Cohort Profile: The Center for cArdiometabolic Risk Reduction in South Asia (CARRS)

Towards equitable AI in oncology

Artificial intelligence (AI) stands at the threshold of revolutionizing clinical oncology, with considerable potential to improve early cancer detection and risk assessment, and to enable more accurate personalized treatment recommendations. However, a notable imbalance exists in the distribution of the benefits of AI, which disproportionately favour those living in specific geographical locations and in specific populations. In this Perspective, we discuss the need to foster the development of equitable AI tools that are both accurate in and accessible to a diverse range of patient populations, including those in low-income to middle-income countries. We also discuss some of the challenges and potential solutions in attaining equitable AI, including addressing the historically limited representation of diverse populations in existing clinical datasets and the use of inadequate clinical validation methods. Additionally, we focus on extant sources of inequity including the type of model approach (such as deep learning, and feature engineering-based methods), the implications of dataset curation strategies, the need for rigorous validation across a variety of populations and settings, and the risk of introducing contextual bias that comes with developing tools predominantly in high-income countries.

Natural History of Type 2 Diabetes in Indians: Time to Progression

OBJECTIVE

To describe the natural history of diabetes in Indians.

RESEARCH DESIGN AND METHODS

Data are from participants older than 20 years in the Centre for Cardiometabolic Risk Reduction in South Asia longitudinal study. Glycemic states were defined per American Diabetes Association criteria. Markov models were used to estimate annual transition probabilities and sojourn time through states.

RESULTS

Among 2,714 diabetes-free participants, 641 had isolated impaired fasting glucose (iIFG), and 341 had impaired glucose tolerance (IGT). The annual transition to diabetes for those with IGT was 13.9% (95% CI 12.0, 15.9) versus 8.6% (7.3, 9.8) for iIFG. In the normoglycemia ↔ iIFG → diabetes model, mean sojourn time in normoglycemia was 40.3 (34.6, 48.2) years, and sojourn time in iIFG was 9.7 (8.4, 11.4) years. For the normoglycemia ↔ IGT → diabetes model, mean sojourn time in normoglycemia was 34.5 (29.5, 40.8) years, and sojourn time in IGT was 6.1 (5.3, 7.1) years.

CONCLUSIONS

Individuals reside in normoglycemia for 35-40 years; however, progression from prediabetes to diabetes is rapid.

Longitudinal associations between ambient PM2.5 exposure and lipid levels in two Indian cities

Background

Exposure to ambient PM2.5 is known to affect lipid metabolism through systemic inflammation and oxidative stress. Evidence from developing countries, such as India with high levels of ambient PM2.5 and distinct lipid profiles, is sparse.

Methods

Longitudinal nonlinear mixed-effects analysis was conducted on >10,000 participants of Centre for cArdiometabolic Risk Reduction in South Asia (CARRS) cohort in Chennai and Delhi, India. We examined associations between 1-month and 1-year average ambient PM2.5 exposure derived from the spatiotemporal model and lipid levels (total cholesterol [TC], triglycerides [TRIG], high-density lipoprotein cholesterol [HDL-C], and low-density lipoprotein cholesterol [LDL-C]) measured longitudinally, adjusting for residential and neighborhood-level confounders.

Results

The mean annual exposure in Chennai and Delhi was 40 and 102 μg/m3 respectively. Elevated ambient PM2.5 levels were associated with an increase in LDL-C and TC at levels up to 100 µg/m3 in both cities and beyond 125 µg/m3 in Delhi. TRIG levels in Chennai increased until 40 µg/m3 for both short- and long-term exposures, then stabilized or declined, while in Delhi, there was a consistent rise with increasing annual exposures. HDL-C showed an increase in both cities against monthly average exposure. HDL-C decreased slightly in Chennai with an increase in long-term exposure, whereas it decreased beyond 130 µg/m3 in Delhi.

Conclusion

These findings demonstrate diverse associations between a wide range of ambient PM2.5 and lipid levels in an understudied South Asian population. Further research is needed to establish causality and develop targeted interventions to mitigate the impact of air pollution on lipid metabolism and cardiovascular health.

Evaluating bias with loss to follow-up in a community-based cohort: empirical investigation from the CARRS Study

BACKGROUND

Retention of participants is a challenge in community-based longitudinal cohort studies. We aim to evaluate the factors associated with loss to follow-up and estimate attrition bias.

METHODS

Data are from an ongoing cohort study, Center for cArdiometabolic Risk Reduction in South Asia (CARRS) in India (Delhi and Chennai). Multinomial logistic regression analysis was used to identify sociodemographic factors associated with partial (at least one follow-up) or no follow-up (loss to follow-up). We also examined the impact of participant attrition on the magnitude of observed associations using relative ORs (RORs) of hypertension and diabetes (prevalent cases) with baseline sociodemographic factors.

