The Association between Weight Loss and Increased Longevity: A Review of the Evidence

1950 Metropolitan Life Insurance Study

The first studies to report a positive benefit of weight loss on mortality were published by Dublin and Marks [1] and by Dublin [2] in the early 1950s. These reports showed similar results from a prospective study conducted by the Metropolitan Life Insurance Company that focused on persons who were initially charged an extra life insurance premium solely because they were too fat to be taken at standard rates [2].

The mortality experience of persons who on subsequent application for insurance had reduced their weight sufficiently and for a long enough period to justify insurance either at standard rates or at a lower rate than they had originally been charged [2] was compared with that of persons who continued to pay an extra premium for life insurance, presumably because they had not lost weight.

Based on the premium paid at the first issuance of insurance, two subgroups were identified among those who had lost weight: those who paid a premium slightly higher than standard (presumably the moderately overweight) and those who paid the highest premium (presumably the severely overweight). No quantitative definitions were given for these two groups.

The mortality experience of these groups was summarized using standardized mortality ratios. The denominator for these ratios was the mortality rate observed for nonoverweight persons who qualified for standard-rate insurance during the same study period. The numerators were the mortality rates for each group originally charged an extra premium because of overweight.

In moderately overweight men who lost weight, the standardized mortality ratio was 113% compared with a ratio of 142% for those who remained moderately overweight and did not qualify for reduced-rate insurance. This finding is equivalent to a mortality relative risk of 0.80 (113%/142%), which indicates that overweight persons who reduced their weight had a mortality rate that was 20% lower ([1.00 0.80] 100%) than that of persons who did not reduce. In moderately overweight women, weight loss was associated with a 37% decrease in mortality rate. In the severely overweight men and women, the mortality reductions were 39% and 16%, respectively.

These findings, however, may be biased toward a positive benefit of weight loss. Before participants could be requalified for insurance at a lower premium, they were required to have an additional medical examination [2]. These reports did not indicate whether persons who had lost enough weight to requalify at a lower insurance rate could have been excluded from the study if they were found to be ill at the second medical examination. If so, persons with prevalent disease may have been removed from the group of participants who lost weight. No additional medical examinations were done, however, on the comparison group who were initially overweight and who did not seek to have their insurance premiums reduced. The comparison group therefore may have been less healthy and may have had higher mortality rates than the weight loss group simply because they had less medical screening.

The study did attempt to control for prevalent disease by excluding persons with other impairments, adverse medical history, or occupational hazards [1], but the exact nature of these exclusions was not described. Because these exclusions were applied at the start of the study, and not later when participants were rerated for insurance, prevalent disease may nonetheless have differentially affected the mortality rates of the comparison group that did not lose weight. Results were also presented after excluding as many as the first 15 years of mortality, but the mortality reductions associated with weight loss remained essentially unchanged [1].

The study did not assess or adjust for possible differences in smoking status between the weight-loss and comparison groups. Thus, the beneficial effects of weight loss may have been partially masked, which would have biased the results in a direction opposite to that caused by unequal medical screening.

No other adjustments were made for potentially confounding factors. Although results were presented in 10-year age groups, it is unclear whether a statistical adjustment was used to account for differences in age between the groups.

1959 Build and Blood Pressure Study

The body-build component of the 1959 Build and Blood Pressure Study [3] was a prospective study of body weight and mortality based on approximately 5 million life insurance policies. Nested within this large cohort was a study similar in design to the 1950 Metropolitan Life Insurance Study. In contrast to the earlier insurance study, some information was given on starting height and weight but only in broad ranges. For example, among men 15 to 39 years old who were of medium height (57 to 510), who initially weighed 195 to 214 pounds, and who subsequently reduced their weight, the standardized mortality ratio was 113%, compared with a ratio of 135% for men of the same height and weight who did not reduce. This result is equivalent to a relative risk of 0.84 for a mortality reduction of 16%. For men of the same height and weight ranges who were 40 to 69 years old, the same mortality reduction was found. Unfortunately, this study was unable to examine the association between weight loss and mortality in women because only 21 deaths were observed in women who had become eligible for standard insurance after weight reduction [3].

The same potential biases found in the 1950 study also apply to this study; specifically, unequal medical screening between the weight loss and comparison groups, lack of control for smoking, and inadequate control for age.

