Introduction In India the urban prevalence of diabetes in adults has risen from 5% in 1984 to just under 15% in 2004 [1],[2]. Markedly lower rural levels of diabetes have been evident for decades, but more recently prevalence appears to have increased from 2% to 6% in rural south India [3]. Underlying these adverse trends in diabetes are increases in obesity affecting urban areas much more than rural areas of India. The 2nd Indian National Family Household Survey in 1998–1999 confirmed the marked rural-urban differences in prevalence of obesity among women [4] and men [5], and also a rising trend between the 2nd and 3rd National Family Household Survey in 2005 [6]. The increasing risks of obesity and diabetes in India and in other low- and middle-income countries have been attributed to increased consumption of saturated fats, sugars, and sedentary behaviour associated with urbanisation and westernisation [7]. However, obesity and diabetes have early life origins that track into adulthood and these may play a critical role in explaining the obesity and diabetes “epidemics” in developing countries [8]. In India urbanisation is caused by urban expansion into peripheral areas and internal migration from rural to urban areas, largely for economic reasons. However, it is not clear how urbanisation increases the risk of obesity and diabetes among people who have had divergent early life experiences, particularly in developing countries. Migration studies are powerful means of identifying environmental causes of common diseases as changes in environment are large and occur at a known time, making causal inferences more feasible. In a major review of the evidence on migration and cardiovascular risk factors and obesity, McKay and colleagues state “it is clear that migrants in general tend to suffer from worse health and display disadvantaged risk factor profiles. In comparison to the host population they are more frequently subject to hypertension, chronic conditions, low birth weight, and obesity etc. Moreover, their ill health and unfavourable risk profiles may worsen with increasing duration of stay” [9]. While the trends of increased risk of obesity and diabetes among both international south Asian migrants are well documented [10],[11], much less is known about the effects of internal rural-to-urban migration in India. The rising rates of diabetes in both urban and rural India indicate that urbanisation is an important but not a sufficient explanation. Examining the experience of rural-to-urban migrants would help understand what is driving these trends. Migrants would be expected to acquire the high risk of the urban population if the disease is largely environmentally determined and for such trends to be dependent on the duration of time spent in a new environment and the extent to which traditional ways of living are lost—as seen for coronary heart disease among the Japanese moving to the US [12]. Conversely, for some common causes of chronic disease in the country of origin, the environmental changes associated with migration may result in lower risk of hypertensive heart disease, for example [13]. Previous studies are conflicting with some suggesting that changes in cardiovascular risk factors (particularly blood pressure) may occur within a few years [14], whilst others indicating that considerable exposure to urban life is necessary [9]. A recent review of the experience of migrants to the US indicates a complex picture with generally better health among migrants but with heterogeneity between different groups, probably reflecting the duration of stay in the US [15]. Based on these findings, the main hypothesis of our study was that rural-urban migrants would have higher rates of obesity and diabetes than rural nonmigrants and secondary hypotheses were that (a) rural-urban migrants' rates of obesity and diabetes would be intermediate to that of rural and life-long urban dwellers and (b) longer stays in the urban environment would increase rates of obesity and diabetes. Methods Using the framework of a cardiovascular risk factor screening study conducted in factories in north, central, and south India [16], we designed a sib-pair comparison study. Details of the design have been reported elsewhere [17]. Briefly, the study was in four Indian factories (Lucknow, Hindustan