Introduction

Weight loss attempts are often triggered by knowledge about the harmful consequences of obesity, including social stigma (1,2) and adverse effects on health (3). However, weight loss attempts are not motivated by overweight alone: many women and girls try to lose weight even in the absence of overweight (4,5). Long-term weight loss is often unsuccessful, because body size, shape, and weight are strongly genetically influenced, with heritability estimates well over 70% in adolescents and young adults (6,7,8), and because permanent environmental and lifestyle changes are hard to implement in our increasingly obesogenic environment.

Despite the high frequency of weight loss and dieting attempts, surprisingly few scientific studies have focused on these behaviors on the population level. Although weight loss is generally seen as beneficial, it may also predispose vulnerable individuals to overeating and eating disorders. Thus, we aimed to explore whether intentional weight loss (IWL)¹ is associated with potentially harmful eating styles in a large and representative population-based sample of young adult twins.

Weight regulation in humans seems highly asymmetrical: weight gain is tolerated but weight loss is strongly defended against (9). However, there are few specific studies on the genetic contributions to weight loss. We aimed to explore the genetic and environmental contributions to IWL in this unique twin sample and to test whether the genetic liability affecting IWL is entirely shared with the genetic liability affecting BMI.

Research Methods and Procedures

Participants

The data reported are from FinnTwin16, a population-based study of five birth cohorts of Finnish twins born between 1975 and 1979 (10). Data collection protocol was submitted to and approved by ethics committees in Finland and Indiana. The baseline assessment was conducted by postal questionnaire at the age of 16 years, with follow-up at 17 and 18.5 years and as young adults. (In the fourth wave of data collection, the birth cohorts of the twins were contacted semiannually from 2000 to 2002, and the mean respondent's age was 24.4 years; range, 22 to 27 years.) The questionnaires assessed personality, social relationships, general health, and health habits. Response rates were high (90%) across all occasions. This study is based on responses in the 16-year and young adult questionnaires, comprised of 4667 twins (2545 women and 2122 men); information on IWL was available from 4662 subjects. Twins' zygosity was determined by standard items included in the baseline questionnaire (11,12) and was, when necessary, supplemented with photographs, fingerprints, and DNA marker studies. The twin pairs were classified as monozygotic, dizygotic, or unknown zygosity. The number of twin pairs where the zygosity of both twins was known was 2009; because of exclusion of opposite sex twin pairs and individuals with chronic weight-affecting illnesses and extreme obesity (exclusion criteria detailed below), our final sample for genetic analyses was 1258 like-sexed twin pairs.

Measures

IWL was assessed in our sample using the following question: "How many times during your life have you intentionally lost >5 kg of weight?" The responders in the category "never" formed the no-IWL group. Individuals responding "once" formed the 1-IWL group, and those responding "2 to 4 times" or "5 times or more" formed the 2-IWL group. The last two categories were combined because of a very small number of male respondents in these categories. For genetic analyses, IWL was dichotomized to "never" or "once or more" because of the skewed distribution of responses.

From genetic analyses, we excluded individuals suffering from chronic weight-affecting illnesses (N = 128), including mental retardation; cerebral palsy and other mobility disorders; malignancies; inflammatory bowel disease, chronic diarrhea, celiac disease; hypo- or hyperthyroidism; diabetes, and systemic lupus erythematosus.

Body Size

At 16, 17, and 22 to 27 years of age, the twins self-reported their current weight and height from which BMI at each age was calculated. Waist circumference was measured by the respondents using a tape measure that was mailed with the questionnaire, together with detailed instructions including a body drawing indicating the site of measurement. Agreement between actual measurements and self-report was 0.96 for height, 0.94 for weight, and 0.88 for waist circumference, as reported previously (7,13). For genetic analyses, BMI was categorized by sex-specific deciles. Because of an apparent nonlinear relationship of IWL and BMI beyond a BMI of 35 kg/m², men (n = 16) and women (n = 25) with BMI 35 kg/m² were excluded from genetic analyses. Women pregnant at follow-up (n = 112) were not excluded, because the weight question was formulated to assess weight before pregnancy.

