PAPER Impact of overweight and obesity on health-related quality of life a Swedish population study

Transcription

1 (2002) 26, ß 2002 Nature Publishing Group All rights reserved /02 $ PAPER Impact of overweight and obesity on health-related quality of life a Swedish population study U Larsson 1, J Karlsson 2 and M Sullivan 2 * 1 The Nordic School of Public Health, Göteborg, Sweden; and 2 Health Care Research Unit, Institute of Internal Medicine, Sahlgrenska University Hospital, Göteborg, Sweden OBJECTIVE: To investigate the impact of overweight and obesity on health-related quality-of-life (HRQL) in the general population in western Sweden. DESIGN: Cross-sectional survey. SUBJECTS: A total of 5633 men and women aged y born in Sweden. MAIN OUTCOME MEASURES: Scale and summary component scores of the SF-36 Health Survey. RESULTS: Obese men aged y rated their HRQL lower than normal-weight men did on all four physical health scales of the SF-36 and on two of the four mental health scales. Obese women in the same age group rated their health worse than normal-weight women on three of the physical health scales. Thus, in younger men and women the analysis indicated a clearer negative association between obesity and physical health than between obesity and mental health. Obese women aged y rated their health worse on all scales than normal-weight women did, while obese men in this age group rated their health worse on only two SF-36 subscales physical functioning and general health perception. The massively obese men and women suffered from a poor level of HRQL. CONCLUSION: Not only does the level of obesity affect HRQL, the impact of overweight and obesity also differs by age and sex. The importance of aspects of both physical and mental health should be fully recognised. (2002) 26, DOI: =sj=ijo= Keywords: health-related quality of life; SF-36; overweight; general population; epidemiology Introduction Health-related quality-of-life (HRQL) has gained increasing interest as an outcome measure in clinical medicine and public health settings. It is particularly useful in studies on chronic diseases where the realistic goal of care is to make life as comfortable as possible. 1 There is consensus on a minimum set of core concepts to be included in the assessment of HRQL: physical functioning, mental health, social functioning, role functioning and general health perceptions. 1,2 There are in principle two paths to follow when choosing an instrument to assess HRQL, the specific and the generic, or a combination of the two. A disease-specific instrument is that are of the disease or the health state that are most *Correspondence: M Sullivan, Health Care Research Unit, Sahlgrenska University Hospital, SE Göteborg, Sweden. healthcare.research@medicine.gu.se Received 25 October 2000; revised 31 August 2001; accepted 16 October 2001 relevant to the patients. This is especially useful in clinical studies where a detailed assessment of the patients HRQL is necessary to fully evaluate a specific therapeutic intervention. On the other hand, a generic instrument is used to measure people s perceived HRQL across a wide range of diseases, complaints and health states and across different sociodemographic groups. It is thus possible to assess the relative burden of different conditions and also interpret that burden in health economic terms. Our knowledge about the impact of obesity on HRQL is increasing due to the recent development of standardised questionnaires with high reliability and validity. One attempt to assess HRQL in the obese is the ongoing Swedish Obese Subjects (SOS) study, where more than 6000 severely obese persons have been recruited to take part in a combined registry and intervention study. 3 HRQL is measured with a battery of instruments comprising established generic measures, an eating questionnaire and a study-specific module on obesity-related problems. The instruments were chosen to cover a broad spectrum of health impairments related to

2 418 obesity. An initial study showed that the obese rated their health and psychosocial functioning lower than a reference group. This study also displayed the consequences of weight reduction on HRQL. Compared to traditionally treated controls, the obese subjects who underwent weight-reduction surgery showed a markedly improved HRQL 2 y after surgery. 4 These results also suggest that obesity causally affects the level of the HRQL. This reasoning is in line with modern thinking on the association between obesity and psychopathology, where the latter now is seen as a consequence rather than a cause. 5 Studies on the relationship between obesity and HRQL usually include obese persons seeking treatment, ie patients. 