Aeronautics Ltd; Nagpur, Indorama Synthetics Ltd; Hyderabad, Bharat Heavy Electricals Ltd; and Bangalore, Hindustan Machine Tools Ltd) situated in the north, centre, and south of the country. Factory workers and their coresident spouses were recruited if they were rural-urban migrants using employer records as the sampling frame. Each migrant worker and spouse was asked to invite one nonmigrant full sibling of the same sex and closest to them in age still residing in their rural place of origin. Precedence was given to gender over age and where multiple same-sex sibs were available the one closest in age was invited. This strategy resulted in rural dwelling sibs being drawn from 20 of the 29 states in India, reflecting the migration patterns of the factory workforce and their spouses. A 25% random sample of nonmigrants was invited to participate in the study. Nonmigrants were also asked to invite a sib who resided in the same city but did not work in the factory. Information sheets were translated into local languages and signed (or a thumb print used if the individual was illiterate), and through this, informed consent obtained. Ethics committee approval was obtained from the All India Institute of Medical Sciences Ethics Committee, reference number A-60/4/8/2004. Field work began in March 2005 and was completed by December 2007. Measurement of Cardiovascular Risk Factors Standing height was measured with mandibular stretch at end expiration using a plastic stadiometer (Leicester height measure; Chasmors Ltd), and weight was measured in light clothes with shoes off using a digital scale (Model PS16). Skinfold thickness was measured three times at the triceps, subscapular, and medial calf using Holtain calipers and the average of the three measures used. Subscapular and triceps skinfolds were used to calculate percent body fat using a standard formula [18]. Waist and hip circumferences were measured using a nonstretch narrow metal tape with a blank lead in (Chasmors metallic tape), taking the average of two readings. Blood pressure was measured using an Omron M5-I automatic machine in the sitting position using the right upper arm and an appropriate sized cuff after a period of 5 min rest. Participants were interviewed using a structured questionnaire to obtain information about tobacco use and alcohol consumption. Obesity- and Diabetes-Related Outcomes Obesity was defined as body mass index (BMI) greater than 25 kg/m2 (Indian adult population standard) [19]. A diagnosis of diabetes was made using the World Health Organization (WHO) fasting plasma glucose criterion of >7.0 mmol/l [20] or report of a doctor diagnosis of diabetes. Homeostasis model assessment (HOMA) scores to estimate insulin resistance were calculated from fasting blood glucose and serum insulin levels using a standard formula of plasma glucose (mol/l) × plasma insulin (mU/l)/22.5), on the basis of the original approach [21]. HOMA has been validated by comparison with biochemical markers of insulin resistance in healthy Indian people, yielding moderate correlations [22]. Dietary Assessment Diet was assessed by an interviewer-administered semiquantitative food frequency questionnaire (FFQ). The questionnaire assessed frequency of intake (daily, weekly, monthly, yearly/never) of 184 commonly consumed food items. In order to assess the reliability of the FFQ, subsamples were asked to complete the questionnaire 1–2 mo (n = 185), as well as 12 mo (n = 305) after completion of the questionnaire during the original period of data collection. Kappa coefficients ranging from 0.26–0.71 were obtained, which are similar to values obtained in other reliability studies [23],[24]. Another 530 participants (53.9% males) were administered a reference method of three 24-h recalls, which was used to validate the FFQ. Most food items yielded validities that were acceptable. Fat intake (g/d) was reliably measured and is presented here as an indicator of dietary change. Physical Activity An interviewer-administered questionnaire was used to assess physical activity of the past month across multiple domains including discretionary leisure time, household chores, work, sleep, sedentary activities, and other common daily activities. For each activity the average amount of time and the frequency were documented. Participants