Model Structure
Harvard Medical School
Ai In Healthcare
Capstone Project,
Dr Bisera Lakinska, MBA (June 2025)
CardioSence
AI -Powered CVD Risk Prediction for the Emirati Population in Abu Dhabi
Leveraging Health Information Exchange data and social determinants of health to transform cardiovascular disease prevention in Abu Dhabi's Emirati population through personalized, AI-driven risk prediction.
Understanding the Landscape
Cardiovascular disease (CVD) is highly prevalent in the Middle East, particularly among the Emirati population. Cultural factors like sedentary lifestyles, smoking, and high-calorie diets contribute significantly to this health crisis. Abu Dhabi's Department of Health (DOH) has prioritized prevention and behavioral change initiatives for Emiratis, who benefit from unlimited access to healthcare services.
As part of a Private-public partnership with Abu Dhabi Health Data Services (ADHDS), the DOH has implemented an advanced Health Information Exchange (HIE) platform, ​Malaffi,​ connecting all healthcare facilities, storing comprehensive health data including radiology images, pharmacogenomic reports, and diagnostic results. Establised in 2019, the HIE now houses 3 billion unique patient records for 10+ million patients . Complementing this, the digital front door ​ patient app, called Sahatna, among the many other featuress connects to the Malaffi HIE, providing users access to their clinical records, enhancing patient engagement and promoting healthier lifestyle choices.
HIE Platform (Malaffi)
Advanced system connecting all healthcare facilities in Abu Dhabi
Reach Data Reposiry
10+ million patients with 3 billion unique records collated over 6 years
Patient portal (Sahatna mobile app)
Mobile platform connecting patients to their health data and wellness resources
Advanced features
Image exchange, Implementation of clinician-focused LLM for improved care decisions
Addressing Critical Healthcare Challenges
Despite unlimited access to healthcare, many Emiratis remain undiagnosed or inadequately managed for cardiovascular disease, resulting in increased morbidity, mortality, and healthcare system burden.
Late Diagnosis
Current assessment methods rely heavily on patient visits and routine exams, missing early intervention opportunities and delaying critical care.
Inconsistent Risk Assessment
Healthcare providers struggle to incorporate all relevant data including lifestyle factors, clinical measurements, and medical history, leading to incomplete risk evaluations.
Limited Preventive Engagement
Insufficient patient engagement and adherence to preventive measures, despite advanced healthcare systems and educational initiatives.
System Overload
The growing CVD burden strains healthcare resources, making effective management and monitoring of at-risk populations increasingly difficult.
Approximately 70% of CVD cases and deaths stem from modifiable risk factors. Our AI-powered predictive model aims to automatically assess cardiovascular risk based on comprehensive patient data, delivering real-time, data-driven insights and personalized recommendations to facilitate earlier interventions.
Our Transformative Approach
Traditional healthcare systems typically provide reactive care, addressing cardiovascular disease only after symptoms manifest. Conventional risk models utilize static data and fail to deliver real-time, personalized insights, significantly limiting opportunities for early intervention.
Our AI-first healthcare solution offers a fresh perspective by integrating real-time data from clinical records and patient-reported information with a roadmap to expand to encopass more information such as SGPD from wearables. This allows for a personalized, proactive risk assessment that predicts cardiovascular events before they occur, enabling earlier intervention. By using AI-driven predictive models, we can provide real-time insights and actionable alerts, empowering both healthcare providers and patients to take preventive measures.
Comprehensive Data Integration
Combine clinical records, wearables data, and patient-reported information for holistic assessment
AI-Powered Risk Analysis
Predictive algorithms identify patterns and risk factors before symptoms appear
Real-Time Alerts
Immediate notifications for healthcare providers when risk factors change
Personalized Interventions
Tailored recommendations based on individual risk profiles
Patient Engagement
Sahatna app delivers personalized guidance to foster behavioral change
Value Proposition and Target Market
CVD prevalence in the UAE, particularly among Emiratis, demands a shift from reactive treatment to proactive prevention. Current approaches fail to leverage the wealth of available data for early risk identification and intervention.
Improved Patient Outcomes
Early intervention leads to better health results
Reduced Healthcare Costs
Prevention is more cost-effective than treatment
Enhanced Clinical Decision-Making
AI-driven insights support evidence-based care
4
Optimized Resource Allocation
Focus resources where they create most impact
Our solution addresses the lack of early diagnosis, poor patient engagement, and inefficient resource utilization in cardiovascular care. We target the Emirati population in Abu Dhabi at risk for CVD, healthcare providers seeking tools for early risk identification, and government/payers looking to optimize healthcare resources and reduce costs.
By providing real-time risk predictions and personalized recommendations, we enable a proactive approach to cardiovascular health that benefits patients, providers, and the healthcare system alike. The model can potentially be replicated for different demographics, extending its impact beyond the initial target population.
Key Stakeholder Analysis
Understanding and addressing the unique needs, goals, and concerns of each stakeholder group is essential for successful implementation and sustained adoption of our cardiovascular risk prediction system.
