This is a community-based study using panel data from the first and second wave of the India Human Development Survey [IHDS-I (2004-05) & IHDS-II (2011-12)]. IHDS is a nationally representative survey covering multiple topics of public health and community basic amenities. Both waves of IHDS were jointly conducted by researchers from the University of Maryland, USA and the National Council of Applied Economic Research (NCAER), New Delhi (https://www.ihds.umd.edu/). The design of IHDS is a panel survey where 41,554 households in 1503 villages and 971 urban neighborhood were interviewed on two occasions, first in 2004-05 and then for the second time in 2011-12. Data used in the study consists of 148,333 individuals of all ages followed from 2004-05 to 2011-12.

Respondents in the first wave (2004-05) were followed up and the same information on their health and morbidities were reassessed again in the second wave (2011-12). The respondents were asked if the doctor had diagnosed them for diabetes and hypertension. Based on their response, a respondent is identified as having diabetes or hypertension or free from these diseases. Diabetes and hypertension are the lifestyle diseases which can be catastrophic to one’s livelihood to a great extent. Behavioural risk factors such as smoking, drinking, consumption of tobacco and sedentary habits have been proved to increase the risk of diabetes and hypertensions [17] [18] [19]. Other physical, social and economic conditions are also found to be significant prognosticators of this lifestyle disease [20] [21] [22]. Besides behavioural risk factors, the study also considers sex, place of residence, level of education, body mass index and income quintile of the household as the prognosticators of the transition from diabetes to hypertension and vice-versa.

Multistate models are often used to describe the progression of diseases [23] - [28]. The multistate model explains how an individual changes its states between a series of states in continuous time. The study considers three states of transition viz. healthy (0), diabetes (1) and hypertension (3) having the state space S ( t ) = { 0 , 1 , 2 } . We assume the probability of transition from state of diabetes to healthy (1 - 0) and from state of hypertension to healthy (2 - 0) is zero as there is no complete cure to diabetes and hypertension rather than controlling its intensity of burden. Figure 1 shows the path of transition between the three states.

Figure 2 shows the transition in the state space S ( t ) at different waves of IHDS. Of all 148,333 respondents who entered in the first wave of IHDS, 224 respondents were diagnosed as having suffered from diabetes and 448 were diagnosed as having suffered from hypertension. Among those who were not diagnosed to be either diabetic or hypertensive in first wave of IHDS, 1468 respondents develop diabetes and 3876 develop hypertension. Among those 224 respondents who were diagnosed as diabetic, 26 respondents develop hypertension. And among those 448 respondents who were diagnosed with hypertension, 58 respondents develop diabetes.

The respondents in the study move through the state space S ( t ) = { 0 , 1 , 2 } which is governed by the transition intensity, q i j ( t , a ( t ) ) ; i , j = 0 , 1 , 2 . Transition intensities are the instantaneous risk of moving from state i to j defined as

q i j ( t , a ( t ) ) = lim δ t → 0 P [ S ( t + δ t ) = j | S ( t ) = i ] δ t

Such that the 3 × 3 transition intensity matrix is, Q = [ q 00 q 01 q 02 q 10 q 11 q 12 q 20 q 21 q 22 ]

where ∑ j 3 q i j = 0 .

Multistate model generally depends on Markov assumption that future movement to next state j depends on the current state i [29].

In a time homogeneous Markov model, where q i j are independent of t, the mean waiting time in each state i is exponentially distributed with mean − 1 / q i j . And the likelihood for the Markov model is calculated from the transition probability matrix, P ( x , t + x ) . The ( i , j ) entry of P ( x , t + x ) is the probability of being in state j at time t+u, given the state at time u is i. This matrix is calculated in terms of Q using Kolmogorov differential equations. If Q is constant over the interval ( x , t + x ) as in a time homogeneous process, then P ( x , t + x ) = P ( t ) and the equation is solved by the matrix exponential of Q scaled by the time interval,

P ( t ) = E x p (tQ)

We used “msm” package in R-Software to produce the desired outputs. The package “msm” was specifically developed to model the progression of diseases [30]. It allows time homogeneous Markov models to be fitted to processes like that of disease progression which can be observed continuously. In a multistate model, the relationship between the time-varying characteristics of individuals to their transition probabilities or intensities is of vital importance. Thus a form of a proportional hazard model was developed to describe this relationship [25]. The package also allows us to model the relationship between individual specific or time dependent characteristics with transition intensities [30]. It also allows us to estimate the transition probabilities, the probability that each state is next and the mean waiting time.

