Transforming Chinese cohort studies through artificial intelligence: a new era of population health research

Abstract

Tien Yin Wong and colleagues explore how artificial intelligence can tackle persistent challenges in Chinese epidemiological studies and analyse the regulatory frameworks and barriers to implementation that must be overcome to ensure equitable, scientifically rigorous population health research

China presents a unique position for conducting large scale cohort studies, with distinct advantages including a population of 1.4 billion, 56 ethnic groups, low international migration, and a rapidly digitalising healthcare system. China has made significant strides in developing large scale cohort studies over the past two decades, exemplified by the establishment of the China Kadoorie Biobank (CKB) in 2004.¹ The National Strategic Science and Technology Innovation Framework (2016-20),² China’s first comprehensive national level initiative, provided unprecedented funding for population scale health research, marking the government’s inaugural systematic commitment to large scale cohort studies.

However, a fundamental challenge facing China’s cohort research is the uneven development of research facilities at the primary healthcare level, particularly in rural and under-resourced regions. Primary healthcare facilities, which comprise 95% of China’s one million medical institutions and manage half of all patient visits (9.55 billion annually), often lack basic research resources, including trained research staff and standardised digital infrastructure, and vary widely in clinical quality.³ This infrastructural deficiency hinders routine data collection, longitudinal tracking, and follow-up—essential elements of scientifically valid cohort studies in the community. As a result, data fragmentation at the community level persists, with only 16.6% of cohort studies achieving follow-up longer than 10 years and just 10.6% (52/470) using standardised data sharing platforms.^{4 5} These limitations contribute to the systematic under-representation of rural populations, thereby reducing both the internal validity and population level applicability of research findings. This imbalance is reflected in the geographical distribution of cohort participants: 76% are drawn from six developed regions (Beijing, Shanghai, Sichuan, Jiangsu, Guangdong, and Hong Kong), whereas rural regions accounting for 35% of the national population contribute only 12% of participants.^{4 6} These infrastructure driven disparities undermine the ability to efficiently identify region specific disease patterns, as shown by Li and colleagues’ cardiovascular study, which revealed significant regional variations in risk factors across rural and urban regions.⁷ As a result, population level interventions may lack validity and effectiveness for rural residents.

Overcoming these challenges requires solutions that not only overcome technical and infrastructural limitations but also respond to the complex and diverse health needs of China’s population. These include the heterogeneous disease patterns across 56 ethnic groups, the rising burden of chronic diseases across urban and rural settings, and the healthcare demands of the world’s largest ageing population. In this article, part of a BMJ Collection focusing on large scale cohort studies in China and internationally, we explore how artificial intelligence (AI) can be deployed to overcome key obstacles in Chinese cohort studies. Our analysis is grounded in China’s evolving national strategic frameworks, its unique healthcare data architecture, recent advances in medical AI and digital health, and ongoing health system reforms aimed at greater integration. By synthesising these developments, we aim to inform researchers, healthcare providers, and policy makers seeking to advance the scientific rigour, equity, and translational impact of cohort studies in China.

National strategic framework for AI enabled cohort research

China has implemented several strategic initiatives targeting healthcare digitisation, data governance, and implementation of AI that can meet the challenges in longitudinal population research (box 1). These frameworks facilitate cohort studies through comprehensive data collection, improved analytical capabilities, and enhanced participant engagement. Sung and colleagues’ analysis shows how epidemiological methods and cohort data can enhance development of AI by providing causal insights.¹¹ This one directional benefit is well illustrated in Chinese research, as exemplified by Cai and colleagues’ multicentre, retrospective cohort study, which used cohort data from multiple healthcare centres to develop AI driven models for accurate diagnosis of ovarian cancer.¹² However, the reverse direction—AI systems specifically designed to improve cohort study methods—remains under-represented in the published literature. Although some AI platforms have been developed for cohort applications, most current AI research focuses on clinical diagnostics rather than enhancing longitudinal population research capabilities. This creates a critical gap in bidirectional AI—the integration of AI into epidemiology research—and underscores an important area for further development.

Box 1. Key policy initiatives supporting integration of artificial intelligence in Chinese cohort studies.

