The implication of this result is unexpected and disturbing: If body weight for optimal survival increases with age, then some weight gain over time is not only permissible but can even be recommended for persons who are not overweight in early adult life. A test of this controversial conclusion is to examine persons on two occasions, to compute their changes in body weight, to follow these persons for specific outcomes, and to relate the observed weight changes to outcome. Such studies have the same potential confounders as those noted previously. Why was weight gained or lost? Although most potentially lethal illnesses lead to weight loss, weight gain is also possible (for example, edema in heart failure and inactivity due to illness). Weight may also be lost, however, in a purposeful program of health promotion that includes increased activity and a healthful diet. Unexplained weight loss in older persons is known to be an ominous symptom just as weight loss in elderly rodents is a harbinger of death.

Despite these complexities, results of studies that quantify weight change must be examined. This review examines only the effects of weight change on all-cause mortality. Studies of the effects of long-term weight change on diabetes, coronary heart disease, cancer, and cause-specific death will not be reported in detail.

Methods

Top Methods Results Discussion Author & Article Info References

Change in weight is sometimes reported in kilograms, sometimes in body mass index (BMI, kg/m²), and sometimes as a percentage. For ease of comparison, data were converted, when feasible, to metric BMI units. It was assumed that the average height of men was 1.75 m (69 in) and of women was 1.63 m [64 in]. This report is limited to all published population studies that 1) assessed change in weight as the independent variable, 2) determined overall mortality rate as the dependent variable, and 3) had not been reported elsewhere in these proceedings. Studies were identified through a comprehensive bibliographic search of the literature. Williamson and Pamuk [1] critically summarized the results of six published studies that specifically purported to show increased longevity in association with long-term weight loss. We report results from 13 other published studies [2–14]. Their salient characteristics are summarized in Table 1. Publications [2–14] should be referred to for more detailed descriptions. The order of presentation was determined alphabetically by author. Descriptors included in the tables are not repeated in the brief summaries that follow.

Each study examined participants at two distinct periods in life. In the first, changes in body weight were determined; in the second, mortality rates were determined. In addition, 7 of the 13 studies included a period of "temporal separation" in the analytic scheme (Table 1). Although monitoring for death started at the end of the weight change period, data from participants who died in the early years of the mortality follow-up period were excluded from the analysis. Thus in the seven studies that included a temporal separation period, persons who had an illness that had caused weight loss and subsequent death were omitted to minimize the effects of serious illness on weight change and death.

Results

Top Methods Results Discussion Author & Article Info References

The Paris Prospective Study [2] of civil servants working in the Paris Police Administration computed BMI at 20 years of age from weight at the age of military service and height measured in middle-age at study entry. Participants were divided into quintiles of change in BMI between 20 and 43 to 53 years of age. Minimal mortality (6.7 deaths per 1000 person-years) occurred in the third quintile of BMI change (gains of 2.5 to 4.4 kg/m²). Highest mortality rates (10.8 and 9.3) occurred in the lowest quintile of weight change (a gain of 0.5 kg/m² and in those who gained the most weight [> 6.5 kg/m²]).

The Dutch Longitudinal Study among the Elderly [3], conducted between 1955 and 1957, examined a "probability sample" of elderly men and women. They were re-examined between 1960 and 1962, and five categories of weight change were then computed. Vital status was ascertained in 1983. Longevity was expressed as the "realized probability of dying" [3]. Additional analyses were limited to only those participants surviving 2 or more years after the end of the weight-change period. Data were analyzed separately for persons 65 to 74 years old and for those 75 years or older. Separate analyses were done for men and women. Both age groups and sexes showed a quadratic relation between weight change and death, but none of these patterns was statistically significant.

The Western Electric Study by Hamm and colleagues [4] was directed primarily at fluctuations in weight and therefore used rather selective and unusual weight-change categories (see Table 1). Of 1959 employees studied, only 178 met the "no weight change" definition, and 133 met the "gain only" criteria. Only these groups provided data pertinent to this report. The weight-gain group had a relative mortality risk of 1.4 compared with the no-change group [95% CI, 1.0 to 2.1]. Weight gains averaged 37% and thus represented a serious degree of increase. No data on weight loss or on lesser degrees of weight gain were described in the report.

The Framingham Heart Study [5] examined residents of Framingham, Massachusetts, and excluded persons who reported smoking cigarettes at any visit. Participants were placed in one of four BMI change groups. Lowest mortality rates occurred in men and women who gained from 0% to 9% in BMI. Men and women who lost 10% or more and men who lost 0% to 9% had significantly increased mortality rates. An analysis of weight change in this population at an earlier age is presented later in this report.

In the Harvard Alumni Study by Lee and Paffenbarger [6], change in weight was monitored after participants had reached ages 35 to 74 years. Weight changes during an earlier phase of the life cycle were reported in a separate report. Participants were divided into five weight-change categories (loss of > 5 kg, loss of 1 to 5 kg, no change [± 1 kg], gain of 1 to 5 kg, and gain of > 5 kg). Relative risks for death (with the no change group set at 1.0) were 1.6, 1.25, 1.0, 1.0, and 1.3 for the five groups, respectively. They further showed similar patterns when analyses were stratified for initial BMI (more and less than 25 kg/m²). When participants were stratified by smoking pattern, nonsmokers showed significantly increased mortality rates in the two weight-loss groups and among those who gained more than 5 kg. Smokers showed the lowest mortality rate with 1 to 5 kg weight gain, and only those who lost more than 5 kg had a significant increase in mortality rate.