Restrictive Eating/Overeating

Restrictive eating and overeating were assessed with the question: "Which of the following best describes you?" The choices were "It's easy for me to eat about the amount I need to" (normal eating, reference category); "I quite often eat more than I actually need" (overeating); "I often try to restrict my eating" (restrictive eating); and "At times, I'm on a strict diet, at others I overeat" (alternating restricting/overeating).

Eating Styles

To assess eating styles of the twins, a short 12-item questionnaire was devised (available from the authors on request): 5 items assessed snacking/grazing styles, 3 assessed health-conscious eating, 2 assessed emotional eating, 1 assessed externally cued eating (eating triggered by seeing food or advertisements of food, etc.), and 1 accessed night eating. Exploratory factor analysis was carried out, and the factors largely corresponded to the a priori groupings of items. Individual items will be referred to throughout this paper, but groupings are based on the factor structure. The factor pattern was similar in both sexes, with the exception of night eating, which emerged rarely in women only and was not included in eating styles explored in this paper.

Statistical Analyses

We studied associations between IWL and both obesity and overweight using cross-tabulations and the Pearson ² test of independence, corrected for clustered sampling of twins within pairs, which is expressed as an F ratio (14). For comparisons of means, correction for clustered sampling was expressed as adjusted 95% confidence intervals (CIs). The association of eating patterns and IWL was assessed using polytomous logistic regression models (15) that controlled for BMI. Odds ratios of all models were adjusted for correlated observations within twin pairs using the statistical software package Stata 8.0 (StataCorp LP, College Station, TX).

Twin Modeling

Sex-specific bivariate twin models were fit to assess the independent contribution of one variable (IWL) after accounting for its shared variance with another predictor variable (BMI). These estimate the genetic covariance of BMI and IWL and the covariance of common environmental influences on BMI and IWL.

From data on twins reared together, up to three of four possible parameters can be simultaneously modeled. We can estimate additive genetic effects (A), genetic effects caused by dominance (i.e., allelic interactions); common environmental effects (C) that are responsible for the environmental similarity within the twin pair; and unique environmental effects (E) that distinguish the environments of each twin from each other. Measurement error was included in the unique environmental effects. Bivariate modeling using Cholesky decompositions was performed using the computer program Mx (16).

In Figure 1, the additive genetic component of BMI (A_BMI) is assumed to be shared between IWL and BMI. This model also specifies an independent additive genetic component that affects IWL specifically (A_IWL). Similar factoring into two components is specified for BMI and IWL (C_BMI and C_IWL; E_BMI and E_IWL). We used the raw data maximum likelihood estimation option in Mx that enabled inclusion of both paired and unpaired twins. In accordance with the principle of parsimony, excess parameters, i.e., either additive genetic effects or common environmental effects, were removed from the model if the model fit (as measured by the likelihood ratio test) did not decrease with statistical significance.

Figure 1.

Sex-specific parameter estimates (with 95% CIs) from bivariate Cholesky decompositions of BMI and IWL. Full models contain the following parameters: A (additive genetic effects), C (common environmental effects), and E (unique environmental effects). In the best-fitting models, C effects could be removed without a significant decrease in model fit.

Full figure and legend (129K)

As mentioned above, very obese individuals (BMI 35 kg/m²) were excluded, because these genetic models are ill-equipped to handle nonlinear relationships. A further reason to exclude extremely obese individuals is that their high BMI may be caused by mutations that are not present in the general population (17), and their inclusion might lead to further biases from the genetic standpoint.

The distributions for men and women on IWL were clearly different, and the mean levels of IWL differed greatly by sex, implying underlying etiologic differences. Because of the low frequency of any IWL in men, the male variance was considerably smaller than the female variance. Were we to model the women in conjunction with the men using sex-limitation models that include information from opposite sex twins, an adjustment of the female data to accommodate the male distributional properties would be required, implying diminished power. Also, in the case of a bivariate Cholesky decomposition, overparametrization may become a problem, because at least one extra parameter is estimated if the sex limitation approach is used. For these reasons, structural equation modeling was accomplished using same-sex pairs only.

