Risk factors for incident cardiovascular events and their population attributable fractions in rural India: The Rishi Valley Prospective Cohort Study

We prospectively determined incident cardiovascular events and their association with risk factors in rural India.

K E Y W O R D S cardiovascular events, cohort, epidemiology, incidence, low-and middle-income countries, population attributable fraction, risk factors

INTRODUCTION
Cardiovascular disease (CVD) is the leading cause of disease burden globally [1], particularly in Asia [2].A critical step in curbing the growing burden of CVD in resource-limited settings is to identify locally relevant risk factors and quantify their relative contribution to cardiovascular events [3].
Most epidemiologic evidence on the contribution of risk factors for CVD was derived from cohorts in high-income countries (HICs) [4][5][6][7].However, findings from HICs might not apply to preventing CVD in low-and middle-income countries (LMICs) because of: (i) uncertainty about the relative contribution of risk factors [3,8]; (ii) differences in cardiovascular risk profiles and incidence of CVD [9]; (iii) variations in ethnicity, culture, and context [10]; (iv) dissimilarities in economic development [11] and (v) diverse healthrelated behaviours [11].This proposition is supported by data from the Prospective Urban Rural Epidemiology (PURE) study, in which cardiovascular event rates were greater in LMICs than in HICs despite a comparatively low burden of cardiovascular risk factors in LMICs [9].This could be due, at least in part, to the greater contribution of non-traditional risk factors in LMICs [10].Such setting-specific differences create uncertainty about which risk factors to target for early prevention and treatment [12].
We hypothesised that non-traditional risk factors contribute to a significant proportion of CVD in rural India.Therefore, we aimed to prospectively determine the associations of various risk factors with the 5-year incidence of a cardiovascular event, and quantify the population attributable fractions (PAFs) of those risk factors.

Study design and sampling
The Rishi Valley Prospective Cohort Study was a populationbased prospective cohort study of CVD, as previously described [13][14][15][16][17].In brief, 7935 residents of the Rishi Valley, Andhra Pradesh, India, aged ≥18 years were surveyed from August 2012 to December 2015.Follow-up, which occurred between June 2019 and November 2020, was limited to 6123 participants due to a local dispute and COVID-19 pandemicrelated government restrictions (Figure S1).

Data collection techniques, procedures and measurements
Trained field workers collected data on socio-demographic and lifestyle factors, and self-reported disease status, using standardised questionnaires, and obtained standard clinical, anthropometric and laboratory measurements [16].
For follow-up, we undertook a house-to-house survey of 6123 residents who had participated in the baseline survey.
For anyone who had died, the next-of-kin was interviewed using a standard verbal autopsy (VA) tool [18,19], supplemented by reviewing death certificates and/or medical records, if available.The validated SmartVA-Analyze system was used to ascertain the cause of death (CoD) using the Tariff method [20][21][22].For those who were alive, the cause of illness for non-fatal cardiovascular events was ascertained using the Rose cardiovascular survey tool [23], and/or selfreported CVD diagnosis, and/or by reviewing the participant's medical records, if available (Supplementary Methods).To assess the robustness of the VA tool for diagnosing COPD, we re-interviewed 48 next-of-kin from November 2022 to January 2023 (Supplementary Methods).

Risk factors
For each participant, we assessed both traditional risk factors (e.g., age, sex, total cholesterol, hypertension, diabetes mellitus, physical inactivity, fruit consumption and current smoking) and non-traditional risk factors (e.g., number of children and hours slept/night), as previously described [13][14][15][16].Hypertension was defined as systolic blood pressure (SBP) ≥140 mmHg, diastolic BP (DBP) ≥90 mmHg, self-reported use of antihypertensive medication or self-reported diagnosis of hypertension.Central obesity was defined according to the International Diabetes Federation (i.e., South-Asian cut-offs for waist circumference of ≥90 cm for men, ≥80 cm for women) [24,25].Fruit intake was classified according to selfreported number of servings of fruit/day.Self-reported night sleeping hours were classified as <6, 6 to <8 or ≥8 h.

Outcome
The primary outcome was a composite cardiovascular event, defined as one or more fatal or non-fatal cardiovascular event: coronary heart disease, ischaemic heart disease, heart failure, myocardial infarction, rheumatic heart disease, cardiomyopathy, intermittent claudication, coronary bypass surgery, angioplasty or stent insertion, stroke, or any selfreported heart disease.