6,7 However, there are reasons to believe that obese people who seek treatment differ in respect to HRQL from those who do not seek treatment. For example, it has been shown that obese patients report more symptoms of certain psychopathologies and more binge eating than obese people who are not in treatment. 8 Therefore, it is uncertain if results from HRQL studies with patient samples are generalisable to all obese persons. So far, four general population surveys using the same psychometrically sound HRQL instrument, SF-36, have examined the association between HRQL and obesity The purpose of the present paper is to further study the impact of overweight and obesity on HRQL in persons from the general population. Using a random sample from the general population, stratified by age and sex, and the generic HRQL instrument, SF-36 Health Survey, the present paper addressed the following questions: Does the impact of overweight and obesity on HRQL among persons in the general population vary with age and sex? Are both physical and mental aspects of HRQL affected? Is HRQL related to the level of obesity? What is the impact of body mass index (BMI) on HRQL after adjusting for age, sex, education, physical activity and sick leave=disability pension? Methods Design A cross-sectional survey was conducted in western Sweden to aid in the planning and organisation of health promotion activities and to support health policy decision-making. A 16-page questionnaire was mailed to a random sample (n ¼ ), aged 16 y or older, during the spring of Statistics Sweden, which is the public authority responsible for the national civil registers, performed the fieldwork. If the questionnaire was not returned within 2 weeks, a reminder was sent. In total three reminders were provided. The fieldwork took about 10 weeks. The questionnaire contained questions on HRQL (SF-36), allergy, smoking and exercise, personal economy, psychosocial and physical working conditions, social network and social support, demography, weight and height, and socioeconomic status including sick leave and disability pension. In addition to the questionnaire, information on age, sex, education, income, marital status, residence, nationality and country of birth was available from the civil registers used for selecting the subjects. The sample was stratified by the 25 municipalities in the area. During the data collection procedure, it was possible to identify and eliminate 204 subjects included in the registers but not in the target population. The majority of these persons were not able to answer the questions because of physical or mental illness (72%) or for other reasons, eg residing abroad. A total of 214 respondents returned the questionnaire with the identification label ripped off, making it impossible to add the registry information to this person s record. Of the persons in the target population receiving the questionnaire, 8751 (72%) returned it, including those with the identification label ripped off. The item non-response, ie returning the questionnaire without completing all questions, was small. In accordance with the SF-36 scoring procedure, imputation was applied for this subgroup of questions. 13 Measures Health-related quality of life, HRQL. HRQL was assessed with the generic and standardised SF-36 Health Survey. The SF-36 taps eight health concepts (scales) representing multiple operational definitions of health, including function and dysfunction, distress and well-being, and favourable and unfavourable self-ratings of one s general health status. 14 The physical functioning (PF) scale measures the ability to perform activities of daily living but also strenuous activities (10 items). Role-physical (RP) reflects the extent that physical health has a limiting effect on work or other activities (four items). Bodily pain (BP) concerns the amount of pain felt and whether it interferes with normal activities (two items). General health (GH) measures perceived general health status (five items). Vitality (VT) includes items on energy, tiredness, etc (four items). Social functioning (SF) concerns how social activities are affected by physical health or emotional problems (two items). Role-emotional (RE) reflects the extent that work or other activities are limited by emotional problems (three items). Mental health (MH) measures emotional well-being (five items). All scale scores range from 0 to 100, with 100 representing optimal physical functioning and well-being. The first four scales (PF, RP, BP, GH) are here referred to as the physical part of the HRQL, while the latter four scales (VT, SF, RE, MH) are referred to as the mental part of the HRQL concept. To reduce the number of outcome measures, two summary components have been extracted from the eight original scales. 13 The physical component summary (PCS) score and the mental component summary (MCS) score together account for 80 85% of the variance in the eight scales. 13 They are standardised through norm-based scoring to a normal distribution with a mean of 50 and a standard