RESULTS

There were 12 270 CARRS cohort members enrolled in Chennai and Delhi at baseline in 2010, and subsequently six follow-ups were conducted between 2011 and 2022. The median follow-up time was 9.5 years (IQR: 9.3-9.8) and 1048 deaths occurred. Approximately 3.1% of participants had no follow-up after the baseline visit. Younger (relative risk ratio (RRR): 1.14; 1.04 to 1.24), unmarried participants (RRR: 1.75; 1.45 to 2.11) and those with low household assets (RRR: 1.63; 1.44 to 1.85) had higher odds of being lost to follow-up. The RORs of sociodemographic factors with diabetes and hypertension did not statistically differ between baseline and sixth follow-up, suggesting minimal potential for bias in inference at follow-up.

CONCLUSION

In this representative cohort of urban Indians, we found low attrition and minimal bias due to the loss to follow-up. Our cohort's inconsistent participation bias shows our retention strategies like open communication, providing health profiles, etc have potential benefits.

The case for precision medicine in the prevention, diagnosis, and treatment of cardiometabolic diseases in low-income and middle-income countries

Cardiometabolic diseases are the leading preventable causes of death in most geographies. The causes, clinical presentations, and pathogenesis of cardiometabolic diseases vary greatly worldwide, as do the resources and strategies needed to prevent and treat them. Therefore, there is no single solution and health care should be optimised, if not to the individual (ie, personalised health care), then at least to population subgroups (ie, precision medicine). This optimisation should involve tailoring health care to individual disease characteristics according to ethnicity, biology, behaviour, environment, and subjective person-level characteristics. The capacity and availability of local resources and infrastructures should also be considered. Evidence needed for equitable precision medicine cannot be generated without adequate data from all target populations, and the idea that research done in high-income countries will transfer adequately to low-income and middle-income countries (LMICs) is problematic, as many migration studies and transethnic comparisons have shown. However, most data for precision medicine research are derived from people of European ancestry living in high-income countries. In this Series paper, we discuss the case for precision medicine for cardiometabolic diseases in LMICs, the barriers and enablers, and key considerations for implementation. We focus on three propositions: first, failure to explore and implement precision medicine for cardiometabolic disease in LMICs will enhance global health disparities. Second, some LMICs might already be placed to implement cardiometabolic precision medicine under appropriate circumstances, owing to progress made in treating infectious diseases. Third, improvements in population health from precision medicine are most probably asymptotic; the greatest gains are more likely to be obtained in countries where health-care systems are less developed. We outline key recommendations for implementation of precision medicine approaches in LMICs.

PM2.5 exposure, glycemic markers and incidence of type 2 diabetes in two large Indian cities

INTRODUCTION

Exposure to fine particulate matter has been associated with several cardiovascular and cardiometabolic diseases. However, such evidence mostly originates from low-pollution settings or cross-sectional studies, thus necessitating evidence from regions with high air pollution levels, such as India, where the burden of non-communicable diseases is high.

RESEARCH DESIGN AND METHODS

We studied the associations between ambient PM2.5 levels and fasting plasma glucose (FPG), glycosylated hemoglobin (HbA1c) and incident type 2 diabetes mellitus (T2DM) among 12 064 participants in an adult cohort from urban Chennai and Delhi, India. A meta-analytic approach was used to combine estimates, obtained from mixed-effects models and proportional hazards models, from the two cities.

RESULTS

We observed that 10 μg/m3 differences in monthly average exposure to PM2.5 was associated with a 0.40 mg/dL increase in FPG (95% CI 0.22 to 0.58) and 0.021 unit increase in HbA1c (95% CI 0.009 to 0.032). Further, 10 μg/m3 differences in annual average PM2.5 was associated with 1.22 (95% CI 1.09 to 1.36) times increased risk of incident T2DM, with non-linear exposure response.

CONCLUSIONS

We observed evidence of temporal association between PM2.5 exposure, and higher FPG and incident T2DM in two urban environments in India, thus highlighting the potential for population-based mitigation policies to reduce the growing burden of diabetes.

Epidemiological pattern of COVID-19 and its association with periodontal health in an urban Indian cohort

Background

Studies have highlighted a possible influence of gingival and periodontal disease (PD) on COVID-19 risk and severity. However, the evidence is based on hospital-based studies and community-level data are sparse.

Objectives

We described the epidemiological pattern of SARS-CoV-2 infection in Delhi and evaluated the associations of gingival and PD with incident COVID-19 disease in a regionally representative urban Indian population.

