This is a preplanned analysis of data from the MOBIL-study (Morbid Obesity treatment, Bariatric surgery versus Intensive Lifestyle intervention, Clinical Trials.gov number NCT00273104), a non-randomized controlled study designed to compare the effects of bariatric surgery and intensive lifestyle intervention on various comorbidities, eating behavior and HRQL.
A total of 228 patients were screened, with 47 found not to be eligible. Of the remaining 181 participants 35 were not enrolled, leaving 146 in the study (Figure , flow of participants).
During the screening procedure all eligible patients underwent a thorough assessment at the Morbid Obesity Center by a multidisciplinary team consisting of an internist, a dietician, a physiotherapist and a trained “obesity” nurse. Patients were provided information about the possible risks and benefits of an operation and also encouraged to incorporate their own values and preferences into the decision-making process. If no contraindication against surgery existed, the patient and the physician together agreed upon the most appropriate choice of therapy; either surgical or conservative [10
A previous report [9
] showed a mean (SD) 1-year weight loss of 30 (8)% of initial body weight in the RYGB group and 8 (9)% in the ILI group. This corresponds to a mean (SD) loss of excess weight above 25 kg/m2
of 67% (18) and 20% (23) (P<0.001) respectively. The patients in the RYGB group lost a mean (SD) of 14.0 (4.1) BMI points and the ILI group 3.7 (4.2) BMI points. The number of subjects in the RYGB group and ILI group who either moved from being inactive to active (12 vs 18), stayed inactive or active (57 vs 32), or moved from being active to inactive (4 vs 5), differed significantly between the groups. Overall, there was a greater increase in the physical activity level of the lifestyle group compared to the surgery group.
The Norwegian Regional Ethics Committee for Medical Research approved the study protocol (S-05175), and the study was performed in accordance with the Declaration of Helsinki. All participants gave informed written consent before enrolment.
Variables, measurement and outcomes
The main outcome in the current analysis was the change in each of the five dimensions of HRQL, conceptualized as encompassing physical, mental and emotional dimensions, as well as the number and burden of obesity-specific symptoms. Since we aimed to compare the effect of two treatment methods, and since entering both weight loss and type of treatment into the same statistical analysis led to multicollinearity (r=.81), weight loss was excluded from the multiple regression analyses of covariance (ANCOVA). Variables that were considered possible confounders included age and BMI prior to intervention, age at the onset of obesity, physical activity and HRQL-score prior to intervention. Three questionnaires were used to measure HRQL: the Medical Outcome Study 36 – Item Short Form Health Survey (SF-36), the Obesity and Weight-Loss Quality of Life (OWLQOL), and the Weight Related Symptom Measure (WRSM).
Medical Outcome Study 36 – Item Short Form Health Survey (SF-36)
SF-36 is a commonly used generic measure of HRQL based on 36 questions or items [11
]. Item 2 is not included in the scoring of the instrument [13
]. The remaining 35 items form eight subscales (physical function, role physical, bodily pain, general health, role emotional, social function, vitality, and mental health) which can be combined into two summary scores; the physical and mental dimensions [14
]. As the validity of the subscales in morbidly obese patients is uncertain [15
] we studied the physical and mental dimensions. The calculations were performed as recommended by the scale authors [14
], using Norwegian norms [17
] and oblique factor scores to account for the correlation between the two HRQL-dimensions. The scores were calculated by multiplying each subject’s SF-36 subscale z score by its respective factor coefficient and then standardizing each to a T score with a mean of 50 and a standard deviation of 10 [14
]. Both scales were set to a range from 0–100, where higher scores indicate better HRQL.
Obesity and Weight-Loss Quality of Life (OWLQOL)
The OWLQOL [18
] primarily measures emotions and feelings [20
] which are believed to result from being obese and trying to lose weight. The instrument consists of 17 statements about weight-related feelings and emotions which are rated on a seven-point scale that ranges from 0 (“not at all”) to 6 (“a very great deal”). The 17 items of the OWLQOL form a scale ranging from 0–102, with higher scores indicating greater emotional HRQL.
Weight Related Symptom Measure (WRSM)
The WRSM [18
] measures 20 obesity-specific symptoms using two different sets of items. The first set assesses whether or not the patient is experiencing specific symptoms. The scoring of this set of items creates an additive scale summing up the number of symptoms, ranging from 0–20. The second set of items concerns the distress of the symptoms, with values from 0 (“not at all”) to 6 (“bothers a very great deal”). They form a symptom distress scale ranging from 0–120, where higher scores indicate greater symptom distress. Both the OWLQOL and the WRSM were obtained with permission from the Seattle Quality of Life Group, University of Washington.
In sum, the three HRQL questionnaires constitute five different measurements of HRQL; physical dimension (SF-36), mental dimension (SF-36), emotional dimension (OWLQOL), number of obesity symptoms (WRSM), and distress of obesity symptoms (WRSM).
