A total of 565 participants took part in the study; 292 from rural and 272 from urban. The median age of participants was 40(26) years and there was no statistical difference between rural and urban participants. The overall median body mass index of participants was 25.6(7.4) with urban dwellers recording a significant higher value. Visceral fat was also significantly higher for urban dwellers compared to rural participants. Calorie, carbohydrate and fibre intakes were significantly higher among rural participants while urban participant consumed higher vitamin A. Other nutrients did not show any significant difference. Rural participants had higher metabolic equivalents compared to their urban counterparts. Table 1 shows socio-demographics, nutritional intake and body composition characteristics of study participants.
Using BMI, there was no difference in the prevalence of obesity among rural and urban participants. Visceral and body fat cut-offs showed a higher prevalence of obesity in urban compared to rural participants. Prevalence of obesity across all parameters was higher among females compared to males with the exception of visceral fat that showed no difference. Table 2 shows the prevalence of obesity by community and gender.
Table 3 shows the community and weight status of study participants by their physical activity category. Across all measures, with the exception of visceral fat, higher proportion of overweight and obese participants did not meet WHO recommendation for physical activity per week. A higher proportion of urban and female participants did not meet the WHO recommendation for physical activity.
Tables 4 and 5 show principal component analysis of household food frequency. A total of four components were extracted for each community. The four components explained 32.4 and 30.9% of the total variance for rural and urban respectively. For the rural community the four components extracted were diverse diet pattern, vegetable convenience pattern, non-convenience pattern and fast food pattern. These contributed 15, 6.1, 5.5 and 5.4% of the 32.4% variance respectively. The diverse pattern consisted of foods from almost all the food groups. The vegetable pattern consisted of diet beverages, fresh and cooked vegetables, restaurant meals and nuts. The non-convenience pattern consisted of tea, coffee, breakfast cereal, sugar, nuts, cooked vegetables and low consumption of raw vegetables and ready to eat meals. The fast food pattern consisted of fried fish, fast foods, restaurant meals fruit and root tubers. For the urban community, the four components were diverse pattern, meat pattern, staple pattern and non-meat pattern. These explained 12.1%. 7.6, 5.9 and 5.4% of the total variance respectively. The diverse pattern as with the rural consisted of foods from almost all the food groups. The meat pattern consisted of ready to eat foods, organ meat, meat, fish, chicken, fruit and legumes. The staple pattern consisted of fried fish, maize, salted fish and raw vegetables and low likelihood for the consumption of ready to eat foods, processed meat and commercial bread. The non-meat pattern consisted of nuts, fish, cereal, legumes and salty snacks with low consumption of organ meat. Tables 6 and 7 show partial spearman correlation adjusted for age, gender and metabolic equivalents between the four components and body composition measures. Among urban participants component 3 (staple pattern) showed a negative correlation with BMI -0.163(0.029) and visceral fat − 0.186(0.013).
Table 8 shows a partial spearman correlation adjusted for age and gender between waist circumference, BMI and body composition measures. All variables showed significant correlations but the strongest correlations were between visceral fat and BMI r = 0.905(p < 0.001), body fat and BMI r = 0.851 (p < 0.001) and BMI and waist circumference r = 0.845(p < 0.001).
Table 9 shows two models of multinomial logistic regression for risk factors of central obesity determined by waist circumference. Multicollinearity was checked and carbohydrate and fibre had strong correlations with energy intake and protein respectively and were therefore excluded from the model. In model 1 the obese group was set as reference for the outcome variable and urban and female were set as reference for the explanatory variables, community and gender respectively. Metabolic equivalent category was added to model 1 to generate model 2. In model 1 males had about 22 times odds of being normal compared to females at p-value < 0.001 while rural dwellers had odds of about 1.7 times of being normal compared to urban participants and this was significant at p < 0.05. The odds of energy intake were low but significant in both models. The predictability of gender for obesity persisted in model 2 but community showed no significant predictability. Participants not meeting WHO recommendation for physical activity had a significant reduced odds of being normal compared to those meeting the recommendation.