Benefits and Adverse Effects of Weight Loss: Observations from the Framingham Study

Methods

The Framingham Study recruited 5209 men and women between ages 30 and 62 years beginning in 1948; observations made at 2-year intervals included measurements of weight and risk factors as well as ascertainment of cardiovascular events and morbidity and death from all causes [3, 1416].

Framingham Study

Men and women between ages 35 and 54 years at the fourth examination of the Framingham cohort (1954-1958) were characterized with respect to change in body mass index (BMI, kg/m²) measured at 2-year intervals during a period of 10 years. Those who did not attend each examination were excluded (18% of those present at examination 4). Individual slopes, based on regressions of BMI for six examinations, were used to define three equal groups (tertiles) of individuals; characterizing approximately those who lost, those who remained relatively stable, and those who gained. Details of examination procedures, laboratory methods, diagnostic criteria, and follow-up methods have been published [3, 1416]. This report defined cardiovascular disease as coronary heart disease, stroke, intermittent claudication, and congestive heart failure. Coronary heart disease was defined as myocardial infarction, coronary insufficiency, and angina pectoris. Hypertension was defined by a blood pressure 160 mm Hg (systolic) or 95 mm Hg (diastolic) or by the use of antihypertensive medication. Pulmonary disease was diagnosed clinically or by self-report of chronic coughing or wheezing. Other clinically diagnosed diseases included mitral or aortic valve disease, arthritis, urinary disease, neurologic disease, and other vascular disease. The term any disease applied to any of these conditions.

Statistical Analysis

Individual slopes of weight, BMI, systolic blood pressure, diastolic blood pressure, and cholesterol were estimated for the 10-year period from examination 4 to examination 9. The coefficient of variation was defined as the standard deviation of BMI divided by the mean BMI for examinations 4 through 9. The variability around the BMI slope for examinations 4 through 9 was calculated as Equation 1

where y = predicted BMI, y = actual BMI, and df = degrees of freedom.

Mean BMIs for examinations 4 through 9 were calculated for the population attending all of these examinations; means for examinations 1 through 3 and 10 through 18 were based on members of this population who attended the examination in question.

Results from the first and third tertiles (loss and gain groups) were compared with those from the no change group using t-tests for continuous variables and chi-square analyses for dichotomous variables. Changes in lipid levels, blood pressure, and smoking status by change in weight were compared using analysis of variance and chi-square tests. Mortality rates were calculated for the cohort during 20 years of follow-up and excluded deaths occurring in the first 4 years. Age-adjusted and multivariate relative risks and 95% confidence intervals (CIs) were calculated from Cox regression coefficients with age, and age plus major risk factors at examination 9 in the models. In these regression analyses, the tertiles of BMI slope were entered as dummy variables and used the middle group as the referent.

Results

Figure 1 shows the age-adjusted mean levels of BMI for the three tertiles of BMI change for the 10-year classification period (examinations 4 through 9), together with mean levels of BMI for 6 years before and 18 years after this period. Each group has been subdivided into those who survived through examination 18 and those who died between examinations 9 and 18. Values of BMI were not available for those who did not return for any of examinations 10 through 18 because of death or for any other reason. In all three groups of BMI change, mean BMIs for those who survived through examination 18 were stable during follow-up.

Among women whose BMIs decreased, the decline was greater and continued beyond examination 9 in the 31% who did not survive to examination 18. The absence of a similar pattern among men whose BMIs decreased may have been due to the higher proportion of these men (53%) who did not survive to examination 18. Twenty-eight percent of deaths were attributed to coronary heart disease in these men, and 15% were sudden, whereas 19% were attributed to coronary heart disease in women whose BMIs decreased, and 6% were sudden.

Among men in the no-change group, mean BMIs were higher in those who died than in those who survived. Mean BMIs were similar, however, in these two groups of women. Among men who gained weight between examinations 4 and 9, those who died before examination 18 were heavier initially, but their BMIs decreased sharply toward the end of the follow-up period. A similar pattern was seen in women, but the discrepancy in BMI was greater. Those who did not survive were heavier until 8 years of follow up, when BMIs decreased sharply in those who did not survive to examination 18.

