Yajie Zhu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Yajie Zhu has a total of 45 co-authors. Here are the top co-authors by number of shared publications:
- Jose Roberto Ayala Solares
University of Sheffield
4 shared publications Kazem Rahimi
The George Institute for Global Health, University of Oxford, Oxford, UK
4 shared publications- Dexter Canoy
Deep Medicine, Nuffield Department of Women’s and Reproductive Health, University of Oxford, Oxford, UK
3 shared publications - Gholamreza Salimi-Khorshidi
Centre for Functional MRI of the Brain (FMRIB), University of Oxford, Oxford, UK
3 shared publications - Abdelaali Hassaine
Computer Science and Engineering Department, College of Engineering, Qatar University, Doha, Qatar
2 shared publications Igor Rudan
Centre for Global Health Research, Usher Institute of Population Health Sciences and Informatics, University of Edinburgh, Edinburgh, UK
2 shared publications
Latest Publications
- Open Access
- 3 October 2025
- Published by International Society of Global Health in Journal of Global Health
- Vol. 15, 04269
Global burden, quality of care, and cross-country inequalities in Alzheimer’s disease and other dementias from 1990 to 2021
Dementia, characterised by the progressive decline in cognitive functions such as memory, language, problem-solving, and other thinking abilities, significantly impairs daily life [1–3]. Alzheimer’s disease and related dementias (ADRDs) are the most prevalent forms of this neurodegenerative condition [4,5]. The World Health Organization (WHO) estimated that approximately 55 million individuals globally were affected by dementia […]
- 3 October 2025
- Published by International Society of Global Health in Journal of Global Health
- Vol. 15, 04269
Open AccessGlobal burden, quality of care, and cross-country inequalities in Alzheimer’s disease and other dementias from 1990 to 2021
Dementia, characterised by the progressive decline in cognitive functions such as memory, language, problem-solving, and other thinking abilities, significantly impairs daily life [1–3]. Alzheimer’s disease and related dementias (ADRDs) are the most prevalent forms of this neurodegenerative condition [4,5]. The World Health Organization (WHO) estimated that approximately 55 million individuals globally were affected by dementia […]
- Open Access
Transfer Learning in Electronic Health Records through Clinical Concept Embedding
Deep learning models have shown tremendous potential in learning representations, which are able to capture some key properties of the data. This makes them great candidates for transfer learning: Exploiting commonalities between different learning tasks to transfer knowledge from one task to another. Electronic health records (EHR) research is one of the domains that has witnessed a growing number of deep learning techniques employed for learning clinically-meaningful representations of medical concepts (such as diseases and medications). Despite this growth, the approaches to benchmark and assess such learned representations (or, embeddings) is under-investigated; this can be a big issue when such embeddings are shared to facilitate transfer learning. In this study, we aim to (1) train some of the most prominent disease embedding techniques on a comprehensive EHR data from 3.1 million patients, (2) employ qualitative and quantitative evaluation techniques to assess these embeddings, and (3) provide pre-trained disease embeddings for transfer learning. This study can be the first comprehensive approach for clinical concept embedding evaluation and can be applied to any embedding techniques and for any EHR concept.
Open AccessTransfer Learning in Electronic Health Records through Clinical Concept Embedding
Deep learning models have shown tremendous potential in learning representations, which are able to capture some key properties of the data. This makes them great candidates for transfer learning: Exploiting commonalities between different learning tasks to transfer knowledge from one task to another. Electronic health records (EHR) research is one of the domains that has witnessed a growing number of deep learning techniques employed for learning clinically-meaningful representations of medical concepts (such as diseases and medications). Despite this growth, the approaches to benchmark and assess such learned representations (or, embeddings) is under-investigated; this can be a big issue when such embeddings are shared to facilitate transfer learning. In this study, we aim to (1) train some of the most prominent disease embedding techniques on a comprehensive EHR data from 3.1 million patients, (2) employ qualitative and quantitative evaluation techniques to assess these embeddings, and (3) provide pre-trained disease embeddings for transfer learning. This study can be the first comprehensive approach for clinical concept embedding evaluation and can be applied to any embedding techniques and for any EHR concept.