Results

IWL was much more common among women than men (Table 1). Overweight (25 BMI < 30 kg/m²) was significantly more common in men (25.7%) than in women (11.9%, p < 0.00001), as reported previously (7), but a much higher percentage of overweight women (76.0%) than overweight men (45.4%) had engaged in IWL (F = 87.9; p < 0.00,001). The rates of obesity (BMI > 30 kg/m²) were similar in women (3.7%) and men (4.1%; F = 0.39; p = 0.53) (7). However, IWL was significantly more common among obese women (81.9%) than obese men (69.0%; F = 6.1; p = 0.015). IWL in underweight twins was relatively uncommon: current BMI was <18.5 kg/m² in 169 (6.6%) women; 13.0% (n = 22) of them belonged to the IWL group. Current BMI was <20.0 kg/m² in 135 (6.4%) men; 5.9% (n = 8) of them belonged to the IWL group. Women and men in the 1-IWL and 2-IWL groups had significantly higher past, current, maximum, and ideal BMI and larger waist circumferences than their no-IWL counterparts; moreover, both women and men in the 2-IWL group had significantly higher current, maximum, and ideal BMI than their counterparts in the 1-IWL group (Table 1).

Table 1. - Means (95% CIs) of weight-related variables in no-IWL, 1-IWL, and 2-IWL individuals.

Full table

Eating Patterns

The frequency of various eating styles in different IWL groups is presented in Table 2; their relation was largely independent of BMI. As detailed in Table 3, where odds ratios adjusted for BMI are given, 1-IWL and 2-IWL individuals exhibited markedly more restricting, overeating, and alternating restricting/overeating than individuals in the no-IWL group. When BMI was controlled for, snacking patterns were nearly equal across all IWL groups; however, replacing meals with snacks was clearly associated with multiple IWL attempts. Avoiding fatty foods and calories was significantly more pronounced in individuals who had engaged in IWL than in the no-IWL group. Eating in response to visual and emotional cues was common in 2-IWL women but much less common in men.

Table 2. - Prevalences of eating styles and proportion of individuals with diverse weight loss histories: no-IWL, 1-IWL, and 2-IWL.

Full table

Table 3. - Sex-specific BMI-adjusted odds ratios (95% CIs) of eating patterns in no-IWL individuals vs. 1-IWL and 2-IWL individuals.

Full table

No statistically significant interactions between BMI and eating styles affecting IWL were found in women. However, in men, alternating restricting/overeating or externally triggered eating styles combined with high BMI were associated with increased IWL.

Bivariate Twin Models

There were no statistically significant differences in IWL or mean BMI between zygosity groups in either sex. Zygosity distributions and twin correlations of IWL are reported in Table 4. The exclusion of individuals with potentially weight-affecting chronic illnesses did not have a large effect on twin correlations (see Table 4), but to avoid potential confounding of heritability estimates of IWL and BMI by chronic illness, only data from healthy individuals were used.

Table 4. - Zygosity distribution of the participants (N = 4662), the frequency of IWL attempts among them, and twin correlations of IWL in the full sample (rIWLall) and after the exclusion of chronic diseases (rIWLfinal).

Full table

In addition to the full models specifying additive genetic effects, common environment, and unique environment or additive genetic effects, genetic dominance, and unique environment, we tested nine nested submodels (analyses not shown, available from authors on request) in each sex separately. In both men and women, simple models specifying only additive genetic effects and unique environment (AE models) provided the most parsimonious fit to our data. The full and best-fitting models for men and women are detailed in Figure 1. Heritability estimates of BMI were similar in both sexes. Overall, genetic factors seemed to exhibit a smaller effect on IWL in men than in women. The genetic covariance between IWL and BMI was 0.45 (95% CI, 0.41 to 0.52) in women and 0.38 (95% CI, 0.28 to 0.47) in men, implying that, although there is some overlap in genetic factors that influence BMI and IWL, there are also substantial unique genetic influences on each trait. In men, environmental effects influencing BMI and IWL were correlated, albeit to a modest degree. In women, genetic effects influencing BMI and IWL were moderately correlated, but the environmental correlation was negligible.