Table 2 and Table 3 present the unadjusted assessment of prevalence of diabetes and hypertension respectively from both waves of the IHDS. Compared to the first wave of IHDS in 2004-05, the prevalence of diabetes and hypertension

Source: Authors computation from Indian human development survey.

Note: ** p-value < 0.01 & * p-value < 0.05; CI – Confidence Interval.

Note: ** p-value < 0.01 & * p-value < 0.05; CI – Confidence Interval.

increased in the second wave in 2011-12. The behavioral risk factors viz. smoking, drinking and chewing tobacco does not show statistically significant association with these lifestyle diseases in the first wave. But, there is an evidence of strong association in the second wave. Comparatively those respondents who drink, smoke and consume tobacco frequently have the higher prevalence of diabetes and hypertension. The prevalence of diabetes in the second wave who drink, consume tobacco and smoke frequently are 9.03, 12.03 and 31.98 per one thousand respectively. Whereas the prevalence of hypertension in the second wave who drink, consume tobacco and smoke frequently are 30.43, 29.90 and 45.48 per one thousand respectively. Sex is not showing significant association with diabetes. But a strong association can be seen between sex and hypertension. The prevalence of hypertension in first wave for males and females are 2.60 and 6.69 per one thousand respectively. And the prevalence of hypertension in second waves for males and females are 18.96 and 39.03 per one thousand respectively. The association between respondent place of residence, level of education, body mass index and income quintile of the household with diabetes and hypertension is highly statistically significant. The prevalence of diabetes and hypertension is higher among urban residents as compared to rural residents. The prevalence of diabetes in rural and urban place of residence in second wave are 7.76 and 19.22 per one thousand respectively. Similarly, the prevalence of hypertension in rural and urban place of residence are 23.25 and 39.65 per one thousand respectively. The prevalence of diabetes and hypertension increases as the level of education increases in the first wave. But the pattern is not the same in the second wave. The prevalence of diabetes in second wave among the illiterates and respondents who have attained primary, secondary and higher education are 11.13, 8.81, 12.15 and 16.01 per one thousand respectively. Similarly, the prevalence of hypertension in second waves among illiterates and respondents who have attained primary, secondary and higher education are 42.20, 24.08, 22.11 and 30.74 per one thousand respectively. In both waves of IHDS, the prevalence of diabetes and hypertension increases as the level of body mass index and the income quintile of the household increase. The prevalence of diabetes and hypertension among the respondents with body mass index greater than 24.9 in second wave are 31.93 and 80.62 per one thousand respectively. While the prevalence of diabetes among the lowest, lower, middle, higher and highest income quintile household in second wave are 7.24, 5.87, 9.51, 13.05 and 19.70 per one thousand respectively. And the prevalence of hypertension among the lowest, lower, middle, higher and highest income quintile household in second wave are 24.56, 19.98, 23.93, 29.94 and 41.76 per one thousand respectively.

Table 4 gives the unadjusted and adjusted fitted transition probability in transitioning from one state to other over the time interval, 2004-05 to 2011-12. The unadjusted probability to remain in a healthy state during the interval is 0.964 compared to the adjusted probability of 0.971. It also gives us the probability of 0.010 and 0.026, for an individual who are currently healthy will be diabetic and hypertensive respectively. Whereas the probability that an individual who is currently healthy will be diabetic and hypertensive after adjusting for several covariates are 0.007 and 0.022 respectively. Also the probability that an individual who is diabetic will be hypertensive and vice versa are 0.058 and 0.044 respectively compared to 0.024 and 0.024 respectively after adjusting for several covariates.

