Evaluation of Madras Diabetes Research Foundation-Indian Diabetes Risk Score in detecting undiagnosed diabetes in the Indian population: Results from the Indian Council of Medical Research-INdia DIABetes population-based study (INDIAB-15)

Introduction

Diabetes mellitus affects more than 463 million people worldwide, with an estimated 4.2 million people dying each year from it due to complications1. While India accounts for one sixth of the world’s population with diabetes (77 million)1, with considerable occurence at younger age (45-64 yr), a significant proportion of the diabetic population experience complications at their productive years with serious economic and social burden. Early identification of individuals with undiagnosed diabetes is therefore critical in reducing the burden of diabetic complications. Several intervention studies have unequivocally demonstrated that type 2 diabetes can be effectively prevented by lifestyle changes in high-risk individuals2,3. The major task for public health administrations is to identify these high-risk individuals who would benefit from intensive lifestyle modification. However, ‘universal screening’, i.e. screening the entire population, is cumbersome and thus a cost-effective tool to selectively screen the high-risk groups would be necessary.

Development of the Indian Diabetes Risk Score (IDRS) developed by Madras Diabetes Research Foundation (MDRF), was a simple tool to aid early detection of undiagnosed diabetes in the community4. The MDRF-IDRS has also been shown to have additional multiple applications5, including predicting incident diabetes6, metabolic syndrome7, coronary artery disease8, non-alcoholic fatty liver disease9, sleep disorders10 and diabetic complications, such as peripheral vascular disease and neuropathy11, and help distinguish type 2 from non-type 2 diabetes12. The MDRF-IDRS has also been validated in several other populations13-25. However, it has not been validated in a large representative population in India. The Indian Council of Medical Research (ICMR)-INdia DIABetes (INDIAB) study provided an opportunity to evaluate the risk score in a large representative population and thereby help detect undiagnosed type 2 diabetes in the community.

Material & Methods

The ICMR-INDIAB study is a cross-sectional, population-based survey of adults aged 20 yr and above26,27. The methodological details of the study have been published elsewhere26. In brief, the study sampled urban and rural residents of all the 30 states/union territories (UTs) of the country such that the total estimated sample of 120,000 individuals was representative of the entire population. The study was conducted in a phased manner. In Phase I, four regions representing the south (Tamil Nadu), north (Chandigarh), east (Jharkhand) and west (Maharashtra) of the country were studied during 2008 to 2010. Between 2011 and 2020, the remaining states were surveyed as follows; Phase II covered Andhra Pradesh (undivided), Bihar, Gujarat, Karnataka and Punjab (survey period, 2012-2013), Phase III included Delhi, Madhya Pradesh, Rajasthan and Uttar Pradesh (survey period, 2017-2018), Phase IV included Kerala, Goa, Puducherry, Haryana and Chhattisgarh (survey period, 2018-2019), North East Phase included Assam, Arunachal Pradesh, Manipur, Meghalaya, Mizoram, Nagaland, Sikkim and Tripura (survey period, 2011-2017) and Phase V included Himachal Pradesh, Uttarakhand, Odisha and West Bengal (survey period, 2019-2020).

A stratified multistage sampling design was followed26, wherein a three-level stratification based on geography, population size, and socio-economic status of each state was done to obtain a truly representative sample of the population from different strata. The primary sampling units were villages in rural areas and census enumeration blocks in urban areas. The ultimate stage units were households in both areas. From each household, one individual was randomly selected using the World Health Organization KISH method28, thereby avoiding selection bias with respect to gender and age.

The calculated sample size of 4000 people (2800 rural + 1200 urban) provided 80 per cent power and a 5 per cent alpha error in each state, assuming a type 2 diabetes prevalence of 10 per cent in urban areas and 4 per cent in rural areas, with a 20 per cent non-response rate26. As a result, the overall sample size for the 30 states/UT was 120,000 people (4000×30); 113,043 individuals responded (94.2% response rate).

The analysis was conducted after excluding individuals with self-reported diabetes) on the pooled data of all the States/UTs (n=98,454), which included 28,150 urban residents and 70,304 rural residents. For the sub-analysis, the States/UTs were divided into six geographical regions as follows: North: Chandigarh, Delhi, Haryana, Himachal Pradesh, Punjab and Rajasthan; South: Andhra Pradesh (undivided), Karnataka, Kerala, Puducherry and Tamil Nadu; East: Bihar, Jharkhand, Odisha and West Bengal; West: Goa, Gujarat and Maharashtra; Central: Chhattisgarh, Madhya Pradesh, Uttar Pradesh and Uttaranchal and the North East: Arunachal Pradesh, Assam, Manipur, Meghalaya, Mizoram, Nagaland, Sikkim and Tripura.

A standardized, structured questionnaire was used to collect information on demographic and socio-economic characteristics, physical activity, smoking, alcohol, medical history and family history of diabetes. Fasting capillary blood glucose was determined using a glucose meter (One Touch Ultra, Lifescan Johnson & Johnson, Milpitas, California, USA) after ensuring at least eight hours of overnight fasting. In subjects with self-reported diabetes, only the fasting glucose was measured and was excluded from the current analysis. Equipment with same specifications was used throughout the study as a measure of quality assurance.

The study and the respective sites was approved by the Institutional Ethics Committee of the Madras Diabetes Research Foundation (MDRF), Chennai, Tamil Nadu, India, and written informed consent was obtained from all the participants.

Definitions: Self-reported/known diabetes was defined by a physician diagnosis of type 2 diabetes and current use of medications for diabetes (insulin or oral hypoglycaemic agents) and were excluded from the analysis. The MDRF-IDRS was developed using multiple logistic regression model based on four simple parameters, namely age, abdominal obesity, family history of diabetes and physical activity as described elsewhere4. The information for these risk factors was obtained based on four questions (Table I) and one anthropometric measurement, namely the waist circumference. Subjects with an MDRF-IDRS value of <30 was categorized as low risk, those between 30 and 50 as moderate risk and those with ≥60 as high risk for diabetes.

Statistical analysis: All statistical analysis was performed using SAS Statistical Package (version 9.4; SAS Institute, Inc., Cary, NC, USA). Estimates were expressed as mean±standard deviation or proportions. Linear regression and Cochran trend test were used used to find the trend across IDRS groups for continuous and categorical variables, respectively. Student’s t test was used to compare continuous variables and χ² tests for differences in proportions. Receiver operating characteristic (ROC) with area under the curve (AUC) was used to assess the performance of MDRF-IDRS to diagnose diabetes; P<0.05 was considered significant.

Results

Table II presents the characteristics of the study population based on the MDRF-IDRS risk scoring as low, moderate and high for developing type 2 diabetes. The results show that females were more likely to have high-risk score for diabetes based on MDRF-IDRS than males, in both urban and rural areas. Individuals with high MDRF-IDRS were older, had significantly higher waist circumference, higher systolic and diastolic blood pressure, were less formally educated, had stronger family history of diabetes, were less physically active and had fewer proportion of smokers and alcohol users. A similar trend was seen when stratified in urban and rural areas.

Table IIIa shows the MDRF-IDRS risk stratification by different regions of the nation. In the overall population, 14.9 per cent were in the low-risk category of MDRF-IDRS, 52.7 per cent in the moderate-risk category and 32.4 per cent in the high-risk category. It varied by regions, with the southern (38.9%) and the northern (38.3%) region having highest proportion of individuals with high MDRF-IDRS and eastern (25.4%) region having the lowest proportion of individuals with high MDRF-IDRS. A similar trend was observed when the regions were stratified as urban and rural areas (Table IIIb).

The ROC curve showing the performance of MDRF-IDRS in diagnosing diabetes among the urban and rural population and by gender is presented in Figures 1 and 2. The ROC-AUC of the MDRF-IDRS for identification of diabetes was 0.697 (95% confidence interval: 0.684-0.709) for urban and 0.694 (0.684-0.704) for rural, as well as 0.693 (0.682-0.705) for males and 0.707 (0.697-0.718) for females.

Close

Fig. 1

ROC curve showing the performance of MDRF-IDRS among the (A) urban and (B) rural population in diagnosing diabetes. ROC, receiver operating characteristic; MDRF-IDRS, Madras Diabetes Research Foundation-Indian Diabetes Risk Score.

Export to PPT

Close

Fig. 2

ROC curve showing the performance of MDRF-IDRS among (A) male and (B) female population in diagnosing diabetes. ROC, receiver operating characteristic; MDRF-IDRS, Madras Diabetes Research Foundation-Indian Diabetes Risk Score.

Export to PPT

An attempt was made to determine how many of the newly diagnosed diabetic subjects found in the study using OGTT were correctly identified by MDRF-IDRS (Table IV). Of all the newly diagnosed individuals with diabetes (by OGTT), 60.2 per cent were identified as having high risk, 35.9 per cent as having moderate risk and 3.9 per cent as having low risk by the MDRF-IDRS. Over 96 per cent of all newly diagnosed diabetic subjects were identified as having either moderate or high risk by MDRF-IDRS. The values ranged from 92.1 per cent in the central region to 97.8 per cent in the southern region. In the overall population, MDRF-IDRS has a sensitivity of 60.2 per cent, specificity of 68.8 per cent, positive predictive value of 8.1 per cent, negative predictive value of 97.4 per cent and accuracy of 68.5 per cent in identifying diabetes. A similar analysis was performed among individuals with self-reported diabetes, where 98.9 per cent of them were categorized by MDRF-IDRS as moderate/high-risk group (Supplementary Table I). Of the individuals with pre-diabetes (diagnosed by OGTT), 45.6 per cent were identified as having high risk, 45.7 per cent as having moderate risk and 8.7 per cent as having low risk for pre-diabetes by MDRF-IDRS. Overall, 91.3 per cent of all individuals with pre-diabetes were identified as having either moderate/high risk by MDRF-IDRS (Supplementary Table II).

Supplementary Table III presents the proportion of newly diagnosed diabetic subjects identified by MDRF-IDRS in the study population by individual states. MDRF-IDRS performed well in all states and did extremely well in the state of Goa where 100 per cent of the individuals, diagnosed as diabetes by OGTT, were categorized as either at high or moderate risk by MDRF-IDRS. MDRF-IDRS performed least in the state of Jharkhand where it identified 85.4 per cent of the cases as high/moderate risk. When stratified as urban or rural areas, MDRF-IDRS identifies 100 per cent of all subjects with diabetes as high/moderate risk in many urban areas of the states, including Andhra Pradesh (undivided), Assam, Goa, Haryana, Manipur, Mizoram, Nagaland, Odisha, Punjab, Sikkim and Tripura, and a few rural areas of the states, such as Goa and Rajasthan.

Discussion

The key findings of the study include the following: (i) based on the MDRF-IDRS, 32.4 per cent of the general population in India falls under high-risk category of developing type 2 diabetes; (ii) of all the newly diagnosed individuals with diabetes (diagnosed by OGTT), MDRF-IDRS identified 96.1 per cent of the population as having high/moderate risk and 3.9 per cent as having low risk for diabetes; (iii) MDRF-IDRS also performed well in urban and rural strata and also in state/ regions.

Diabetes risk scores are risk assessment tools use a combination of demographic and clinical information. Although globally several such risk scores exist29-38, their applicability to other populations is uncertain in different ethnic groups and sociocultural settings. The IDRS developed by MDRF4 used routinely collected, minimal information making the risk score simple, fast and more importantly, non-invasive. The variables included in MDRF-IDRS are age, waist circumference, physical activity levels and family history of diabetes. Through this approach, targeted screening with non-invasive test is done for the identification of high risk groups, followed by glucose testing only in high risk individuals.

Although several risk scores available, there is no consensus on the gold-standard tools that should be used in practice. Each of these risk scores uses varying combinations of risk factors. Diabetes risk scores have also been used to predict diabetes in various populations, such as Finnish Diabetes Risk Score (FINDRISC), Finland29, Danish Diabetes Risk Score, Denmark30, Data from the Epidemiological Study on the Insulin Resistance Syndrome, France31, American Atherosclerosis Risk in Communities (ARIC), four US communities32 and Cambridge Risk Score (CRS), England33 among others34,35. The sensitivity and specificity of these scores varied widely: FINDRISC29 (78% sensitivity and 77% specificity in the 1987 cohort and 81% sensitivity and 76% specificity in the 1992 cohort), DANISH30 (76% sensitivity and 72% specificity), ARIC32 (52% sensitivity and 86% specificity) and CRS33 (77% sensitivity and 72% specificity). The variables included in the risk scores are age, sex, ethnicity, body mass index (BMI), waist circumference, history of antihypertensive drug treatment, use of corticosteroids, hypertension, physical activity, family history of diabetes, smoking status, daily consumption of fruits, berries or vegetables29-35.

Risk scores developed in one population may not perform well when applied to other population. There are a few other diabetes risk scores specific for Indian population. The diabetes risk score by Ramachandran et al36 included age, BMI, waist circumference, physical activity levels and history of diabetes and yielded sensitivity of around 70 per cent and specificity of around 60 per cent in three tested cohorts in Chennai, Tamil Nadu. Chaturvedi et al37 included age, waist circumference, blood pressure and family history of diabetes, which had 73 per cent sensitivity but only 56 per cent specificity. Oommen et al38 included age, waist circumference and family history of diabetes and reported 59.5 per cent sensitivity and 60.5 per cent specificity. The applicability of the risk score to a different population is uncertain given the differences in risk factor distribution, such as varying degrees of obesity, lifestyle factors and other cultural differences. As cited above, all the risk scores had similar sensitivity and specificity and used similar clinical characteristics.

The applicability of MDRF-IDRS has also been extensively validated in several Indian populations13-25 (Supplementary Table IV). Some studies16,21 reported that the MDRF-IDRS had good predictive value for detecting undiagnosed diabetes in the community and also suggested alternative score for MDRF-IDRS, with better sensitivity and specificity. Some studies14,15,17-20,25 used MDRF-IDRS in classifying the study population at risk as low, moderate and high. Despite such studies validating the applicability of MDRF-IDRS, a number of issues limit their utility, such as small sample size, lack of representative sample questioning the generalizability of the results to the larger population, differences in study design/methodology used, lack of external validation and referral bias. The ICMR-INDIAB study, by using standardized protocol and uniform data collection procedures, provided precise details on the variation in the sensitivity in different regions of India – with the highest sensitivity being recorded in central India (67%) and lowest in eastern region (51%). Apart from the differences in the characteristics of the population studied in the different regions/states, the modest performance of the MDRF-IDRS risk score in the INDIAB population could be attributed to the data quality on the variable, ‘family history of diabetes’. Although it is known that the knowledge of family history of diabetes may influence lifestyle behaviours of the individual, the reliability of the family history data is unclear. The level of awareness of individual’s family history is unknown in our population, which is one of the limitations of this study.

Modifications to the risk variables included in the MDRF-IDRS were also tested. In a recent study by Venkatrao et al23, the MDRF-IDRS score was tested by replacing the waist circumference component of the score with both BMI and a composite of BMI and waist circumference using data from a randomized cluster sample survey including 7496 adults at high risk for type 2 diabetes. It was reported that MDRF-IDRS using waist circumference, BMI and both BMI and waist circumference had high sensitivity (87, 88 and 82%, respectively) in detecting diabetes and use of both BMI and waist circumference in MDRF-IDRS improved its specificity. In the ICMR-INDIAB study and earlier investigations on MDRF-IDRS, waist circumference was preferred over BMI, as Indian population express Asian Indian phenotype, including increased insulin resistance and greater abdominal adiposity, i.e. increased waist circumference despite having a lower BMI compared to other ethnic groups.

In a country like India with a population of 1.4 billion, where more than 50 per cent of the people with type 2 diabetes remain undiagnosed, if no risk score is used, we have to screen the entire population. The logistics of such an exercise and cost can well be imagined. Hence, adding a preliminary step of using a screening score to identify at-risk individuals would have the potential to make such programs more cost-effective.

Our study has several strengths; it included states and union territories in India, and a represented both urban and rural areas. Moreover, a representative sampling frame and robust methodology were used utilizing OGTT for identification of diabetes in a large sample of 113,043 individuals with a response rate of 94.2 per cent. The high-level quality control measures implemented in the field survey ensured data reliability, allowing results to be generalized at the country level. However, there were also some limitations to our study. First, the cross-sectional nature of the study did not allow for inferences of causality to be made and the validation of the score with incident diabetes was not performed. Although cross-sectional data are adequate for detecting undiagnosed diabetes using risk score, prospective data would provide additional useful insight. Second, some of the differences could be attributed to the time lag in data collection between various phases of the study, which, however, is inevitable when sampling a country as big as India. However, as the objective of this current study was to evaluate the use of MDRF-IDRS in different parts of the country and the country as a whole, this time lag in the survey period does not affect the results. Our results also do not provide information on the prevalence of diabetes in individuals younger than 20 yr because this was beyond the scope of the study. Finally, our study methodology did not allow differentiating between type 2 diabetes and type 1 diabetes, latent autoimmune diabetes of adults (LADA) and maturity onset diabetes of the young (MODY) .

In conclusion, validation of MDRF-IDRS in Indian population has clinical implications. Through this study, the performance of MDRF-IDRS was evaluated across the nation. The study findings underscore the needs for using this validated score-tool in diabetes prevention intervention programmes. Such an approach may aid in an economic way of screening, conducted at the individual as well as the healthcare provider level. As the score can be self-assessed, it provides an inexpensive, easy and secure way to comprehend risk factors and empower the high-risk population for screening appropriate help. In addition, as the predictors included in the score keep changing with time and changes in lifestyle, periodic usage of the score in the same population will add value. Thus, identifying the high-risk group for diabetes will pave the way for effective intervention programmes, which is a desirable goal in public health policy and clinical practice.