点击显示 收起
1 From the Julius Center for Health Sciences and Primary Care, University Medical Center, Utrecht, Netherlands (SM, YTvdS, DEG, and PHMP); the Netherlands Institute for Health Sciences, Erasmus University Medical School, Rotterdam, Netherlands (SM); and the Regional Cancer Registry, Utrecht, Netherlands (CHFG)
2 This study was conducted as part of MSc training by S Maru. 3 Reprints not available. Address correspondence to PHM Peeters, Julius Center for Health Sciences and Primary Care, University Medical Center, Utrecht, PO Box 85500, Room D01.335, 3508 GA Utrecht, Netherlands. E-mail: p.h.m.peeters{at}jc.azu.nl.
ABSTRACT
Background: The question of whether weight loss increases a persons longevity is important given the high prevalence of obesity and the large number of people trying to lose weight in industrialized countries.
Objective: This study assessed the effect of body mass index (BMI) and weight change (within 1 y of baseline) on subsequent mortality.
Design: This prospective cohort study had a median follow-up of 17 y. Subjects were 8100 women aged 5066 y who participated in a population-based breast cancer-screening project in the Netherlands between 1974 and 1977. Weight and height were measured during the first and second screening visits. All women were followed until 1996 for causes of death. Cox proportional hazard analyses were conducted for mortality from all causes (n = 1269), cardiovascular disease (CVD; n = 402), and cancer (n = 492).
Results: During follow-up we observed a statistically significant increased risk of mortality from all causes (hazard ratio: 1.4; 95% CI: 1.2, 1.6), cardiovascular disease, and cancer for women in the highest quartile of BMI (in kg/m2), 27.77. Mortality from all causes, CVD, and cancer did not increase significantly in women with a weight gain of 15%. Weight loss was also not significantly related to mortality from all causes, CVD, and cancer.
Conclusions: Mortality was highest for women in the highest BMI quartile (27.77). Extreme weight gain or weight loss within 1 y was not statistically significantly associated with subsequent higher mortality.
Key Words: Body mass index cancer mortality cardiovascular mortality cohort study overall mortality weight change
INTRODUCTION
Overall, there is agreement about the health risks of overweight and obesity. In fact, research has shown that excess body weight increases the risk of diabetes mellitus, ischemic heart disease, certain types of cancer, and premature mortality (1, 2). Obesity is a growing problem in industrialized countries. Several studies reported that the prevalence of obesity in adults is 1025% in most countries of Western Europe and 2025% in some states in the United States (3-5). The estimated annual number of deaths attributable to obesity (280 000) among US adults underlines the magnitude of this public health problem (6). Because the adverse effects of obesity are well known, weight loss efforts are being recommended and widely practiced. In 1996, 44% of adult US women reported that they were currently trying to lose weight (7). The long-term effects of weight loss, however, are not yet well known. Studies showed that successful weight loss improves several cardiovascular risk factors and diabetic control (8-10) but report little about the effects on mortality. Most observational cohort studies tend to show that higher mortality rates occur not only in adults who have gained excessive weight but also in adult who lose weight. The lowest mortality rates are generally associated with modest weight gains (11). Explanations for this finding were sought in the differential effects of weight change in healthy or diseased adults (12-17), young or older people (18, 19), intentional or unintentional weight loss (20), stable or fluctuations in weight loss (14, 21), and obese or nonobese people who were trying to lose weight (21-24).
The purpose of the present investigation was to study prospectively the long-term effects of body mass index (BMI) and weight change on all-cause and cause-specific mortality in a well-defined cohort of 8100 healthy Dutch women, aged 5066 y at baseline, in whom 1269 deaths occurred in a median follow-up period of 17 y.
