¹German Institute of Human Nutrition, Department of Epidemiology, Potsdam-Rehbruecke, Germany

²University of South Carolina, Department of Epidemiology and Biostatistics, Norman J Arnold School of Public Health, Columbia, South Carolina, USA

Correspondence to: A Kroke, German Institute of Human Nutrition, Arthur-Scheunert-Allee 114-116, D-14558 Bergholz-Rehbruecke, Germany. E-mail: kroke@mail.dife.de

Objective: To evaluate the influence of recent weight changes (weight gain, loss and cycling) on subsequent weight changes.

Design: Prospective cohort study with 2 y of follow-up. Data analysis with a polytomous logistic regression model.

Subjects: A total of 18 001 non-smoking subjects, 6689 men and 11 312 women, from the general population.

Measurements: Body height and weight measurements and interview data on lifestyle habits and medical history at baseline. For follow-up, self-administered questionnaires for assessment of body weight and incident diseases.

Results: Recent changes in body weight, that is weight gain, weight loss and weight cycling, were significant predictors of subsequent weight changes in both men and women after controlling for age, baseline BMI and several lifestyle and behavioural characteristics as potential confounding factors. Weight cycling before baseline was the strongest predictor of subsequent large weight gain (2 kg) with an odds ratio (OR) of 4.84 (95% confidence interval (CI) 3.34-7.02) in men. In women, prior weight loss was the strongest predictor of subsequent large weight gain (OR 4.77; 95% CI 3.63-6.03), followed by weight cycling (OR 3.02; 95% CI 2.15-4.25).

Conclusion: These data indicate the need for thorough weight history assessment to identify those who are most likely to gain weight. Effective weight control before the development of obesity or after intentional weight loss due to obesity should be a primary goal in the management of obesity.

International Journal of Obesity (2002) 26, 403-409. DOI: 10.1038/sj/ijo/0801920

weight change; weight cycling; prospective study

Introduction

The relation of body weight and weight change to chronic disease development is complex. A large body of evidence shows that obesity in adulthood is an established risk factor for several diseases such as diabetes, hypertension, cardiovascular disease and certain types of cancer.^1,2 These consistent findings, however, are modified by observations regarding further weight development. Several prospective studies have shown that weight loss is linked to an increased all-cause mortality,^3,4,5,6,7 and a recent prospective study observed the effects of weight change, in either direction, on mortality to be independent of the level of attained body weight after the weight change.⁸ Weight change was also found to be associated with an increased disease risk for myocardial infarction, stroke and diabetes,⁹ coronary heart disease,¹⁰ cholecystectomy,¹¹ or colon polyps.¹² Additionally, frequent changes in body weight, that is weight cycling, were observed to be associated with an increased all-cause mortality, and the development of chronic diseases.^{9,10,13,14,15}

For an effective prevention of obesity on the one hand, and of the excess disease risk associated with changes in body weight on the other hand, it is important to study the characteristics of those who experience changes in body weight. Previous analyses have identified several behavioural and lifestyle factors as predictors of weight change, such as alcohol consumption,^16,17 socio-demographic factors,^17,18,19 physical activity,^20,21,22,23 stress,²⁴ smoking habits,^17,25 dietary factors,^20,26 voluntary weight loss²⁷ or dieting behaviour.^28,29

The focus of the present analysis was to evaluate the influence of recent weight changes on subsequent weight change. We applied a polytomous logistic regression model based on generalised logits to estimate the effect of predictors for weight gain and for weight loss, with stable weight as the reference. The data used for this prospective analysis came from the EPIC (European Prospective Investigation into Cancer and Nutrition) Study Centre in Potsdam, Germany. The current cohort consisted of 18 001 middle-aged, non-smoking men and women with a mean follow-up time of 2 y.

Methods and materials

Study design

The EPIC-Potsdam Study³⁰ represents one study centre of the European multi-centre, prospective cohort study EPIC.³¹ Baseline recruitment in Potsdam, Germany, started in 1994 and ended in 1998. In total, 27 548 subjects, 16 644 women and 10 904 men, from the general population formed the final cohort population that is now contacted for follow-up every 2 y.³².