reported frequencies to fixed categories of “daily,” “once a week,” “2–4 times a week,” “5–6 times a week,” “once a month,” and “2–3 times a month.” Metabolic equivalent tasks (METs) were estimated as the ratio of resting metabolic rate where 1 MET is equivalent to the energy expenditure value of sitting quietly. When all the activities reported did not cumulatively account for 24 h, a standard MET of 1.4 was applied to the residual time [25]. For manual occupational activity an integrated energy index (IEI) of the activity was applied instead of the absolute MET value. IEI take into account “rest” or “pause” periods, which individuals are likely to take when engaged in these manual activities [26]. Validation of the questionnaire was conducted in 49 rural and 45 urban participants by making comparisons with uni-axial accelerometers and a 24-h activity diary. Physical activity showed acceptable validity with these reference methods with little evidence of bias although correlations were only modest (accelerometer r = 0.28; p 25 kg/m2) was greatest in urban women (53.5%, 95% CI 50.5–56.5) and lowest in rural men (18.0%, 95% CI 17.0–21.0), with migrants in an intermediate position (see Figure 1). The age, occupation, and factory adjusted odds of obesity were between 3- and 4-fold greater in migrant than rural men and women (Table 3). Percentage body fat estimated from skinfold thicknesses showed markedly higher values among women than men, with similar levels among urban and migrant groups, but lower levels among rural dwellers. 10.1371/journal.pmed.1000268.g001 Figure 1 Age-, factory-, and occupation-adjusted percent prevalence (95% CI) of obesity, BMI >25 kg/m2, by migrant group and sex, Indian migration study 2005–2007. Including number of participants with information about obesity and number of obese. 10.1371/journal.pmed.1000268.t002 Table 2 Risk factors for cardiovascular disease. Risk Factors Men Women Urban Migrants Rural p for Trend Test Urban = Migrant Urban Migrants Rural p for Trend Test Urban = Migrant BMI (kg/m2) 24.3 (24.1–24.5) 24.0 (23.8–24.2) 21.9 (21.7–22.1) 7 mmol/l 2.33 (1.46–3.73) 2.38 (1.51–3.76) 1 0.0006 0.92 2.38 (1.18–4.80) 2.26 (1.13–4.51) 1 0.02 0.83 Regular alcohol 1.42 (1.08–1.88) 1.38 (1.05–1.73) 1 0.007 0.70 0.31 (0.11–0.86) 0.63 (0.28–1.42) 1 0.02 0.15 Current smoker 0.82 (0.66–1.03) 0.61 (0.49–0.75) 1 0.03 0.01 0.28 (0.09–0.89) 0.66 (0.27–1.63) 1 0.02 0.11 Physically inactive 2.00 (1.66–2.41) 1.62 (1.33–1.97) 1 7 mmol/l b 1.78 (1.10–2.88) 1.86 (1.16–2.98) 1 0.03 0.82 1.62 (0.80–3.31) 1.61 (0.80–3.25) 1 0.25 0.97 a Geometric mean. b Odds ratios (95% CI) for the risk of disease in a sibling compared to a rural sibling, adjusted for BMI, occupation, age, age group, and factory with an individual-specific random effect of sib-pair. LDL, low-density lipoprotein; SBP, systolic blood pressure. Smoking and drinking alcohol were rare among women; among men, migrants reported the lowest prevalence of smoking and rural men the highest. Alcohol use was highest among migrant men and lowest in rural men. Odds of hypertension (i.e., doctor diagnosis, on blood pressure lowering drugs, or blood pressure >140/90) in urban and migrant men were almost twice those of rural men and evidence of increased odds were also seen in women. Blood cholesterol and triglycerides were similar in urban and migrant groups but values were lower in rural men (p trend 7.0 mmol/l) was higher in urban and migrant groups than the rural group (see Figure 2). Both urban and migrant men and women had over 2-fold increased odds of diabetes compared with rural participants. 10.1371/journal.pmed.1000268.g002 Figure 2 Age-, factory-, and occupation-adjusted percent prevalence (95% CI) of diabetes (diagnosed, on treatment, or fasting glucose >7 mmol/l) by type of migrant and sex, Indian migration study 2005–2007. Including number of participants with information about diabetes and number of diabetics. Further adjustment for BMI in men weakened the associations between place of origin and systolic blood pressure, hypertension, total cholesterol, and triglycerides but did not reduce the strength of associations with fasting blood glucose, HOMA, or the prevalence of diabetes (see Table 4). In women, further adjustment for BMI weakened the associations between place of origin and