Statistical analysis
Analyses were restricted to participants without a selfreported history of CVD at baseline and were undertaken using Stata version 17.0 (StataCorp, College Station, TX, USA) or R (version 4.1.2).Quantitative variables are reported as mean ± standard deviation (SD) and compared using Student's unpaired t test.Categorical variables are presented as frequency (percentage) and compared using chisquared.Missing values were not imputed [26].
The time-to-event was defined as the time from the baseline assessment to the date of diagnosis of a cardiovascular event, death, interview, or loss to follow-up, whichever came first.Participants who did not develop a cardiovascular event were censored at the date of their last follow-up interview, while those who died from non-CVD causes were censored at the date of death.Person-years (PYs) of follow-up for each participant were computed as the difference between the date of the baseline assessment and the date of diagnosis of a cardiovascular event, the date of death or the last follow-up interview, whichever came first.We calculated the crude and sex-specific incidence rate (per 1000 PYs) of a cardiovascular event and the proportion of incident cardiovascular events attributable to a given risk factor (PAF) [27].The average PAF for mutually adjusted traditional or non-traditional risk factors was calculated using the 'averisk' package in R and 95% CIs were computed using Monte Carlo simulation [28].
To develop the model, we first assessed the assumption of proportionality.Then, we assessed the strength of association between each risk factor and a cardiovascular outcome using univariable Cox proportional hazards regression (Table S1).In the multivariable Cox proportional hazards models, we initially included variables with p ≤ 0.25 in the univariable analysis or those with known association with cardiovascular events such as age, sex, and family history of CVD.Three separate multivariable Cox proportional hazards regression models were developed (see below).
Model 1 comprised traditional risk factors such as smoking, diabetes, hypertension, central obesity and fruit consumption, adjusted for age and sex.Model 2 comprised non-traditional risk factors such as sleeping hours, number of children, selfreported health status, cooking location, smoke in the kitchen, marital status, income, and educational status, adjusted for age and sex.A stepwise approach was used to select the best-fitting models from groups of traditional (Model 1) and nontraditional cardiovascular risk factors (Model 2).We used the Akaike information criterion and log-likelihood ratio tests to compare model performance.Model 3 comprised a mutually adjusted best-fitting model which included both traditional and non-traditional risk factors.We further stratified Model 3 by sex and age.p ≤ 0.05 (two-tailed) was considered statistically significant.Separate hazard ratios were calculated for nontraditional risk factors that were not dichotomised (Table S2).When variables were statistically significant (p ≤ 0.05), we calculated the adjusted average PAF.

Socio-demographic characteristics
Of 6123 participants in the baseline survey, 38 could not be located and 1054 only had a brief survey at baseline, leaving 5031 (63.4%) eligible for inclusion (Figure S1).We excluded a further 162 (3.2%) participants with a self-reported history of CVD, 54 (1.1%) participants due to a local political dispute precluding follow-up and 6 (0.1%) participants without complete time to event information (Supplementary Results, Figure S1).There were slight differences in the baseline characteristics of those who were and were not followed up (Supplementary Results, Table S3).A total of 4809 participants with 23,180 PYs of follow-up met the inclusion criteria, of whom 4471 (93.0%) were alive and 338 (7.0%) had died (Supplementary Results, Tables S4 and S5).Of these, the mean age was 45.3 years (SD = 16.0) and 2774 (57.7%) were women (Table 1).Compared with women, men had a greater prevalence of diabetes (53%), hypertension (18%) and current smoking (96-fold).Similarly, SBP and DBP were greater in men than women.Men and women had similar BMI and similar prevalence of central obesity, physical inactivity and family history of CVD.

CoD in the Rishi Valley Prospective Cohort Study
Non-communicable diseases were the leading CoD (77.8%), followed by injuries (11%) and communicable diseases (6%) (Table S5).Deaths from chronic respiratory disease were more frequent than expected in this region, so we undertook further investigation of CoD in this group (Supplementary Results, Tables S6 and S7).These additional data indicate that any potential overestimation of deaths from COPD did not confound the assessment of deaths due to CVD, the focus of this work.
Over two thirds of cardiovascular events occurred before the age of 65 years.Incidence was 96% greater in smokers, 163% greater in those with diabetes, 205% greater in those with hypertension, 57% greater in those who were physically inactive and 110% greater in those who were centrally obese, compared with those without these traditional risk factors.Incidence was progressively greater in those reporting fewer hours of sleep per night, poor self-reported health status, and lesser education.The incidence was lower in those with one or two children compared with those with no children or ≥3 children.The incidence of a cardiovascular event was greater in hypertensive men than women and in those with daily fruit intake of <2 servings, but no statistically significant sex-dependent differences were observed for any of the other risk factors assessed.