Subjects

For this analysis, data from those men and women of the EPIC-Potsdam cohort who responded to the first follow-up questionnaire until 13 September 2000, were used (n=26 121 (95%)). From this group, subjects who were current smokers at baseline, who gave up smoking during the 2 y before baseline, those with prevalent cancer, those with incident diseases at follow-up (cancer, stroke, myocardial infarction, chronic colitis, diabetes mellitus, immobilising fractures), and those taking appetite suppressants were not considered for the present analysis. Women who were pregnant or post-partum at baseline or during follow-up were also excluded. This resulted in a primary data set of 18 767 subjects for this analysis. After exclusions due to missing values in the exposure variables, the final data set consisted of 18 001 subjects, 6689 men and 11 312 women.

Data assessment

Baseline information on lifestyle and health-related variables, including questions on weight history and recent weight changes, was obtained from personal, PC-guided interviews by trained and quality-monitored personnel.³³ Self-administered questionnaires provided information on reproductive history and socio-economic status. Anthropometric measurements of body height and weight were performed by trained personnel.³⁴ Body height was measured with a flexible anthropometer to the nearest mm. Body weight was measured on a digital scale to the nearest 100 g with the subjects wearing only light underwear.

Weight changes before the baseline examination were assessed with questions about weight gain and weight loss of more than 5 kg during the last 2 y. For these changes in body weight it was asked whether the changes were intentional or unintentional. Using this information, weight cycling was defined as unintentional weight gain and intentional weight loss of more than 5 kg each during the 2 y prior to baseline assessment. Weight gain and weight loss were defined correspondingly, with these three groups of weight change being mutually exclusive. Physical activity was assessed with questions about engagement in activities such as cycling, walking, gardening, work, housework and sports as hours per week. The daily physical activity level (PAL) of each subject was calculated by multiplying each activity time by the metabolic equivalents published by Ainsworth and colleagues.³⁵

Follow-up information on incident diseases and current body weight was obtained from a self-administered questionnaire that was mailed to the cohort members approximately 2 y after the baseline examination.³⁶ Weight change per year of follow-up was calculated as reported body weight at follow-up minus body weight at baseline divided by years of follow-up.

Statistical analysis

Statistical analyses were performed with SASÒ, release 6.12 and 8.0. All analyses were performed separately for men and women. Means and standard deviations, or frequency distributions respectively, were used to describe the lifestyle and weight related characteristics of the study population. Results were considered to be statistically significant at the two-tailed <0.05.

For the outcome variable, five mutually exclusive categories of weight change during follow-up were defined: stable body weight was defined as £±1 kg/y, small weight gain as >1-<2 kg/y, large weight gain 2 kg/y, small weight loss as <-1->-2 kg/y, and large weight loss as £-2 kg/y. Polytomous logistic regression using the 'catmod' procedure in SASÒ was applied to model generalised logits for the four weight change categories with the stable weight group as the reference category. This approach allows the estimation of the effects of predictor variables on each of the above-mentioned weight change categories simultaneously, and avoids the crucial assumption of proportional odds needed for applying the 'logistic' procedure alternatively.