• The National 13th Five-Year Plan for S&T Innovation (2016) ²—China's first national level initiative, which provides funding for million person cohorts, generating large scale datasets that serve as test beds for artificial intelligence (AI) tools

• National Data Governance and Integration Strategy (2024-26) ⁸—Established legal and technical infrastructure for cross-institutional sharing of health data. This marks the first nationwide effort to standardise medical data exchange protocols and facilitate integration of longitudinal datasets, while ensuring privacy protections

• Promoting the construction and development of medical consortiums (2023) ⁹—Established structured pathways for data sharing between urban and rural institutions, enhancing cohort recruitment and follow-up in underrepresented regions

• “AI+” initiative (2024) ¹⁰—Prioritised development of healthcare specific foundation models trained on medical data in China, advancing capabilities in cohort data analysis. Supported multi-scenario deployment of AI that enhanced recruitment of participants, monitoring of follow-up, and harmonisation of data across diverse healthcare settings

The nationwide “AI+” initiative reflects the government’s systematic approach to integrating AI across healthcare research. This initiative may accelerate the adoption of AI in healthcare research, with broader implementation of AI research platforms in hospitals—including those supporting cohort studies. Recently, many of these platforms are powered or complemented with DeepSeek as the backbone large language model.¹³ From December 2023 to December 2024, the government issued 44 major policies related to digital health and AI, emphasising standardisation and systematic development of the healthcare system. These policies establish a regulatory foundation that could theoretically support more standardised cohort research methods. However, significant challenges to implementation persist. The regulatory mechanisms governing AI enabled research platforms lack comprehensive evaluation standards and quality control processes. This indicates that although broader healthcare initiatives using AI contribute valuable technological infrastructure, cohort specific methodological challenges require more targeted policies that meet the distinct needs of longitudinal population research.

The translation from policy framework to research practice faces substantial barriers. The centralised policy approach, despite enabling theoretically uniform protocol deployment, encounters practical limitations including variable institutional technical capacity, inconsistent research workforce training across regions, and heterogeneous implementation environments. These factors raise questions about whether deployment of AI will achieve the intended methodological standardisation or even inadvertently amplify existing research inequities through uneven technological adoption in regions with less developed research resources.

Building robust data infrastructure for AI powered cohort studies: progress and challenges

A robust research data infrastructure forms the cornerstone of effective cohort research by providing systematic frameworks for data organisation, integration, and management across diverse sources. China’s recent data infrastructure developments are intended to meet these methodological requirements. The establishment of the National Data Bureau in 2023 and the subsequent National Data Governance and Integration Strategy (2024-26)⁸ have created a comprehensive framework for integration of healthcare data. These, in principle, facilitate the development of cohort studies. For cohort methodology, this infrastructure enables standardised pre-processing, advanced analytics, and quality control protocols for longitudinal data analysis. For example, Zhou and colleagues showed the practical application of this approach by developing an automated data collection tool for real world cohort studies of chronic hepatitis B by using optical character recognition and natural language processing technologies.¹⁴ When tested on more than 4000 patient records from two hospitals, the system maintained accuracy comparable to manual collection (98.66%) while reducing processing time from 63.64 to just 3.57 minutes per patient—a 17-fold improvement. The tool integrates with REDCap, features data security measures, and has been open sourced to support large scale, multicentre hepatitis B studies. This innovation of data infrastructure improved the efficiency and accuracy of data collection across diverse healthcare settings, offering a technological solution to overcome China’s fragmented healthcare data landscape and, in theory, can enhance cohort research representativeness.

To overcome concerns about privacy and data security, China has also implemented comprehensive protection measures including data encryption, de-identification algorithms, and controlled access protocols. Shanghai’s launch of a “medical data space” in September 2024 introduced federated learning approaches that theoretically allow data collaboration while maintaining compliance with local privacy regulations,¹⁵ featuring comprehensive datasets from 10 clinical departments and 20 major diseases.¹⁵ These datasets encompass various modalities including lung nodules, diabetic retinopathy, computed tomography fractional flow reserve, and breast ultrasonography, as well as specialised datasets for conditions such as spermatogenic dysfunction and genetic mutations for acute leukaemia with transplant prognosis. The federated learning approach allows simultaneous model training across multiple data sources while maintaining local data custody, preserving patients’ privacy while enabling broader analytical capabilities.

However, as indicated above, these research platforms and initiatives remain concentrated in major urban centres, with limited implementation in primary healthcare and community settings where longitudinal cohort research is most needed. As a comparison, the European Electronic Health Records for Clinical Research (EHR4CR) project illustrates the transformative potential of comprehensive data infrastructure when properly implemented.¹⁶ Through systematic standardisation, EHR4CR creates the essential foundations for cohort research excellence—ensuring data quality, facilitating secure cross-institutional sharing, and enabling seamless integration of baseline measurements with longitudinal follow-up data. Although China has adopted similar principles of standardisation and privacy protection in its data infrastructure initiatives, these efforts ought to be expanded equitably to rural areas so as to overcome existing geographical disparities. Furthermore, the lack of standardised regulations across regions threatens consistency of implementation, potentially reinforcing rather than resolving current imbalances in cohort representation.

