Stroke risk prediction using machine learning: a prospective cohort study of 0.5 million Chinese adults

OBJECTIVE

To compare Cox models, machine learning (ML), and ensemble models combining both approaches, for prediction of stroke risk in a prospective study of Chinese adults.

MATERIALS AND METHODS

We evaluated models for stroke risk at varying intervals of follow-up (<9 years, 0-3 years, 3-6 years, 6-9 years) in 503 842 adults without prior history of stroke recruited from 10 areas in China in 2004-2008. Inputs included sociodemographic factors, diet, medical history, physical activity, and physical measurements. We compared discrimination and calibration of Cox regression, logistic regression, support vector machines, random survival forests, gradient boosted trees (GBT), and multilayer perceptrons, benchmarking performance against the 2017 Framingham Stroke Risk Profile. We then developed an ensemble approach to identify individuals at high risk of stroke (>10% predicted 9-yr stroke risk) by selectively applying either a GBT or Cox model based on individual-level characteristics.

RESULTS

For 9-yr stroke risk prediction, GBT provided the best discrimination (AUROC: 0.833 in men, 0.836 in women) and calibration, with consistent results in each interval of follow-up. The ensemble approach yielded incrementally higher accuracy (men: 76%, women: 80%), specificity (men: 76%, women: 81%), and positive predictive value (men: 26%, women: 24%) compared to any of the single-model approaches.

DISCUSSION AND CONCLUSION

Among several approaches, an ensemble model combining both GBT and Cox models achieved the best performance for identifying individuals at high risk of stroke in a contemporary study of Chinese adults. The results highlight the potential value of expanding the use of ML in clinical practice.

Tea consumption and risk of stroke in Chinese adults: a prospective cohort study of 0.5 million men and women

BACKGROUND

Many cohort studies have explored the relation between tea consumption and stroke risk; however, the conclusions have been inconsistent. In addition, evidence is lacking in China, where the patterns of tea consumption and main types of tea consumed differ substantially from those in high-income countries.

OBJECTIVE

We aimed to systematically assess the association of tea consumption with the risk of stroke based on a Chinese large-scale cohort study.

METHODS

A total of 487,377 participants from the China Kadoorie Biobank were included in the present study. Detailed information about tea consumption (including frequency, duration, amount, and tea type) was self-reported at baseline. After ∼4.3 million person-years of follow-up, 38,727 incident cases of stroke were recorded, mainly through linkage with mortality and morbidity registries and based on the national health insurance system.

RESULTS

Overall, 128,280 adults (26.3%) reported drinking tea almost daily (41.4% men, 15.9% women), predominantly green tea (86.7%). Tea consumption had an inverse and dose-response relation with the risk of stroke (Ptrend < 0.001). Compared with nonconsumers, those who consumed tea occasionally, weekly, and daily had adjusted HRs and 95% CIs of 0.96 (0.94, 0.99), 0.94 (0.90, 0.98), and 0.92 (0.89, 0.95) respectively, with little difference by stroke type. Among those who consumed tea daily, the HRs for stroke decreased with the increasing duration and amount of tea consumed (all P < 0.001). These inverse associations were significant for green tea but not for other types of tea. Among men, but not women, the inverse relations could be detected, and similar inverse associations could be found for male noncurrent alcohol-consumers and noncurrent smokers as well.

CONCLUSIONS

Among Chinese adults, higher consumption of tea, especially green tea, was associated with a lower risk of ischemic and hemorrhagic stroke.

Association Between Active Commuting and Incident Cardiovascular Diseases in Chinese: A Prospective Cohort Study

Background Active commuting is related to a higher level of physical activity but more exposure to ambient air pollutants. With the rather serious air pollution in urban China, we aimed to examine the association between active commuting and risk of incident cardiovascular disease in the Chinese population. Methods and Results A total of 104 170 urban commuters without major chronic diseases at baseline were included from China Kadoorie Biobank. Self-reported commuting mode was defined as nonactive commuting, work at home or near home, walking, and cycling. Multivariable Cox regression was used to examine associations between commuting mode and cardiovascular disease. Overall, 47.2% of the participants reported nonactive commuting, 13.4% reported work at home or work near home, 20.1% reported walking, and 19.4% reported cycling. During a median follow-up of 10 years, we identified 5374 incidents of ischemic heart disease, 664 events of hemorrhagic stroke, and 4834 events of ischemic stroke. After adjusting for sex, socioeconomic status, lifestyle factors, sedentary time, body mass index, comorbidities, household air pollution, passive smoking, and other domain physical activity, walking (hazard ratio, 0.90; 95% CI, 0.84-0.96) and cycling (hazard ratio, 0.81; 95% CI, 0.74-0.88) were associated with a lower risk of ischemic heart disease than nonactive commuting. Cycling was associated with a lower risk of ischemic stroke (hazard ratio, 0.92; 95% CI, 0.84-1.00). No significant association was found of walking or cycling with hemorrhagic stroke. The associations of commuting mode with major cardiovascular disease were consistent among men and women and across different levels of other domain physical activity. Conclusions In urban China, cycling was associated with a lower risk of ischemic heart disease and ischemic stroke. Walking was associated with a lower risk of ischemic heart disease.

Prenatal famine exposure, adulthood obesity patterns and risk of type 2 diabetes

Background

Prenatal exposure to famine and adulthood obesity have been independently related to the risk of type 2 diabetes; however, little is known about the joint effects of these risk factors at different stages of life on adulthood diabetes risk.