Discussion

Weight loss is a balancing act between maintaining strict control and losing control altogether. In our large sample of young adults, individuals engaging in IWL attempted to restrict food intake and avoid fatty and calorie-rich foods, but simultaneously exhibited disordered and unhealthy eating patterns. Similar associations have been reported in other populations (18). In this sample, individuals who had engaged in IWL were consistently heavier and had larger waist circumferences than their no-IWL counterparts. The direction of causality is, however, unknown. Although it is possible that a predisposition to higher BMI prompted IWL, it is also (and perhaps equally) plausible that repeated IWL attempts resulted in higher BMI.

As previously observed in many studies of adolescents and adults (19,20,21,22), IWL was much more common in women than men in our sample. Sex differences in dieting frequency and IWL may be true findings, but, alternatively, they may be biased for a number of reasons. We tried to minimize bias by giving a clear definition of IWL (5 kg) instead of assessing the harder-to-operationalize concept of "dieting." However, the perception of the IWL question may be sex-specific, and there may also be sex differences in reporting failed IWL attempts. Finally, the motives of IWL may be sex-specific: in women, dieting behavior has become almost normative (23), and IWL was much more associated with emotional aspects of eating than it was in men (24). In men, health concerns or lifestyle changes (such as divorce or becoming widowed) may trigger dieting (25). Thus, environmental factors that influence dieting may be very different in men and women.

To our knowledge, this is the first population study to address the components of genetic and environmental variance in liability for IWL. The results show that less than one-half of the genetic factors affecting the liability for IWL and BMI are shared. Perhaps, the genetics of weight homeostasis and weight loss are more complex than previously assumed. Our finding has implications for searches for genes associated with weight loss: finding genes that influence BMI may not be enough to establish genetically justified treatments of obesity, because other genetically driven processes are likely also involved.

Our BMI heritability estimates were consistent with previous literature (6,7). Individual environmental factors had a larger effect on IWL than on BMI. This is perhaps self-evident, given that dieting or restrictive eating is a learned behavior. Although permanent weight loss is notoriously difficult to maintain, these findings suggest substantial environmental etiology and are encouraging for interventions aimed at altering environment (e.g., decreasing unhealthy dieting behavior).

This study has several notable strengths: genetic and environmental influences on IWL were examined in a large population sample with an excellent participation rate. The studies of adult twins have shown that the BMI of twins and non-twins is comparable (26,27), and the conclusions derived from this study are applicable to the general population.

There are also several important limitations to this study. Our findings are cross-sectional, and causal inferences cannot be easily drawn. Despite excellent population coverage, some self-selection bias is possible in our study. As a probable result of multiple waves of participation by mailed questionnaire, women and individuals with a higher education (more than the mandatory 9 years) were somewhat overrepresented in the fourth follow-up sample. Furthermore, the reliability of dieting and food-related measures is uncertain because of underreporting or biased reporting (28): our findings are based on self-report and, thus, are liable to self-reporting bias. The presence of these potential sources of bias needs to be considered when interpreting our results.

Our definition of IWL is not fully interchangeable with the concept of "dieting"—surely there are individuals who have attempted to diet but have not managed to lose 5 kg. Nor does our definition of IWL reveal anything about how successful attempts to lose weight may have been; some individuals remain overweight despite multiple weight loss attempts. We do not know whether our measure truly distinguishes intentional from unintentional weight loss. In certain circumstances, these two can be indistinguishable: for instance, it is relatively common for young women to lose some weight after a break-up from a relationship—some of the weight loss may be caused by depressive mood and a lack of appetite, but weight loss in this context may also be used to bolster self-confidence in the process of getting ready for a new relationship (29).