In the Baltimore Longitudinal Study of Aging [7], change in BMI among community-dwelling volunteers was computed as a slope for each participant from four consecutive measurements made during a period that averaged 3.9 years. A significant (P = 0.05) negative association was noted between weight change and mortality rate; that is, weight loss was associated with increased mortality rate. To test for a quadratic (U-shaped) association, we used further analyses to show that, when participants were divided by quintiles of BMI change (from a loss of > 1.1 kg/m² to a gain of > 0.8 kg/m²), a corresponding decrease was seen in relative risk for death. The values were 1.00 (referent), 0.94, 0.90, 0.78, and 0.75 for the five groups, respectively. A test for linear trend (orthogonal polynomials) showed the results analyzed by quintiles to be of borderline significance (P = 0.058).

The Gothenburg prospective studies [8] combined two separate population studies. In women, a multiple logistic regression analysis showed that change in BMI was negatively associated with death (weight loss predicted death) (P < 0.03); in men, results were similar (P < 0.001). Addition of smoking to the model did not change either result.

In addition to the weight-change analysis of persons 55 to 65 years old presented by Harris and colleagues [5], an analysis of the Framingham data based on weight changes occurring between 25 and 44-76 years of age has been published [9]. Highly significant effects of the slope of BMI change with time on total mortality rate were present in men and in women (P < 0.001 for both groups). These findings persisted despite the inclusion of five other risk factors for cardiovascular disease: smoking, serum cholesterol level, systolic blood pressure, glucose tolerance, and physical activity. Results remained statistically significant with temporal separation periods of either 4 or 6 years.

The Harvard Alumni Study by Paffenbarger and coworkers [10] measured the height and weight of incoming Harvard freshmen during the years 1916 to 1959. Participants were enrolled in a follow-up study at ages 35 to 74 years, when weight was obtained by questionnaire. They were then placed into quintiles according to change in BMI. The lowest quintile of BMI change included men who lost weight, did not change, or gained no more than 2.1 kg/m². The other three cut points were 2.8, 3.5, and 4.2 kg/m². The relative risks for death were 1.00 (referent), 0.84, 0.71, 0.77, and 0.77 for the five groups, respectively. Thus, the highest mortality rate occurred in Harvard men who failed to gain at least 6 kg from the time of their matriculation.

In the Honolulu Heart Study [11], men of Japanese descent were identified from draft registration files for the period 1939 to 1945. Baseline examinations were conducted between 1965 and 1968. The authors presented the effect of change in BMI on mortality rate for quartiles of BMI at 25 years of age. The quartiles of change in BMI were a decrease of more than 1.13, –1.13 to 1.12, 1.13 to 3.75, and 3.76 or more. Annual deaths per 1000 persons for the average of all men, regardless of initial BMI, was computed to be 11.2, 7.8, 7.8, and 8.4 for the quartiles of BMI change, respectively. Thus, mortality rate was lowest in those whose weight did not change or who gained up to 3.75 kg/m² (approximately 10 kg). Weight loss or higher weight gain was associated with increased mortality rate.

The Glostrup (Denmark) Population Study [12] limited the analysis of the effect of weight change to a multivariate Cox regression analysis that showed very few significant correlates of mortality. Change in weight had no effect, but a quadratic effect (high mortality rate at the extremes of weight change) would have been missed by this method of analysis.

The Kaiser Permanente Medical Care Program [13] analyzed persons who were thin or of average weight. Participants were asked to recall their "greatest adult weight," and weight change could therefore only be zero or negative. The mean period of weight change was not given. Persons were excluded if at enrollment they had evidence of cancer, heart attack, angina, diabetes, recent involuntary weight loss, limited or disabled worker status, use of heart medicine, proteinuria, or myocardial infarction as shown by electrocardiogram.

"Thin" persons were defined as those in the lowest BMI decile (< 22.3 in men and < 20.1 in women). The mean BMIs of those in combined deciles four and five, defined as those of average weight, were 25.1 for men and 22.8 for women. In both groups of men and in thin women, a progressive increase was seen in the mortality rate (per 1000 person-years) as weight loss increased (from 5.5 to 12.5 kg in thin men, from 6.5 to 13.6 kg in average-weight men, and from 2.9 to 9.8 kg in thin women). The lowest mortality rate occurred in those who were within 0.6 kg/m² of their maximum weights at enrollment. Average-weight women showed no consistent relation between weight loss on mortality rate.

