E D I T O R I A L
Blood Lipid Levels in Type 2 Diabetes What are the effects of diet? acrovascular disease is a major cause of death among people with diabetes (1). Diabetes is recognized as an independent risk factor for cardiovascular disease and appears to confer increased risk beyond that associated with lipid levels. However, elevated blood lipid levels are associated with an increased risk of coronary heart disease for people with diabetes, as they are for people without diabetes (2). In this issue of Diabetes Care, Mayer-Davis and associates use data from two observational epidemiologic studies, the San Luis Valley Diabetes Study (SLVDS) and the Insulin Resistance Atherosclerosis Study (IRAS), to examine the effects of diet on lipid and lipoprotein levels among people with type 2 diabetes (3). Observational epidemiologic studies have many strengths. They can examine naturally occurring exposures, some of which may be difficult or impossible to assign experimentally, in a free-living population over a long period. They are more easily carried out than experimental interventions or clinical trials and can provide a larger sample size. However, because these studies are observational rather than experimental, they lack randomization to control for confounding factors and they also lack the direct control over the exposure of interest that would be found in a clinical trial or experimental intervention. Thus, observational studies may be subject to biases related to sample selection, ascertainment of exposures and outcomes, and measured or unmeasured confounding variables. This distinction between observational and experimental studies may be particularly crucial for studies of dietary intake and health outcomes (4). Measuring nutrient intake accurately in observational epidemiologic studies is a formidable task. Compared with behaviors such as cigarette smoking or alcohol consumption (both difficult to measure accurately), dietary intake is highly complex. The number of distinct food items available for consumption runs into the thousands. Most people eat a variety of food items every day. Dietary intake does not appear to be a highly salient behavior (in other words, people do not
M
always pay close attention to what they eat), and people are not very good at remembering or reporting exactly what they ate, particularly over a long time period. Collecting complete and accurate data on the varied diets consumed by respondents poses a challenge. The two studies used by Mayer-Davis et al. used two different approaches to the problem of collecting dietary information from participants. In the SLVDS, respondents were asked to provide a detailed account of the preceding day’s dietary intake. The IRAS used a food frequency questionnaire (FFQ), in which people were asked to report the frequency with which they consumed each of 114 specified food categories over the previous year and to give a general indicator of portion sizes. As estimates of usual diet, both methods (the daily recall and the FFQ) are problematic. In theory, a single-day recollection provides an accurate, detailed, and highly specific description of the foods consumed on the previous day; however, a single day’s dietary intake, or even a few days’ worth of intake, may not be a very precise estimate of the person’s usual dietary intake. In theory, the FFQ, although it includes much less detail, provides a better picture of usual intake because it covers a longer time period. However, in practice, a main source of error for both is the difficulty that respondents have in accurately reporting intake. In recall data, respondents may omit foods that were eaten, include foods that were not eaten, and estimate portion sizes inaccurately. Compared with the recall method, the FFQ poses yet more difficult cognitive tasks for participants, who have to try to estimate accurately the relative frequency with which many food categories were consumed over a lengthy time period. Difficulties in estimating frequency accurately appear to be the main reason for disagreements between FFQ data and reference methods (5). The FFQ also relies heavily on data from other sources. Rather than collecting data on specific foods and on specific portion sizes eaten by the respondent, it is assumed that within each food category, all respondents eat the same
DIABETES CARE, VOLUME 22, NUMBER 10, OCTOBER 1999
