Thomas L. Halton, Sc.D., Walter C. Willett, M.D., Dr.P.H., Simin Liu, M.D., Sc.D., JoAnn E. Manson, M.D., Dr.P.H., Christine M. Albert, M.D., M.P.H., Kathryn Rexrode, M.D., and Frank B. Hu, M.D., Ph.D.
November 9, 2006
Abstract
Background. Low-carbohydrate diets have been advocated for weight loss and to prevent obesity, but the long-term safety of these diets has not been determined.
Methods. We evaluated data on 82,802 women in the Nurses' Health Study who had completed a validated food-frequency questionnaire. Data from the questionnaire were used to calculate a low-carbohydrate-diet score, which was based on the percentage of energy as carbohydrate, fat, and protein (a higher score reflects a higher intake of fat and protein and a lower intake of carbohydrate). The association between the low-carbohydrate-diet score and the risk of coronary heart disease was examined.
Results. During 20 years of follow-up, we documented 1994 new cases of coronary heart disease. After multivariate adjustment, the relative risk of coronary heart disease comparing highest and lowest deciles of the low-carbohydrate-diet score was 0.94 (95% confidence interval [CI], 0.76 to 1.18; P for trend=0.19). The relative risk comparing highest and lowest deciles of a low-carbohydrate-diet score on the basis of the percentage of energy from carbohydrate, animal protein, and animal fat was 0.94 (95% CI, 0.74 to 1.19; P for trend=0.52), whereas the relative risk on the basis of the percentage of energy from intake of carbohydrates, vegetable protein, and vegetable fat was 0.70 (95% CI, 0.56 to 0.88; P for trend=0.002). A higher glycemic load was strongly associated with an increased risk of coronary heart disease (relative risk comparing highest and lowest deciles, 1.90; 95% CI, 1.15 to 3.15; P for trend=0.003).
Conclusions. Our findings suggest that diets lower in carbohydrate and higher in protein and fat are not associated with increased risk of coronary heart disease in women. When vegetable sources of fat and protein are chosen, these diets may moderately reduce the risk of coronary heart disease.
----------------------------------
Obesity in the United States has reached epidemic proportions. Leading research and medical societies advocate a low-fat, high-carbohydrate, energy-deficient diet to manage weight.1-4 Despite these recommendations, diets high in fat and protein and low in carbohydrate remain popular, and several best-selling books endorse this strategy for weight loss.5-9
The long-term safety of carbohydrate-restricted diets remains controversial. Most such diets tend to encourage increased consumption of animal products and therefore often contain high amounts of saturated fat and cholesterol. This may cause unfavorable changes in serum lipid levels and increase the risk of coronary heart disease. Several professional organizations have cautioned against the use of low-carbohydrate diets.10-13
We devised a system to classify women who participated in the Nurses' Health Study according to their relative levels of fat, protein, and carbohydrate intake and created a simple summary score designated the “low-carbohydrate-diet score.” We then examined prospectively the association between the low-carbohydrate-diet score and the risk of coronary heart disease in this cohort.
Methods
STUDY POPULATION
The Nurses' Health Study was initiated in 1976, when 121,700 female registered nurses 30 to 55 years of age completed a mailed questionnaire. Since 1976, information on disease status and lifestyle factors has been collected from this same cohort every 2 years. Diet was assessed by means of a semiquantitative food-frequency questionnaire in 1980, 1984, 1986, 1990, 1994, and 1998; 98,462 women completed the 1980 questionnaire.
For this investigation we excluded all women at baseline who left 10 or more food items blank or had implausibly high (>3500 kcal) or low (<500 kcal) daily energy intakes on the food-frequency questionnaire. We further excluded women with a history of diabetes, cancer, or cardiovascular disease before 1980, because these diagnoses may cause alterations in diet. After these exclusions, 82,802 women remained in this investigation. The study was approved by the Human Research Committee of Brigham and Women's Hospital in Boston; the completion of the self-administered questionnaire was considered to imply informed consent.
ASSESSMENT OF DIET AND GLYCEMIC LOAD
The 1980 food-frequency questionnaire included 61 food items and was revised in 1984 to include about twice that number.14,15 Study participants reported average frequency of consumption of specific foods throughout the previous year. The validity and reproducibility of the questionnaire have been documented elsewhere.14,15
To calculate the intake of specific foods, a commonly used portion size for each food was specified (e.g., one egg or one slice of bread) and participants were asked how often, on average, during the previous year they had consumed that amount. The possible responses ranged from never or less than once per month to six or more times per day.