Since no variable selection method is available in the 'catmod' procedure, all possible predictors of weight change during the 2 y of follow-up period were tested separately in a model that included age and body mass index (BMI) at baseline. The initially tested predictors included age, BMI, educational attainment, physical activity level, energy intake, energy-adjusted fat intake, life satisfaction, health satisfaction, dietary changes due to overweight, prevalent diseases (diabetes, myocardial infarction, stroke, thyroid diseases, colitis, hypertension, gastritis, rheumatism, osteoporosis, gout), drugs assumed to influence body weight (neuro- and psychopharmacological drugs, corticoids, insulin), and additionally menopausal status among women. All significant predictors from this foregoing procedure were then added to a full model. In this full model all variables remained that were significant predictors for at least one of the four simultaneous parameter estimations related to the weight change categories. Excluding those variables no longer significantly related to all weight change categories resulted in the final multivariate model. To account for regression to the mean, baseline BMI was included in all models. In men, the polytomous logistic regression model included age, baseline BMI, educational attainment, energy intake, energy-adjusted fat intake, dietary changes due to overweight before baseline, dietary changes during follow-up, prevalent stroke, and use of anti-depressive or anti-epileptic drugs and insulin therapy during follow-up. The additional predictors of the final model for women were educational attainment, physical activity, energy intake, energy-adjusted fat intake, dietary changes due to overweight before baseline, dietary changes during follow-up, life contentment, prevalent diabetes, psychopharmacological drug use before baseline, pharmacological thyroid treatment, corticoid use and anti-depressive drug use during follow-up.

In order to remove possible effects of diagnosis and start of treatment of chronic diseases just before baseline, we examined separately the effect of chronic diseases diagnosed within the 2 y prior to the baseline assessment.

Odds ratios and 95% confidence intervals were calculated from the maximum likelihood estimates of the final models with an adaptation of the procedure described by Hosmer and Lemeshow.³⁷.

Results

The mean age of the study population at baseline was 52 y (range 24-69) in men and 49 y (range 19-70) in women. BMI at baseline was 27 in men and 26 in women, ranging from 17 to 55 and from 15 to 54, respectively. During the 2 y of follow-up men lost 0.2 kg body weight and women lost 0.4 kg of body weight on average. The frequency distribution of men and women across the five categories of weight change is presented in Table 1. According to the information about weight changes during the 2 y prior to the baseline assessment, 298 men and 446 women were classified as weight cyclers. In both men and women, slightly less than 60% maintained a stable body weight when baseline body weight was compared to the reported follow-up weight, whereas about 16% of either sex were categorised into a large weight change group.

Table 2 presents the means or frequencies of baseline characteristics for the five categories of weight change. In both sex groups, those who gained weight were slightly younger than those who lost weight and weight cyclers were most frequent among those with large weight gain. Those who lost weight prior to the baseline assessment were most frequently found among those with large weight gain at follow-up. Recent weight gainers at baseline were about twice as frequent among those who had a large weight loss at follow-up than among those who experienced large weight gain. However, a considerable number of women who gained weight before baseline experienced further weight gain during follow-up.

The polytomous logistic regression demonstrated that recent changes in body weight, that is weight gain, weight loss and weight cycling, were significant predictors of subsequent weight changes in both men and women. In the basic model, adjusting only for age and baseline BMI, the strongest predictor of subsequent large weight gain in men was weight cycling with an odds ratio (OR) of 5.68 (95% confidence interval (CI) 3.97-8.13), whereas prior weight loss was the strongest predictor of large weight gain in women (OR 5.29; 95%CI 4.19-6.69), followed by weight cycling (OR 3.32; 95% CI 2.38-4.62).

Adjustment for the various confounding variables reduced the magnitude of effects slightly to the estimates presented in Tables 3 and 4. In men, weight cycling before baseline remained the strongest predictor of subsequent weight gain with an OR of 4.84 (95% CI 3.34-7.02). In women, prior weight loss remained the strongest predictor of subsequent large weight gain (OR 4.77; 95% CI 3.70-6.15), followed by weight cycling (OR 3.02; 95% CI 2.15-4.25). Prior weight gain predicted subsequent large weight gain in women, but not in men. Large weight loss at follow-up was predicted in both sexes only by weight gain prior to the baseline assessment. Among men, weight cycling before baseline was a significant predictor of subsequent large weight loss in the basic model (OR 1.52, 95% CI 1.04-2.24). This effect, however, was removed after the adjustment for other confounding variables. In the final, multivariate models, weight cycling and prior weight loss showed no significant associations with small or large weight loss during follow-up in either sex.