In the Lipid Research Clinics Follow-up Study [14], participants included a 15% random sample of the Lipid Research Clinics Prevalence Study plus all persons with hyperlipidemia detected in that study [14]. Exclusionary factors included nonparticipation; limitation to white participants; use of medications (to treat hyperlipidemia, hypertension, or diabetes); use of gonadal hormones; pregnancy; apparent cardiovascular disease; extreme obesity (BMI > 50); and age less than 30 years at the beginning of the follow-up period. The detailed clinical "clean-up" exclusions at study entry "were an attempt to eliminate persons with overt disease that might have led to a change in their obesity status" before entry.

The effect of change in weight on mortality risk was analyzed in several ways. Cox regression coefficients were computed with adjustment for multiple variables. Weight change had no significant effect on mortality rate in women but was inversely related to mortality rate in men (P < 0.05) when quadratic analysis was done. Mortality rate in men decreased monotonically (but curvilinearly) from values 80% higher in those who lost 10% than in those who gained 15% in weight. More parsimonious models that adjusted only for age and smoking yielded essentially the same results. Furthermore, when adjustment was made for BMI at the beginning of follow-up (linear and quadratic terms), weight change had no effect on mortality rate in women; men, however, continued to show an inverse relation.

Discussion

Top Methods Results Discussion Author & Article Info References

Each of the 13 studies reviewed here have measured the weight of participants at two separate times during adult life so that change in weight could be computed. Participants were subsequently followed for a period of years to quantify the mortality rate.

The segment of the adult life during which weight change occurred varied widely: Several studies assessed change from early adulthood to middle age, whereas others studied changes occurring entirely during middle age or even at old age. Also, the duration of the weight change period varied widely (from as short as 3.7 years to as long as several decades). The period of follow-up also varied but not as widely (8 to 22 years). Seven studies introduced a "temporal separation" between the periods of weight change and of follow-up.

To reduce the effect of illness, a clinical evaluation was done at the end of the weight change period, that is, the beginning of the mortality follow-up period (see Table 1). The effectiveness of this technique was unclear in many of the studies. Some reports explicitly described the exclusion criteria, whereas others did not.

Cigarette smoking was assessed as a confounding variable in 10 of the 13 studies (see Table 1). Smoking is associated with lower body weight, and cessation of smoking generally results in weight gain. Smoking is also powerfully associated with death from many causes. Although a control for smoking was frequently included in the reported studies, stratified analyses for smokers and nonsmokers were rarely reported. Studies that have accounted for smoking habits have generally failed to show any significant effects of smoking on the association between weight change and death.

It is beyond the scope of this article to review associations of long-term changes in weight with the development of specific diseases; we note, however, that the following references provide data for disease incidence and mortality: Cardiovascular disease [2–911, 12, 14]; cerebrovascular disease [3, 6-8, 11]; cancer [2–911, 14]; diabetes mellitus [8]; pneumonia/influenza [3].

The issue of involuntary weight loss, generally acknowledged to be an important symptom of significant illness (compared with voluntary weight loss, either self-motivated or in response to physician advice), is important. The reported studies addressed this problem in several ways. First, clinical evaluations were done at the end of the weight change period (the beginning of the mortality follow-up period), and persons with specified identifiable illnesses were excluded from the study (see Table 1). Second, the "temporal separation" period of several years duration at the beginning of the mortality follow-up phase served also to exclude persons with significant [but perhaps undetected] illness at the time of enrollment. Weight changes were thus likely to be limited to those resulting from multiple societal, psychological, behavioral, and individual influences in population groups. Categorization into distinct voluntary and involuntary weight-change groups cannot be made based on the studies reported here.

Weight cycling or weight variability as an additional influence on mortality rate, independent of change in weight, has been reported in four studies [4, 7–9]. In the Western Electric Study [4], only 98 men (5% of those enrolled) were classified in a "gain and loss" category. They showed a gain of at least 10% and a loss of at least 10% during separate 5-year follow-up periods. Mortality rate was 50% higher in this group than in the weight stable group (P = 0.05). In the Baltimore Longitudinal Study [7], weight variability was assessed from the root mean square error of the slope of weight over time. In 846 men, weight variability had no effect on mortality rate (P > 0.96). In the Gothenburg studies of men and women [8] and in the Framingham Study [9], weight variability was measured by the coefficient of variation of the data points surrounding the mean value. Thus, a progressive, unidirectional change in weight would be computed to have a high coefficient of variation (that is, to show high variability). Those studies showed statistically significant increases in mortality rate in both men and women. The number of participants and the mortality follow-up periods are given in Table 1. The strength of the effect on mortality rate is difficult to determine in the Gothenburg studies [8]. In the Framingham report [9], when the mean BMI level and the slope of BMI with time were both considered, the relative risks of high variability compared with low variability in weight (upper and lower thirds of the population) were 1.74 for men and 1.45 for women. Thus, three of the four studies in men and both of the studies in women showed an effect on mortality rate. Better quantification of weight cycling is needed in future studies.

The current analysis of 13 studies of weight change now provides support for the concept that some degree of weight gain during adulthood is associated with lower all-cause mortality rates. The results of these analyses show that persons who gain some weight during adulthood survive longer, on average, than do those who maintain or lose weight Table 2 and that long-term weight loss, even of a mild or moderate degree, is generally associated with high mortality rate (Table 3).