The study was a facility based cross sectional study that applied quantitative methods of data collection and analysis.
The study was conducted in two purposively selected high capacity diabetic clinics of Kampala i.e. Mulago and St. Francis Hospital Nsambya diabetic clinics. Mulago diabetic clinic is part of Mulago hospital which serves as a public and Uganda’s National Referral Hospital and is situated in Kawempe division of Kampala. The clinic runs its T2DM clinic day every Wednesday on a weekly basis. It is managed by various nurses, doctors and specialised diabetologist plus other intern doctors and clinicians. On the other hand, St. Francis Hospital Nsambya diabetic clinic is part of St. Francis Hospital Nsambya which is a private not for profit hospital, based in Makindye division in Kampala. It runs its T2DM clinic day every Monday on a weekly basis. It is managed by diabetic nurses, intern doctors and diabetologists. This study was conducted between March and May 2016.
The study involved patients aged 18 to 75 newly diagnosed, attending and receiving treatment in a diabetic clinic of Kampala. In this study, newly diagnosed patients were defined as those whose T2DM was diagnosed within the past 24 months from the time of the study interview.
All patients aged 18 to 75 years newly diagnosed with T2DM, attending and receiving treatment in Mulago or St. Francis Hospital Nsambya diabetic clinics. Only participants with confirmed diagnosis of T2DM from medical records (according to the International Diabetes Federation criteria ) were enrolled in the study. Patients who met the inclusion criteria but did not provide written informed consent, had history of known mental illness, were debilitated by diabetes or any other pre-existing condition, or were pregnant were excluded from the study.
The sample size was calculated basing on the formula for cross sectional studies , n=(Za)2
2; where the estimated proportion of obesity among type two diabetic patients was 21.3%  and the estimated sampling error was 6%. When a non-response of 10% was assumed, a total of 200 patients were selected.
Participants from each of the study site were selected consecutively; where all patients meeting the eligibility criteria were selected.
A structured pretested questionnaire was used to capture several sociodemographic characteristics (age, sex, marital status, highest level of education and occupation), lifestyle habits and family history (current smoking status, physical activity, current alcohol use, family history of diabetes and obesity, and use of concomitant medication) and several clinical measurements including weight, height, blood pressure and fasting blood glucose.
Sociodemographic, lifestyle and family history data
Marital status was classified into four categories (single, married, divorced/separated, and widowed). Highest level of education was also classified into four categories (no formal education, primary, secondary and tertiary). Occupation was classified into four levels which were: office based employees, manual labourers, students, and not employed/retired.
The measurement of smoking was based on current and past tobacco smoking status according to questions adapted from the Centre for Disease Control’s Global Adult Tobacco Survey (GATS) tool . Physical activity was assessed for all participants in the study using the WHO reference measure of physical activity for adults  with three main physical activity categories which included: light, moderate and vigorous activity. Alcohol use was assessed using the Dietary Guidelines for Americans alcohol intake recommendations . Family history of diabetes and obesity were assessed basing on presence or absence of diabetes and obesity in 1st and 2nd degree relatives on both paternal and maternal lineages. To assess for concomitant medication use, patients were asked whether they take any other medication in addition to that of diabetes.
Anthropometric, clinical and laboratory measurements
Weight was measured in kilograms and recorded to the nearest 0.1 kg using a pre calibrated Seca® scale with the patients in light clothing and shoes removed. Two measurements were taken and their average was considered. Height was measured in centimetres and recorded to the nearest 0.1 cm using a standard height metre when the participant was in an upright standing position without shoes. Blood pressure was measured using the AccuMed® wrist digital blood pressure monitor with patients seated in a calm environment. Two measurements were taken five minutes apart and their average was considered. To assess fasting blood glucose, a laboratory technician took off a small amount of blood via a needle prick from the patient’s finger. The blood sample was then tested for fasting blood glucose using a glucometer (Accu-Check® Active, Roche diagnostics, India) after 8 h of overnight fast.