Transforming cohort research: implementation of AI across Chinese healthcare settings

Integration of AI in epidemiological research can affect all elements of the cohort study, from the design and conduct to the analysis of cohort studies. The supplementary table summarises the major AI platforms implemented in Chinese healthcare settings and their specific applications for cohort research. At the study design stage, AI applications are enhancing generation of research questions and recruitment of participants through large language models specifically adapted for the Chinese healthcare context. Yidu Cloud, Alipay’s Anzhener, and JD Health’s AI Research Assistant, implemented across millions of patient visits, support cohort study design by analysing disease prevalence patterns to identify priority research areas and relevant participant inclusion criteria.¹⁷ These systems theoretically support researchers in designing more representative cohorts, particularly for conditions with regional variation.

Continuous monitoring systems powered by AI overcome traditional challenges in longitudinal data collection, particularly in regions with limited healthcare infrastructure. The integration of AI powered mobile health applications with wearable devices has created opportunities for continuous, real time health monitoring in cohort participants. Guo and colleagues’ study of 246 541 people using Huawei wristwatches showed the feasibility of continuous home monitoring for atrial fibrillation screening.¹⁸ This approach shows potential for long term cohort follow-up by collecting reliable health data without requiring participants to regularly visit healthcare facilities.

AI platforms for analysis of cohort data incorporate machine learning algorithms designed to handle the complexity and scale of longitudinal datasets. Yidu Cloud's “Data Platform + AI Platform” has been implemented across more than 2500 hospitals, providing cohort researchers with integrated tools for data governance, quality control, and advanced analytics.¹⁷ Yidu’s dual platform model enables hospitals to independently develop customised AI applications for their specific research needs. This is exemplified by Liu’s AI model for nasopharyngeal carcinoma patient services that enhances cohort recruitment, data collection, and analysis.¹⁹ These platforms are particularly valuable for analysing the complex, multimodal data that are increasingly common in modern cohort studies. By automating pattern recognition across diverse data types, they enable researchers to identify subtle relations that might be missed by traditional statistical association analyses.

AI enhanced data sharing platforms aim to tackle the fragmentation of cohort data across multiple institutions. Shanghai’s Medical Data Space exemplifies this approach by providing secure mechanisms for cohort researchers to share and integrate data while maintaining regulatory compliance.¹⁵ The Beijing International Big Data Exchange Center further demonstrated practical applications in cerebrovascular medical device development.²⁰ As listed in the supplementary table, platforms such as Tencent Health, Yidu Tech, and SenseTime provide specialised data sharing capabilities that enable multi-institutional collaboration while ensuring data security and participant confidentiality. These implementations illustrate how AI driven data infrastructure can facilitate the translation of cohort research into clinical applications while maintaining data security and participant privacy.

Healthcare system reform: bridging the rural-urban divide in cohort research

The development of integrated county level medical consortiums (ICMCs) represents a transformative approach to enhancing longitudinal cohort research in China, particularly in overcoming rural-urban healthcare disparities. Following the Promoting the Construction and Development of Medical Consortiums,⁹ these networks create comprehensive linkages between county hospitals, township health centres, and village clinics, providing the infrastructure needed for truly representative cohort recruitment and retention.

For cohort researchers, ICMCs can standardise data collection methods across diverse healthcare settings, creating the foundation for more scientifically valid cohort studies with enhanced generalisability to China’s population. The ICMC framework also directly tackles the problem of limited follow-up in Chinese cohort studies by establishing formal information sharing mechanisms between different levels of healthcare facilities. This infrastructure potentially enables researchers to track participants across China.

Furthermore, AI powered tools can automate participant reminders, standardise data collection across primary care visits, and ensure consistent documentation throughout all facility levels—directly tackling the data quality challenges arising from heterogeneous documentation practices. Recent developments, particularly the DeepSeek release, have accelerated the willingness to adopt AI in clinical settings.¹³ However, several barriers may threaten the realisation of this potential. Limited computational infrastructure, insufficient technical expertise, and prohibitive deployment costs constrain rural healthcare facilities’ capacity to use these AI systems effectively.