Characteristics of the three BMI change groups during the classification period (examinations 4 through 9) are shown in Table 1. Weight loss ranged from losses of 24.4 and 24.0 kg (54 and 53 lb) to gains of 2.2 and 7.7 kg (5 and 17 lb) for men and women, respectively, in the BMI loss tertile. Men and women whose BMIs decreased were older; their mean BMIs were 27.1 and 25.7 at examination 4 and 25.3 and 24.2 at examination 9. They were the heaviest groups at the beginning and the lightest groups at the end of this period; BMIs averaged for the six examinations were identical for the three groups of men but were lowest for women whose BMI did not change. Change in BMI correlated weakly with BMI at examination 4 (r = 0.2 and 0.12),with mean BMI for examinations 4 through 9 (r = 0.05 and 0.11), with the coefficient of variation (r = 0.06 and 0.19), and with variability around the BMI slope (r = 0.04 and 0.12)for men and women, respectively. The coefficient of variation (variability around the mean) and variability around the slope of BMI were greater for those whose BMIs decreased or increased (Table 1).

Compared with those whose BMIs were unchanged, men and women whose BMIs decreased had higher systolic blood pressures and cholesterol levels initially but not at the end of the baseline period. Rates of cigarette smoking at examinations 4 and 9 were higher among those whose BMIs decreased, and the percentages of smokers who quit were lower among those whose BMIs decreased than among those whose BMIs changed only slightly. Rates of smoking cessation were twice as high in men and women who gained as in those who lost weight. Levels of physical activity were similar in the three groups of men but were slightly lower in the BMI loss group of women. At examination 7 or 8, 16% of men and women who lost weight were on weight control diets; 20% of women and 9% of men who gained and 11% of women and 7% of men whose weight remained unchanged reported being on similar diets. Average annual changes in weight and the major cardiovascular risk factors during the 10-year period are also shown in Table 1. Among men of average height (1.7 m [68 in]), the average annual weight change for the three tertiles of BMI change ranged from a loss of 0.51 kg (1.14 lb) to a gain of 0.60 kg (1.33 lb). Among women of average height (1.6 m [63 in]), average annual changes in weight varied from a loss of 0.39 kg (0.86 lb) to a gain of 0.71 kg (1.58 lb). Women with the most stable weights gained an average of 0.16 kg (0.36 lb) per year. Systolic and diastolic blood pressures increased least in those whose BMI decreased and increased most in those whose BMIs increased. Changes in cholesterol levels ranged from a decrease of 0.02 mmol/L (0.76 mg/dL) per year for men who lost weight to increases of 0.01 to 0.02 mmol/L (0.38 to 0.76 mg/dL) per year for those whose weight increased or remained unchanged. Changes in cholesterol were all positive in women, ranging from increases of 0.03 mmol/L (1.1 mg/dL) per year for those who lost weight to increases of 0.06 mmol/L (2.3 mg/dL) for other women. In contrast, rates of smoking cessation were more beneficial for health, that is, higher in those whose BMIs increased.

Prevalence rates of cardiovascular disease and coronary heart disease were significantly higher at examination 9 among men whose BMIs decreased (see Table 1); diabetes, pulmonary disease, and the aggregate of all diseases were similarly more prevalent among men in this group than among those whose BMIs did not change. Diabetes was the only condition that was significantly more frequent in those whose BMIs increased than in those whose BMIs remained unchanged. Among women, prevalence rates of hypertension, pulmonary conditions, and any disease were significantly higher at examination 9 among those whose BMIs decreased than among those whose BMIs did not change. Hypertension was also more prevalent in women whose BMIs increased. Although not statistically significant, rates of cancer were higher among men and women who lost weight than among those with stable weights; rates of coronary heart disease and diabetes were highest in this group of women.

To separate those with disease at the beginning of the classification period from those who developed it during this period, we calculated age-adjusted prevalence rates at examination 4 for the three groups of men and women. Except for pulmonary disease or symptoms, these rates were all lower at examination 4 than at examination 9; the only significant difference was observed in the prevalence rates of hypertension in women, which were 13%, 9%, and 11%, for the loss, no change, and gain groups, respectively. The trend for hypertension was similar in men. Most persons with cardiovascular conditions, cancer, or diabetes at examination 9 developed these disorders during the classification period.

Figure 2 shows the 20-year mortality rates for the three tertiles of change in BMI according to tertile of BMI at the start of the classification period. Deaths in the first 4 years of follow-up were excluded to reduce the effect of preexisting disease on the analysis. Total death rates were generally highest for those who lost weight, regardless of baseline BMI, but they were slightly higher for those who were leanest initially. Men who were in the heaviest BMI tertile initially had higher mortality rates than did those with lower initial BMIs, provided that they were not in the BMI loss group.