SUBJECTS AND METHODS
Study population
Between December 1974 and September 1977, 14 697 (72%) of the 20 555 women born between 1911 and 1925 and living in Utrecht, Netherlands, were enrolled in the DOM (Diagnostisch Onderzoek Mammacarcinoom, or Diagnostic Investigation into Breast Cancer) cohort as part of a population-based breast cancer-screening program (25). The screening examination was conducted by trained technicians, who measured, among other things, height and weight. The screening examination was repeated from 1 to 5 times, and some women dropped out of the study between screening rounds. Self-reported questionnaires were also used to gather information on smoking habits (never, past, and currently <10, 1020, and >20 cigarettes/d) and on medication use during the previous 12 mo for the treatment of diabetes, hypertension, or heart disease. Of the 14 697 women enrolled, 2832 (19%) had at least one missing value for height, weight, or smoking habit at either the first or the second screening visit and were, therefore, excluded from the present study. An additional 3765 women were excluded because they reported using medication for hypertension, cardiovascular disease (CVD), or diabetes or for being on a restricted diet (salt-free, diabetic) at first or second examination. These exclusions resulted in a final study population of 8100 healthy women. The study was performed according to the responsible committee on human experimentation.
Outcome ascertainment
Vital status of the DOM cohort members was obtained from the Utrecht Municipal Registry on a monthly basis until 1 January 1996. Causes of death were ascertained from the subjects general practitioners and were coded according to the ninth revision of the International Classification of Diseases (ICD). The primary endpoint of the present study was total mortality (death from any cause). In addition, cause-specific mortality analyses were conducted for total CVD (ICD 390459) and total cancer (ICD 140208).
Predictors of mortality
Body mass index
Participants wore light indoor clothing and no shoes when weighed. BMI (in kg/m2) quartiles were used in the analyses as follows: quartile 1 (23.30), quartile 2 (23.3125.41), quartile 3 (25.4227.76), and quartile 4 (27.77).
The normal range of BMI for adults (not age-specific) is 18.525, according to the World Health Organization Expert Committee on Physical Status, which corresponds to the first quartile and most of the second quartile in our study population (26). The World Health Organization defines overweight in adults to be BMI = 25.029.9 and obesity to be BMI > 30.0. Of the 8100 women included in our study 939 (11.6%) had a BMI > 30, accounting for 46% of the fourth quartile.
Weight change
Body weight was measured at baseline and at the second screening visit, on average 12 mo later (range: 641). Weight change was calculated as the difference between weights measured at the first (baseline) and the second visit. Weight change was categorized into 7 empirically based categories: <5% change (stable; reference), 59% gain (moderate), 1014% gain (severe), 15% gain (extreme), 59% loss (moderate); 1014% loss (severe), and 15% loss (extreme). No information was available on how weight loss was achieved (eg, diet, physical activity, stress, and morbidity) or whether the loss was intended.
Statistical analysis
To assess the effect on mortality, we used Cox proportional hazard regression models to calculate hazard ratios (HRs) and 95% CIs for the BMI quartiles (reference: quartile 1) and for the weight-change categories (reference: stable with <5% weight change). Death from any cause was the endpoint. Losses to follow-up as a result of moving outside the study area and women alive at the end of follow-up (1 January 1996) were treated as censoring events. In the cause-specific mortality analyses (ie, cardiovascular), all other causes of death (ie, noncardiovascular) were treated as censoring events.
Adjustments were made for baseline information on age (continuous) and smoking (never, past, and currently <10, 1020, and >20 cigarettes/d). We also evaluated confounding by number of children, age at first childbirth, hormonal replacement use, oral contraceptive use, and menopausal status. None of these variables meaningfully changed the ß values, and they were, therefore, not included in the final models.
Tests for linear trends were performed for all-cause and cause-specific mortality. To control for type I error, we applied Bonferroni corrections to the level of significance ().
To test the hypothesis that effects of weight loss might be beneficial only for overweight and obese women, we conducted a separate analysis for women above the normal range of baseline BMI (BMI 25). We also tested interaction between weight change and BMI (continuous) by including their product term as an interaction term in the models. All analyses were done with SPSS software package for WINDOWS, version 10.1. (SPSS Inc, Chicago).