Methods

In a prospective study nested within the Centre for Cardiometabolic Risk Reduction in South-Asia (CARRS) study, participants with clinical gingival and periodontal status available at baseline (2014–16) (n = 1,727) were approached between October 2021 to March 2022. Information on COVID-19 incidence, testing, management, severity was collected as per the WHO case criteria along with COVID-19 vaccination status. Absolute incidence of COVID-19 disease was computed by age, sex, and oral health. Differences in rates were tested using log-rank test. Poisson regression models were used to evaluate independent associations between gingival and PD and incidence of COVID-19, adjusted for socio-demographic and behavioral factors, presence of comorbidity, and medication use.

Results

Among 1,727 participants, the mean age was 44.0 years, 45.7% were men, 84.5% participants had baseline gingival or PD and 89.4% participants had received at least one dose of COVID-19 vaccine. Overall, 35% (n = 606) participants were tested for COVID-19 and 24% (n = 146/606) tested positive. As per the WHO criteria total number of cases was 210, constituting 12% of the total population. The age and sex-specific rates of COVID-19 were higher among men and older participants, but women aged >60 years had higher rates than men of same age. The incidence rate did not differ significantly between those having gingival or PD and healthy periodontium (19.1 vs. 16.5/1,000 person-years) and there was no difference in risk of COVID-19 by baseline oral disease status.

Conclusion

Gingival and PD were not associated with increased risk of COVID-19.

Estimated glomerular filtration rate trajectories in south Asians: Findings from the cardiometabolic risk reduction in south Asia study

Background

Few longitudinal data characterize kidney function decline among South Asians, one of the world's largest population groups. We aimed to identify estimated glomerular filtration rate (eGFR) trajectories in a population-based cohort from India and assess predictors of rapid kidney function decline.

Methods

We used 6-year longitudinal data from participants of a population-representative study from Delhi and Chennai, India who had at least two serum creatinine measures and baseline CKD-EPI eGFR> 60 ml/min/1.73m² (n=7779). We used latent class trajectory modeling to identify patterns of kidney function trajectory (CKD-EPI eGFR) over time. In models accounting for age, sex, education, and city, we tested the association between 15 hypothesized risk factors and rapid kidney function decline.

Findings

Baseline mean eGFR was 108 (SD 16); median eGFR was 110 [IQR: 99-119] ml/min/1.73m². Latent class trajectory modeling and functional characterization identified three distinct patterns of eGFR: class-1 (no decline; 58%) annual eGFR change 0.2 [0.1, 0.3]; class-2 (slow decline; 40%) annual eGFR change -0.2 [-0.4, -0.1], and class-3 (rapid decline; 2%) annual eGFR change -2.7 [-3.4, -2.0] ml/min/1.73m². Albuminuria (>30 mg/g) was associated with rapid eGFR decline (OR for class-3 vs class-1: 5.1 [95% CI: 3.2; 7.9]; class-3 vs. class-2: 4.3 [95% CI:2.7; 6.6]). Other risk factors including self-reported diabetes, cardiovascular disease, peripheral arterial disease, and metabolic biomarkers such as HbA1c and systolic blood pressure were associated with rapid eGFR decline phenotype but potential 'non-traditional' risk factors such as manual labor or household water sources were not.

Interpretation

Although mean and median eGFRs in our population-based cohort were higher than those reported in European cohorts, we found that a sizeable number of adults residing in urban India are experiencing rapid kidney function decline. Early and aggressive risk modification among persons with albuminuria could improve kidney health among South Asians.

Funding

The CARRS study has been funded with federal funds from the National Heart, Lung, and Blood Institute, National Institutes of Health, under Contract No. HHSN2682009900026C and P01HL154996. Dr. Anand was supported by NIDDK K23DK101826 and R01DK127138.

Association of family history of cardiometabolic diseases (CMDs) and individual health behaviours: Analysis of CARRS study from South Asia

Objectives

Family history is considered as an important predictor of cardiovascular diseases (CVDs) and diabetes. Available research findings suggest that family history of chronic diseases is associated with perceived risk of disease and adoption of healthy behaviours. We examined the association between family history of cardio-metabolic diseases (CMDs) and healthy behaviours among adults without self-reported CMDs.

Methods

Cross-sectional data of 12,484 adults, without self-reported CMDs, from the baseline survey of Centre for cArdiometabolic Risk Reduction in South-Asia (CARRS) cohort study were analysed.

Results

Family history was positively associated with non-smoking and high fruits & vegetables consumption in the age group of 45–64 years and moderate to high physical activity in the age group ≥65 years after adjusting for sex, education, wealth index, city and body mass index.

Conclusions

Understanding perceived risks and cultural or psychological factors related to family history through ethnographic studies may deepen understanding of these associations.