Methods

The analysis included 88 830 participants of the China Kadoorie Biobank, who were born around the time of the Chinese Great Famine and without diabetes, cardiovascular diseases, or cancer at baseline. We defined famine exposure subgroups as nonexposed (born between 1 October 1962 and 30 September 964), fetal-exposed (born between 1 October 1959 and 30 September 1961) and early-childhood exposed (born between 1 October 1956 and 30 September 1958). General obesity was assessed by body mass index (BMI: overweight ≥ 24.0, obesity ≥ 28.0) and abdominal obesity assessed by waist-to-hip ratio (WHR, men/women: moderate ≥ 0.90/0.85, high ≥ 0.95/0.90).

Results

During a median 7.3 years (642 552 person-years) of follow-up, we identified 1372 incident cases of type 2 diabetes. Compared with nonexposed and early-childhood exposed participants combined as a single comparison group, fetal-exposed participants showed an increased risk of diabetes in adulthood [hazard ratio (HR) = 1.25; 95% confidence interval (CI): 1.07-1.45]. The association between general obesity and diabetes was consistent across subgroups according to famine exposure (P for interaction > 0.05). A stronger association between abdominal obesity and diabetes was observed in the fetal-exposed subgroup than in other subgroups (P for interaction = 0.025 in the whole population). This interaction was more obvious in women (P = 0.013) but not in men (P = 0.699). Compared with normal-BMI and -WHR participants, those with both general (BMI ≥ 24.0) and abdominal (WHR ≥ 0.90/0.85) obesity in adulthood had 5.32 (95% CI: 3.81-7.43)-, 3.13 (2.48-3.94)- and 4.43 (3.45-5.68)-fold higher risks if these were carried during, before and after times of famine, respectively.

Conclusions

Coexistence of prenatal experience of undernutrition and abdominal obesity in adulthood was associated with a higher risk of type 2 diabetes.

Self-Rated Health Status and Risk of Ischemic Heart Disease in the China Kadoorie Biobank Study: A Population-Based Cohort Study

BACKGROUND

Self-rated health (SRH) is a strong predictor of mortality in different populations. However, the associations between SRH measures and risk of ischemic heart disease (IHD) have not been extensively explored, especially in a Chinese population.

METHODS AND RESULTS

More than 500 000 adults from 10 cities in China were followed from baseline (2004-2008) through December 31, 2013. Global and age-comparative SRH were reported from baseline questionnaires. Incident IHD cases were identified through links to well-established disease registry systems and the national health insurance system. During 3 423 542 person-years of follow-up, we identified 24 705 incident cases of IHD. In multivariable-adjusted models, both global and age-comparative SRH was significantly associated with incident IHD. Compared with excellent SRH, the hazard ratios for good, fair, and poor SRH were 1.02 (95% confidence interval [CI], 0.98-1.07), 1.32 (95% CI, 1.27-1.37), and 1.76 (95% CI, 1.68-1.85), respectively. Compared with better age-comparative SRH, the hazard ratios for same and worse age-comparative SRH were 1.23 (95% CI, 1.19-1.27) and 1.78 (95% CI, 1.70-1.86), respectively. The associations persisted in all subgroup analyses, although they were slightly modified by study location, education, and income levels.

CONCLUSIONS

A simple questionnaire for self-assessment of health status was significantly associated with incident IHD in Chinese adults. Individuals and healthcare providers can use SRH measures as a convenient tool for assessing future IHD risk.

Associations of blood glucose and prevalent diabetes with risk of cardiovascular disease in 500 000 adult Chinese: the China Kadoorie Biobank

AIMS

To examine the relationship of self-reported diabetes, and of random blood glucose levels among individuals without known diabetes, with the prevalence of cardiovascular disease in Chinese adults.

METHODS

We examined cross-sectional data from the China Kadoorie Biobank of 0.5 million people aged 30-79 years recruited from 10 diverse regions of China in the period 2004-2008. Logistic regression was used to estimate the odds ratios of prevalent cardiovascular disease associated with self-reported diabetes, and with measured random blood glucose levels among participants with no history of diabetes, adjusting simultaneously for age, sex, area, education, smoking, alcohol, blood pressure and physical activity.

RESULTS

A total of 3.2% of participants had self-reported diabetes (men 2.9%; women 3.3%) and 2.8% had screen-detected diabetes (men 2.6%; women 2.8%), i.e. they had no self-reported history of diabetes but a blood glucose level suggestive of a diagnosis of diabetes. Compared with individuals without a history of diabetes, the odds ratios associated with self-reported diabetes were 2.18 (95% CI 2.06-2.30) and 1.88 (95% CI 1.75-2.01) for prevalent ischaemic heart disease and stroke/transient ischaemic attack, respectively. Among participants without self-reported diabetes there was a positive association between random blood glucose and ischaemic heart disease and stroke/transient ischaemic attack prevalence (P for trend <0.0001). Below the diabetic threshold (<11.1 mmol/l) each additional 1 mmol/l of random blood glucose was associated with 4% (95% CI 2-5%) and 5% (95% CI 3-7%) higher odds of prevalent ischaemic heart disease and stroke/transient ischaemic attack, respectively.

CONCLUSIONS

In this adult Chinese population, self-reported diabetes was associated with a doubling of the odds of prevalent cardiovascular disease. Below the threshold for diabetes there was still a modest, positive association between random blood glucose and prevalent cardiovascular disease.