Structural equation modeling to identify the risk factors of diabetes in the adult population of North India

Study design

This study analyses data from a large household survey conducted in the state of Punjab, India using WHO-STEPwise approach to surveillance (STEPS) approach [8].

Study setting

The survey was carried out in Punjab, which is a prosperous state in the northern part of India bordering Pakistan with a population of 27 million according to 2011 national census. It ranks higher than most other states in terms of Human Development Index with a per capita income twice that of the national average [9, 10].

Study sampling strategy

A state-wide non-communicable disease (NCD) risk factor survey based on WHO-STEPS approach was undertaken in Punjab in 2014–2015. The survey adopted a multistage stratified sampling approach using the 2011 census sampling frame. A three-stage design was employed in urban areas whereas in rural areas a two-stage sampling design was followed. A total of 100 primary sampling units (PSUs) were selected (60 villages from rural areas and 40 Census Enumeration Blocks from urban areas) by probability proportional to size (PPS) method. From each selected PSU, 54 households were selected using systematic random sampling. The ultimate sampling units were the households and one individual in the age group of 18–69 years residing in the selected household was selected using the KISH method. The details of the methodology can be found in another paper [11].

Data collection instrument

A culturally adapted, Punjabi (local language) translated and pre-tested version of the WHO STEPS questionnaire (version 3.1) was used with minor adaptations [12]. As part of the household survey, sociodemographic and behavioral information on tobacco and alcohol use, diet, physical activity, history of chronic diseases, family history of chronic conditions, health screening, and health care expenditure were collected in step 1. Physical measurements such as height, weight, blood pressure, and waist circumference were done in step 2. Biochemical tests were conducted to measure fasting blood glucose, total cholesterol, and triglycerides in step 3.

Data collection and operational definitions used

A team of trained investigators collected the survey data. SECA adult portable stadiometer was used to measure height after removing shoes, socks, slippers, and any head gear. It was measured in centimeters up to 0.1 cm. SECA digital weighing scale was used to measure weight of the individuals. The scale was regularly calibrated against a standard weight. The participants were asked to remove footwear and socks, and weight was recorded in kilograms up to 0.1 kg. Waist circumference was measured using a SECA constant tension tape to the nearest 0.1 cm at the level of the midpoint between the inferior margin of the last rib and the iliac crest in the mid-axillary plane. The measurement was taken at the end of a normal expiration with the arms relaxed at the sides.

One serving of vegetable was considered to be one cup of raw green leafy vegetables or 1/2 cup of other vegetables (cooked or chopped raw). One serving of fruit was considered to be one medium size piece of apple, banana, or orange; 1/2 cup of chopped, canned fruit; or 1/2 cup of fruit juice.

Physical activity was assessed using the Global Physical Activity Questionnaire (GPAQ), which has been developed by the World Health Organization. This questionnaire assesses physical activity behavior in three different domains: work, transport, and during leisure time. Activities are classified into three intensity levels: vigorous, moderate, and light based on the physical effort it requires. Participants were classified as sufficiently active who exceed the minimum duration of physical activity per week recommended by WHO, i.e., 150 min of moderate intensity physical activity or 75 min of vigorous intensity physical activity or an equivalent combination of moderate- and vigorous-intensity physical activity achieving at least 600 MET-minutes per week with each activity performed in bouts of at least 10-min duration [13]. Body mass index (BMI) was calculated as weight in kilograms/height in meters squared. Show cards (pictorial, adapted to the local context) were used to explain to the participants the type of physical activity, servings of fruits and vegetables, and salty food intake. Obesity was defined as a BMI ≥ 27.5 kg/m² for both genders (based on the World Health Organization Expert Consultation for Asian populations) [14]. Abdominal obesity was defined as a waist circumference ≥ 90 cm for men and ≥ 80 cm for women [15].

For blood pressure measurement, electronic equipment (OMRON HEM 7120, Omron Corporation, Kyoto, Japan) was used. After resting for 5 min, blood pressure was recorded in the sitting position in the right arm supported at the level of the heart. Three blood pressure measurements were taken at 3 min interval each. The final reading was recorded as the average of last two readings.