3 deviation of 10. The Swedish version of the SF-36 has been shown to possess sound psychometric properties in different groups of people and is now widely used. Selected variables. Background variables analysed included BMI, sex, age, education level, physical activity in leisure time and sick leave more than 6 months or disability pension. BMI (kg=m 2 ) was calculated from questionnaire data and classified according to standards proposed by WHO. 18 Data on sex, age and educational level was obtained from the civil registers. Subjects The analysis was restricted to the age group y. Respondents over 64 were excluded since physical health status declines with increased age and thus interferes with the study purpose to evaluate the effect of obesity on HRQL. Non-response rates were also higher among the elderly. This subsample of the target sample comprised 9410 subjects and 6618 (70%) returned the questionnaire. The prevalence of overweight and obesity in this group is given in Table 1. Reference values for the Swedish population are also given. Compared with a Swedish nationwide interview survey (personal interview, self-reported weight and height) on living conditions 1996=1997, the prevalence of overweight (BMI kg=m 2 ) differed between 0.1 and 1.2 percentage points for men and women and the two age groups from our sample. The prevalence of obesity and massive obesity differed slightly more between the two studies. Analysis of data showed that the prevalence of overweight and obesity is larger in people born outside of Sweden and also that they suffer from lower levels of HRQL (data available from the authors upon request). Thus, to eliminate potentially confounding results due to ethnicity, individuals born outside of Sweden were excluded (n ¼ 881). Underweight persons were also excluded (n ¼ 104), leaving 5633 subjects in the study with valid values on sex, age and BMI. Statistical methods Different technigues were applied to adjust for non-response and selection bias. 19 Since auxiliary information (distribution of persons over strata categorised by sex, age and municipality) was available from the registers at the population level, it was possible to adjust not only for non-response bias but also for selection bias. Different techniques were adopted to eliminate and minimise the influence of confounders. After excluding persons born outside of Sweden and underweight persons, the remaining subjects were stratified into two age strata, following the practice used in the Swedish SF-36 manual. 13 That resulted in 1084 men and 1027 women in age stratum y and 1711 men and 1811 women in age stratum y. Dichotomising age in this manner was considered relevant based on the performance of physical health scales in relation to age in the normative database. When testing for differences between means, age was standardised in the y group and age and education was standardised in the y group. Education was not standardised in the younger group since many were still too young to continue their education on higher levels. The z-test was used to assess differences between groups. One-way analysis of variance (ANOVA) with contrasts was applied to test for a linear trend in the SF-36 scale and summary component scores across the three BMI categories (normal-weight, overweight and obesity). 20 In this test, data was not standardised for age (16 34 y) or age and education (35 64 y). Also, since the two role-functioning scales (RP, RE) deviate substantially from the assumption of normality, ANOVA analyses were not performed on these two scales. Fourteen men and 16 women were massively obese (BMI 40.0 kg=m 2 ) in the ages y (mean age 44 y). To be able to compare the HRQL in this small group with those of normal-weight (BMI kg=m 2 ), the method of matching was used. 21 Matching was carried out taking into account the potentially confounding factors sex, age and educational level. 22 A group of 924 normalweight individuals was selected and stratified into 14 strata 419 Table 1 Estimated prevalence of BMI categories by sex and age; random sample Men (%) Women (%) y y y y Classification BMI (kg=m 2 ) (n ¼ 1209) ULF a (n ¼ 2020) ULF a (n ¼ 1204) ULF a (n ¼ 2185) ULF a Underweight < Normal-weight b b b b Overweight Obesity c c c c Massive obesity > Total a Adopted from ULF, Swedish nationwide interview survey on living conditions 1996=97 (Johansson S-E, unpublished data, Statistics Sweden). b Underweight þ normal-weight. c Obesity þ massive obesity.