However, unintentional weight loss caused by underlying disease or the aging processes is rare in this age group. We also excluded the few morbidly obese subjects: the conclusions from this study apply only to the segment of the population with BMI <35 kg/m²; weight fluctuations in morbid obesity have likely different genetic origins. Generalizations of our results to older, more overweight populations may, thus, be inappropriate; more research about the genetic epidemiology of IWL is clearly warranted, particularly in older adults.

Finally, this paper has explored the magnitude of genetic and environmental influences on IWL and the overlap of genetic factors influencing IWL and BMI, but based on these analyses, the extent to which the genetic and environmental factors observed are the same across men and women cannot be confidently stated.

IWL is associated with an increased risk of disordered eating patterns in women and men, although these risks may be outweighed by the health-conscious eating patterns also associated with IWL. However, distinct sex differences also exist in eating styles associated with IWL: men are much less affected by the emotional aspects of eating than women are. For women, food has many other functions than the strictly nutritional one. Thus, weight loss programs that address psychological needs otherwise fulfilled by eating may be particularly important for women.

The genetic architecture of IWL is also sex-specific. More importantly, genetic factors affecting IWL differ substantially from those affecting BMI: the majority of genetic factors affecting BMI are not shared with those affecting IWL in either sex. This finding implies that the search for susceptibility genes associated with weight loss needs to be extended beyond genetic factors affecting BMI alone.

Notes

¹ Nonstandard abbreviations: IWL, intentional weight loss; CI, confidence interval; A, additive genetic effects; C, common environmental effects; E, unique environmental effects.