RESULTS
Baseline characteristics of the study population are provided in Table 1. Moderate weight loss was more frequent in the women in the highest BMI quartile (9%, compared with 56% in women in the lowest 3 quartiles; P = 0.001 chi-square test). In contrast, moderate weight gain was more frequent in the leanest women (10%, compared with 67% in women in the upper 3 quartiles of BMI; P = 0.001 chi-square test). The percentage of current smokers decreased with increasing weight: 35% in the first, 29% in the second, 28% in the third, and 25% in the fourth BMI quartile (P = 0.001 chi-square test).
View this table:
TABLE 1. Baseline characteristics of the study population according to BMI quartiles in 8100 healthy women aged 5066 y in 19741977
A total of 1201 (14.8%) women had moved outside the study area by the end of follow-up (1 January 1996). In addition, during 17 y of median follow-up (maximum: 21) or 140 468 women-years, 1269 deaths were ascertained (Table 2). Table 2 shows the crude mortality rates of total, CVD, and cancer mortality according to baseline BMI quartiles. After adjustment for age at baseline (continuous), smoking changes, and weight changes, mortality rates were highest in the fourth BMI quartile (Table 2). These women showed 40% higher overall mortality (HR: 1.4; 95% CI: 1.2, 1.6), with a greater number of deaths as a result of CVD (HR: 1.8; 95% CI: 1.3, 2.3) and cancer (HR: 1.5; 95% CI: 1.2, 2.0) than did women in the first quartile. Trend tests for total, CVD, and cancer mortality were all statistically significant.
View this table:
TABLE 2. Follow-up results and incidence density rates of cause-specific mortality, according to BMI quartiles
The mortality HRs according to categories of weight change, adjusted for age, smoking habits, and BMI at entry are shown in Table 3. Total mortality and deaths from CVDs and cancer were all not statistically significantly increased in women with weight gain of 15%. Weight loss was also not statistically significantly related to overall mortality, CVD mortality, and cancer mortality.
View this table:
TABLE 3. Mortality hazard ratio (and 95% CI) of death in 8100 women with regard to 1-y weight change1
Analyzing whether an effect of weight change could be observed in women with baseline BMI 25 did not essentially change the result (data not shown). The interaction between weight loss and BMI (continuous) at baseline was also tested. The P values for these terms for overall, CVD, and cancer mortality were 0.24, 0.15, and 0.61, respectively. This finding indicates that the absence of an effect of weight loss on mortality was observed in women with normal baseline weights as well as in women with weights above normal.
DISCUSSION
The present study showed statistically significant higher risks of all-cause, CVD, and cancer mortality for women who were in the highest BMI quartile at age 5066 y. In this study these women had a BMI 27.77. Extreme weight gain (15%) did not show significantly higher risks of all-cause, cardiovascular, and cancer mortality. Weight loss, compared with stable weight, was also not associated with a significantly increased risk of dying of CVD or a decreased risk of dying of cancer.
This study has some strong methodologic points, ie, a prospective design with a long follow-up period (17 y) for a large group (8100) of healthy women. For all women 2 anthropometric measurements were available taken by medical technicians with an interval of approximately 1 y. Although the range of the interval varied from 6 to 41 mo, 90% of all women were measured for the second time within a period of 1014 mo and 97% within 16 mo. Furthermore, results were adjusted for smoking, an important determinant of mortality. We did not take into account changes in smoking habits between first and second measurement, because only 67 women reported to have stopped smoking in between the 2 examinations. No one reported taking up smoking.
Frequently in cohort studies, deaths occurring in the first years of follow-up are excluded from analyses as an approach to adjust for confounding by occult disease present at baseline. It was shown that this is not necessarily an effective strategy for dealing with this type of confounding (27); therefore, we did not exclude any deaths. An analysis excluding deaths that occurred within the first 4 y of follow-up did not change the results.
We had no information on the stability of the observed weight change over a longer period of time. It was reported that most people who intentionally lose weight eventually regain it (28, 29). Moreover, some of the women who lost (or gained) weight could belong to the group of women whose weight fluctuates. Several studies showed that individuals with a large fluctuation in body weight are at higher risk of hypertension and of overall and CVD mortality (30-33). There are, however, populations that do not show this association (34, 35). Another study showed that the incidence of, and mortality from, CVDs is not higher with fluctuating body weight but that glucose tolerance is decreased (36).
