Obesity is a multifactorial disorder deriving from genetic and metabolic factors as well as environmental factors, socio-economic and behavioural(Reference Hill and Melanson1, Reference Marti, Moreno-Aliaga, Hebebrand and Martinez2). These factors differ in their respective contributions to the obesity epidemic in recent decades(Reference Popkin and Doak3). Since genetic factors and their influence on the energy balance have not changed substantially(Reference De Castro4), they are unlikely to be responsible for the increase in obesity. Behavioural factors, however, have undergone important modifications that might account for the epidemic in obesity(Reference French, Store and Jeffery5).
Among the behavioural factors, sedentariness and eating play a major role(Reference Gutiérrez-Fisac, Royo-Bordonada and Rodríguez-Artalejo6). Nonetheless, analysis of the influence of energy intake and of the percentage of energy intake from specific nutrients has not yielded consistent results. Obesity has risen in countries where energy intake has increased sharply in recent years, as well as in others where it has decreased; similarly, it has risen in countries where fat intake has increased and in others in which it has decreased(Reference Gutierrez-Fisac, Banegas, Rodríguez-Artalejo and Regidor7, Reference Heini and Weinsier8). Consequently, research has focused on new factors related to eating behaviour. Pre-eminent among these are: food consumption frequency; temporal distribution of the meals throughout the day; skipping one of the main meals, particularly breakfast; and frequency of meals eaten away from home (‘eating out’). However, research on the influence of these factors on obesity is not conclusive: whereas some studies have reported no association between frequency of meals and obesity(Reference Summerbell, Moody, Shanks, Stock and Geissler9, Reference Kant, Schaztkin, Graubard and Ballard-Barbash10), others reported a high frequency of food-consumption episodes as having a protective effect on obesity(Reference Fabry and Tepperman11–Reference Sánchez-Villegas, Martinez-Gonzalez, Toledo, Irala-Estevez and Martinez15). Furthermore, individuals who skip breakfast register a higher frequency of obesity(Reference Ma, Bertone, Stanek, Reed, Hebert, Cohen, Merriam and Ockene13), though this could due to the fact that having no breakfast is associated with worse diet quality(Reference Sjoberg, Hallberg, Hoglund and Hulthen16). Indeed, having no breakfast is an ineffective way of losing weight(Reference Cho, Dietrich, Brown, Clark and Block17).
One of the factors with greatest interest is eating out. Some studies have shown a relationship between body weight and the frequency of food consumption at restaurants, particularly fast-food establishments(Reference Kant and Graubard18, Reference Duffey, Gordon-Larsen, Jacobs, Williams and Popkin19). Among the reasons given for such association were the higher energy intake due to the larger size of portions, or the high energy density of certain foods served in many restaurants(Reference French, Harnack and Jeffrey20, Reference Diliberti, Bordi, Conklin, Roe and Rolls21).
Almost all previous studies on these issues have been conducted in Anglo-Saxon countries, particularly the USA. The model of socio-economic development that underlies some of the changes in eating behaviour, such as the increase in food consumption away from home, is common to the entire Western world. However, the influence of these factors on obesity could vary between populations, given the enormous differences in dietary patterns between countries, for instance between the Mediterranean and those of Northern Europe or America.
This study examines the association between several eating behaviours and obesity in a representative sample of the adult population of Spain; in addition, it analyzes the influence of socio-demographic and lifestyle factors on such association.
Participants and methods
Data were drawn from the 1999 Survey on Disabilities, Impairments and Health Status (Encuesta Sobre Discapacidades, Deficiencias y Estado de Salud) which covered a representative sample of the non-institutionalised Spanish population. Study participants were selected through two-stage stratified sampling. First, census sections were randomly selected, stratified by town size and socio-economic level of the households. Second, random-start systematic sampling was used to select family dwellings, where one person was chosen at random to answer the questionnaire. A total of 69 555 interviews were conducted at the participant's households by trained personnel. We restricted our analyses to the 35 190 individuals aged 25–64 years. After excluding 216 subjects with missing data on some variable of interest, the final number of participants included in the analyses was 34 974.
The dependent variable was obesity. This was estimated from the BMI, calculated as weight in kilograms divided by the square of the height in metres (kg/m2). Self-reported weight and height were obtained with the following question: ‘What is your weight and height without shoes and clothes on?’ Obesity was defined as BMI ≥ 30 kg/m2.
The main independent variables were some eating habits in the 6 months preceding the interview. For each of the three main meals (breakfast, luncheon, dinner), information was obtained on whether the meal was eaten regularly, and whether it was eaten at home or away from home. Accordingly, each of the three meals was classified as: eaten regularly at home; eaten out; not eaten. Also, data were collected on eating frequency, which refers to the number of meals per day, categorised as follows: three or four meals (including the three main meals and afternoon tea); two meals (two of the main meals); one meal (one main meal); and several times per day (eating small amounts of food many times over the course of the day).