Healthcare providers prioritize trustworthy predictions and seamless system integration. Patients seek clear communication about their risks and personalized guidance. Payers focus on evidence of return on investment and long-term cost reduction through preventive care.
Our implementation strategy addresses these diverse perspectives by ensuring the system delivers clear value to each stakeholder group while proactively mitigating their specific concerns through targeted education, transparent communication, and continuous refinement based on feedback.
Scientific Foundation
Our cardiovascular risk prediction model is grounded in robust scientific evidence regarding CVD patterns in the UAE and proven intervention approaches for the Emirati population.
In the UAE, cardiovascular diseases are among the four main non-communicable diseases causing death in both males and females, accounting for 34% of all deaths in 2021. This significant mortality burden underscores the critical need for improved prevention and early intervention strategies.
Research indicates that approximately 70% of CVD cases and deaths in the region are attributed to modifiable risk factors, including tobacco use, physical inactivity, unhealthy diet, increased sodium intake, and alcohol consumption. These findings form the scientific foundation of our predictive model, which prioritizes these modifiable factors to identify intervention opportunities.
Model Structure
Prediction Target
CVD risk score indicating the likelihood of a cardiovascular event within a specific period (high risk, medium risk, low risk)
Input Features
Clinical data (blood pressure, cholesterol, etc.), lifestyle factors (physical activity, diet, stress), SDH (family history, socioeconomic status), and wearables data (if available).
Model Type
Supervised supervised machine learning model (e.g., logistic regression, random forests).

Model Type
  • Logistic Regression (for binary outcomes like "CVD event" or "no event").
  • Random Forests or Gradient Boosting Machines (GBM) (for handling nonlinear relationships and feature interactions).
  • Neural Networks (when working with large and complex datasets, such as continuous data from wearables in a later phase).
Model Training: The model will be trained on historical patient data from Malaffi Emirati patients
Model Evaluation: Cross-validation will be used to ensure the model's robustness, and the model's performance will be evaluated using metrics like accuracy, precision, recall, and AUC-ROC.
Outcome:
A personalized CVD risk score to guide early intervention, preventive care, and personalized recommendations for both clinicians and patients in a later phase
1
High-risk patients
may receive recommendations for immediate action, such as further testing or referral to a cardiologist.
2
Low-risk patients
may be recommended for routine check-ups or lifestyle counseling.
Comprehensive Data Strategy
To ensure our model delivers accurate, unbiased predictions, we've developed a focused data strategy targeting the Emirati population, which has specific cultural CVD risk factors including sedentary lifestyle, caloric diet, and smoking prevalence.
1
HIE Clinical Data
ICD-10 codes, procedures, vital signs, lab results, medication history
Demographic Information
Age, gender, socioeconomic status (UAE nationals only)
Self-Reported SDH Data
Lifestyle factors collected through Sahatna surveys
Ethical Validation
Bias detection, fairness assessment, transparency measures

Our label focuses on ICD-10 diagnosis codes in the I00-I99 range, classifying patients with diseases of the circulatory system. To address ethical concerns, we will implement rigorous measures including diverse training data representative of the target population, regular fairness audits, collaboration with clinical experts, and transparent model explainability.
The solution is designed as a support tool with required human oversight to prevent potential harm. We maintain ethical transparency by clearly communicating how the system works, obtaining informed consent, and adhering to all privacy and data protection regulations. This comprehensive approach ensures our model serves as a trustworthy clinical resource rather than a replacement for medical judgment.
Implementation Roadmap
Our systematic implementation approach ensures seamless integration of the cardiovascular risk prediction system into existing clinical workflows while maximizing adoption and effectiveness.
System Development and Integration
The algorithm will be developed on top of the existing HIE database infrastructure, with direct access to comprehensive patient data. Risk prediction will be triggered automatically by new relevant data inputs such as blood pressure or cholesterol readings.
Clinical Workflow Optimization
The Clinical Decision Support module will send real-time notifications to healthcare providers, suggesting specific follow-up actions including additional testing or lifestyle modification recommendations based on risk assessments.
Training and Adoption
Comprehensive training for healthcare providers will focus on interpreting CVD predictions, responding to notifications, and communicating effectively with patients. Ongoing education will address model evolution and new features.
Evaluation and Refinement
Pilot testing will generate feedback to refine both the model and user interface. Success metrics include prediction accuracy (AUC, precision, recall, F1 score), clinician adoption rates, and patient outcome improvements.
Our implementation timeline spans three phases: initial planning and data collection (3 months), clinical workflow integration and testing (3-6 months), and ongoing deployment with continuous improvement (6-12 months). This phased approach allows for iterative refinement based on real-world performance.
Implementation step-by-step
1
Data Collection through Sahatna
Collect patient-reported data via Sahatna surveys (smoking, diet, physical activity, alcohol consumption, etc.).
2
API Integration
Automatically sync data from Sahatna into Malaffi and clinical records for each patient via APIs.
3
Model Training
Use data from Sahatna (alongside clinical data) to train the AI model on risk prediction.
4
Model Validation
Use a separate validation dataset (including Sahatna data) to evaluate and refine the model.
5
System Integration
Clinical Workflow Modifications
6
Pilot & Evaluation
Conduct pilot testing and collect feedback to refine the model and user interface.