Nutrient values were computed by multiplying the frequency of consumption of each food by the nutrient content of the portion and then adding these products across all food items. All food-composition values were obtained from the Harvard University food-composition database, which was derived from U.S. Department of Agriculture sources16 and supplemented with information from the manufacturer. The validity of estimated nutrient intake as assessed by the questionnaire has previously been evaluated with the use of multiple diet records. The correlation between the 1986 questionnaire and the average of six 1-week diet records collected in 1980 and 1986 was 0.73 for carbohydrate, 0.67 for total fat, and 0.56 for protein.15
The method used to assess glycemic load in the Nurses' Health Study has been described elsewhere.17 Briefly, we calculated the total dietary glycemic load by multiplying the carbohydrate content of each food by its glycemic index (the glycemic index of glucose is 100) and then multiplied this value by the frequency of consumption and summed these values for all foods. Dietary glycemic load, therefore, represents both the quality and quantity of carbohydrate consumed. Each unit of glycemic load represents the equivalent blood glucose–raising effect of 1 g of pure glucose.
CALCULATION OF THE LOW-CARBOHYDRATE-DIET SCORE
We divided the study participants into 11 strata each of fat, protein, and carbohydrate intake, expressed as a percentage of energy (Table 1). For fat and protein, women in the highest stratum received 10 points for that macronutrient, women in the next stratum received 9 points, and so on down to women in the lowest stratum, who received 0 points. For carbohydrate, the order of the strata was reversed; those with the lowest carbohydrate intake received 10 points and those with the highest carbohydrate intake received 0 points. We used the percentage of energy consumed instead of absolute intake to reduce bias due to underreporting of food consumption and to represent dietary composition.
The points for each of the three macronutrients were then summed to create the overall diet score, which ranged from 0 (the lowest fat and protein intake and the highest carbohydrate intake) to 30 (the highest protein and fat intake and the lowest carbohydrate intake). Therefore, the higher the score, the more closely the participant's diet followed the pattern of a low-carbohydrate diet. Thus, the score was termed the “low-carbohydrate-diet score.”
We also created two additional low-carbohydrate-diet scores. One was calculated according to the percentage of energy as carbohydrate, the percentage of energy as animal protein, and the percentage of energy as animal fat, and the other was calculated according to the percentage of energy as carbohydrate, the percentage of energy as vegetable protein, and the percentage of energy as vegetable fat (Table 1).
MEASUREMENT OF NONDIETARY FACTORS
In 1976, women provided information regarding parental history of myocardial infarction. Beginning in 1976, participants also provided information every 2 years on the use of postmenopausal hormones, smoking status, body weight, and other covariates. They provided information on aspirin use repeatedly throughout the follow-up. The correlation coefficient between self-reported body weight and measured weight was 0.96.18 Physical activity was assessed in 1980, 1982, 1986, 1988, 1992, 1996, and 1998, and we calculated the cumulative average number of hours per week spent in moderate or vigorous physical activity.19
OUTCOME
The outcome of this study was incident coronary heart disease, including nonfatal myocardial infarctions or fatal coronary events. Each participant contributed follow-up time from the date of returning the 1980 questionnaire to the date of the first end point (death or nonfatal myocardial infarction) or until the censoring date of June 1, 2000.
We requested permission to examine the medical records of all participants who reported a diagnosis of coronary heart disease on one of the follow-up questionnaires that were completed every two years. A myocardial infarction was considered to be confirmed if it met the World Health Organization criteria of symptoms and either typical electrocardiographic changes or elevated cardiac-enzyme levels.20 Infarctions that necessitated a hospital admission and for which confirmatory information was obtained by interview or letter but for which no medical records were available were designated as probable and were included in the analysis.
Deaths were identified from state vital records and the National Death Index or reported by the participants' next of kin or the U.S. Postal Service.21 Fatal coronary heart disease was confirmed by an examination of autopsy or hospital records, by a listing of coronary heart disease as the cause of death on the death certificate, and by the availability of evidence of previous coronary heart disease. Those deaths in which coronary heart disease was the underlying cause on the death certificate but for which no medical records were available were designated as deaths from presumed coronary disease.