Risk factors for an incident cardiovascular event
In the fully adjusted Cox model, the risk of a cardiovascular event was significantly associated with older age, diabetes, hypertension and central obesity, but not male sex, smoking, or fruit intake (Table 3).For non-traditional risk factors, the risk of a cardiovascular event was significantly associated with self-reported sleep of <6 h/night, fewer children, and lesser education.For the most part, patterns of risk appeared to be slightly different in men and women (Table 4, Supplementary Results) and across the various age-bands (Table 5).For example, the associations of a cardiovascular event with sleeping hours and diabetes were more apparent in women than men, while the association between hypertension and central obesity was more apparent in men than women (Table 4).

PAF of cardiovascular risk factors
In the fully adjusted multivariable analysis, the greatest contributors to incident cardiovascular events were hypertension, accounting for 18% of the average PAF, followed by 12% for self-reported sleep of <8 h, 9% for central obesity and 4% each for diabetes mellitus and having ≥3 children (Figure 1).Non-traditional risk factors contributed 16% of the PAF for CVD, while traditional risk factors contributed 31%.The unadjusted PAF for each risk factor was greater than the mutually adjusted average PAF (Figure S2).

DISCUSSION
Our analysis provides three novel and important findings.First, the incidence of cardiovascular events in rural India was greater than expected compared with some, but not all, studies from resource-limited settings.Second, in a community-based cohort in rural India, we found that a substantial proportion of incident cardiovascular events were attributable to traditional risk factors such as hypertension, diabetes and central obesity.Third, the non-traditional risk factors of sleeping hours and number of children were significantly associated with the risk of an incident cardiovascular event, contributing half as much of the PAF as traditional risk factors.However, the patterns of risk were slightly different between men and women and across the various age-bands.Therefore, age-and sex-specific interventions targeting both traditional and non-traditional risk factors could substantially reduce the incidence of cardiovascular events in rural India.

Incidence of cardiovascular events
We observed a greater incidence of cardiovascular events in rural India than in some [12,29], but not all [30,31], previous studies in resource-limited settings.For example, the incidence rate we observed was two-fold greater than in similar rural Indian settings [12,32].Data generated in the PURE study indicate substantial variation in ageand sex-adjusted incidence rates in South Asian countries (i.e., Bangladesh: 17.07/1000 PYs, Pakistan: 9.7; and India: 4.4) [32].This variation could be explained by differences in settings, methods, and population characteristics among various cohorts.For example, the definition of a composite cardiovascular event in our study was broader than for other studies from India [12,32].Furthermore, the incidence of a cardiovascular event is greater in rural than urban settings [32].These factors may partly explain the greater incidence rate we observed.Further larger prospective studies are required to better quantify the incidence of cardiovascular events in rural India and how it varies geographically and demographically.

Traditional risk factors
A substantial proportion of incident cardiovascular events was attributable to modifiable traditional risk factors such as hypertension, diabetes, and central obesity, as has been reported in other similar settings [12,33,34].Our findings are comparable to those from recent larger studies such as the PURE study [3,32] and a recent systematic review and meta-analysis [35].Thus, better prevention and management of traditional risk factors could reduce the burden of cardiovascular events in resource-limited settings, but this will require global and local political and scientific priority and action [36,37].For this, the World Health Organization (WHO) aims to reduce the prevalence of hypertension and diabetes mellitus by 25% by 2025 [38].Indeed, previous evidence indicates that for every 10 mmHg reduction in SBP, the risk of a cardiovascular event is reduced by 20% [39], so population-wide reductions in these risk factors will likely contribute to the WHO effort.Additionally, integrated management of hypertension and diabetes in primary care by introducing accessible and free clinics could provide one way to reduce these cardiovascular risk factors [40,41].
Compared with HICs, Asian countries have lesser prevalence of overweight or obesity as assessed by BMI but have greater prevalence of central obesity [42].Therefore, as recently recommended [43], central obesity could be a target for early prevention of cardiometabolic diseases even in those with normal weight in these regions [44].