1979 Build Study

The most recent prospective mortality study of U.S. life insurance policy holders is the 1979 Build Study [4], which included approximately 4 million policies. This study had a much lower power to examine the issue of weight reduction and subsequent mortality than did the earlier studies: The male experience involved approximately 1900 policies with 35 terminated by death, but there were fewer than 100 policies in the experience on female lives [4]. The number of deaths among women who reduced their weight was not given. Although the report states that the total experience from date of issue of the original policies among overweight men in this study who reduced was quite favorable, no standardized mortality ratios were given for overweight men who did not reduce their weight [4]. Thus, we are unable to estimate the relative risk and the mortality reduction associated with weight loss for either men or women in this study.

Cancer Prevention Study I

The American Cancer Society's Cancer Prevention Study I was unique in that respondents were asked whether their reported weight change was voluntary. Weight change was assessed as follows: The subjects were asked whether they had had a change in weight, and, if so, how many pounds they had gained or lost and over what period of years. Those who had a change in weight were also asked whether they had tried to bring about the change [5]. The authors reported, however, that to our surprise, the relationship between change in weight within a period of five years and later death rates from coronary heart disease and stroke was about the same for subjects who tried to bring about the change as for those who did not try to do so [5]. The separate effects of voluntary and involuntary weight loss were therefore not reported in the study.

Three groups were defined on the basis of their reported weight loss (<10 pounds, 10 to 19 pounds, and 20 pounds) during the 5 years before the start of the study. It is unclear, however, whether those whose weight remained stable or who gained weight were included among participants who lost 10 pounds or less. An internal standard was used to define relative body weight at the start of the weight-loss. This standard was expressed as a percentage of the mean weight observed in each 5-year age group of the same sex and height and was categorized as less than 90%, 90% to 109%, 110% to 119%, and 120% or more. Thus, the study included a total of eight separate weight-loss groups (for example, starting relative weight category of 120% and weight loss of 10 to 19 pounds).

Effects on total mortality were not examined in the study. The effect of weight change on mortality from coronary heart disease and stroke was expressed as a relative risk (the age-adjusted death rate of a specific weight-change group was the numerator, and the age-adjusted death rate observed in the comparison group who lost less than 10 pounds was the denominator). In six cases, the authors found that weight loss was associated with a reduction in mortality related to coronary heart disease or stroke. All of the protective effects occurred in persons whose relative weight was 10% or more above average at the start of their weight loss. In four of these cases, mortality was reduced by only 1% to 2%. The remaining two relative risks were 0.89 (11% reduction in coronary heart disease mortality) in men whose relative weight category was 110% to 119% and who had lost 10 to 19 pounds and 0.75 (25% reduction in stroke mortality) in women whose relative weight category was 120% or more and who had lost 10 to 19 pounds. No protective effects of weight loss were observed in men or women who had lost 20 pounds or more at any level of starting relative weight.

The authors attempted to control for prevalent disease by excluding persons who reported having been diagnosed with cardiovascular disease or cancer within the previous 5 years. They also excluded persons who reported they were sick at the time of the interview, but no indication was given whether this term refers to persons with a serious illness or to those with a minor complaint such as a cold. Early deaths were not excluded.

The results were only presented for a very broad age range: 40 to 79 years of age. The effects of weight loss on 6-year mortality are probably different for persons in their 40s than for persons in their 70s. We cannot judge whether the effects of weight loss on mortality from coronary heart disease and stroke differed between younger and older adults because age-stratified results were not presented.

British Regional Heart Study

In 1990, Wannamethee and Shaper [6] published the results of a 4-year prospective study of weight change and mortality in middle-aged men enrolled in the British Regional Heart Study. These men were selected at random from general medical practices and were interviewed, examined, and weighed between 1978 and 1980. Five years later, the surviving men were recontacted, sent a detailed questionnaire on health status, and asked to report their current weight.

Among men who were initially overweight (body mass index 28 kg/m²), the authors found that the men who lost 4% or more of their starting weight and became not overweight had a reduction in total mortality of 10% and a reduction in cardiovascular disease mortality of 50% compared with the weight-stable comparison group. In contrast, men who lost 4% or more of their weight but remained overweight had a 50% increase in their cardiovascular disease mortality. None of these mortality changes, however, was statistically significant.

The authors further analyzed their data on cardiovascular disease mortality with respect to hypertension status. They reported their results in terms of incidence rates, and we converted the rates to relative risks. The authors found that among overweight hypertensive men, those who were not overweight after weight loss had a cardiovascular disease mortality relative risk of 0.2 (P < 0.05), an 80% reduction in mortality. They also found, however, that men who lost weight but who remained overweight had a relative risk of 2.3, a 130% increase in cardiovascular disease mortality. The authors did not indicate whether this finding was statistically significant.