4 420 according to the above mentioned criteria (sex, age and education). To further investigate the relationship between HRQL and obesity, multiple regression analysis was used without classifying BMI or age. In the total sample the PCS and MCS scores were used as dependent variables and BMI, age, sex, physical activity and sick leave=disability pension were incorporated as independent variables. Since sex, physical activity and sick leave=disability pension are not continuous variables, they were represented with dummy variables. Regarding physical activity, subjects were divided into two groups: whether they were sedentary in leisure time (n ¼ 635) or not (n ¼ 4998). Among the 5633 subjects in the study, 283 had been on sick leave for more than 6 months or received disability pensions. To test for interaction effects between the main variables, 10 first-order (11) and 10 second-order (111) interaction terms were constructed. Age intervals were set to 10 y instead of 1 y in order to facilitate interpretation. Partial F-tests were used to test the significance of coefficients of the independent variables. Results The mean BMI was slightly less in the younger age group than in the older. In the age stratum y mean BMI was 24.1 kg=m 2 for men and 23.0 kg=m 2 for women. In the age stratum y mean BMI was 25.9 kg=m 2 for men and 24.7 kg=m 2 for women. Age stratum y Mean scores for the eight scales and the two summary components of SF-36 are displayed in Tables 2 and 3, for normal-weight, overweight and obese persons. As shown in Table 2, overweight men and women, aged y rated their health worse than normal-weight men and women. This was particularly true for PF, GH and the PCS. Compared with the normal-weight persons, obese men and women suffered from lower levels of HRQL on all of the physical health scales, although the difference was not significant for women on RP. Obese men also scored lower on VT and SF. Table 2 Mean SF-36 scale and summary scores (standard error) by sex and BMI category; standardised for age; y BMI Linear trend b (kg=m 2 ) (kg=m 2 ) (kg=m 2 ) F P-value Men n a a a Physical functioning (PF) 97.7 (0.3) 95.8** (0.8) 94.4** (1.5) Role-physical (RP) 91.1 (0.9) 89.5 (1.8) 83.2** (3.7) NA NA Bodily pain (BP) 85.1 (0.8) 82.6 (1.4) 78.2* (3.6) General health (GH) 84.0 (0.6) 80.0*** (1.2) 73.5*** (2.5) Vitality (VT) 69.9 (0.7) 67.8 (1.2) 61.0*** (2.9) Social functioning (SF) 91.0 (0.6) 91.6 (1.1) 84.4** (2.9) Role-emotional (RE) 89.4 (1.0) 87.9 (1.7) 82.8 (4.2) NA NA Mental health (MH) 81.3 (0.6) 80.9 (1.1) 79.4 (2.1) Physical component (PCS) 55.1 (0.2) 53.8** (0.5) 51.9*** (0.9) Mental component (MCS) 52.5 (0.4) 52.4 (0.6) 50.0* (1.4) Women n a a a Physical functioning (PF) 95.9 (0.4) 91.5*** (1.4) 90.8*** (1.9) Role-physical (RP) 85.8 (1.2) 84.4 (2.5) 79.4 (4.0) NA NA Bodily pain (BP) 80.8 (0.9) 77.6 (1.9) 72.5** (3.2) General health (GH) 80.6 (0.7) 76.0** (1.7) 73.0*** (2.5) Vitality (VT) 63.4 (0.8) 62.0 (1.7) 60.5 (2.7) Social functioning (SF) 85.8 (0.7) 88.0 (1.4) 84.2 (2.8) Role-emotional (RE) 82.3 (1.1) 85.1 (2.1) 82.5 (4.5) NA NA Mental health (MH) 77.2 (0.6) 76.1 (1.4) 76.8 (2.4) Physical component (PCS) 54.0 (0.3) 52.2** (0.7) 50.0*** (1.0) Mental component (MCS) 49.8 (0.4) 50.6 (0.8) 50.5 (1.6) *P compared with the normal-weight group ( kg=m 2 ). **P compared with the normal-weight group ( kg=m 2 ). ***P compared with the normal-weight group ( kg=m 2 ). a The number of individuals for different scales=components differ due to non-response. b Analysis of variance (ANOVA) test for linear trend by the use of a contrast. NA ¼ not applicable.

5 Table 3 Mean SF-36 scale and summary scores (standard error) by sex and BMI category; standardised for age; y BMI Linear trend b (kg=m 2 ) (kg=m 2 ) (kg=m 2 ) F P-value 421 Men n a a a Physical functioning (PF) 92.3 (0.6) 88.8*** (0.6) 87.0*** (1.4) Role-physical (RP) 85.5 (1.2) 81.8** (1.2) 89.3 (2.5) NA NA Bodily pain (BP) 77.4 (0.9) 74.4** (0.9) 78.6 (2.2) General health (GH) 76.7 (0.7) 73.6*** (0.7) 73.3* (1.8) Vitality (VT) 69.8 (0.8) 68.0 (0.8) 69.7 (1.9) Social functioning (SF) 89.7 (0.7) 88.9 (0.7) 89.4 (1.9) Role-emotional (RE) 90.9 (1.0) 88.1** (1.0) 88.9 (2.4) NA NA Mental health (MH) 82.3 (0.7) 81.3 (0.6) 81.6 (1.6) Physical component (PCS) 51.3 (0.3) 49.8*** (0.3) 50.6 (0.8) Mental component (MCS) 54.0 (0.4) 53.6 (0.3) 54.1 (0.9) Women n a a a Physical functioning (PF) 89.4 (0.5) 84.4*** (0.8) 79.3*** (2.0) Role-physical (RP) 82.2 (1.1) 76.8*** (1.7) 71.1*** (3.7) NA NA Bodily pain (BP) 74.2 (0.8) 69.1*** (1.2) 59.7*** (2.8) General health (GH) 76.4 (0.7) 70.9*** (1.0) 59.9*** (2.3) Vitality (VT) 66.6 (0.7) 62.0*** (1.1) 55.1*** (2.3) Social functioning (SF) 87.1 (0.7) 85.2 (1.0) 78.9*** (2.3) Role-emotional (RE) 87.8 (0.9) 86.4 (1.4) 81.8* (3.1) NA NA Mental health (MH) 80.4 (0.6) 79.7 (0.8) 73.0*** (2.1) Physical component (PCS) 50.5 (0.3) 47.8*** (0.5) 44.8*** (1.2) Mental component (MCS) 52.9 (0.3) 52.9 (0.5) 50.1** (1.2) *P compared with the normal-weight group ( kg=m 2 ). **P compared with the normal-weight group ( kg=m 2 ). ***P compared with the normal-weight group ( kg=m 2 ). a The number of individuals for different scales=components