Changes in scores between two time-points or groups can be statistically significant. An important follow-up question is whether the changes are clinically relevant. There are different approaches to addressing this. Here we have chosen the effect size (ES) to grade the efficiency of surgical versus nonsurgical treatment [22
Physical activity was assessed through structured interviews performed by registered dieticians. Time spent performing light (e.g. casual walking), moderate (e.g. brisk walking) and vigorous (e.g. jogging) intensity aerobic physical activities for periods of 10 minutes or more was recorded. Participants who performed 150 minutes or more per week of moderately intense aerobic physical activities were considered to be physically active, as were those participants who performed 60 minutes or more per week of vigorously intense aerobic physical activities [24
A total of 139 patients completed the MOBIL-study (Figure ). At baseline, all patients in both the RYGB group (n=76) and the ILI group (n=63) completed the three HRQL instruments. At 1 year follow up 62 (82%) participants in the RYGB group and 48 (76%) in the ILI group had completed the questionnaires. In order to assess the representativeness of the sample at the end of the study we used an independent samples t-test to compare differences between patients not completing the questionnaires at the end of the study versus completers. Patients who did not complete the questionnaires after 1 year (n=29) were comparable with those who did (n=110) with regards to baseline HRQL, gender, age, body weight, employment status, and weight loss after 1 year (data not shown).
During follow-up, patients allocated to RYGB were examined by a bariatric surgeon 6 weeks after surgery, while patients were seen by a dietician quarterly, usually in groups of 12–16. The patients in the ILI group were admitted to a rehabilitation center specializing in the care of morbidly obese patients. The aim was to attain a sustained 1-year weight loss ≥10%. Each patient was encouraged to increase their physical activity and to normalize eating habits. The program intended to increase each patient’s self-efficacy in dealing with their weight problem, as well as an improvement in self-esteem.
The 1-year lifestyle program comprised four stays at the rehabilitation center – three 5-day stays in weeks 1, 26, and 51, and a four-week stay from weeks 13–17 (Figure ). The daily schedule was divided between organized daily physical activity (3–4 hours) and various psychosocially-oriented interventions combined with a motivational approach both in group sessions and individual sessions (3–4 hours). These sessions were supervised by a medical doctor, nutritionists, physiotherapists and mental health-trained nurses. No special diet or weight-loss drugs were prescribed, but patients were encouraged to follow the guidelines of the Norwegian National Council of Nutrition [25
], which recommends that the daily intake of protein, fat, carbohydrate and alcohol should account respectively for 10–20, <30, 50–60, and <5% of energy consumed. In addition, the patients were asked to reduce their daily total energy intake, but not using calorie-counting. Outside of these stays patients were contacted by phone once every 2 weeks. They were also encouraged to self-monitor their eating habits and physical activities in a pre-fabricated diary, as well as to consult their general practitioner for weight measurement and follow-up every four weeks.
Schedule of stays during the 1-year intensive lifestyle intervention program at the rehabilitation centre.
Data are presented as mean (SD) or n (%) unless otherwise stated. Skewed data were transformed to approximate normality using natural logarithms. To assess the reliability of the HRQL-scales we calculated Cronbach’s alpha coefficients.
After applying Little’s test of randomness of missing data, missing values (SF-36:23.5%, OWLQOL:24.5%, WRSM:23.7%) were imputed using multiple imputation. The imputation model consisted of the HRQL-scores, physical activity at baseline and 1 year, and age of onset of obesity as predictor and imputation variables, and treatment, gender, age, baseline BMI, marital status, employment, and education as predictor variables. Through a fully conditional specification model, applying linear regression as the prediction method for scale variables and two-way interactions for categorical variables, we generated twenty complete datasets for each of the HRQL-scores with 10 iterations per dataset. The statistical analyses were performed on each complete dataset, and thereafter the multiple analyses results were combined to achieve single estimates. The combined estimates are presented. Observing the fraction of missing information, relative increase variance, and relative efficiency, the imputed data-sets (n=139) were comparable with the original data-set (n=110) in terms of the imputed variables (data not shown).
Within-group analyses in both groups were performed using paired samples t-test. Between-group comparisons at baseline were analyzed using independent samples t-test for continuous variables and χ2 for categorical variables.
Within groups ES was calculated as the mean HRQL change score between 1 year and baseline divided by the standard deviation of the baseline HRQL. Between groups ES was calculated as the difference in mean HRQL change score between groups at 1 year divided by the standard deviation of baseline HRQL [22
]. An ES from .20–.49 was considered small, .50–.79 as moderate, and greater than .80 as large [22
In order to avoid problems of regression towards the mean [26
], we applied one-way between-group analyses of covariance (ANCOVA) to compare the effect of RYGB and lifestyle intervention on five dimensions of HRQL. Age at baseline, age at the onset of obesity, BMI at baseline, physical activity at baseline, and baseline HRQL-scores were used as covariates in each of the five analyses [28
]. Assessments of normality, linearity, homogeneity of variance and regression slopes were conducted to ensure assumptions for the ANCOVA. The unadjusted changes from baseline in the RYGB group and ILI group, together with the adjusted between group differences (95% CI), are reported. To account for the percentage explained variance in the dependents, calculations of partial eta squared (ηp2
) were performed. To test the effect of weight reduction (instead of treatment choice) on HRQL, multiple linear regression analyses were conducted with each of the 12 months HRQL changes (physical, mental, emotional, number of obesity symptoms, and symptom distress) as dependents, with gender, age at baseline, age at the onset of obesity, BMI at baseline, physical activity at baseline, and weight change in per cent of baseline weight as independents. Throughout, we report two-tailed P values, with P<.05 was considered to be statistically significant. The statistical analysis was conducted using SPSS v.18.0.