mix of foods, and portion sizes are taken from a different data set from different respondents of the same sex and age. The assumptions built into the method may cause the nutrient estimates to be inaccurate for a given individual. A certain degree of error is inherent in self-reported dietary assessments. The classical measurement error model assumes that the error term is additive, unbiased, and independent of the true value and the outcome. The expected effect of this type of error in a single independent variable would be to make an association appear weaker. This is the only potential effect of measurement error in dietary data mentioned by Mayer-Davis and associates. However, measurement errors in dietary data generally do not correspond to the classic measurement error model. As described by Kipnis et al. (6), “reporting errors in dietary studies are usually biased, correlated with true nutrient intakes and with each other, heteroscedastic and nonnormally distributed.” The “flattened slope” syndrome, in which large amounts are underreported and small amounts are overreported, has been observed in a number of studies. In addition, since different nutrients are estimated from the same foods, errors in nutrient estimates are often correlated with each other. The effects of this more complex error structure on study findings are difficult to predict. Because of the limitations and uncertainties inherent in dietary data, observational epidemiologic studies are not well suited either to confirm or to refute highly specific hypotheses about nutrient intake. They serve better to suggest new hypotheses and possible lines of investigation. Results from observational epidemiologic studies that suggest new findings need to be confirmed by more detailed experimental studies. For instance, many epidemiologic studies suggested a beneficial effect of fruit and vegetable consumption on cancer risk, which was hypothesized to be due to a specific constituent of fruits and vegetables, b-carotene. However, subsequent clinical trials and intervention studies in which b-carotene was administered directly 1605
Editorial showed either no effect of b-carotene on cancer risk or an adverse effect, causing at least one clinical trial to be stopped before completion (7). The specific effect of bcarotene hypothesized from observational epidemiologic studies was not borne out in practice when more rigorous experimental interventions were conducted. The broad findings of the study by Mayer-Davis et al., of a relationship between dietary fat intake and serum cholesterol level, are consistent with other research (8). As noted by the authors, some of their specific findings, such as the finding of a positive association between oleic acid intake and LDL cholesterol levels in one cohort, are less consistent with published evidence showing neutral or beneficial effects of oleic acid on LDL cholesterol levels. The authors also found an association of carbohydrate with serum triglycerides in some subgroups but not others. These detailed results should be interpreted cautiously. Some of the heterogeneity in the results of the study by Mayer-Davis et al. may be related to differences in the dietary methods and to the error structure in the dietary data. Some end points in observational epidemiologic studies, such as cancer incidence or mortality, would require a
1606
large sample size and a long time frame to examine in a randomized clinical trial or metabolic study. However, for blood lipid levels, the outcome used in the study by Mayer-Davis and associates, dietary effects can be examined in short-term experimental studies (9). Thus, for example, the finding from the observational study by Mayer-Davis and associates that oleic acid may adversely affect blood lipids among people with diabetes should be considered as suggesting an interesting hypothesis that would benefit from confirmation by experimental results. KATHERINE M. FLEGAL, PHD From the National Center for Health Statistics, Centers for Disease Control and Prevention, Hyattsville, Maryland. Address correspondence to Dr. Katherine M. Flegal, National Center for Health Statistics, 6525 Belcrest Rd., Room 900, Hyattsville, MD 20782. E-mail:
[email protected].
References 1. Fagan TC, Sowers J: Type 2 diabetes mellitus: greater cardiovascular risks and greater benefits of therapy (Editorial). Arch Intern Med 159:1033–1034, 1999 2. Lehto S, Rönnemaa T, Haffner SM, Pyörälä
3.
4. 5.
6.
7. 8.
9.
K, Kallio V, Laakso M: Dyslipidemia and hyperglycemia predict coronary heart disease events in middle-aged patients with NIDDM. Diabetes 46:1354–1359, 1997 Mayer-Davis EJ, Levin S, Marshall JA: Heterogeneity in associations between macronutrient intake and lipoprotein profile in persons with type 2 diabetes. Diabetes Care 22:1632–1639, 1999 Flegal KM: Evaluating epidemiologic evidence of the effects of food and nutrient exposures. Am J Clin Nutr 69:1339S–1344S, 1999 Flegal KM, Larkin FA: Partitioning macronutrient intake estimates from a food frequency questionnaire. Am J Epidemiol 131:1046–1058, 1990 Kipnis V, Freedman LS, Brown CC, Hartman AM, Schatzkin A, Wacholder S: Effect of measurement error on energy-adjustment models in nutritional epidemiology. Am J Epidemiol 146:842–855, 1997 Albanes D: Beta-carotene and lung cancer: a case study. Am J Clin Nutr 69:1345S–1350S, 1999 Hegsted DM, Ausman LM, Johnson JA, Dallal GE: Dietary fat and serum lipids: an evaluation of the experimental data. Am J Clin Nutr 57:875–883, 1993 Lichtenstein AH, Ausman LM, Jalbert SM, Schaefer EJ: Effects of different forms of dietary hydrogenated fats on serum lipoprotein cholesterol levels. N Engl J Med 340:1933–1940, 1999
DIABETES CARE, VOLUME 22, NUMBER 10, OCTOBER 1999