differ due to non-response. b Analysis of variance (ANOVA) test for linear trend by the use of a contrast. NA ¼ not applicable. A test for linear trend was used to see if the magnitude of obesity was related to HRQL. For men and women, the results of the ANOVA tests for a linear trend strengthened the impression of a negative association between obesity and HRQL. The association was more pronounced for the physical than mental health aspects. Thus, in y old men and women, the magnitude of obesity appeared to be related to impaired physical health status. For many of the scales, the decline in SF-36 scores was slightly larger in comparison between overweight and obese men than between overweight women and obese women. Age stratum y Results regarding the older age group, y, are shown in Table 3. Overweight men scored lower on the four physical scales (PF, RP, BP, GH) and the mental scale RE than the normal-weight men. In contrast, the obese men scored higher than the overweight men on all scales but PF and GH. Consequently, linear trends were found only for PF and GH and PCS. Overweight women rated their physical health (PF, RP, BP, GH) and VT lower than the normal-weight women. In contrast, obese women rated both their physical and mental health lower than the normal-weight women did. In particular, large differences between obese and overweight women were seen in BP and GH. The evidence for a linear trend, and thus for a correspondence between the magnitude of obesity and HRQL, was strong on all scales except for MH. In sum, the negative impact of obesity on HRQL was stronger in women than in men. The massively obese The massively obese (14 men and 16 women) were compared with a sex, education and age-matched sample of normalweight persons. Table 4 shows that the massively obese men and women reported poorer physical and mental HRQL status compared to the reference group. The massively obese persons had the poorest levels of HRQL of all groups in Tables 2 4.

6 422 Table 4 Mean SF-36 scale and summary scores (standard error); matched by sex, age and education; y (kg=m 2 ) 40.0 (kg=m 2 ) (n ¼ ) a (n ¼ 30) Physical functioning (PF) 92.8 (0.4) 69.6 (5.4)*** Role-physical (RP) 85.0 (1.0) 67.4 (9.9)* Bodily pain (BP) 77.0 (0.8) 59.8 (6.4)*** General health (GH) 77.4 (0.6) 55.0 (4.4)*** Vitality (VT) 66.5 (0.7) 54.3 (4.4)*** Social functioning (SF) 87.2 (0.7) 68.5 (6.3)*** Role-emotional (RE) 86.1 (1.0) 74.5 (6.5)* Mental health (MH) 79.8 (0.6) 69.4 (4.4)** Physical component (PCS) 52.0 (0.3) 42.6 (2.3)*** Mental component (MCS) 51.9 (0.3) 47.9 (2.2)* *P compared with the normal-weight group ( kg=m 2 ). **P compared with the normal-weight group ( kg=m 2 ). ***P compared with the normal-weight group ( kg=m 2 ). a The number of individuals for different scales=components differ due to nonresponse. Regression analysis Regression analysis, used to uncover the association between HRQL and BMI adjusted for age, sex, physical activity and sick leave=disability pension, confirmed the negative association between PCS and BMI (Table 5). Age, a sedentary leisure time and sick leave=disability pension were negatively related to PCS. In the total sample, women scored 1.6 units lower than men on average. The model explained 24% of the variance in PCS. Interaction effects suggested that physical health (PCS) was more impaired in women than men at increasing BMI levels (data not shown). With MCS as dependent variable, the coefficient for BMI was not significant, suggesting no relationship between mental health and weight. Being a woman, having a sedentary leisure time and being on sick leave or disability pension contributed negatively, while age was positively related to MCS. However, the proportion of explained variance was low (5%). Discussion In this paper we report the impact of overweight and obesity on HRQL in a general Swedish population. Our results suggest that overweight and obesity in young (16 34 y) men and women mostly have a negative affect on physical health, but not on mental health. Among the middle-aged (35 64 y), however, obese women reported impairments on all of the eight physical and mental scales, while obese men reported impairments on only two of the physical scales: physical functioning and general health perceptions. The relationship between obesity and mental health has been a subject of considerable debate for decades. In a review, Friedman and Brownell conclude that obesity is not associated with increased psychopathology in the general population. 23 However, they suggest that increased risk for developing psychopathology may be present for particular groups in the obese population. This hypothesis was confirmed in a recent population study that investigated the relationship between HRQL and obesity using the SF-36 and questions on chronic illness. 