Table 5 gives the unadjusted and adjusted mean waiting time for the three state of transition. The unadjusted waiting time to remain in the healthy state, diabetic state and hypertensive state are 27.13 years (95% CI = [26.41, 27.87]), 16.23 years (95% CI = [5.09, 11.07]) and 21.8 years (95% CI = [5.18, 8.74]) respectively. Whereas the mean waiting time after adjusting for several covariates to remain in the healthy state, diabetic state and hypertensive state are 32.23 years (95% CI = [32.60, 36.43]), 39.90 years (95% CI = [2.82, 20.52]) and 40.77 years (95% CI = [3.52, 7.45]) respectively.

Table 6 presents the transition specific hazard ratio of different socioeconomic and demographic characteristics on different paths of transition. Compared to respondents who never drink alcohol, respondents who drink alcohol but not

Note: ^#Adjusting for several behavioral, individual and household level characteristics. State 0: Healthy; State 1: Diabetes & State 2: Hypertension.

Note: ^#Adjusting for several behavioral, individual and household level characteristics. State 0: Healthy; State 1: Diabetes & State 2: Hypertension.

Note: **p-value < 0.01 & *p-value < 0.05; HR: Hazard Ratio; CI: Confidence Interval. Transition from healthy to diabetes (0 - 1); Transition from healthy to hypertension (0 - 2); Transition from diabetes to hypertension (1 - 2); Transition from hypertension to diabetes (2 - 1).

frequently are more likely to have diabetes (HR = 2.42; 95% CI = [1.49, 3.94]). While among the respondents who are diabetic, respondents who drink alcohol but not frequently (HR = 4.56; 95% CI = [1.73, 12.02]) and who drink alcohol frequently (HR = 9.85; 95% CI = [2.35, 41.22]) are more likely to have hypertension. Similarly, compared to respondents who had never consumed tobacco, respondents who consumed tobacco but not frequently (HR = 2.11; 95% CI = [1.62, 2.76]) and who consumed tobacco frequently (HR = 1.78; 95% CI = [1.51, 2.10]) are more likely to have hypertension. Among those who are hypertensive, respondents who consumed tobacco but not frequently are more likely to have diabetes (HR = 2.59; 95% CI = [1.38, 4.85]). Compared to respondents who had never smoked, respondents who smoked but not frequently are more likely to have diabetes (HR = 1.77; 95% CI = [1.03, 3.04]). While respondents who smoked but not frequently (HR = 1.93; 95% CI = [1.39, 2.68]) and respondents who smoked frequently (HR = 2.31; 95% CI = [1.94, 2.74]) are more likely to have hypertension. Sex of the respondents is also significantly associated with transition from healthy state to diabetic state and hypertensive state. Compared to male, female are less likely to suffer from diabetes (HR = 0.72; 95% CI = [0.52, 0.98]) but more likely to suffer from hypertension (HR = 2.59; 95% CI = [2.28, 2.93]). Compared to illiterate respondents, respondents who have attained secondary (HR = 3.08; 95% CI = [1.86, 5.11]) and higher education (HR = 5.93; 95% CI = [3.42, 10.28]) are more likely to have diabetes. Also respondents who are diabetic and have attained higher education are more likely to have hypertension (HR = 4.70; 95% CI = [1.48, 14.91]). But respondents who have attained primary education are less likely to have hypertension compared to illiterate respondents (HR = 0.84; 95% CI = [0.75, 0.95]). Similarly respondents who are hypertensive and have attained higher education are less likely to have diabetes (HR = 0.25; 95% CI = [0.08, 0.83]). Compared to respondents with body mass index less than 18.5, respondents having higher body mass index are more likely to have hypertension. But, respondents who are hypertensive and having body mass index 18.5 to 24.9 are less likely to have diabetes (HR = 0.39; 95% CI = [0.18, 0.82]). Respondents living in urban are more likely to suffer from diabetes (HR = 2.16; 95% CI = [1.70, 2.75]) and hypertension (HR = 1.26; 95% CI = [1.15, 1.38]) than those living in rural areas. Compared to respondents from household with lowest income quintile, respondents from household with higher income quintile (HR = 1.59; 95% CI = [1.05, 2.41]) and highest income quintile (HR = 1.89; 95% CI = [1.18, 3.05]) are more likely to suffer from diabetes. Similarly, respondents from household with higher income quintile (HR = 1.19; 95% CI = [1.06, 1.36]) and highest income quintile (HR = 1.52; 95% CI = [1.32, 1.74]) are more likely to suffer hypertension.