Among women, decreases in BMI were associated with increased mortality rate, as were increases in BMI among women who were heavy at study entry. Death rates for all causes were generally lowest among those with the lowest BMIs initially, provided that they did not lose weight. Death rates for cardiovascular disease and coronary heart disease showed similar patterns to those for all causes, except that increased mortality rate was more strongly related to being heavy initially and to further weight gain in women.

Relative risks for death from all causes, from cardiovascular disease, and from coronary heart disease were estimated for tertiles of BMI change using the no-change group as the standard for comparison (Table 2). For each cause of death in men, a decrease in BMI was associated with increased risk, ranging from 44% for all causes to 65% for cardiovascular disease. Adjustment for BMI and risk factors at examination 9 as well as for age showed a persistent, statistically significant increase in mortality rate among those whose BMIs decreased for all three categories of death. Weight gain was associated with relative risks less than 1.0; however, none of these estimates differed significantly from the reference value in those whose BMIs were unchanged, regardless of adjustments for age, BMI, and risk factors at examination 9.

Similar trends were noted in higher relative risks for death from all causes, from cardiovascular disease, and from coronary heart disease in women who experienced a loss in BMI, but only the 38% increase in the age-adjusted all-cause mortality rate was statistically significant. Women who gained weight had higher relative risks for each of the three causes of death than did women whose weight did not change during the 10-year classification period. However, none of the increased relative risks was statistically significant (see Table 2). In general, controlling for BMI and risk factors at examination 9 reduced, but did not eliminate, the increased relative risks associated with weight loss or gain.

Discussion

Weight loss was associated with preexisting and incident disease in Framingham and was a factor that undoubtedly contributed to the increased mortality rates observed among those who lost weight, although deaths occurring early in the follow-up period were excluded. It is unknown whether weight loss between examinations 4 and 9 was due to clinical or preclinical disease itself or was a component of treatment for disease. It is also impossible to distinguish voluntary from involuntary weight loss in the Framingham cohort. Interpretation of the data on weight loss requires consideration of age, sex, the level and variability of BMI before and during weight loss, the amount of weight loss, distribution of body fat, levels of other risk factors, and the reasons for weight loss. Unfortunately, much of this information is lacking because most studies were not designed to address these issues.

Hazards of overweight are well known and include elevated total and low-density lipoprotein cholesterol levels; decreased high-density lipoprotein cholesterol levels; hypertension; and increased prevalence and incidence of cardiovascular disease, diabetes mellitus, and other diseases [15]. The J-shaped or U-shaped curve showing increased mortality rate at the low and the high ends of the BMI (or weight) distribution is also well known [4, 5, 1719]. Explanations include associations between leanness and cigarette smoking, infectious diseases, cancer, digestive, chronic respiratory diseases, and other abnormalities.

In their analysis of Framingham data, Lissner and colleagues [9] found that participants whose weights varied most had relative risks of 1.3 to 2.0 for total mortality and coronary heart disease morbidity and mortality compared with participants whose weights varied least. These authors classified participants by weight change during a period of 14 years as well as by reported weight at age 25 years. The coefficient of variation of BMI was more highly correlated with the slope of BMI (r = 0.65 in men and 0.59 in women) in their analysis than in that reported here. Regression coefficients for the relation of slope of BMI to total mortality rate and to coronary heart disease mortality and morbidity rates were negative and highly significant. The authors concluded that fluctuations in body weight may have negative consequences beyond the risk associated with obesity and the trend in body weight over time.

In most studies, weight histories are not available, and it is unknown whether lean persons have been lean throughout their lives, whether weights have fluctuated, or whether recent weight loss has occurred. The Framingham data suggest that maintained stable weights and leanness in the absence of weight loss are associated with low rates of mortality. These data confirm previously recognized paradoxical effects of weight loss showing that major risk factors (except smoking) improve but that mortality rates increase. Weight loss for any reason improves risk factor profiles, but any long-term benefits are overshadowed by the adverse effects of weight loss due to smoking and disease in middle-aged populations.

Although weight loss is associated with prevalent and preclinical disease, which are powerful determinants of future morbidity and mortality, weight loss may be voluntary and associated with other changes in lifestyle that are aimed at improving health, fitness, or appearance. The data presented here indicate that leanness and maintenance of stable weight remain appropriate goals until further evidence is obtained about the long-term benefits and adverse effects of voluntary weight loss.