Seeking Insights into Disease Patterns, Risks & Treatment Effects
WRH Research Group operating within our department's Data Science theme, lead by Prof. Kazem Rahimi
Deep Medicine
Deep Medicine takes advantage of large datasets, pioneering and established data science approaches, and digital trials to identify solutions that will help tackle some of the major causes of death and disability in the UK and worldwide.
The insights generated by the Deep Medicine team enable clinicians and health service providers to predict the risk of developing chronic disease, better assess its consequences, and identify the best management practices and interventions to improve health outcomes.
Why is the project important?
Recent advances in medicine have led to an unprecedented increase in life expectancy, but have also contributed to a rise in chronic diseases that affect people’s quality of life. Indeed, chronic diseases (such as arthritis, depression, diabetes, valvular heart disease, and heart failure) are estimated to account for almost 60% of the global disease burden.
More concerningly, it is the disadvantaged and underrepresented patient groups that are carrying a large burden of potentially preventable diseases.
In order to better understand these challenges and to provide solutions, our interdisciplinary research team has been harnessing emerging opportunities in data science and large-scale clinical studies.
Our group is recognised for its innovative applications of machine learning and Artificial Intelligence (AI) to electronic health records (EHR), conduct of large-scale individual-participant data meta-analyses, and digital clinical trials.
Deep Medicine is led by Prof. Kazem Rahimi and is in partnership with The Oxford Martin School.
We are devoted to generating insights into complex disease patterns, risk trajectories and treatment effects.
- Prof Kazem Rahimi
How the project will help
Findings from this research will progress how chronic diseases are evaluated, managed, and treated, and bolster strategies that either prevent their occurrence in the first place, or support critical interventions.
Clinicians and health systems will be better able to pinpoint which groups are likely to benefit most from which treatment and what factors could be contributing to poor health, so improving quality of life and survival rates.
VIDEO: ESC Congress
Prof. Kazem Rahimi presents on ESC TV programme recorded at ESC Congress 2020. Blood Pressure Lowering for Prevention of Cardiovascular Events across Different Levels of Blood Pressure.
Deep Medicine Projects
The Antihypertensive Treatment Evaluation in Multimorbidity and Polypharmacy Trial (ATEMPT)
Co-Leads: Kazem Rahimi, Milad Nazarzadeh, Shishir Rao
The ATEMPT trial aims to evaluate the effects of blood pressure lowering in older, multi-ethnic populations with multiple long-term conditions and frailty, who are typically underrepresented in clinical trials. This large-scale randomised controlled trial builds on the ATEMPT pilot study, which successfully demonstrated that an IT-enabled, home-based approach to recruiting and monitoring participants can effectively reduce blood pressure while minimising the burden of participation and trial cost.
The ATEMPT trial will assess both cardiovascular outcomes and functional independence (measured by walking speed and cognitive function) to determine the balance of benefits and harms of adding antihypertensive drugs compared to usual care. By using digital tools and focusing on reducing barriers for participation, the trial aims to improve healthcare equity and generate robust evidence to guide treatment in this vulnerable patient group, addressing the uncertainties associated with blood pressure-lowering treatment in multimorbid and frail individuals.
Key Findings
- The ATEMPT pilot trial demonstrated that remotely delivered antihypertensive treatment significantly reduced blood pressure in older adults with multimorbidity or polypharmacy, without increasing the risk of serious adverse events.
- The successful implementation of a remote trial design in this underrepresented population supports the feasibility of future larger-scale clinical trials focusing on cardiovascular events, safety, physical functioning, and cognitive outcomes.
Future Direction
The main phase of the ATEMPT trial is planned after securing the funding to evaluate long-term effects on cardiovascular outcomes, cognitive function, physical activity, and serious adverse events.
Elevated blood pressure and hypertension
Co-leads Milad Nazarzadeh, Kazem Rahimi
Elevated blood pressure is a major global health challenge, driving significant cardiovascular disease (CVD) burdens and contributing to millions of premature deaths annually. The Blood Pressure Lowering Treatment Trialists’ Collaboration (BPLTTC), led by researchers at the University of Oxford, is the world’s largest collaboration of clinical trialists focused on the effects of blood pressure-lowering treatments.