Non-traditional risk factors
Non-traditional risk factors comprised a substantial proportion of the PAF for incident cardiovascular events.Self-reported short sleeping hours at night were associated with cardiovascular events in rural India, as previously reported in LMICs [45].However, there is a growing body of evidence that long sleep duration is also associated with the risk of a cardiovascular event [46][47][48].This is supported by the findings of a recent meta-analysis in which both short (<7 h) and long sleep duration (>9 h) were associated with increased risk of a cardiovascular event [49].Our inability to detect an association in those with long sleep duration could be due to our crude categorisation of sleep (<6, 6-8 and >8 h).Overall, evidence exists that both insufficient and excessive sleep durations are related to an increased risk of a cardiovascular event.However, while mechanisms underlying the link between sleep duration and a cardiovascular event remain unclear [49], sleep duration could be influenced by socio-cultural, psychological, behavioural, pathophysiological and environmental factors [50], all factors that may affect cardiovascular risk.Nevertheless, maintaining optimal sleep duration (i.e., 7-9 h per night) could be an important strategy to overcome the development of cardiovascular events and promote cardiovascular health while preventing associated risk factors [51].Furthermore, the quality of sleep is also necessary for better cardiovascular health, as factors such as frequent awakenings or not reaching deep sleep stages, poor-quality sleep, and sleep disorders can also negatively affect cardiovascular health [52,53].Therefore, public education on the importance of healthy sleep habits, including that individuals ensure sufficient, but not excessive, sleep, could be an important strategy to enhance both primordial and primary cardiovascular prevention at the population level [54].However, further research is required to better understand the mechanisms underlying the link between sleep duration and cardiovascular events and to explore the role of sleep in preventing CVDs and its risk factors.Such knowledge could potentially lead to more targeted interventions and treatments [54,55].Similar to reports from other LMICs, having ≥3 children was associated with the development of cardiovascular events [56].Some [56,57], but not all [58], evidence indicates a J-or U-shaped association between number of children and cardiovascular events.Having multiple children may result in greater socio-economic burden and this could increase the risk of CVD [35].Given that the risk appears to be similar in men and women, the underlying mechanism could be associated with socio-economic status.For example, in recent evidence from Europe and Israel lower educational status, regional variations, and socioeconomic position were the strongest predictors of the association between the number of children and a cardiovascular event in both men and women [59,60].Moreover, there may also be biological, socio-cultural, or behavioural and lifestyle mediators that explain this association.Further research is required to reveal the underlying mechanisms.
In general, in addition to the traditional risk factors, non-traditional risk factors such as short or long sleep duration and having ≥3 children contribute a substantial proportion of the PAF for incident cardiovascular events.Therefore, targeting these non-traditional cardiovascular risk factors or considering them as indicators of cardiovascular health could be an important strategy for preventing CVD.Moreover, public education on the significance of healthy sleep habits and creating awareness about the link between the number of children and a cardiovascular event might assist in enhancing both primordial and primary prevention mechanisms at the population level, thereby reducing the CVD burden in LMICs.However, the relationships between these non-traditional risk factors and a cardiovascular event are probably complex and influenced by several factors, so merit further investigation.
The patterns of risk and strength of association for cardiovascular events were slightly different between men and women and across the various age-bands.This difference in the strength of association is consistent with prior work in both LMICs [29] and HICs [61].A plausible explanation could be differences in the cardiovascular risk profile between men and women and between various age-bands.For example, in our study, men were older and had greater prevalence of hypertension than women.These gender differences could be at least partly related to sex-hormones [62].Therefore, as also indicated in a recent Lancet Commission paper, age-and sex-specific strategies in modifying risk factors could help prevent cardiovascular events in rural India [63].