Discussion

By using a polytomous logistic regression model this study identified recent short-term (2 y) weight changes as significant predictors of subsequent weight gain and weight loss. In this study population, weight loss was effected by unintentional prior weight gain, whereas weight gain was effected by prior intentional weight loss. An additional strong predictor of weight gain was weight cycling, defined as self-reported unintentional weight gain and intentional weight loss of more than 5 kg each during the 2 y prior to baseline assessment. This was observed in both men and women, and after adjustment for age, baseline BMI and other lifestyle and health-related variables.

Very few studies so far have focused on recent weight history as a predictor for subsequent weight change. Partial support for our findings comes from a prospective study that investigated the influence of dietary intake and physical activity on weight change.²⁶ Recent prior weight change was the strongest predictor of subsequent weight change in that study. Another study looked at characteristics of weight maintainers over a 5 y period and found normal-weight subjects to have had less variation in body weight during that period at annual examinations, whereas obese subjects were more frequently among those who fluctuated in their body weight during that time.³⁸ In contrast, Bild et al³⁹ studied correlates of weight loss in young adults and observed in univariate analyses weight loss and regain to be associated with subsequent weight loss.

With respect to weight cycling, relatively few studies have addressed the effects of weight cycling on weight development. Our results are consistent with the observations made in one prospective study which also reported that weight cyclers gained significantly more weight during follow-up than non-cyclers.⁴⁰

Changes in body composition related to weight cycling were reported from a cross-sectional analysis of 87 normal-weight women, where weight cycling status was significantly positively correlated with waist-to-hip ratio (WHR), and from a follow-up study of 32 obese subjects after a weight loss programme that reported changes in body composition after weight regain.⁴¹ In the latter study weight regain in the visceral fat area was smaller, whereas subcutaneous abdominal and hip fat areas were higher than before the weight loss.

Data on negative health effects of weight gain and obesity, respectively, are very consistent. Our results implicate weight cycling as a strong predictor of weight gain. Therefore, irrespective of the still unclear picture regarding the health effects of weight cycling, the assessment of weight cycling might help to identify those at high risk for weight gain. Since we defined only those subjects with intentional weight loss and unintentional weight gain as weight cyclers, it can be assumed that these subjects were unsuccessful in their attempt to maintain their lost weight.

A limitation of our study is that our weight change variable was derived from the difference between the measured weight at baseline and the self-reported weight at follow-up. The validity of self-reported body weight has been investigated in various studies. Most studies reported a high validity of self-reported as compared to measured weight.^42,43 However, under-reporting of self-reported weight was a frequent finding in these studies.

In the East German population among men no major changes in average BMI during the time between 1991 and 1998 have been observed, whereas among women a slight decrease in BMI was observed.^44,45 In our study population we observed a small decrease in body weight over time among both sexes¾either a true effect due to the specific population recruited into the cohort (healthy cohort effect) or due to under-reporting. If the latter is assumed the under-reporting might have resulted in some degree of misclassification. Those with larger weight gains might have been classified into the small weight gain or stable weight group. Weight losses might also have been exaggerated, putting people into the large weight loss group instead of the small weight loss group. To be more robust against misclassification errors, we used a categorical instead of a continuous variable as the dependent variable in our regression model, and have focused our attention on the extreme groups of large weight gain and loss. Since we have observed strong effects of predictors especially in the large weight gain group and under-reporting of weight would have decreased the true number of individuals to be classified into that group, our estimates might actually underestimate the magnitude of effect. In addition, in most instances, the significant findings for the large weight gain and weight loss group were supported by the observed associations in the corresponding small weight change groups.

As a strength of our study we regard the selection of a polytomous logistic regression model that enabled us to use the stable weight group as the reference category. In previous analyses, weight gainers were often compared to the rest of the study population that consisted of people with stable weight and weight loser. Alternatively, separate binomial logistic regression models were used. Application of a polytomous logistic regression is advantageous because it allows the simultaneous parameter estimation for a polytomous outcome variable and thereby a direct testing of hypotheses concerning the comparison of different parameters.