Information was also obtained for a number of potential confounders. Among these, there were socio-demographic variables, such as sex, age, size of town of usual residence ( < 10 000, 10 001–50 000, 50 001–500 000 and >500 000 inhabitants), and education, classified into low level (no formal education and primary education) and high level (secondary and university education). Also, there were lifestyle variables, such as smoking, with the following categories: non-smokers (neither smoke nor have ever done so previously), ex-smokers (do not smoke but did so previously) and smokers (smoke daily or occasionally); alcohol consumption, with participants classified as: abstainers (do not consume alcohol), occasional drinkers (consume alcohol once per week or less), frequent drinkers (consume alcohol two to six times per week) and daily drinkers (consume alcohol daily); and leisure-time physical activity, in two categories: sedentary (no physical exercise during leisure time) and active (some physical activity occasionally or several times per week or month). Finally, self-perceived health was classified as good (very good or good) or poor (fair, poor or very poor).
The association between the principal independent variables and obesity was summarised with OR and their 95 % CI obtained from logistic regression. Four types of models were built: a crude model; an age-adjusted model; a model adjusted for age, health status and lifestyle variables; and a saturated model, which, in addition to the above variables, also adjusted for socio-demographic factors. Separate analyses were conducted for men and women.
Statistical significance was set at two-tailed P < 0·05. Analyses were performed with the SPSS version 12.0 software (SPSS Inc., Chicago, IL, USA).
Results
Table 1 shows the prevalence of obesity according to socio-demographic and lifestyle characteristics. Obesity increased with age, rising to 17 % in men and 20 % in women in the 55–64 age group. Obesity was more prevalent among sedentary individuals, ex-smokers, subjects reporting poor health status, and those with a low educational level.
Number of individuals, number of obese subjects and prevalence of obesity according to demographic and lifestyle characteristics in Spanish men and women aged 25–64 years*
* For details of procedures, see Participants and methods.
Table 2 shows the distribution of obesity by the eating behaviours studied. Obesity was more prevalent among subjects who usually skipped breakfast, ate their midday meal at home, and reported not having dinner. Furthermore, among men, obesity was more prevalent in those who had two meals per day. Among women, in contrast, prevalence of obesity was higher in those who ate several times per day.
Number of individuals, number of obese subjects and prevalence of obesity according to eating habits in Spanish men and women aged 25–64 years*
* For details of procedures, see Participants and methods.
Tables 3 and 4 show the OR of obesity according to eating behaviour. Among men, absence of breakfast was associated with obesity, so that in the saturated model those skipping breakfast had an OR of obesity of 1·58 (95 % CI 1·29, 1·93) as compared with those having breakfast at home (Table 3). Similar results were obtained for women, with an OR 1·53 (95 % CI 1·15, 2·03) in the saturated model (Table 4). Eating breakfast out versus at home showed no association with obesity in either sex.
OR and 95 % CI of obesity according to eating habits in Spanish men aged 25–64 years
* Physical activity, smoking, alcohol consumption and health status.
† Educational level, size of town of residence and marital status.
OR and 95 % CI of obesity according to eating habits in Spanish women aged 25–64 years
* Physical activity, smoking, alcohol consumption and health status.
† Educational level, size of town of residence and marital status.
No relationship was observed between the main midday meal and obesity. Only in women who regularly went out for lunch was an inverse association found, with an OR of 0·52 (95 % CI 0·43, 0·73) in the crude analysis, and an OR of 0·78 (95 % CI 0·64, 0,95) in the age-adjusted model (Table 4). However, this association did not remain significant after adjustment for lifestyle (OR 0·88; 95 % CI 0·72, 1·07) and socio-demographic characteristics (OR 1·03; 95 % CI 0·84, 1·26).
As for dinner, no association with obesity was found in men. In contrast, women who had no dinner showed a higher prevalence of obesity than those who had dinner at home (age-adjusted OR 1·76; 95 % CI 1·29, 2·41). This association also reached statistical significance in the saturated model (OR 1·66; 95 % CI 1·20, 2·29) (Table 4). Dining out versus at home showed no relationship with obesity across the sexes.
Lastly, we found an association between food frequency and obesity. Compared to having three or four meals per day, having only two showed an age-adjusted OR of obesity of 1·67 (95 % CI 1·41, 1·99) in men and 1·35 (95 % CI 1·09, 1·67) in women. This association also held after additional adjustment for lifestyle and socio-demographic characteristics in men (OR 1·63; 95 % CI 1·37, 1·95) and women (OR 1·30; 95 % CI 1·05, 1·62). Eating several smaller-sized meals per day likewise displayed an association with obesity, with an age-adjusted OR of 1·50 (95 % CI 1·06, 2·12) in men and 1·63 (95 % CI 1·28, 2·09) in women. Among women, this association remained significant in the saturated model (OR 1·51; 95 % CI 1·17, 1·95). Among men, the association still held on adjusting for lifestyle (OR 1·43; 95 % CI 1·01, 2·03) but not after additional adjustment for socio-demographic characteristics (OR 1·42; 95 % CI 0·99, 2·01).