Strengths and limitations
To the best of our knowledge, this is the first examination of the relative contributions of both 'traditional' and 'nontraditional' risk factors in the development of cardiovascular events in rural India.Our sample size comprised 4809 participants with 23,180 PYs of follow-up and 202 cardiovascular events, providing robust estimates of risk.
There are also some limitations.Our data are not nationally representative, so the findings might not be generalisable to people from other regions in India.Losses to follow-up due to COVID-19 might also affect generalisability, particularly for age-and sex-specific risk estimates, so should be interpreted with caution.Larger prospective studies with a longer follow-up period could provide better quantification of the sex-and age-specific risk factors for cardiovascular events in rural India.However, our findings are comparable to those of the larger PURE study conducted in urban and rural settings in LMICs [3,32], adding credence to our findings.Furthermore, self-reported CVD status for identifying non-fatal cardiovascular events and the VA technique for identifying fatal cardiovascular events could over-or underestimate incidence.However, previous research in a similar setting [64], as well as from a HIC setting [65], has shown that cardiovascular events are unlikely to be missed, supporting the validity of self-reported CVD status for epidemiological studies [65].Moreover, we used a combination of questions to increase validity [66], including the validated Rose questionnaire [67] and available medical records, in addition to self-reported identification of cardiovascular events.We also might have introduced bias by dichotomising non-traditional risk factors for the purpose of generating PAFs.Additionally, self-reported sleep duration was not measured objectively, although the use of actigraphy F I G U R E 1 Population attributable fraction and 95% confidence intervals from the multivariable analysis of potential risk factors for a cardiovascular event: (a) adjusted for hypertension, diabetes, central obesity, night sleeping hours and ≥3 children; (b) adjusted for all the variables in "a" plus potential confounders (i.e., age, sex, current smoking, educational status and self-reported health status).h, hours.Central obesity is defined as waist circumference ≥90 cm for men and ≥80 cm for women.
and polysomnography might not be feasible in such large samples [45].Herein, even though death from chronic respiratory disease was not our major objective, deaths due to COPD were surprisingly greater than expected in this region, a finding which should be interpreted with caution (see Supplementary Discussion).

CONCLUSIONS
Both traditional and non-traditional risk factors accounted for substantial proportions of incident cardiovascular events, with non-traditional risk factors contributing almost half that of traditional risk factors.Therefore, along with targeting modifiable traditional cardiometabolic risk factors, age-and sex-specific targeted prevention strategies for nontraditional risk factors could play a pivotal role in reducing the incidence of cardiovascular events in LMICs.
Baseline characteristics and number of new cardiovascular events among participants without a history of CVD in the Rishi Valley Prospective Cohort Study.Data are presented as mean (SD) or frequency (%).p Values compared men and women and were derived from the independent sample t test (continuous variables) or chisquared test (categorical variables).The p value for cardiovascular events was derived from a two-sample test of proportions.Central obesity: waist circumference ≥90 cm for men and ≥80 cm for women.Sex-specific crude incidence rates per 1000 person-years (PYs) of a cardiovascular event by participant characteristics in the Rishi Valley Prospective Cohort Study.
T A B L E 1Note: Abbreviations: BMI, body mass index; CVD, cardiovascular disease; DBP, diastolic blood pressure; FH, family history; PY, person-year; SBP, systolic blood pressure; SD, standard deviation.a 1-73 missing observations.b 138-371 missing observations.c 827-1064 missing observations.T A B L E 2 T A B L E 3 Associations between traditional and/or non-traditional cardiovascular risk factors and incident cardiovascular events in the Rishi Valley Prospective Cohort Study 2012-2020.Model 1 was adjusted for age, sex, current smoking, diabetes, hypertension, central obesity and fruit intake; Model 2 was adjusted for age, sex, sleeping hours, number of children, self-reported health status and educational status; and Model 3 is adjusted for both Model 1 and Model 2. Central obesity is defined as waist circumference ≥90 cm for men and ≥80 cm for women.Abbreviations: CI, confidence interval; CVD, cardiovascular disease; HR, hazard ratio. Note: Multivariable a -adjusted hazard ratios (HRs) and 95% CI of incident cardiovascular event stratified by sex in the Rishi Valley Prospective Cohort Study 2012-2020.Central obesity is defined as waist circumference ≥90 cm for men and ≥80 cm for women.Abbreviations: CI, confidence interval; CVD, cardiovascular disease; HR, hazard ratio.
T A B L E 4Note: a Multivariable analyses are adjusted for all other variables in the column.
T A B L E 5 Multivariable a -adjusted hazard ratios (HRs) and 95% CI of incident cardiovascular event stratified by age in the Rishi Valley Prospective Cohort Study 2012-2020.Central obesity is defined as waist circumference ≥90 cm for men and ≥80 cm for women.Abbreviations: CI, confidence interval; CVD, cardiovascular disease; HR, hazard ratio.
Note:a Multivariable analyses are adjusted for other variables in the column.TROPICAL MEDICINE & INTERNATIONAL HEALTH