An additional strength of the current analysis is that at baseline we have collected information on intentionality of weight gain and weight loss. This allowed us to define our weight change and weight cycling variables by excluding unintentional weight loser and intentional weight gainer. As a limiting factor we have to point out that data on intentionality of weight changes were not available for the changes of weight during follow-up. However, in order to minimise the confounding effects of unintentional weight loss due to severe illnesses, we excluded the affected subjects from our analysis. We furthermore looked for effects of prevalent diseases diagnosed within the 2 y prior to baseline; however, no effects were found (data not presented).

In summary, our data indicates the need for thorough weight history assessment to identify those who are most likely to gain weight. Given the severe health risks associated with weight gain and obesity, avoidance of weight cycling, that is effective weight control before the development of obesity or after intentional weight loss due to obesity, should be a primary goal in the management of obesity.

Acknowledgements

The Potsdam part of the EPIC study was mainly supported by the Federal Ministry of Science, Germany, Grant No. 01 EA 9401, and funds from the 'Deutsche Krebshilfe'. This project was further financially supported by the 'Europe against Cancer' programme funded by the European Community, grant no. SOC 95 201408 05F02. We would like to thank all interviewers and technical assistants for their daily work in data assessment.

1 NRC . Diet and health¾implications for reducing chronic disease risk. National Academy Press: Washington, DC, 1989.

2 WCRF/AICR . Food, Nutrition, and the prevention of cancer: a global perspective. American Institute of Cancer Research: Washington, DC, 1997.

3 Lee IM, Paffenbarger RS Jr. Change in body weight and longevity. [See comments.]. JAMA 1992; 268: 2045-2049. MEDLINE

4 Manson JE, Willett WC, Stampfer MJ et al. Body weight and mortality among women. [See comments.]. New Engl J Med 1995; 333: 677-685. MEDLINE

5 Dyer AR, Stamler J, Greenland P. Associations of weight change and weight variability with cardiovascular and all-cause mortality in the Chicago Western Electric Company Study. Am J Epidemiol 2000; 152: 324-333. MEDLINE

6 Yaari S, Goldbourt U. Voluntary and involuntary weight loss: associations with long term mortality in 9,228 middle-aged and elderly men. Am J Epidemiol 1998; 148: 546-555. MEDLINE

7 French SA, Jeffery RW, Folsom AR, Williamson DF, Byers T. Relation of weight variability and intentionality of weight loss to disease history and health-related variables in a population-based sample of women aged 55-69 y. Am J Epidemiol 1995; 142: 1306-1314. MEDLINE

8 Mikkelsen KL, Heitmann BL, Keiding N, Sorensen TI. Independent effects of stable and changing body weight on total mortality. Epidemiology 1999; 10: 671-678. MEDLINE

9 French SA, Folsom AR, Jeffery RW, Zheng W, Mink PJ, Baxter JE. Weight variability and incident disease in older women: the Iowa Women's Health Study. Int J Obes Relat Metab Disord 1997; 21: 217-223. MEDLINE

10 Lissner L, Odell PM, D'Agostino RB, et al. Variability of body weight and health outcomes in the Framingham population. [See comments]. New Engl J Med 1991; 324: 1839-1844. MEDLINE

11 Syngal S, Coakley EH, Willett WC, Byers T, Williamson DF, Colditz GA. Long-term weight patterns and risk for cholecystectomy in women. Ann Intern Med 1999; 130: 471-477. MEDLINE

12 Bird CL, Frankl HD, Lee ER, Haile RW. Obesity, weight gain, large weight changes, and adenomatous polyps of the left colon and rectum. Am J Epidemiol 1998; 147: 670-680. MEDLINE

13 Folsom AR, French SA, Zheng W, Baxter JE, Jeffery RW. Weight variability and mortality: the Iowa Women's Health Study. Int J Obes Relat Metab Disord 1996; 20: 704-709. MEDLINE