12 Consistent with our results, they conclude that overweight and obesity have a greater impact on physical than mental health. Further, the presence of obesity in conjunction with other chronic illnesses was associated with a significant deterioration in both physical and mental health, while obese persons with no concurrent conditions reported only a slight deterioration in physical health. After adjusting for the number of comorbid conditions, an independent association between obesity and impacts on physical but not mental health was established. Doll et al suggested that poor emotional well-being among the obese may be due to comorbidity rather than obesity per se. 12 In the present study, concomitant conditions were not controlled for and it is possible that the poorer HRQL among middle-aged women is attributable to a higher prevalence of comorbidities. On the other hand, our results indicate that gender and age are also important determinants of HRQL in the obese population. As would be expected, a sedentary leisure time and long-term sick leave or disability pension further substantially contribute to lower physical health. 13 In a general population study of y old women, Brown et al found that the obese (BMI kg=m 2 ) had lower scores on all of the eight SF-36-scales than their normal-weight counterparts (BMI kg=m 2 ). 9 However, the effect of obesity on the mental health scales was slightly weaker than in the present study. In a study of obese subjects seeking treatment (mean age 45 y, mean BMI 38 kg=m 2 ), it was found that women suffered from poorer Table y Regression models with physical component summary (PCS) and mental component summary (MCS) of the SF-36 as dependent variables; Coefficients for independent variables and standard error (s.e.) Dependent variable n Intercept BMI Age a Sex Physical activity Sick r 2 PCS *** (0.76) *** (0.03) *** (0.09) *** (0.22) *** (0.35) *** (0.54) 0.24 MCS *** (0.95) 0.02 (0.04) 1.11*** (0.11) *** (0.27) *** (0.43) *** (0.68) 0.05 *P < 0.05; **P < 0.01; ***P < a Age is measured by 10 y. Coding of dummy variables sex: men ¼ 0, women ¼ 1; physical activity in leisure time: non-sedentary ¼ 0, sedentary ¼ 1; sick ¼ sick leave > 6 months or disability pension: no ¼ 0, yes ¼ 1.

7 psychosocial health than did men. 24 It is generally believed that young women are particularly vulnerable to psychosocial distress due to their obesity since societal pressures against obesity are especially focused during adolescence and young adulthood, particularly among women. The reason why our study did not indicate any such detrimental effects in young women may be that mental health in general, as measured by the SF-36, is not heavily influenced by specific weight-related distress, such as body image disparagement. In line with this, Wadden et al found that adolescent obese girls were more dissatisfied with their weight than their non-obese peers, but did not display greater anxiety or depression. 25 An earlier population study of the impact of obesity on HRQL failed to demonstrate negative psychosocial consequences. 10 A possible explanation for this may be that the study sample was not stratified by sex or age. Thus, the effects of gender and age may have confounded the results. The general finding that women report more problems and poorer well-being than men is well known from other epidemiological studies. 26 In another population study (20 59 y), Han et al concluded that overweight subjects did not suffer from more non-physical problems, such as poor social functioning, adverse mental health, or role limitations due to emotional problems, than other subjects. 11 One possible reason why these results diverge from ours might be that Han et al did not distinguish between the overweight and obese and thus mixed the impact of overweight with that of obesity on HRQL. Our regression analysis provided further evidence for the strong relationship between physical health (PCS), and BMI even after adjusting for age, sex, physical activity and sick leave=disability pension. The independent variables were chosen from those shown in previous research to be related to HRQL. 13,27 The conclusion that BMI is less related to mental health (MCS) was confirmed. Age was negatively associated with physical health (PCS) and positively related to mental health (MCS). These results, however, may be due in part to a scoring artefact, where MCS is artificially inflated by low physical health scale scores. 