Measurement of dietary intake
To capture usual dietary intake, a pretested semi structured 24-h dietary recall questionnaire was used. Qualified and experienced dieticians carried out face to face interviews to ask patients questions on food intake within the previous 24 h. Food models, utensils to estimate portion sizes and food images to scale were used to assist participants to recall food portions and quantities eaten. Participants were asked if the intake they reported represented their usual daily diet intake amounts in case their data were to be considered usable. The reported food intake from the 24-h dietary recall was entered into DietOrganizer® Software which converted the food intake information into nutrient intake. Local Ugandan foods that were not in the DietOrganizer® database were incorporated using the HarvestPlus food composition tables . These composition tables have all foods and recipes from eastern and central Uganda. In this study, our emphasis was on the intake of proteins, carbohydrates, total fat, saturated fats, monounsaturated fats, polyunsaturated fats, and fibre; which were used as dietary intake variables. We then compared dietary intake of all the selected nutrients for the participants of our study with the Diabetes and Nutrition Study Group recommendations (DNSG) .
Data collected from the study were entered in Epi info™ software and transferred to STATA® v. 13.0 for analysis. Dietary intake data were entered and analysed in Diet Organiser® software and later transferred to STATA® v. 13.0 for the final analysis.
At univariate level, we used descriptive statistics including means ± standard deviation (SD), percentages, proportions, and median to summarise sociodemographic, lifestyle, clinical characteristics and dietary intake of the patients as appropriate. At bivariate level, we compared sociodemographic, lifestyle, anthropometric characteristics and dietary intake of men and women using independent sample t tests or Man Whitney U test and chi square tests or Fisher exact tests as appropriate for continuous and categorical variables respectively. Owing to the fact that we used self-reported measures of dietary intake and we wanted to obtain more precise estimates; we applied energy adjusted measures of nutrient intake through nutrient density models for all dietary intake analyses . Energy adjusted measures of nutrient intake were expressed as percentage of total energy (%E) for protein, carbohydrate, total fat, saturated fat, polyunsaturated fatty acids, and monounsaturated fatty acids and grams per 1000 kcal (g/1000 kcal) for dietary fibre.
The primary dependent variable was BMI, calculated as weight (kg)/height (m2). It was measured and analysed on a continuous scale. The primary independent variable was dietary intake measured for six different nutrients which were: protein, carbohydrates, total fat, saturated fat, polyunsaturated fatty acids, monounsaturated fatty acids and dietary fibre.
To establish dietary intake of the patients, we compared patients’ intake in this present study with the DNSG recommendations and ascertained whether they meet, are above or below recommendations. To establish the association between dietary intake and BMI, patients were divided according to quintile of nutrient intake. Multiple linear regression at the 95% C.I was used to establish the differences in average BMI between the quintiles of nutrient intake while adjusting for age, marital status, current alcohol drinking status, current smoking status, occupation and education level. Further multiple linear regression models were run for only those nutrients found to be statistically associated with BMI in the first regression model while adjusting for more other variables. We observed the strength of association between the different nutrients and BMI with different levels of adjustment. The results for both men and women were presented together in the regression models as the dietary intake for majority of the nutrients did not differ between the two sexes. For all tests, a p value <0.05 was considered statistically significant. Multicollinearity between the independent variables was tested and for each model, collinear variables were eliminated.
All data collection tools and equipment were pretested and calibrated respectively before data collection. Prior to the main study, we conducted interviews on participants from a similar respondent group to pre-test the study questionnaire. The main aim of the pre-test was to identify ambiguities in questions asked, to examine participants understanding of the different questions, to assess if the questionnaire was able to capture the required data and to assess the feasibility of the study procedures. Necessary adjustments to the questionnaire were made to clear any discrepancies after which the tool was passed for final data collection. The dieticians were also trained for a period of 3 days before data collection commenced.