STATISTICAL ANALYSIS
We divided women into 10 categories (deciles) according to their low-carbohydrate-diet score. To represent long-term intake and reduce measurement error, we calculated the cumulative average low-carbohydrate-diet score based on the information from the 1980, 1984, 1986, 1990, 1994, and 1998 questionnaires.22 For example, the low-carbohydrate-diet score from the 1980 questionnaire was related to the incidence of coronary heart disease between 1980 and 1984, and the low-carbohydrate-diet score from the average of the 1980 and 1984 questionnaires was related to the incidence of coronary heart disease between 1984 and 1986. Incidence rates for coronary heart disease were calculated by dividing cases by the person-years of follow-up for each decile of the low-carbohydrate-diet score. Relative risks of coronary heart disease were calculated by dividing the rate of occurrence of coronary heart disease in each decile by the rate in the first (lowest) decile. We used Cox proportional-hazards models23 to adjust for potentially confounding variables. Because low-carbohydrate diets may decrease subsequent energy intake,24 we did not control for total energy intake in multivariate models. However, further adjustment for caloric intake was performed in a secondary analysis. We also examined the association between each macronutrient and the risk of coronary heart disease in multivariate nutrient-density models.22 All P values are two-sided.
Results
The cumulative average low-carbohydrate-diet score ranged from a median of 5.0 in the 1st decile to a median of 26.0 in the 10th decile (Table 2). The mean daily carbohydrate intake ranged from 234.4 g in the 1st decile to 116.7 g in the 10th decile. At the midpoint of follow-up (1990), women who had a higher score were more likely to smoke and had a higher body-mass index, a lower dietary glycemic load, a lower caloric intake, and a higher intake of saturated fat. On average, body-mass index increased by approximately 2.5 units from baseline to the end of follow-up, regardless of the low-carbohydrate-diet score.
Because the Nurses' Health Study did not routinely collect data on blood lipid levels, the effect of a low-carbohydrate diet on lipids could not be assessed for the entire study cohort. However, a subgroup of women from the study (466 women) had blood drawn in 1990 for determinations of lipid levels. In this subgroup, the low-carbohydrate-diet score was not associated with the total cholesterol level or with the levels of high-density lipoprotein (HDL) cholesterol or low-density lipoprotein (LDL) cholesterol after adjustment for age, smoking status, and other covariates. The low-carbohydrate-diet score was inversely associated with the triglyceride level (126.5 mg per deciliter in the lowest quintile and 99.3 mg per deciliter in the highest quintile of the low-carbohydrate-diet score, P for trend=0.05).
During 20 years of follow-up (1,584,042 person-years), we documented 1994 cases of coronary heart disease. In age-adjusted analyses, the relative risk comparing women in the 10th decile with those in the 1st decile of the low-carbohydrate-diet score was 1.29 (95% confidence interval [CI], 1.04 to 1.60). After further adjustment for smoking status, the relative risk of coronary heart disease was 1.11 (95% CI, 0.89 to 1.38) comparing women in the same deciles of the low-carbohydrate-diet score (P for trend=0.54) (Table 3). After controlling for potential confounders, the relative risk was 0.94 (95% CI, 0.76 to 1.18; P for trend=0.19). Further adjustment for total calories did not appreciably alter the results (relative risk, 0.96; 95% CI, 0.77 to 1.20; P for trend=0.27). When body-mass index was removed from the multivariate model, the results did not change significantly.
In stratified analyses, there was no evidence that the relationship between the low-carbohydrate-diet score and coronary heart disease was modified as a result of body-mass index, level of physical activity, smoking status, or the presence or absence of diabetes, hypertension, or hypercholesterolemia. Specific data on blood lipid levels were not available for most of the cohort. As a result, it was not feasible to adjust or stratify our analysis for this factor.
We also created a second low-carbohydrate-diet score according to the percentages of energy from carbohydrate, animal protein, and animal fat (Table 1). The multivariate relative risk of coronary heart disease was 0.94 (95% CI, 0.74 to 1.19) for the comparison of the 10th with the 1st decile (P for trend=0.52) (Table 3). We also created a third low-carbohydrate-diet score according to the percentages of energy from carbohydrate, vegetable protein, and vegetable fat (Table 1). For the comparison of the 10th with the 1st decile, the multivariate relative risk of coronary heart disease was 0.70 (95% CI, 0.56 to 0.88; P for trend=0.002) (Table 3).