28 Conclusions about mental health in the obese should therefore be drawn with caution. A fairly good proportion of explained variance, 24%, in the PCS regression model is noteworthy. All independent variables contributed substantially: age, being a woman, having sedentary leisure time and being on longterm sick leave or disability pension contributed negatively. Since the inclusion of interaction terms did not increase the percentage of variance explained combined with the difficulties in interpreting such results, the interaction models were not shown. An important question is whether the obese in the general population differ from obese patients regarding HRQL. No studies have yet investigated this question; however, some light may be shed on this subject by comparing our population data with patient data from Fontaine et al. 6 The massively obese (BMI 40 kg=m 2 ) patients reported lower HRQL than in our study, particularly on PF, RP, BP and VT. Comparisons of this kind are, however, difficult to interpret due to differences between studies, eg magnitude of obesity, sociodemographic status and cultural context. Still, the comparison suggests that massively obese persons seeking treatment report poorer HRQL. This observation is consistent with differences in health status found between outpatients with rheumatoid arthritis and women with rheumatoid arthritis in the population. 29 The prevalence rates in the present investigation give the impression that the population of western Sweden is fairly slim compared to many European countries and the US. 30,31 In order to determine if our sample was representative, we compared our prevalence rates with those obtained in the Swedish nationwide survey on living conditions. The difference between the two investigations was found to be very small. This fact in itself indicates the validity of the measurement of BMI in the two studies. The validity of self-reported height and weight has been confirmed in several studies. The bias resulting from self-report was in general small, and values of self-reported weight and height were highly correlated with true values. Weight underestimates were particularly prevalent among overweight and obese subjects, especially women and the elderly. One study found selfreport to underestimate BMI by 0.7 units among women (20 84 y) and by 0.4 units among men (16 84 y) compared to objective methods. 33 The prevalence of obesity was underestimated by 1.7 percentage units among women and by 4.2 percentage units among men. Although the possibility that subjects were misclassified into BMI classes cannot be entirely negated, this would have at most marginal influence on the HRQL results in this study. Generic instruments have the advantage of enabling HRQL comparisons across different diseases and sociodemographic groups. A disadvantage, however, is that they may not cover all essential health aspects pertinent to one particular disease. It was shown in our study that SF-36 had a high degree of responsiveness to obesity, in the sense that it managed to discriminate between people in different categories of overweight and obesity. Our results also stressed the importance of using a multidimensional instrument embracing different health aspects that do not necessarily correlate. To reduce bias related to potential confounding factors, sociodemographic variables such as ethnicity, age, sex and education were controlled for. However, it has also been shown that obesity is associated with diseases like diabetes mellitus, hypertension, coronary heart disease, respiratory disease and osteoarthritis. 35 In the present study we did not control for comorbidity or discriminate between the effect of obesity on HRQL and obesity-related comorbidity on HRQL. We did confirm, however, that persons on long-term sickleave or receiving disability pension for any reason had profoundly poorer physical and mental health. Our analyses showed that the impact on HRQL of overweight and obesity, with or without concomitant ailments, 423

8 424 differs between the young and the middle-aged, and between men and women. Longitudinal population studies with stratified samples would be of considerable value to learn more about the causal pathways between obesity and HRQL, and to what extent certain groups in society suffer. Acknowledgements The authors are indebted to the county council of Västra Götalandsregionen for permission to use the data. We gratefully acknowledge helpful comments on the manuscript from Charles Taft. We also extend our thanks to Helen A Doll, IJO referee of this paper, for challenging and constructive criticism. This article was made possible by support from the Medical Faculty, Göteborg University. References 1 Sullivan M, Karlsson J, Taft C. How to assess quality of life in medicine: rationale and methods. In: Guy-Grand B, Ailhaud G (eds). Progress in obesity research: 8. Libbey: London; pp Ware JE. Standards for validating health measures: definition and content. J Chron Dis 1987; 40: Sullivan M, Karlsson J, Sjöström L, Backman L, Bengtsson C, Bouchard C, Dahlgren S, Jonsson E, Larsson B, Lindstedt S, Näslund I, Olbe L, Wedel H. Swedish obese subjects (SOS) an intervention study of obesity. Baseline evaluation of health and psychosocial functioning