References

1. Harris, M. B. (1983) Eating habits, restraint, knowledge and attitudes toward obesity. Int J Obes Relat Metab Disord. 7: 271–286.
2. Turnbull, J. D., Heaslip, S., McLeod, H. A. (2000) Pre-school children's attitudes to fat and normal male and female stimulus figures. Int J Obes Relat Metab Disord. 24: 1705–1706. | Article | PubMed |
3. Pi-Sunyer, F. X. (1991) Health implications of obesity. Am J Clin Nutr. 53: 1595S–1603S. | PubMed |
4. Strauss, R. S. (1999) Self-reported weight status and dieting in a cross-sectional sample of young adolescents: National Health and Nutrition Examination Survey III. Arch Pediatr Adolesc Med. 153: 741–747. | PubMed | ISI | ChemPort |
5. Wardle, J., Johnson, F. (2002) Weight and dieting: examining levels of weight concern in British adults. Int J Obes Relat Metab Disord. 26: 1144–1149. | Article | PubMed | ChemPort |
6. Maes, H. H., Neale, M. C., Eaves, L. J. (1997) Genetic and environmental factors in relative body weight and human adiposity. Behav Genet. 27: 325–351. | Article | PubMed | ISI | ChemPort |
7. Schousboe, K., Willemsen, G., Kyvik, K. O., et al. (2003) Sex differences in heritability of BMI: a comparative study of results from twin studies in eight countries. Twin Res. 6: 409–421. | Article | PubMed | ISI |
8. Bulik, C. M., Sullivan, P. F., Kendler, K. S. (2003) Genetic and environmental contributions to obesity and binge eating. Int J Eat Disord. 33: 293–298. | Article | PubMed |
9. Blundell, J. E., Gillett, A. (2003) Control of food intake in the obese. Obes Res. 9: (Suppl 4) 263S–270S.
10. Rose, R. J., Dick, D. M., Viken, R. J., Kaprio, J. Gene-environment interaction in patterns of adolescent drinking: regional residency moderates longitudinal influences on alcohol use. Alcohol Clin Exp Res. 25: 637–643. | PubMed |
11. Sarna, S., Kaprio, J., Sistonen, P., Koskenvuo, M. (1978) Diagnosis of twin zygosity by mailed questionnaire. Hum Hered. 28: 241–254. | PubMed | ISI | ChemPort |
12. Sarna, S., Kaprio, J. (1980) Use of multiple logistic analysis in twin zygosity diagnosis. Hum Hered. 30: 71–80. | PubMed | ChemPort |
13. Silventoinen, K., Sammalisto, S., Perola, M., et al. (2003) Heritability of adult body height: a comparative study of twin cohorts in eight countries. Twin Res. 6: 399–408. | Article | PubMed | ISI |
14. Rao, JNK, Scott, A. J. (1984) On chi-squared tests for multiway contingency tables with cell proportions estimated from survey data. Ann Stat. 12: 46–60. | Article |
15. Hosmer, D. W., Lemeshow, S. (2000) Applied Logistic Regression 2nd ed Wiley New York.
16. Neale, M. C., Boker, S. M., Xie, G., Maes, H. H. (2004) Mx: Statistical Modeling 6th ed. Department of Psychiatry, Virginia Commonwealth University Richmond VA.
17. Farooqi, I. S., Keogh, J. M., Yeo, G. S., Lank, E. J., Cheetham, T., O'Rahilly, S. Clinical spectrum of obesity and mutations in the melanocortin 4 receptor gene. N Engl J Med. 348: 1085–1095. | Article | PubMed | ISI | ChemPort |
18. Ackard, D. M., Neumark-Sztainer, D., Story, M., Perry, C. (2003) Overeating among adolescents: prevalence and associations with weight-related characteristics and psychological health. Pediatrics. 111: 67–74. | Article | PubMed |
19. Field, A. E., Austin, S. B., Taylor, C. B., et al. (2003) Relation between dieting and weight change among preadolescents and adolescents. Pediatrics. 112: 900–906. | Article | PubMed | ISI |
20. Neumark-Sztainer, D., Sherwood, N. E., French, S. A., Jeffery, R. W. (1999) Weight control behaviors among adult men and women: cause for concern? Obes Res. 7: 179–188.
21. Neumark-Sztainer, D., Rock, C. L., Thornquist, M. D., Cheskin, L. J., Neuhouser, M. L., Barnett, M. J. (2000) Weight-control behaviors among adults and adolescents: associations with dietary intake. Prev Med. 30: 381–391.
22. Meltzer, A. A., Everhart, J. E. (1995) Self-reported substantial 1-year weight change among men and women in the United States. Obes Res. 3: (Suppl 2) 123s–134s. | PubMed |
23. Rolls, B. J., Fedoroff, I. C., Guthrie, J. F. (1991) Gender differences in eating behavior and body weight regulation. Health Psychol. 10: 133–142. | PubMed |
24. Lluch, A., Herbeth, B., Mejean, L., Siest, G. (2000) Dietary intakes, eating style and overweight in the Stanislas Family Study. Int J Obes Relat Metab Disord. 24: 1493–1499. | Article | PubMed | ChemPort |
25. Sobal, J., Rauschenbach, B., Frongillo, E. A. Marital status changes and body weight changes: a US longitudinal analysis. Soc Sci Med. 56: 1543–1555. | Article | PubMed | ISI |
26. Korkeila, M., Kaprio, J., Rissanen, A., Koskenvuo, M. (1991) Effects of gender and age on the heritability of body mass index. Int J Obes Relat Metab Disord. 15: 647–654.
27. Rissanen, A., Heliövaara, M., Aromaa, A. (1988) Overweight and anthropometric changes in adulthood: a prospective study of 17,000 Finns. Int J Obes Relat Metab Disord. 12: 391–401. | ChemPort |
28. Lissner, L., Heitmann, B. L., Bengtsson, C. (2000) Population studies of diet and obesity. Br J Nutr. 83: (Suppl 1) S21–S24. | PubMed |
29. Jeffery, R. W., Rick, A. M. (2002) Cross-sectional and longitudinal associations between body mass index and marriage-related factors. Obes Res. 10: 809–815. | PubMed | ISI |

Acknowledgments

Data collection was supported by NIAAA Grants AA12502 and AA08315, with supplementary funds from the Academy of Finland (44069 and 201461), European Union Fifth Framework Program (QLRT-1999-00916 and QLG2-CT-2002-01254), the State Endowment for Helsinki University Central Hospital (EVO), and the Yrjö Jahnsson, Jalmari and Rauha Ahokas, Helsingin Sanomat, Biomedicum, and Finnish Cultural Foundations.