- Top Cited PapersOpen Access
- 1 May 2020
- Published by Elsevier in The Lancet Global Health
- Vol. 8 (5) , e721-e729
Global and regional prevalence, burden, and risk factors for carotid atherosclerosis: a systematic review, meta-analysis, and modelling study
- 1 May 2020
- Published by Elsevier in The Lancet Global Health
- Vol. 8 (5) , e721-e729
Top Cited PapersOpen AccessGlobal and regional prevalence, burden, and risk factors for carotid atherosclerosis: a systematic review, meta-analysis, and modelling study
- Top Cited PapersOpen Access
- 28 April 2020
- Published by Springer Nature in Scientific Reports
- Vol. 10 (1) , 1-12
BEHRT: Transformer for Electronic Health Records
Today, despite decades of developments in medicine and the growing interest in precision healthcare, vast majority of diagnoses happen once patients begin to show noticeable signs of illness. Early indication and detection of diseases, however, can provide patients and carers with the chance of early intervention, better disease management, and efficient allocation of healthcare resources. The latest developments in machine learning (including deep learning) provides a great opportunity to address this unmet need. In this study, we introduce BEHRT: A deep neural sequence transduction model for electronic health records (EHR), capable of simultaneously predicting the likelihood of 301 conditions in one’s future visits. When trained and evaluated on the data from nearly 1.6 million individuals, BEHRT shows a striking improvement of 8.0–13.2% (in terms of average precision scores for different tasks), over the existing state-of-the-art deep EHR models. In addition to its scalability and superior accuracy, BEHRT enables personalised interpretation of its predictions; its flexible architecture enables it to incorporate multiple heterogeneous concepts (e.g., diagnosis, medication, measurements, and more) to further improve the accuracy of its predictions; its (pre-)training results in disease and patient representations can be useful for future studies (i.e., transfer learning).
- 28 April 2020
- Published by Springer Nature in Scientific Reports
- Vol. 10 (1) , 1-12
Top Cited PapersOpen AccessBEHRT: Transformer for Electronic Health Records
Today, despite decades of developments in medicine and the growing interest in precision healthcare, vast majority of diagnoses happen once patients begin to show noticeable signs of illness. Early indication and detection of diseases, however, can provide patients and carers with the chance of early intervention, better disease management, and efficient allocation of healthcare resources. The latest developments in machine learning (including deep learning) provides a great opportunity to address this unmet need. In this study, we introduce BEHRT: A deep neural sequence transduction model for electronic health records (EHR), capable of simultaneously predicting the likelihood of 301 conditions in one’s future visits. When trained and evaluated on the data from nearly 1.6 million individuals, BEHRT shows a striking improvement of 8.0–13.2% (in terms of average precision scores for different tasks), over the existing state-of-the-art deep EHR models. In addition to its scalability and superior accuracy, BEHRT enables personalised interpretation of its predictions; its flexible architecture enables it to incorporate multiple heterogeneous concepts (e.g., diagnosis, medication, measurements, and more) to further improve the accuracy of its predictions; its (pre-)training results in disease and patient representations can be useful for future studies (i.e., transfer learning).
- Top Cited Papers
- 1 January 2020
- Published by Elsevier in Journal of Biomedical Informatics
- Vol. 101, 103337
Deep learning for electronic health records: A comparative review of multiple deep neural architectures
- 1 January 2020
- Published by Elsevier in Journal of Biomedical Informatics
- Vol. 101, 103337
Top Cited PapersDeep learning for electronic health records: A comparative review of multiple deep neural architectures
- Top Cited PapersOpen Access
- 1 December 2019
- Published by American Medical Association (AMA) in JAMA Pediatrics
- Vol. 173 (12) , 1154-1163
Global Prevalence of Hypertension in Children
Hypertension, also known as elevated blood pressure (BP), is a well-recognized risk factor for cardiovascular diseases and chronic kidney disease worldwide.1-3 Hypertension also substantially contributes to mortality and disability.3 Globally, more than 1 billion adults were living with hypertension in 2015, among whom most were in low- and middle-income countries.2,4
- 1 December 2019
- Published by American Medical Association (AMA) in JAMA Pediatrics
- Vol. 173 (12) , 1154-1163
Top Cited PapersOpen AccessGlobal Prevalence of Hypertension in Children
Hypertension, also known as elevated blood pressure (BP), is a well-recognized risk factor for cardiovascular diseases and chronic kidney disease worldwide.1-3 Hypertension also substantially contributes to mortality and disability.3 Globally, more than 1 billion adults were living with hypertension in 2015, among whom most were in low- and middle-income countries.2,4
- 1 October 2019
- Published by Oxford University Press (OUP) in European Heart Journal
- Vol. 40 (Supplement)
P1548Long-term past, current and usual systolic blood pressure and incident cardiovascular disease: risk prediction using large-scale, routinely recorded clinical data