hypertension. Adjustment for percent body fat computed from skinfold thicknesses instead of BMI produced similar effects on associations (unpublished data). Between Sib-Pair Differences Data in Table 5 correspond to Tables 2– 4 but focus on the estimated contrasts between migrant and rural sibs. The values shown represent the average difference between the migrant and nonmigrant rural sib. For example, the BMI of the male migrant sib group was 2.10 kg/m2 (95% CI 1.84–2.37 kg/m2) greater than that of the rural sib group of the same age. Among men the migrant sib group had consistently more adverse measures of obesity, lipids, and diabetes than the rural sib group. The between sib group differences were modest, 4.08% (95% CI 3.7–4.47) difference in body fat, 2.2 mmHg (95% CI 1.0–3.4 mmHg) in systolic blood pressure, and 0.20 mmol/l (95% CI 0.11–0.28 mmol/l) in total cholesterol. Comparing the migrant with the rural male sibs, HOMA scores were 1.25- (95% CI 1.16–1.34) fold higher, triglycerides were 1.08- (95% CI 1.04–1.12) fold higher, and fasting glucose was 1.03- (95% CI 1.02–1.05) fold higher. Adjustment for BMI attenuated these small differences between migrant and urban sibs. Differences were less marked between migrant and rural women. 10.1371/journal.pmed.1000268.t005 Table 5 Estimated contrast (95% CI) between migrant and rural sibling for men and women adjusted for age, age group, and factory including a random effect of sibling pair. Risk Factors Men Women Adjusting For Age Group, Occupation, and Factory Adjusting For Age Group, Occupation, Factory, and BMI Adjusting For Age Group, Occupation, and Factory Adjusting For Age Group, Occupation, Factory, and BMI BMI (kg/m2) 2.10 (1.84–2.37) — 2.65 (2.25–3.06) — Standing height (cm) 0.32 (−0.12 to 0.77) — 0.41 (−0.10 to 0.92) — Waist∶hip ratio 0.03 (0.02–0.03) — 0.01 (0.01–0.02) — Percent body fat 4.08 (3.70–4.47) — 3.29 (2.82–3.76) — SBP (mmHg) 2.21 (1.02–3.40) 0.08 (−1.14 to 1.30) −0.02 (−1.55 to 1.52) −1.90 (−3.47 to −0.33) Total cholesterol (mmol/l) 0.20 (0.11–0.28) 0.08 (−0.00 to 0.17) 0.06 (−0.05 to 0.16) −0.01 (−0.12 to 0.10) LDL cholesterol (mmol/l) 0.16 (0.08–0.23) 0.08 (0.01–0.16) 0.08 (−0.02 to 0.17) 0.02 (−0.07 to 0.12) Triglycerides a (mmol/l) 1.08 (1.04–1.12) 1.01 (0.97–1.05) 0.99 (0.95–1.03) 0.94 (0.91–0.98) Fasting blood glucose a (mmol/l) 1.03 (1.02–1.05) 1.02 (1.00–1.03) 1.01 (0.99–1.03) 1.00 (0.98–1.02) Fasting insulin a (mU/l) 1.23 (1.15–1.31) 1.02 (0.96–1.09) 1.10 (1.01–1.19) 0.97 (0.90–1.06) HOMA score a 1.25 (1.16–1.34) 1.03 (0.96–1.11) 1.09 (1.00–1.18) 0.96 (0.88–1.05) MET h/day −2.01 (−2.35 to −1.67) −1.84 (−2.19 to −1.49) −1.04 (−1.37 to −0.71) −0.92 (−1.26 to −0.58) Fat intake a (g/day) 1.22 (1.19–1.26) 1.18 (1.14–1.21) 1.21 (1.16–1.25) 1.17 (1.13–1.22) In column 2 and 4 also adjusting for BMI. No adjustments made for variables related to BMI. a Relative difference. LDL, low-density lipoprotein; SBP, systolic blood pressure. The prevalence of obesity and of diabetes was examined by stratifying years since migration (>10 y versus ≤10 y). In men, but not women, there was weak evidence for linear trends in both obesity and diabetes from rural, more recent migrants, longer-term migrants, and urban dwellers (see Figures 1 and 2; Table 6). However, there was no strong statistical evidence of differences in odds of obesity or diabetes between the two migrant groups in either men or women. 10.1371/journal.pmed.1000268.t006 Table 6 Odds ratios (95% CI) for the risk of disease in a sibling compared to a rural sibling, adjusted for occupation age, age group, and factory with an individual-specific random effect of sib-pair. Condition Men Women Urban Migrants >10 y Migrants ≤10 y Rural p for Trenda Urban Migrants >10 y Migrants ≤10 y Rural p for Trenda Obese 3.85 (2.96–5.01) 3.24 (2.52–4.17) 2.04 (1.08–3.87) 1 25 kg/m2, Indian standard) among employed people of 20% in urban areas and 6% in rural areas [5], which is markedly lower than our prevalence of over 50% and 20% in urban and rural areas, respectively. In a large survey of six cities, an age-adjusted diabetes prevalence of 12% was reported in 2000 [47], which is lower than our urban prevalence estimate of around 15%. A recent study in urban India reported a 15% prevalence of diabetes, comparable with our estimate [48]. In comparisons with the 3rd National