in the first 1743 subjects examined. Int J Obes Relat Metab Disord 1993; 17: Karlsson J, Sjöström L, Sullivan M. Swedish obese subjects (SOS) an intervention study of obesity. Two-year follow-up of health-related quality of life (HRQL) and eating behavior after gastric surgery for severe obesity. Int J Obes Relat Metab Disord 1998; 22: Wadden TA, Stunkard AJ. Psychosocial consequences of obesity and dieting. In: Stunkard AJ, Wadden TA (eds). Obesity: theory and therapy. Raven Press: New York; pp Fontaine KR, Cheskin L, Barofsky I. Health-related quality of life in obese persons seeking treatment. J Fam Pract 1996; 43: Kolotkin RL, Head S, Hamilton M, Tse CJ. Assessing impact of weight on quality of life. Obes Res 1995; 3: Fitzgibbon ML, Stolley MR, Kirschenbaum DS. Obese people who seek treatment have different characteristics than those who do not seek treatment. Hlth Psychol 1993; 12: Brown WJ, Dobson AJ, Mishra G. What is a healthy weight for middle aged women? Int J Obes Relat Metab Disord, 1998; 22: Le Pen C, Levy E, Loos F, Banzet MN, Basdevant A. Specific scale compared with generic scale: a double measurement of the quality of life in a French community sample of obese subjects. J Epidemiol Community Health 1998; 52: Han TS, Tijhuis MAR, Lean MEJ, Seidell JC. Quality of life in relation to overweight and body fat distribution. Am J Public Hlth 1998; 88: Doll HA, Petersen SEK, Stewart-Brown SL. Obesity and physical and emotional well-being: associations between body mass index, chronic illness, and the physical and mental components of the SF-36 questionnaire. Obes Res 2000; 8: Sullivan M, Karlsson J, Ware J. SF-36 Hälsoenkät, manual och tolkningsguide. Sektionen för vårdforskning, Medicinska fakulteten, Sahlgrenska Sjukhuset: Göteborg; Ware JE, Snow KK, Kosinski M, Gandek B. SF-36 health survey manual and interpretation guide. New England Medical Center, The Health Institute: Boston, MA; Sullivan M, Karlsson J, Ware J. The Swedish SF-36 health survey I. Evaluation of data quality, scaling assumptions, reliability and construct validity across general populations in Sweden. Soc Sci Med 1995; 41: Persson L-O, Karlsson J, Bengtsson C, Steen B, Sullivan M. The Swedish SF-36 health survey II. Evaluation of clinical validity: results from population studies of elderly and women in Gothenburg. J Clin Epidemiol 1998; 51: Sullivan M, Karlsson J. The Swedish SF-36 health survey III. Evaluation of criterion-based validity: results from normative population. J Clin Epidemiol. 1998; 51: WHO Expert Committee. Physical status: the use and interpretation of anthropometry, WHO Technical Report Series no WHO: Geneva; Särndal C-E, Swensson B, Wretman J. Model assisted survey sampling. Springer: New York; Kleinbaum DG, Kupper LL. Applied regression analysis and other multivariable methods. Duxbury Press: Boston, MA; Rothman KJ, Greenland S. Modern epidemiology, 2nd edn. Lippincott-Raven: New York; Seidel JC, Flegal KM. Assessing obesity: classification and epidemiology. Br Med Bull 1997; 53: Friedman MA, Brownell KD. Psychological correlates of obesity: moving to the next research generation. Psychol Bull 1995; 1: Mannucci E, Ricca V, Bariculli E, Di Bernardo M, Travaglini R, Cabras PL, Rotella CM. Quality of life and overweight: the obesity related well-being (Orwell 97) questionnaire. Addict Behav 1999; 3: Wadden TA, Foster GD, Stunkard AJ, Linowitz JR. Dissatisfaction with weight and figure in obese girls: discontent but not depression. Int J Obes 1989; 13: Jenkinson C, Coulter A, Wright L. Short form 36 (SF36) health survey questionnaire: normative data for adults of working age. Br Med J 1993; 306: Laforge RG, Rossi JS, Prochaska JO, Velicer WF, Levesque DA, McHorney CA. Stage of regular exercise and health-related quality of life. Prev Med 1999; 28: Taft C, Karlsson J, Sullivan M. Do SF-36 summary component scores accurately summarize subscale scores? Quality Life Res 2001; 10: Ahlmén EM, Bengtsson CB, Sullivan BM, Bjelle A. A comparison of overall health between patients with rheumatoid arthritis and a population with and without rheumatoid arthritis Scand J Rheumatol 1990; 19: Seidell JC. Time trends in obesity: an epidemiological perspective. Horm Metab Res 1997; 29: Flegal KM, Carroll MD, Kuczmarski RJ, Johnsson CL. Overweight and obesity in the United States: prevalence and trends, Int J Obes Relat Metab Disord 1998; 22: Stewart AW, Jackson RT, Ford MA, Beaglehole R. Underestimation of relative weight by use of self-reported height and weight. Am J Epidemiol 1987; 125: Kuskowska-Wolk A, Karlsson P, Stolt M, Rössner S. The predictive validity of body mass index based on self-reported weight and height. Int J Obes 1989; 13: Roberts RJ. Can self-reported data accurately describe the prevalence of overweight? Public Hlth 1995; 109: Pi-Sunyer FX. Medical hazards of obesity. Ann Intern Med 1993; 119: