Prevalence of diabetes and pre-diabetes in rural Tehri Garhwal, India: influence of diagnostic method

We randomly selected five villages from 58 villages located on the west side of Chamba, in the mountainous Tehri District in the Garhwal region of the state of Uttarakhand in northern India (May to July 2015), using random number generation. We undertook a cross-sectional survey of adults in each village for between 8 and 10 days, with the aim of obtaining approximately 100 participants per village. As we did not have a census for each village, randomization of individual participants was not possible. Health staff, including a nurse and health workers, recruited participants following full informed consent. Ethical approval was obtained from the Garhwal Community Development and Welfare Society, and written informed consent was obtained before any data were collected from each participant.

Diabetes and pre-diabetes were defined according to the American Diabetes Association’s cut points [9]. Participants with 5.7% ≤ HbA_1c ≤ 6.4% were classified as pre-diabetic and those with HbA_1c ≥ 6.5% were classified as diabetic. Pre-diabetes was also defined as 100 mg/dL ≤ FBG ≤ 125 mg/dL or 140 mg/dL ≤ PBG ≤ 200 mg/dL. Diabetes was defined as a FBG > 125 mg/dL or PBG > 200 mg/dL. Levels of HbA_1c and FBG were recoded into normal, pre-diabetes and diabetes based on these definitions. Moderate to severe anemia was defined as hemoglobin< 11.4 mg/dL adjusted for high altitude according to the World Health Organization’s criteria [18]. Hemoglobin concentration levels above the cut-off were classified as normal to mild anemia.

All analyses were conducted using Stata version 14 (StataCorp., Texas). Age, weight, height, waist, hip, body mass index (BMI) and monthly income per capita (household monthly income/household size) are presented as medians and quartiles (Q1, Q3). Categorical variables (sex and education) are presented in percentages.

We assessed the crude prevalence of pre-diabetes and diabetes among the participants using HbA_1c measured during the initial survey (time 1). Age and gender specific prevalence of diabetes were compared using Pearson’s Chi-square test. Similarly, we compared age specific percentages of diabetes and pre-diabetes based on the FBG and/or PBG obtained in time 2 using Pearson’s Chi-square test.

The high prevalence of pre-diabetes at the initial survey prompted us to consider that anemia may have contributed to this surprising finding. To test whether HbA_1c over-diagnoses pre-diabetes and diabetes, we compared the proportions of the two conditions at time 1 with those determined by glucose tests taken at time 2 (August – October 2015) using McNemar’s Chi-square test. In the majority of cases, FBG was used (n = 117, 84%), but in those instances where this was unavailable (n = 23, 16%), PBG was used. We excluded one case who began treatment for diabetes between time 1 and time 2 (Fig. 1).

To further understand the influence of hemoglobin concentration on the level of HbA_1c, we used McNemar’s test to compare the proportions of diabetes and pre-diabetes based on HbA_1c and glucose-based tests.

Of the 500 participants, 61% (303) were female and 39% (197) male. The median age of all participants was 50 years (IQR 42–60 years). More than half of the participants (61.4%) had no formal schooling or no more than 6 years of education. The median monthly income per capita was 1000 rupees (500–3000 rupees, n = 400 due to missing values for the number of people in the household). One person had missing information on HbA_1c (Fig. 1). There was a similar number of people recruited to each village (Additional file 1: Table S1).

There were notable differences in most of the anthropometric measures between individuals who were non-diabetic and those who were either pre-diabetic or diabetic according to HbA_1c. Kruskal Wallis tests (set at 0.05 significance level) show that those with pre-diabetes or diabetes were progressively heavier, with greater BMI, waist and hip circumference, and waist-hip ratio (Additional file 2: Table S2).

Based on HbA_1c, the overall prevalence of diabetes was 10% (95% CI 7.6–13%) and pre-diabetes 56.4% (95% CI 52–60.7%). Prevalence of pre-diabetes and diabetes was lowest in the youngest age group (Table 1).

There was a strong association between diabetes status and age (χ² = 13.3; p = 0.01), but we were unable to detect a statistically significant association between gender and prevalence of pre-diabetes or diabetes (χ² = 1.07; p = 0.59).

Among the 140 people we followed-up, we could detect no significant change in anthropometric variables between the times of measurement of HbA_1c and FBG/PBG (Table 2). Of these 140 people, 124 (88.6%) were pre-diabetic according to HbA_1c cut-offs compared to 56 (40.0%) using FBG/PBG (Table 3). This corresponds to a 48.6% greater percentage of pre-diabetes using HbA_1c than by FBG/PBG (95% CI: 39.3 to 57.7%) (McNemar’s chi-square = 66.1; p < 0.001).

There was no evidence that diabetes was over-diagnosed when HbA_1c was used as the diagnostic criterion. The difference in proportions of diabetes assessed by HbA_1c and FBG/PBG was − 0.02 (95% CI: − 0.07 to 0.05; McNemar’s χ² = 0.29; p = 0.79).

In both people with and without moderate to severe anemia, a much smaller proportion were diagnosed with pre-diabetes by FBG/PBG than HbA_1c (Table 4). The difference between the two diagnostic criteria was consistent across the two categories of hemoglobin concentration [no anemia or mild anemia: 0.48 (0.37–0.61); McNemar’s χ² = 44.1; p < 0.01; moderate to severe anemia: 0.48 (0.31–0.64), McNemar’s χ² = 22; p < 0.01]. Similarly, the proportions categorized by HbA_1c and FBG/PBG as diabetic did not differ markedly according to category of blood hemoglobin concentration (Table 4).

Our findings indicate that HbA_1c has comparable performance to FBG/PBG with regard to diagnosis of diabetes in this rural population. HbA_1c and glucose tests both detected very similar proportions of diabetes in the sample we studied. Our finding is consistent with those of two recent studies, one in an out-patient clinic in China and another in a tertiary care setting in India [20, 21]. However, substantial disparities between diagnoses based on HbA_1c and glucose-based tests have previously been reported [4, 22‐26]. In some cases HbA_1c diagnosed much larger proportions of diabetes than did FBG [4, 23‐25] while in other studies HbA_1c diagnosed smaller percentages of diabetes than did FBG [22, 26]. These studies varied in settings ranging from general communities as well as sea-level and high-altitude settings, covering various ethnic groups including Arabs, Asian-Americans, Native Hawaiians and Peruvians. Given that there is evidence of HbA_1c levels being associated with both genetic factors [11, 27, 28] and geographical settings [23] further investigation of HbA_1c-FBG relationship in rural mountainous regions are required.

In contrast to the diagnosis of diabetes, 49% of study subjects who were diagnosed with pre-diabetes based on HbA_1c were classified as normal by measurement of FBG/PBG. Our findings are consistent with those of two recent studies which indicate considerable proportions of individuals who are pre-diabetic based on HbA_1c criteria have normal FBG results [24, 25]. However, others found that smaller proportions of pre-diabetics were identified by HbA_1c than glucose-based methods [26, 29]. The divergence is unlikely to reflect device measurement errors, given the evidence that the point-of-care device used in our study has comparable performance to a laboratory based HbA_1c test [30, 31]. Despite the fact that FBG and OGTT have both been widely used as the gold standards for diagnosing diabetes, there is still no uniform definition of pre-diabetes [32‐35]. Marked discordance between pre-diabetes by HbA_1c and FBG/PBG criteria in the present study therefore does not necessarily imply false positives. In fact, there is some evidence that HbA_1c and glucose-based diagnoses identify different populations within the hyperglycemic category [27, 29]. Both tests may offer uniquely important prognostic information.

The presence of a large proportion of individuals diagnosed as having pre-diabetes based on HbA_1c but with normal glucose-based results may have important prognostic implications. HbA_1c’s ability to predict major clinical complications such as cardiovascular disease is of major prognostic significance [36‐38]. For example, HbA_1c levels were able to predict lipid profile, a key determinant of cardiovascular heart disease [37, 38]. Indeed, in a community-based study of non-diabetic, middle-aged adults in four U.S. communities, Selvin and colleagues observed that elevated HbA_1c (≥6%) was strongly associated with the risks of cardiovascular disease, all-cause mortality and ischemic stroke [38]. The associations remained strong and significant after adjusting for baseline FBG levels. Moreover, in individuals with stable coronary artery disease, HbA_1c values of 6.3% or greater were linked with adverse cardiovascular outcomes [36]. There is also evidence that reducing HbA_1c by 0.2% can lead to a 10% reduction in risk of mortality within 12 months [37]. Whether the increased risk of cardiovascular diseases associated with elevated HbA_1c is due to pre-diabetic conditions or the subsequent development of diabetes remains unclear [39, 40]. However, the additional clinical information provided by a measurement of HbA_1c in the pre-diabetic range may render HbA_1c a more cost-effective option than FBG or OGTT, particularly in areas where comprehensive medical tests are less accessible and populations are more predisposed to cardiovascular diseases. Further research on this subject should be pursued in order that this additional prognostic benefit can be fully realized.

Another possible prognostic implication of pre-diabetes diagnosed by HbA_1c but not FBG/PBG lies in the prediction of progression to diabetes. In a systematic review of studies investigating the performance of HbA_1c in predicting progression to diabetes among adults aged 18 years and over, Zhang and colleagues observed that HbA_1c ≥ 6.0% was associated with a very high risk of subsequent development of diabetes [41]. Pre-diabetes identified by fasting glucose and oral glucose tolerance tests, on the contrary, had limited ability to predict progression to diabetes [42]. On the other hand, others found similar rates of progression to diabetes in those identified as pre-diabetic by a single FPG test compared to a HbA_1c test [43, 44]. Regardless, our current findings indicate that the prevalence of pre-diabetes in rural Garhwal is much higher than the national rural prevalence [4]. Given that timely lifestyle changes can be effective in preventing or delaying progression to diabetes [33], the advantages of using HbA_1c for mass screening of pre-diabetes in low-income rural areas may outweigh its disadvantages in the long run.

Consumptive anemia can confound the relationship between HbA_1c level and glycemic control [14, 45]. Thus, it could potentially account for the observed difference in diagnoses of pre-diabetes by HbA_1c and FBG. However, we found that both the pattern and magnitude of differences in the proportions of pre-diabetes and diabetes diagnosed by HbA_1c compared with FBG/PBG, were independent of anemia status. There is prior evidence of substantial changes in HbA_1c levels in the presence of anemia [14, 15]. In a recent systematic review of the effects of anemia and abnormalities of erythrocyte indices on HbA_1c, the authors suggested that iron deficiency, and particularly iron deficiency anemia, may lead to an increase in HbA_1c [14]. In addition, relatively larger divergence was found between diagnoses of pre-diabetes and diabetes based on HbA_1c and OGTT, in a subgroup of young anemic adults who were deficient in iron, B12 and folic acid, when compared to a reference group [15]. The association between HbA_1c and hemoglobin concentration was also found to vary according to the forms of anemia. While iron deficiency anemia was associated with elevated HbA_1c [15], some forms of anemia were found to be linked with diminished HbA_1c level. A typical example is hemolytic anemia, in which the lifespan of erythrocytes is shortened, causing a drop in HbA_1c [14]. Other conditions which may decrease levels of HbA_1c include acute hemorrhage and hemoglobinopathies [46]. However, so far there is no consistent evidence regarding the influence of anemia and hemolytic disorders on HbA_1c.

Because of the difference in timing between HbA_1c and FBG tests, caution should be applied in the interpretation of these data. Participants were told their level of HbA_1c immediately after the test. Thus, some of the disparity between diagnosis of pre-diabetes by HbA_1c and FBG/PBG may be due to altered behavior between the two time points. However, lifestyle changes, if any, were unlikely to have contributed to the large disparity, given the minor differences in weight, body mass index and waist-hip ratio between time 1 and time 2. Furthermore, because we did not follow-up people who were categorized as having a normal HbA_1c at baseline, we will have missed detecting anyone who was negative for diabetes and pre-diabetes at the first assessment, but who were then positive using the glucose-based definition. This may mean that the differences that we observed are overestimated. We also acknowledge that our conclusions may not be generalizable across the whole region of Tehri-Garhwal because only five villages were surveyed. However, our findings provide preliminary evidence for a particularly high prevalence of diabetes and poor glycemic control in the sub-Himalayan region. They therefore provide the impetus for larger studies, including sampling of participants in a manner that represents the wider population of the region, to fully characterize the prevalence of diabetes in mountainous areas of North India.