Biochemical measurements (step 3): every alternate individual (50%) of the initial sample was subjected to biochemical assessment. For blood glucose, dry chemistry method was used by blood glucose measurement device (Optium H, Freestyle). For lipid profile, i.e., cholesterol and triglycerides measurements, blood samples were drawn on individuals after 10–12 h fasting. 5 ml of venous blood was taken in sitting position, was centrifuged immediately to separate serum, and was transferred under cold chain condition to the Central Reference Laboratory of Department of Biochemistry, Post Graduate Institute of Medical Education and Research, Chandigarh, India which is a tertiary medical care institute.

Laboratory measurement of total cholesterol and triglyceride was made on Modular P 800 autoanalyzer (Roche Diagnostics, Germany) using commercially available kits (Roche Diagnostics, Germany). Hypertriglyceridemia was defined as serum triglyceride levels ≥ 150 mg/dl (≥ 1.7 mmol/l) [16].

Sample size

Taking the estimated prevalence of physical activity as 50%, 5% margin of error and 95% confidence interval, a design effect of 1.5, a sample size of 4609 was derived which was adequate to present results by two age groups (18–44, 45–69), both sexes (male, female) and residence (urban, rural). Assuming a response rate of 85%, sample size was raised to 5400 for this study. Every second individual was subjected to step 3, i.e., biochemical assessment. Out of 2700 respondents eligible for step 3, 2499 (93%) gave consent to blood sampling for biochemical assessment.

Statistical analysis

The conceptual a priori model that specifies the relations among variables operationalized in this study is based on the model proposed by Bardenheier et al [17]. We used structural equation modeling with path analysis, which includes the direct and indirect effects of factors previously reported to be associated with diabetes (Fig. 1). Direct effects are depicted as an arrow originating from an independent variable (exposure) leading and pointing to a dependent variable (outcome). For example, see the arrow between waist circumference and systolic blood pressure. An indirect effect is not only depicted as a mediating variable having an arrow pointing to it from an independent variable but also pointing to yet another dependent variable. For example, waist circumference mediates the effect of alcohol intake on blood sugar levels. A confounder, according to the use of these arrows, is depicted as a variable with direct effects on both the exposure and the dependent variable.

In this study, we report standardized path coefficients, their standard errors and p values. As indices of the models’ statistical fit to the data, we used standard criteria, including comparative fit index (CFI) > 0.90, root mean square error of approximation (RMSEA) < 0.08, and the standardized root mean square residual (SRMSR) < 0.06. Model building and estimation was done using STATA/IC version 12 (StataCorp LP, USA).

Prevalence of diabetes

SEM model

The following risk factors deemed to have a direct statistically significant effect on blood sugar status: family history of diabetes (4.5), urban residence (3.1), increasing age (0.05) and waist circumference (0.18), systolic blood pressure (0.11), and triglycerides (0.46) (Figure 1 and Table 2). There are specific indirect effects of waist circumference (0.028, p = < 0.001), alcohol use (1.43, p = 0.001), weekly fruit intake (− 0.009, p = 0.034), increasing age (0.156, p < 0.001), and family h/o diabetes (0.844, p = 0.001) on fasting blood glucose (Table 2).

Waist circumference, alcohol intake, and increasing age impact blood sugar levels mediated by raised blood pressure. Waist circumference mediates the indirect effects of age, family h/o of diabetes, alcohol intake, and weekly fruit intake on blood sugar levels. Triglycerides also mediate the indirect effect of age on diabetes status (Fig. 2).

Strengths and limitations

The strengths of the study include a large multistage stratified community-based sample, high participant response rate, use of standardized STEPS questionnaire, and adherence to STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines for reporting the findings of the study [32]. The study also employed a robust statistical technique, i.e., SEM which allows measurement of both direct and indirect effects of variables and incorporate models with multiple dependent variables by using several regression equations simultaneously.

There were a few limitations in this study. First, the main outcome measure, i.e., fasting blood glucose which defined diabetes status was assessed at one single point in time when in reality it can naturally fluctuate. Second, capillary blood glucose estimation was done in place of the ideal venous plasma glucose estimation due to logistic constraints such as non-availability of quality-controlled laboratories, storage and transport of blood specimens, varied methods of estimation, and poor compliance to venous blood collection due to its invasive nature. However, it has been reported that capillary blood collection is a feasible and valid alternative to venous blood collection for screening in large epidemiological studies [33, 34]. Third, the cross-sectional nature of the survey prevents us from making causal inferences about the outcome.