We had no information on how weight loss was achieved. It could be unintentional as a consequence of underlying illnesses, although we tried to prevent this possibility by excluding women on medication. Unintentional weight changes also could be caused by (random) fluctuations in the sizes of fat mass and fat-free mass (37). If weight loss was intentional, it could have been achieved by a restricted diet or by increased physical activity. A study by Wood et al (38) showed beneficial effects on plasma lipoproteins from losing weight either by diet or by increased physical activity. Other studies reported that exercise is associated with unique favorable effects on CVD, cancer, and diabetes, independent of weight loss (39, 40). We were not able to study this effect, because no data on diet or physical activity were available for our population. A lack of fitness could be as important as (or even more important than) fatness, because lean physical unfit men show higher overall and CVD mortality than obese men who are fit (41-43).
The absence of information on whether or not weight change was intentional might be important (44), because unintentional weight loss is usually caused by illness. In a population-based study of American women aged 5569 y, 69% reported having at least one period in adulthood in which they had intentional weight loss, whereas 29% experienced an unintentional weight loss (45). A study also showed that intentional weight loss resulted in no effects on cardiovascular mortality, whereas unintentional weight loss was related to higher mortality (46, 47).
Although mortality from CVD with weight loss was not statistically significantly increased in our population, weight loss and subsequent higher mortality were reported in several populations (12, 14, 15, 18, 20, 48-55). Some of those studies seem to indicate that the effect of weight loss is differential with regard to baseline health characteristics. People who are obese and suffer from obesity-related health problems (ie, hypertension, diabetes) and, thus, have a high risk of dying of CVDs (or other obesity-related diseases) could benefit from weight loss. Clinical trials showed that intentional weight loss by obese individuals results in reduced blood pressure, reduced total serum cholesterol, and improved glucose tolerance (8, 9). Other studies also showed that weight loss greatly improves glycemic control in obese type 2 diabetic subjects (10, 56) and that such improvements can last for 13 y even if some weight is regained (38). However, we were not able to show beneficial effects of weight loss in women who were overweight at baseline.
The relation of weight loss and mortality can be influenced also by age (57, 58). Weight gain during (early) adulthood is detrimental, whereas overweight, although associated with mortality in all age classes (59), can be less of a problem for the elderly (19, 60, 61). Some older women can benefit from a slight overweight. The reason can be found in the association between leanness (or weight loss) and mortality from causes other than CVD (62). Of the 12 000 women participating in the Adventist Health Study, weight loss substantially related to higher risk of fatal respiratory disease (63-65). BMI was positively associated with overall mortality, although mortality was also higher in lean older women not on hormonal replacement treatment (65). Those researchers hypothesized that lean older women can experience a higher mortality risk because they have lower levels of adipose tissue-derived estrogen (65). In our population, only 720 women (8.9%) reported the use of hormonal treatment therapy, and adjustment did not change the results.
In conclusion, we observed that women who were classified in the upper quartile of BMI aged between 50 and 70 y showed highest risks of dying during subsequent years. Weight gain of 15% within 1 y was not statistically significantly associated with increased risk of mortality. Weight loss was also not significantly associated with CVD or cancer deaths.
ACKNOWLEDGMENTS
We thank B Slotboom for data processing.
None of the authors had any financial or personal interest in any company or organization sponsoring the research. S Maru analyzed the data and helped write and revise the manuscript. YT van der Schouw helped supervise the statistical analysis of the data and helped write and revise the manuscript. CHF Gimbrère provided advise throughout the study and helped write the manuscript. DE Grobbee provided advise on the statistical analysis and helped revise the paper. PHM Peeters designed the research question, oversaw the work, helped supervise the statistical analysis of the data, and helped write and revise the paper. S Maru conducted this study as part of MSc training.
REFERENCES