14 Iribarren C, Sharp DS, Burchfiel CM, Petrovitch H. Association of weight loss and weight fluctuation with mortality among Japanese American men. [See comments.]. New Engl J Med 1995; 333: 686-692. MEDLINE

15 Reynolds MW, Fredman L, Langenberg P, Magaziner J. Weight, weight change, mortality in a random sample of older community-dwelling women. J Am Geriatr Soc 1999; 47: 1409-1414. MEDLINE

16 Liu S, Serdula MK, Williamson DF, Mokdad AH, Byers T. A prospective study of alcohol intake and change in body weight among US adults. Am J Epidemiol 1994; 140: 912-920. MEDLINE

17 Rissanen AM, Heliovaara M, Knekt P, Reunanen A, Aromaa A. Determinants of weight gain and overweight in adult Finns. Eur J Clin Nutr 1991; 45: 419-430. MEDLINE

18 Martikainen PT, Marmot MG. Socioeconomic differences in weight gain and determinants and consequences of coronary risk factors. Am J Clin Nutr 1999; 69: 719-726. MEDLINE

19 Lahmann PH, Lissner L, Gullberg B, Berglund G. Sociodemographic factors associated with long-term weight gain, current body fatness and central adiposity in Swedish women. Int J Obes Relat Metab Disord 2000; 24: 685-694. MEDLINE

20 Klesges RC, Klesges LM, Haddock CK, Eck LH. A longitudinal analysis of the impact of dietary intake and physical activity on weight change in adults. Am J Clin Nutr 1992; 55: 818-822. MEDLINE

21 Owens JF, Matthews KA, Wing RR, Kuller LH. Can physical activity mitigate the effects of aging in middle-aged women? Circulation 1992; 85: 1265-1270. MEDLINE

22 Sherwood NE, Jeffery RW, French SA, Hannan PJ, Murray DM. Predictors of weight gain in the Pound of Prevention study. Int J Obes Relat Metab Disord 2000; 24: 395-403. MEDLINE

23 Williamson DF, Madans J, Anda RF, Kleinman JC, Kahn HS, Byers T. Recreational physical activity and ten-year weight change in a US national cohort. Int J Obes Relat Metab Disord 1993; 17: 279-286. MEDLINE

24 Korkeila M, Kaprio J, Rissanen A, Koshenvuo M, Sorensen TI. Predictors of major weight gain in adult Finns: stress, life satisfaction and personality traits. Int J Obes Relat Metab Disord 1998; 22: 949-957. MEDLINE

25 Taylor CB, Jatulis DE, Winkleby MA, Rockhill BJ, Kraemer HC. Effects of life-style on body mass index change. Epidemiology 1994; 5: 599-603. MEDLINE

26 Colditz GA, Willett WC, Stampfer MJ, London SJ, Segal MR, Speizer FE. Patterns of weight change and their relation to diet in a cohort of healthy women. Am J Clin Nutr 1990; 51: 1100-1105. MEDLINE

27 Coakley EH, Rimm EB, Colditz G, Kawachi I, Willett W. Predictors of weight change in men: results from the Health Professionals Follow-up Study. Int J Obes Relat Metab Disord 1998; 22: 89-96. MEDLINE

28 French SA, Jeffery RW, Forster JL, McGovern PG, Kelder SH, Baxter JE. Predictors of weight change over two years among a population of working adults: the Healthy Worker Project. Int J Obes Relat Metab Disord 1994; 18: 145-154. MEDLINE

29 Korkeila M, Rissanen A, Kaprio J, Sorensen TI, Koskenvuo M. Weight-loss attempts and risk of major weight gain: a prospective study in Finnish adults. Am J Clin Nutr 1999; 70: 965-975. MEDLINE

30 Boeing H, Wahrendorf J, Becker N. EPIC-Germany¾a source for studies into diet and risk of chronic diseases. European Investigation into Cancer and Nutrition. Ann Nutr Metab 1999; 43: 195-204. MEDLINE