Background: The impact of long-term exposure to elevated systolic blood pressure (SBP) on future cardiovascular disease (CVD) in “real-world” settings, and its relevance to risk prediction, are less investigated. Purpose: To examine the risk of incident CVD in relation to long-term past, current, and usual SBP, and compare their predictive performance, using evidence from large-scale electronic health records (EHR). Methods: Using data extracted from UK primary care linked EHR, we applied a landmark cohort study design, by including patients aged 40 (N≈64,000), 50 (N≈80,000) and 60 (N≈67,000) years (y) at study entry who had recorded SBP and with no prior CVD or previous antihypertensive or lipid-lowering prescriptions at baseline. We estimated past SBP (mean, time-weighted mean, and variability recorded up to 10 years prior to baseline) and usual SBP (correcting current values for past time-dependent SBP variability). We used Cox regression to estimate hazard ratio (HR), and applied Bayesian analysis within a machine learning framework in developing and validating models. To evaluate predictive performance of the models, we used discrimination (area under the curve [AUC]) and calibration metrics. The outcome was incident CVD (first hospitalisation for or death from coronary heart disease or stroke/transient ischaemic attack). Analyses were conducted separately for each age cohort. Results: After a mean follow-up of 8 years, the numbers of patients who developed incident CVD were over 1000 (40y), 3000 (50y) and 5000 (60y). Higher past, current and usual SBP values were separately and independently associated with increased incident CVD risk. Per 20-mmHg rise in SBP, the HR (95% credible interval [CI]) for current SBP for ages 40, 50 and 60 years were 1.18 (1.08 to 1.26), 1.22 (1.18 to 1.30) and 1.22 (1.19 to 1.24); the corresponding HR were stronger in magnitude for past SBP (mean and time-weighted mean) and usual SBP (HR ranged from: 40y=1.31 to 1.41, 50y=1.39 to 1.45 and 60y=1.32 to 1.48). For each age cohort, the AUC (95% CI) for the model that included current SBP, sex, smoking, deprivation, diabetes and lipid profile in the validation sample were: 40y=0.739 (0.730 to 0.746), 50y=0.750 (0.716 to 0.810), and 60y=0.647 (0.642 to 0.658). Adding past SBP mean, time-weighted mean or variability to this model were associated with modest increases in the AUC and all models showed good calibration. Small improvements in the AUC were similarly observed when evaluating models separately for men and women within each age cohort. Conclusion: Using multiple SBP recordings from patients' EHR showed stronger associations with incident CVD than a single SBP measurement, but their addition to multivariate risk prediction models had negligible effects on model performance. Acknowledgement/Funding: Oxford Martin School and National Institute for Health Research Oxford Biomedical Research Centre
- 1 October 2019
- Published by Oxford University Press (OUP) in European Heart Journal
- Vol. 40 (Supplement)
P1548Long-term past, current and usual systolic blood pressure and incident cardiovascular disease: risk prediction using large-scale, routinely recorded clinical data
Background: The impact of long-term exposure to elevated systolic blood pressure (SBP) on future cardiovascular disease (CVD) in “real-world” settings, and its relevance to risk prediction, are less investigated. Purpose: To examine the risk of incident CVD in relation to long-term past, current, and usual SBP, and compare their predictive performance, using evidence from large-scale electronic health records (EHR). Methods: Using data extracted from UK primary care linked EHR, we applied a landmark cohort study design, by including patients aged 40 (N≈64,000), 50 (N≈80,000) and 60 (N≈67,000) years (y) at study entry who had recorded SBP and with no prior CVD or previous antihypertensive or lipid-lowering prescriptions at baseline. We estimated past SBP (mean, time-weighted mean, and variability recorded up to 10 years prior to baseline) and usual SBP (correcting current values for past time-dependent SBP variability). We used Cox regression to estimate hazard ratio (HR), and applied Bayesian analysis within a machine learning framework in developing and validating models. To evaluate predictive performance of the models, we used discrimination (area under the curve [AUC]) and calibration metrics. The outcome was incident CVD (first hospitalisation for or death from coronary heart disease or stroke/transient ischaemic attack). Analyses were conducted separately for each age cohort. Results: After a mean follow-up of 8 years, the numbers of patients who developed incident CVD were over 1000 (40y), 3000 (50y) and 5000 (60y). Higher past, current and usual SBP values were separately and independently associated with increased incident CVD risk. Per 20-mmHg rise in SBP, the HR (95% credible interval [CI]) for current SBP for ages 40, 50 and 60 years were 1.18 (1.08 to 1.26), 1.22 (1.18 to 1.30) and 1.22 (1.19 to 1.24); the corresponding HR were stronger in magnitude for past SBP (mean and time-weighted mean) and usual SBP (HR ranged from: 40y=1.31 to 1.41, 50y=1.39 to 1.45 and 60y=1.32 to 1.48). For each age cohort, the AUC (95% CI) for the model that included current SBP, sex, smoking, deprivation, diabetes and lipid profile in the validation sample were: 40y=0.739 (0.730 to 0.746), 50y=0.750 (0.716 to 0.810), and 60y=0.647 (0.642 to 0.658). Adding past SBP mean, time-weighted mean or variability to this model were associated with modest increases in the AUC and all models showed good calibration. Small improvements in the AUC were similarly observed when evaluating models separately for men and women within each age cohort. Conclusion: Using multiple SBP recordings from patients' EHR showed stronger associations with incident CVD than a single SBP measurement, but their addition to multivariate risk prediction models had negligible effects on model performance. Acknowledgement/Funding: Oxford Martin School and National Institute for Health Research Oxford Biomedical Research Centre