Discussion
In the present study, skipping breakfast and eating two times or less per day were associated with obesity in both sexes, while snacking was associated with obesity in women. However, no association was observed between obesity and having any of the main meals away from home.
As for breakfast, the present results coincide with those in other populations, where skipping breakfast has been associated with a higher BMI and obesity(Reference Ma, Bertone, Stanek, Reed, Hebert, Cohen, Merriam and Ockene13, Reference Cho, Dietrich, Brown, Clark and Block17). Of note is that we have observed this association in a country, such as Spain, in which breakfast is much lighter than in the USA, where most previous research was done. With respect to the mechanisms of this association, some studies in the USA and Finland have shown that having breakfast is usually accompanied by a better dietary macronutrient composition and certain healthy habits, such as regular physical exercise or alcohol abstention, which would reduce the risk of obesity(Reference Sjoberg, Hallberg, Hoglund and Hulthen16, Reference Song, Chun, Obayashi, Cho and Chung22). Individuals who skip breakfast also have an inadequate energy intake and a certain tendency to compensate for the energy needs not supplied at breakfast time with nutrient-poor, fat-rich foods(Reference Ruxton and Kirk23). Furthermore, in obesity treatment programmes, eating breakfast reduces dietary fat content and frequency of snacking, which in turn leads to weight loss(Reference Wyatt, Grunwald, Mosca, Klem, Wing and Hill24). The effect of breakfast protecting from obesity, coupled with the decreasing trend in breakfasting in some countries with a high prevalence of obesity(Reference Kant and Graubard25), suggests that a recommendation to have breakfast should be included in programmes addressing obesity in developed countries.
In the present study, women who did not have dinner were more likely to be obese. Other studies have reported a positive relationship between eating at night and obesity, especially when dinner is eaten late(Reference Ma, Bertone, Stanek, Reed, Hebert, Cohen, Merriam and Ockene13). An explanation for this association has been sought in the accumulation of energy in the form of glycogen after eating a carbohydrate-rich dinner late at night, which would prevent such energy from being rapidly used and, thus, favour its accumulation(Reference Keim, Van Loan, Horn, Barbieri and Mayclin26). It is likely that the present results are due to the cross-sectional study design. Specifically, obese women might reduce their daily intake, especially during dinner, with the intention of controlling their weight. In men, we did not find an association between obesity and not having dinner, possibly because men are less concerned than women about body image and overweight. This is highlighted by the lower percentage of men who undergo diets and other slimming treatments(Reference Bish, Blanck, Serdula, Marcus, Kohl and Khan27).
In comparison with having three or four meals per day, having only two meals was associated with obesity in both sexes. Frequent intake over the course of the day has shown a certain protective effect against obesity in some studies(Reference Fabry and Tepperman11, Reference Bellisle, McDevitt and Prentice12, Reference Drummond, Crombie, Cursiter and Kirk14). It has been suggested that frequent meals would lead to a relatively higher intake of carbohydrates and, by extension, a lower fat intake (a higher carbohydrate:fat ratio), which would reduce weight. The present results could also be due to the cross-sectional design, because individuals with a higher BMI could restrict the number of meals to control weight. Nevertheless, restriction of intake could paradoxically increase BMI, because it might coexist with episodes of unrestrained eating(Reference De Lauzon Guillain, Basdevant, Romon, Karlsson, Borys, Charles and FLVS Study Group28). In any case, the association between meal frequency and obesity is not consistent across the literature, since there are also studies that fail to find this association(Reference Summerbell, Moody, Shanks, Stock and Geissler9) or even report conflicting results between cross-sectional and longitudinal analyses of their data(Reference Kant, Schaztkin, Graubard and Ballard-Barbash10).
We observed a clear association between obesity and eating several meals involving smaller quantities of food than those consumed at the main meals. This pattern, which could be likened to snacking, has frequently been associated with obesity. Snacking would lead to a higher intake of saturated fats and of total energy, which would not be offset by a reduction in main meals(Reference Sánchez-Villegas, Martinez-Gonzalez, Toledo, Irala-Estevez and Martinez15, Reference Zizza, Siega-Riz and Popkin29). The association between obesity and snacking could also be due to the irregularity of this eating pattern, because there is recent experimental evidence that following an irregular diet had prejudicial effects on thermogenesis, fasting lipids and postprandrial insulin profile(Reference Farshchi, Taylor and Macdonald30). There are also studies, however, that report no association whatsoever between snacking and obesity(Reference Hampl, Heaton and Taylor31, Reference Drummond, Crombie and Kirk32). Hampl et al. (Reference Hampl, Heaton and Taylor31) failed to observe that snacking was accompanied by higher BMI(Reference Zizza, Siega-Riz and Popkin29), despite the greater energy intake among those indulging in snacking. Yet their study did not take physical activity into account, which might explain the unexpected results.