The ongoing third cycle of BPLTTC leverages data from over 300,000 participants worldwide, enabling detailed analyses to inform clinical practice and health policy.
Key Findings
- Efficacy Across Populations: Blood pressure-lowering treatments significantly reduce major cardiovascular events, such as heart attacks and strokes, across diverse groups, including varying ages, baseline risks, and comorbidities. Even modest reductions yield substantial benefits for both primary and secondary prevention.
- Personalised Strategies: Patients at higher cardiovascular risk benefit most from treatment, supporting tailored, risk-based approaches that maximise effectiveness and optimise healthcare resource use.
- Policy Implications: Findings advocate for prioritising treatment based on overall cardiovascular risk rather than high baseline blood pressure alone, influencing global guidelines such as the 2024 ESC Hypertension Guidelines and promoting equitable access to care.
Future Directions
BPLTTC’s next projects will focus on:
- Identifying optimal medication combinations and long-term treatment effects.
- Exploring risk-based patient selection to minimise adverse events and costs while maximising benefits.
- Cost-effectiveness and economic analysis of blood pressure reduction strategies.
Valvular Heart Disease
Co-leads Milad Nazarzadeh, Kazem Rahimi
Valvular heart disease (VHD) is a growing health challenge, particularly in ageing populations, where it is a leading cause of morbidity and mortality. Despite its impact, invasive surgery remains the only treatment option, and preventive medical therapies are underexplored. Current global guidelines rely heavily on expert opinion due to a lack of robust clinical evidence.
This project addresses gaps in VHD research by leveraging large-scale clinical datasets, omics data, and advanced analytical methods to better understand risk factors, disease mechanisms, and therapeutic targets. Our aim is to improve diagnosis, management, and treatment outcomes for patients with VHD.
Key Findings
Risk Factors and Mechanisms
- Identification of genetic variants and metabolic pathways linked to early valvular degeneration.
- Causal links established between VHD and cardiovascular risk factors like hypertension and lipid abnormalities.
Drug Repurposing
- Existing drugs with potential benefits for VHD identified using omics data and bioinformatics.
- Digital trials facilitated by Zeesta are planned to evaluate the safety and efficacy of these candidates.
Future Directions
- Validate drug repurposing candidates and expand understanding of VHD genetics and molecular mechanisms.
- Identifying genetic, metabolic, and environmental contributors to disease onset and progression.
Heart failure
co-leads Shishir Rao, Kazem Rahimi
Heart failure is a common, costly and severe condition that carries a risk of premature death that is higher than that for most types of cancer. With funding from Horizon EU and Novo Nordisk, we are using advanced AI models based on BEHRT (the first AI Transformer for electronic heath records) to analyse medical records and molecular data in heart failure patients.
Relying on a unified data-driven approach, we have developed several Transformer-based methods for disentangling the complexity of heart failure including models for clinical outcome prediction, for heart failure subtyping, and for conducting more robust causal inference – all to help us identify more informative subgroups of heart failure, their unique risk factors, and potential drug treatments that could be repurposed for each group.
Perinatal risk
co-leads Shishir Rao, Kazem Rahimi
Most pregnancies are healthy, but current risk assessment tools often miss dangerous cases. This leads to over-intervention in all pregnancies through early deliveries and C-sections, which can harm both mothers and babies while straining healthcare resources. Funded by Oxford University Hospital Trust in collaboration with Dr. Lawrence Impey and Dr. Samuel Dockree, to address this challenge, our team is developing AI-based dynamic risk models using routine clinical data to predict potential complications throughout pregnancy.
This approach aims to better identify truly high-risk cases, allowing healthcare providers to focus resources where needed while avoiding unnecessary intervention in low-risk pregnancies.
Musculoskeletal disease
co-leads Shishir Rao, Kazem Rahimi
While genetic research holds promise for developing new medicines, converting these findings into effective treatments has been challenging. This is particularly true for musculoskeletal (MSK) diseases, which cause more disability than any other type of illness. Despite this, MSK conditions remain understudied due to their complex nature involving cells, tissue structure, and physical forces.