Family Household Survey [5] and the 2001 Census [49], our study population had lower proportions of illiterate individuals and higher proportions of individuals with access to household facilities and assets, indicating a generally wealthier and more educated population than the national average in both rural and urban areas. This finding was expected given that our sample was drawn from employed people and their relatives. Our participants reflect those in the vanguard of social and epidemiologic change. Our findings confirm a previous report of higher levels of serum insulin in urban as compared to the rural participants [50]. This suggests that some of the effects of urbanisation may be mediated through biological factors that result in increased secretion of insulin due to tissue resistance to its actions. Our findings are consistent with other studies of migrants where high levels of serum insulin have been reported in Asian Indians living abroad [51], in populations from other developing countries experiencing rapid urbanisation, and in migrant populations elsewhere [52]. Our response rates were lower than anticipated largely because of the logistic complexity of the sib-pair design. In a majority of cases these logistics involved at least a day to travel to the study centre and a day to travel back for the rural sib; in extreme cases up to 3 d travelling each way was involved. The differences in smoking prevalence between responders and nonresponders and nonconsenters were consistent with the play of chance. The prevalence of cardiovascular disease in the nonresponders was lower than in the responders and the nonconsenters. However, when considering those who took part with those who did not take part, no strong evidence of difference is apparent. While the response rate was suboptimal, from the data we have, there does not appear to be any major bias in health status or health behaviour. Response bias would influence our findings if there was differential nonresponse by health status and place of origin. Responders did report more cardiovascular disease than nonresponders but there was no difference in place of origin. This finding would be unlikely to alter the substantial differences we observed in prevalence of obesity and diabetes in the urban compared with rural samples. A further limitation is the cross-sectional design that does not permit longitudinal measurement to examine how cardiovascular risk and diabetes evolve over time in relation to migration. It is sometimes feasible to recruit participants into migration studies prior to migration (e.g., Luo [34], Tokelu islanders [35], Yi [36] studies), which have generally demonstrated that changes in risk factors are not explained by selection effects [14]. The forced migration of large populations living in the Three Gorges dam project in China is providing an opportunity to evaluate the effects of migration longitudinally on whole populations without any selection of who migrates [53],[54], but the process has been carefully planned and will not necessarily be generalisable to the effects of more typical migration experiences. If migration effects on health outcomes in India are as rapid as appears to be the case, establishing prospective studies in areas with high rural outflow to cities would be feasible. Migrants (particularly in the workplace) and their families are a readily identifiable group who might be more motivated to take part in health promotion activities and treatment of risk factors than the general population. The scale of obesity and diabetes among these factory workers, their spouses, and rural sibs is very large, arguing for much wider adoption of population prevention activities as proposed by the WHO [55]. Supporting Information Figure S1 Flow chart for participation in Indian migration study 2005–2007. *, Rural nonmigrants excluded for these analyses as they were factory workers living in rural areas and commuting to urban factory site. (0.31 MB TIF) Click here for additional data file. Figure S2 Distribution of years spent in urban setting by migrant migrants by sex, Indian Migration Study 2005–2007. (0.32 MB TIF) Click here for additional data file. Table S1 Participant (factory worker or spouse) characteristics by responder status. (0.02 MB RTF) Click here for additional data file.