In addition, the authors stated that in obese men who were not hypertensive weight reduction had little effect on cardiovascular mortality [6]. Their data showed, however, that no deaths related to cardiovascular disease occurred among overweight nonhypertensive men who lost weight but who remained overweight. This result is equivalent to a relative risk of 0, implying that weight loss completely eliminated cardiovascular disease mortality in this subgroup of men. The authors did not indicate whether this effect was statistically significant nor did they estimate the effect of weight loss on total or cancer-related mortality after stratification by hypertension status.

Although the authors indicated that information was collected at both the baseline examination and at the 5-year interview on physician diagnosis for 12 conditions, they gave no indication that this information was used in the mortality analysis. Early deaths were not excluded from the analysis.

Because of the relatively young age of the participants (40 to 59 years) and the short mortality follow-up period (4 years), the study results apply only to the effects of weight loss on premature mortality. Stronger and more consistent protective effects of weight loss may have been found had the study been able to follow this middle-aged sample for a longer period of time, thus increasing the number of deaths observed in the cohort.

Aberdeen Diabetic Study

In 1990, Lean and colleagues [7] published a report in which they concluded that for the treatment of diabetes, the present study provides for the first time evidence that weight loss improves life expectancy [7]. This retrospective study used data from the medical records of deceased diabetics who had survived for at least 1 year after diagnosis.

The authors used linear, multiple regression analysis to estimate the effect of weight change 1 year after diagnosis (independent variable) on survival time (dependent variable). The analysis was done on both sexes combined and was statistically adjusted for sex, age, presence of ischemic heart disease, plasma glucose level, and use of oral hypoglycemic drugs.

The authors stated that on average, each 1 kg weight loss over the first year from diagnosis was associated with about 3-4 months increased survival [7]. From these results the authors inferred that, a 10 kg weight loss, which would be attainable by most patients, would predict the restoration of about 35% in life expectancy [7].

Unfortunately, these findings cannot be derived from the tables in the paper. In their Table 4 [7], the authors report the linear regression coefficients. To satisfy the underlying statistical assumption of normality in the dependent variable's residual values, the authors transformed the survival time variable by expressing it in terms of its square root. For the main effect of weight loss (kg), they report a statistically significant (P < 0.001) effect of 0.036 for the slope coefficient. The interpretation of this coefficient is that for every 1 kg loss of weight the survival time increases by 0.036 years¹/2. To convert this value to actual years, 0.036 must be squared, yielding a result of 0.0013 years or 0.015 months, rather than the 3 to 4 months reported by the authors in the text.

Discussion

Taken as a whole, the evidence from these six studies that weight loss in obese persons increases longevity is equivocal. Although each of the studies reported that weight loss was associated with decreased mortality, two of the studies did not provide data to support this finding [4, 7]. Two other studies found that mortality actually increased among some subgroups of persons who lost weight [5, 6]. Only the data from the first two life insurance studies consistently supported a positive benefit of weight loss on longevity [1-3].

The life insurance studies, however, may have had two potentially countervailing biases: greater medical screening for those who lost weight and no control for differences in smoking status between those who did and did not lose weight. In addition, the life insurance studies did not provide information on the amount or duration of weight loss that was associated with the reported mortality reductions. Also, little information was provided on the levels of obesity in the weight-loss and comparison groups; perhaps those who lost weight were initially less obese than those who did not lose weight. None of the six studies provided information on the types of weight loss methods used by the participants.

In our judgment, none of the studies adequately controlled for differences in prevalent disease between the comparison and weight-loss groups or adequately differentiated between voluntary and involuntary weight loss. This bias would reduce the likelihood of finding beneficial effects of weight loss. Even if volition of weight loss were assessed, however, it remains important to control for prevalent disease and smoking status because some successful dieters may lose weight (and keep it off) because they are ill, whereas others may voluntarily lose weight because they have been diagnosed with a weight-related illness.

Randomized, controlled trials would seem theoretically best to study this issue. However, because of the high prevalence of voluntary weight-loss attempts in overweight persons and because of ethical considerations [10, 11], it will be exceedingly difficult to ensure that, given a randomized trial of adequate duration to detect changes in mortality, the control group will not try to lose weight. Although observational studies require exceedingly careful design and interpretation, they will probably remain the most practical approach for the study of this important public health issue.