As the AI methods group as a part of the Functional Genomics Cluster funded by Medical Research Council (MRC) and led by Professor Dominic Furniss, we at Deep Medicine are developing AI models (BEHRT-based) to disentangle different subtypes of MSK diseases. By analysing patient records and multi-omics data, we aim to more sensitively subtype patients, ultimately to understand which treatments will work best for specific patient groups.
Eventually, we plan to implement these models in clinical settings, using real-time patient data to provide better-matched, precise treatment recommendations.
Latest updates
- We have published the findings of our pilot trial, ATEMPT, in The Lancet Healthy Longevity. Read here
- Professor Kazem Rahimi played a seminal role in redefining European Society of Cardiology’s 2024 Hypertension guidelines with many notable BPLTTC papers cited as evidence. Read here
- We have built TRisk model for revolutionising incident CVD prediction in primary prevention population and the pre-print is out! Read here
How we use machine learning
Data mining
We adopt when necessary traditional data modelling and mining techniques to extract valuable information and insights from our large datasets using various statistical and computational techniques. Data mining allows us to discover patterns, relationships and trends within our data.
Machine learning
We use the most advanced techniques in machine learning, such as restricted Boltzmann machines (RBM). These techniques provide further insights into trends and compile real—time data which can be learned from and processed quickly.
Deep learning
Convolutional neural networks (CNNs) are ones of the most important methodologies in Deep Learning that have not been extensively applied to neuroimaging and cardiac imaging. Their application in computer vision has been hugely successful.
For scientists and researchers across healthcare, AI, and digital health domains - Deep Medicine provides:
- Novel methodologies for analysing complex healthcare – both clinical and “Omics” data at unprecedented scales to understanding chronic diseases including cardiometabolic, musculoskeletal, perinatal conditions.
- A revolutionary platform (Zeesta.ai) for conducting decentralised clinical trials, reducing costs and increasing efficiency
- Advanced AI-based Transformer modelling tools like BEHRT, Targeted-BEHRT, TRisk that enable more sophisticated analysis of electronic health records and extensions for Omics data
- Methods for integrating genetic information with clinical data for more comprehensive disease understanding
- High-impact post-hoc analyses on trials including individual level patient data meta-analyses of randomised evidence
For healthcare companies and research institutions - these innovations offer opportunities to:
- Accelerate drug development and clinical trial processes
- Improve patient risk stratification and personalised treatment approaches
- Develop more targeted interventions based on comprehensive data analysis
- Implement more efficient and cost-effective research methodologies
Useful links
AI4hf
Trustworthy Artificial Intelligence for Personalised Risk Assessment in Chronic Heart Failure
BPLTTC
Blood Pressure Lowering Treatment Trialists Collaboration
Data Science Theme
Data science, a cross-cutting theme, extends across all four research pillars/themes.
Latest publications
Efficacy of decentralised home-based antihypertensive treatment in older adults with multimorbidity and polypharmacy (ATEMPT): an open-label randomised controlled pilot trial.
Majert, Jeannette, et al.
The Lancet Healthy Longevity 5.3 (2024): e172-e181.
Bidel, Z., Nazarzadeh, M., Canoy, D., Copland, E., Gerdts, E., Woodward, M., Gupta, A. K., Reid, C. M., Cushman, W. C., Wachtell, K., Teo, K., Davis, B. R., Chalmers, J., Pepine, C. J., Rahimi, K., & Blood Pressure Lowering Treatment Trialists’ Collaboration (2023).
Hypertension (Dallas, Tex. : 1979), 80(11), 2293–2302.
Sodium-based paracetamol: impact on blood pressure, cardiovascular events, and all-cause mortality.
Rao, Shishir, et al.
European Heart Journal 44.42 (2023): 4448-4457.
The research team
-
Shishir Rao
Postdoctoral Researcher in Machine Learning for Cardiovascular Epidemiology
-
Milad Nazarzadeh
Postdoctoral Research Fellow in Cardiovascular and Genetic Epidemiology