31 Riboli E, Kaaks R. The EPIC Project: rationale and study design. European Prospective Investigation into Cancer and Nutrition. Int J Epidemiol 1997; 26: S6-14. MEDLINE

32 Boeing H, Korfmann A, Bergmann MM. Recruitment procedures of EPIC-Germany. European Investigation into Cancer and Nutrition. Ann Nutr Metab 1999; 43: 205-215. MEDLINE

33 Kroke A, Bergmann MM, Lotze G, Jeckel A, Klipstein-Grobusch K, Boeing H. Measures of quality control in the German component of the EPIC study. European Prospective Investigation into Cancer and Nutrition. Ann Nutr Metab 1999; 43: 216-224. MEDLINE

34 Klipstein-Grobusch K, Georg T, Boeing H. Interviewer variability in anthropometric measurements and estimates of body composition. Int J Epidemiol 1997; 26: (Suppl 1) S174-180. MEDLINE

35 Ainsworth BE, Haskell WL, Leon AS et al. Compendium of physical activities: classification of energy costs of human physical activities. [See comments.] . Med Sci Sports Exerc 1993; 25: 71-80. MEDLINE

36 Bergmann MM, Bussas U, Boeing H. Follow-up procedures in EPIC-Germany¾data quality aspects. [In Process Citation.]. Ann Nutr Metab 1999; 43: 225-234. MEDLINE

37 Hosmer D, Lemeshow S. Applied logistic regression John Wiley and Sons: New York, 1989.

38 St Jeor ST, Brunner RL, Harrington ME et al. Who are the weight maintainers? Obes Res 1995; 3: (Suppl 2) 249s-259s. MEDLINE

39 Bild DE, Sholinsky P, Smith DE, Lewis CE, Hardin JM, Burke GL. Correlates and predictors of weight loss in young adults: the CARDIA study. Int J Obes Relat Metab Disord 1996; 20: 47-55. MEDLINE

40 Field AE, Byers T, Hunter DJ et al. Weight cycling, weight gain, and risk of hypertension in women. Am J Epidemiol 1999; 150: 573-579. MEDLINE

41 van der Kooy K, Leenen R, Seidell JC, Deurenberg P, Hautvast JG. Effect of a weight cycle on visceral fat accumulation. Am J Clin Nutr 1993; 58: 853-857. MEDLINE

42 Niedhammer I, Bugel I, Bonenfant S, Goldberg M, Leclerc A. Validity of self-reported weight and height in the French GAZEL cohort. Int J Obes Relat Metab Disord 2000; 24: 1111-1118. MEDLINE

43 Stevens J, Keil JE, Waid LR, Gazes PC. Accuracy of current, 4-year, and 28-year self-reported body weight in an elderly population. Am J Epidemiol 1990; 132: 1156-1163. MEDLINE

44 Bergmann E, Grimm J, Hoffmeister H et al. Die Gesundheit der Deutschen. RKI-Hefte 1995; 7: 117-128.

45 Bergmann KE, Mensink GBM. Anthropometric data and overweight. Das Gesundheitswesen 1999; 61: S115-S120. MEDLINE

46 Willett WC, Stampfer MJ. Total energy intake: implications for epidemiologic analyses. Am J Epidemiol 1986; 124: 17-27. MEDLINE

Table 1 Frequency distribution of men and women in five categories of prospective weight change; EPIC-Potsdam, Germany, n=18 001

Table 2 Means (STD), and frequencies,^a respectively, of selected baseline characteristics by categories of prospective weight change; women (n=11 312) and men (n=6689); EPIC-Potsdam, Germany

Table 3 Odds ratios^a (95% confidence interval) from one multivariate polytomous logistic regression model for categories of weight change with stable body weight as reference; men (n=6689); EPIC-Potsdam Study, Germany

Table 4 Odds ratios^a (95% confidence interval) from one multivariate, polytomous logistic regression model for categories of prospective weight gain and weight loss with stable body weight as reference; women (n=11 312); EPIC-Potsdam, Germany