Future directions and implementation challenges

The integration and deployment of AI in cohort studies mark a significant evolution in epidemiological research methods. The synergy between AI and cohort studies represents a paradigm shift in health research, disease prevention, and personalised medicine, with China positioned as a potential key player in this global trend. However, as discussed above, China faces unique challenges owing to its highly fragmented healthcare data infrastructure across 1.03 million medical institutions.

Four critical barriers threaten to undermine the potential benefits of AI in Chinese cohort studies (fig 1). Firstly, geographical concentration presents a fundamental challenge, with implementation of AI remaining concentrated in major cities and tertiary hospitals in Beijing, Shanghai, Guangdong, Zhejiang, and Jiangsu (representing 27% of China’s population).¹⁷ These five regions disproportionately account for more than 80% of the nation’s medical AI enterprises, creating significant disparities in AI enhanced cohort research capabilities.¹⁷ This concentration aligns with existing participant distribution biases, potentially reinforcing rather than resolving current research inequities. Secondly, gaps in expertise severely limit implementation capacity, as successful AI-cohort integration requires research professionals with combined expertise in epidemiology, data science, and AI systems—capabilities that are particularly lacking in rural areas and primary healthcare facilities. Thirdly, non-standardisation of data persists despite development of infrastructure, with inconsistent documentation practices across healthcare settings creating barriers to valid longitudinal analysis. Fourthly, China lacks comprehensive regulatory frameworks with adequate validation standards for AI applications in cohort research. This regulatory gap is concerning, as research by Mittermaier and colleagues and Chen and colleagues shows that AI systems can potentially amplify existing biases in cohort data, particularly the under-representation of rural and minority populations.^{21 22}

Fig 1.

Key barriers and strategic priorities for implementing artificial intelligence (AI) in Chinese cohort studies. The figure outlines four critical barriers to integration of AI in cohort research: geographical concentration, gaps in expertise, data standardisation challenges, and lack of validation standards and guideline (left panel), alongside corresponding strategic priorities for overcoming these challenges (right panel). Solving these problems through targeted investments, interdisciplinary training, national data standards, and robust validation frameworks is essential to achieving equitable and scientifically rigorous cohort research in China

Overcoming these barriers to implementation requires coordinated strategic action across multiple dimensions. Firstly, geographical expansion efforts must prioritise extending AI capabilities to under-represented regions through targeted infrastructure investment and mobile health technologies. Secondly, drawing inspiration from successful international models such as the EU’s AI4Health initiative or the UK’s Health Data Research Alliance, development of interdisciplinary expertise should focus on sustainable training programmes that integrate epidemiology, data science, and AI.²³ Thirdly, standardisation efforts must establish national frameworks, such as the TRAIN Initiative or FUTURE-AI,^{24 25} to enable more efficient cohort data collection and ensure interoperability across China’s diverse healthcare systems. Finally, regulatory development must deliver comprehensive validation standards that uphold scientific rigour and mitigate the risk of bias amplification in under-represented populations.

In summary, the integration of AI into cohort studies demands not only robust frameworks that uphold fairness, safety, validation, and bias mitigation but also a broader commitment to transparency, explainability, and reproducibility. Technology should reinforce sound methods and enhance population representativeness. If implemented thoughtfully and equitably, AI has the power to transform cohort studies and population health research in China, making it more inclusive, methodologically rigorous, and globally impactful. Realising this vision will require sustained commitment to overcoming systemic barriers while leveraging China’s unique and diverse technological strengths to drive the future of epidemiological science.

Key messages.

• Artificial intelligence (AI) offers transformative potential to overcome major barriers in Chinese cohort studies, including rural-urban disparities in research infrastructure, fragmented health data systems, and limited longitudinal follow-up

• National policies such as the “AI+” initiative, healthcare data governance frameworks, and integrated medical consortia provide key foundations for scaling cohort research

• These platforms and frameworks are reshaping the cohort research landscape in China—from improving participant recruitment and follow-up to enabling large scale, multimodal data analysis and secure data sharing

• To fully harness the potential of AI in cohort research, China must systematically tackle key limitations, including geographical disparities in access, shortages in interdisciplinary expertise, lack of standardised data protocols, and the absence of robust validation frameworks

• These efforts are essential to ensure that integration of AI strengthens scientific rigour and promotes equity in population health research

Web extra.

Extra material supplied by authors

Web appendix: Supplementary table