One of the most interesting results is the absence of association between obesity and regularly eating away from home, since it contradicts previous research. In such studies, the higher prevalence of obesity in those eating out may be due to a number of mechanisms. In the USA and UK, regularly eating away from home is associated with a higher energy and fat intake and a lower fibre intake(Reference Kant and Graubard18, Reference Nielsen, Siega-Riz and Popkin33). This results from the consumption of energy-dense foods, served in larger portions than those at home(Reference Diliberti, Bordi, Conklin, Roe and Rolls21). The higher energy intake also results from a greater social stimulus for food intake at restaurants, because individuals tend to eat more when in the presence of others(Reference De Castro34).
Most of the evidence on this association is based on eating at fast-food restaurants. The food of these establishments has been associated with high energy density(Reference Prentice and Jebb35), larger-sized portions, and obesity and its consequences(Reference French, Harnack and Jeffrey20, Reference Pereira, Kartashov, Ebbeling, Van Horn, Slattery, Jacobs and Ludwing36). Indeed, two studies have investigated the effect of food consumption in fast-food restaurants in a Mediterranean cohort, and they did observe an association between BMI and obesity(Reference Schroder, Fito and Covas37, Reference Bes-Rastrollo, Sánchez-Villegas, Gómez-García, Martínez, Pajares and Martínez-González38). Yet, despite the sharp rise in the number of these restaurants in Spain in recent years(39), fast-food restaurants might still be poorly frequented compared to the more traditional establishments, where the cuisine comes closer to the model of the Mediterranean diet, which has shown a protective effect against obesity(Reference Schroder40). This may explain the absence of association between eating out and obesity in Spain.
The present study has some limitations. First, because it was a cross-sectional study, causal inference is limited. For instance, skipping breakfast could be both a cause and consequence of obesity, since obese subjects might eliminate breakfast to lose weight.
Second, the questions on eating location have not been validated. However, the questions used are simple enough to expect not many problems with classification, because people should easily remember the place where they usually eat. Also, we are not aware of reasons why such potential classification errors might vary between the obese and the non-obese, or according to socio-demographic and behavioural characteristics. Thus, had such errors occurred, their influence on the present results is probably small.
Third, because data on eating behaviours were self-reported we cannot exclude some classification bias. There is no reason, however, to believe that such bias might have been different between the obese and the non-obese subjects. Indeed, obese subjects might well tend to conceal more frequent food consumption, so that the observed effect of snacking on obesity could be underestimated.
Fourth, we have not measured individuals' intake. However, controlling the study association for food and nutrient intake could be interpreted as over-adjustment, because food and nutrient intake is one of the mediators of the association between obesity and eating away from home. Had food and nutrient data been available, we would have been able to explore some mechanisms of the study association. Yet, we believe that the inability to examine such mechanisms does not reduce the importance of finding that eating out is not associated with obesity in Spain, in contrast to several Anglo-Saxon countries.
Fifth, we did not obtain data on household composition, despite it surely influencing eating behaviours. Nevertheless, we made an attempt to account for it, because our analyses adjusted for many socio-demographic (age, education, marital status) and lifestyle variables which are correlated with household composition. Accordingly, we expect that the effect of household composition on the present results should be relatively small.
Lastly, we acknowledge that parity is a predictor of overweight in women. In Spain, parity is also strongly associated with marriage, education and age. Our logistic models are already adjusted for these variables. Because further adjustment for parity did not materially change the results, we decided to exclude parity from the multivariate analysis. Moreover, the fact that ‘eating out’ and obesity were neither associated in men, supports that parity and other sex-linked variables are not crucial for the study association.
Notwithstanding these limitations, the present study is important because it is the first to investigate the relationship between eating out and obesity in a Mediterranean country. The absence of association between obesity and eating out would suggest the enormous variability in the impact of certain eating habits on obesity across countries.
Acknowledgements
This study was supported by FIS grant 06/0366. The funding body had no role in data extraction and analysis, writing of the manuscript, or in the decision to submit the paper for publication. There are no conflicts of interest. J. L. G.-F. designed the study and coordinated the writing of the article. A. C. M.-G. contributed to the analysis of this study and to the drafting of the paper. P. G.-C., J. R. B. and F. R.-A. contributed to the interpretation of the results and to the drafting of the paper. All authors contributed to the final version of the article.
