Of the 1,250 distributed questionnaires, 1,021 (81.7%) were returned. In 978 cases, body weight and height were measured by the GPs. 347 (34.0%) of the 1,021 included patients were male and 674 (66.0%) were female. The comparison of patients who returned their questionnaire with the non-responders did not reveal significant differences regarding the following characteristics which could be retrieved form the medical file: sex, age, duration of OA and number of comorbidities as well as number of prescriptions. 278 (80.1%) men and 296 (43.9%) women were married or lived with a partner. Completely retired from work were 233 (67.1%) men and 482 (71.5%) women. Most of the missing data referred to sociodemographic variables and could be completed by means of the patients' medical files. Details about the study sample, separated by BMI, are shown in Table . The displayed p-values are the result of group comparisons between the normal weight and the overweight group and between the overweight and the obese group. As can be seen, the groups did not differ regarding age and duration of disease. Both the radiological grading according to Kellgren and the number of comorbidities increased significantly with an increasing BMI. The educational level decreased from normal weighted patients to obese patients.
Baseline characteristics of study sample
Table provides information about the association of patients' comorbidities and the BMI. As can be seen, the prevalence of high blood pressure was significantly higher in the overweight group (compared to normal weight; p < 0.001) and also significantly higher in the obese group than in the overweight group (p = 0.002). Similar findings could be revealed for the prevalence of diabetes since regarding coronary vessel disease the only the differences between the normal weighted patients and the other groups achieved significance but not when we compared over weighted and obese patients. 21 men and 16 women reported about a history of cancer or current cancer disease, significant differences between the groups did not occur.
Association of comorbidities with obesity (n)
Regarding OA specific QoL (Table ), differences between overweight and normal weight patients were not significant in any dimension, including the PHQ-9 score which was used to assess depression. Significant differences occurred in the lower body scale, the symptom scale, the affect scale and the PHQ-9 score when the BMI surpassed 29.9 m/kg2 in comparison to the overweight as well as to the normal weight group. The upper body scale did not differ between the three groups, a finding which is most likely due to the study sample that consisted only of patients with OA to the knee or hip. Also, there were no significant differences between all groups regarding scores of the social scale, which reflects social network and support, and the work scale. However, it has to be acknowledged that the work scale was only applicable in 263 cases since most of the patients were already retired.
Impact of BMI on OA related quality of life
Table displays the comparison of PA between the three weight groups by means of ANCOVA (adjusted for age, disease duration and comorbidities). As can be seen, PA decreased significantly from patients with normal weight to overweight and to obese patients.
Physical activity according to IPAQ scoring, separated by BMI
The health service utilization (HSU) patterns of the study sample are displayed in Table . In unadjusted analysis, visits to GPs significantly increased with the BMI. Since visits to GPs may often be related to other reasons than OA, we adjusted the ANCOVA for comorbidities (as displayed in Table ). Interestingly, the significant difference between normal weight and overweight patients faded. However, the difference between obese and normal weight patients remained significant (p = 0.002) even after adjusting for the number of comorbidities. Visits to orthopaedics as well as performed x-rays remained significantly associated with the BMI after adjustment.
Health service utilization according to BMI