increase in fruit and vegetable consumption, diet reported at enrollment might not accurately reflect usual diet during the relevant time of exposure. It is also possible that time to exposure is an important factor since investigators for 2 of these cohorts reported in the early 1990s that high levels of vegetable intake significantly reduced the risk of breast cancer.^{11 - 12}

Ultimately, all available sources of information must be used to understand the complexities of diet, including fruit and vegetable intake, and the associations with cancer. In the case of fruit and vegetable intake and breast cancer risk, the "dose" or amount of vegetables and fruits and the potential to confer protection is unknown. The analyses by Smith-Warner and colleagues suggest that the association is not linear, although a threshold effect cannot be ruled out. As the authors acknowledge, the ability to detect an association was reduced "if breast cancer risk was lower only above a threshold of intake, and if only a subset of the studies had a substantial number of women consuming above this threshold." The difficulty of detecting a threshold effect is highlighted by the inconsistent classification of intake among the cohorts. For instance, women in the Sweden Mammography Cohort were ranked in the upper quartile if they consumed 331 g/d of fruits and vegetables, while women in the Nurses Health Study (b) were ranked in the lowest quartile when their intake was less than 328 g/d of fruits and vegetables. Ability to detect a threshold effect could be hampered by inconsistent classification of intake between cohorts. Protective associations have been observed in some diet and cancer studies only at high levels of consumption or when the nutrients were in the form of supplements,^{13 - 14} adding support to the likelihood of a threshold effect.

Given all this uncertainty, what recommendations should be made about fruit and vegetable intake and cancer risk reduction? The National Cancer Institute initiated its "5-A-Day for Better Health" program in 1991 to increase fruit and vegetable consumption.¹⁵ This recommendation was based on numerous studies supporting direct associations between higher fruit and vegetable intake and lower risk of cancer. However, results of recent studies appear to be challenging the underpinnings of the 5-a-day program. Since there is potential for bias in all studies, no single report can resolve the question, whereas all well-conducted studies should contribute to a more definitive answer.

Until we know for sure, one can ask: is the 5-a day program harmful? The answer is "probably not." The benefits to health from eating at least 5 servings of fruits and vegetables per day likely include a reduction in risk for heart disease, other cancers (if not breast cancer), diabetes, and obesity.^{2 ,16 - 17} Should the 5-a-day program be reevaluated? Possibly. The current evidence almost uniformly suggests that vegetables are more protective than fruits in reducing cancer risk. Therefore, recommendations that focus on specific types of vegetables or fruits should be considered. Furthermore, although 5-a-day is achievable, higher levels of fruit and vegetable intake may be more beneficial for cancer risk reduction. In the meantime, an apple (or broccoli stalk) each day is probably not a bad idea.