Use metrics like prediction accuracy (AUC, precision, recall, and F-1 score), clinician adoption, and patient outcomes to evaluate success.
7
Ongoing Data Collection
Continue collecting data from Sahatna through surveys and patient interactions.
8
Continuous Model Evaluation
Monitor AI model predictions and improve based on real-world clinical feedback and patient data.
Regulatory and Compliance Framework
Implementing an AI-based cardiovascular risk prediction system in Abu Dhabi requires careful navigation of multiple regulatory domains to ensure compliance, protect patient privacy, and establish appropriate clinical usage guidelines.
Data Privacy Protection
Full compliance with UAE Data Protection Law is essential for securing patient information. Our implementation leverages the existing privacy and security frameworks of Malaffi and Sahatna, which already meet stringent requirements.
Medical Device Classification
Consultation with the Department of Health and Ministry of Health and Prevention will determine if our Clinical Decision Support System requires medical device classification and associated regulatory approvals.
Ethical and Liability Framework
Our implementation includes comprehensive protocols addressing potential model bias, clearly defined liability parameters, and robust informed consent processes for data usage in AI applications.
Healthcare Setting Integration
System deployment will adhere to all local healthcare facility regulations and clinical practice guidelines, with documentation of validation processes and quality assurance measures.
We've established a dedicated regulatory affairs team to monitor evolving AI healthcare regulations in the UAE and ensure continuous compliance. This proactive approach includes regular system audits, documentation updates, and stakeholder education about regulatory requirements.
Model outline
Our cardiovascular disease prediction model combines multiple data sources with advanced machine learning techniques to deliver personalized risk assessments.
1
Prediction Target
CVD risk score indicating the likelihood of a cardiovascular event within a specific period (high risk, low risk)
2
Input Features
Clinical data (blood pressure, cholesterol), lifestyle factors (physical activity, diet, stress), SDH (family history, socioeconomic status), and wearables data (future roadmap)
3
Model Types
Supervised machine learning models including Logistic Regression for binary outcomes, Random Forests or Gradient Boosting Machines for nonlinear relationships, and Neural Networks for complex datasets
4
Model Training and Evaluation
The model will be trained on historical patient data from Malaffi Emiriati patients, which includes over 2 billion unique patient records. Cross-validation ensures robustness, with performance evaluated using metrics like accuracy, precision, recall, and AUC-ROC.
5
Outcomes & Recommendations (future roadmap)
A personalized CVD risk score guides early intervention, preventive care, and personalized recommendations. High-risk patients receive recommendations for immediate action, while low-risk patients are advised on routine check-ups or lifestyle counseling.
Future Roadmap
1
Phase 1: Initial Rollout
Deploy the basic CVD prediction model with risk assessments
0-12 months
2
Phase 2: Actionable Advice
Implement actionable advice for doctors and patients, integrate Sahatna, and begin model tweaking
12-18 months
3
Phase 3: Expansion & Wearables
Expand to more demographics, integrate wearable data, and enhance patient engagement via Sahatna
18-24 months
4
Phase 4: Provide personalized medication recommendation
Provide personalized medication recommendation based on the integrated pharmacogenomics reports in Malaffi
24-28 months
5
Phase 5: Refinement & Scalability
Refine AI model, implement longitudinal risk analysis, and ensure scalability for broader populations
28 + months
6
Phase 6: Personalization & Global Expansion
Offer personalized treatment and care; scale globally, with continuous adaptation of the system
28+ months
Transform Cardiovascular Care in Abu Dhabi
Proactive Prevention
Shift from reactive treatment to predictive intervention
Clinical Excellence
Empower healthcare providers with AI-driven insights
Patient Empowerment
Engage Emiratis in their cardiovascular health journey
System Optimization
Reduce healthcare burden through timely interventions
Join us in revolutionizing cardiovascular care for the Emirati population. By combining advanced AI with Abu Dhabi's comprehensive health information exchange, we can identify risks earlier, intervene more effectively, and significantly reduce the burden of cardiovascular disease.
Together, we can create a healthier future for Abu Dhabi through data-driven, personalized cardiovascular care that respects cultural contexts while leveraging cutting-edge technology.
Strategic Implementation Timeline
Our comprehensive rollout strategy balances rapid value delivery with systematic capability expansion to maximize impact and adoption across the Emirati healthcare ecosystem.
6
Initial Deployment Months
Phase 1: Deploy basic CVD prediction model with risk assessments and recommendations
12
Integration Period Months
Phase 2: Implement actionable advice, integrate Sahatna, and refine models
18
Expansion Timeframe Months
Phase 3: Extend to more demographics, integrate wearable data, enhance engagement
24+
Full Maturity Timeline Months
Phases 4-5: Implement longitudinal analysis, personalized treatment, global scaling
Our implementation begins with core prediction capabilities for immediate clinical value, then progressively incorporates enhanced features like Sahatna integration and wearable data connectivity. Later phases focus on model refinement, expanded demographic coverage, and longitudinal risk analysis.
This phased approach allows for continuous learning and adaptation based on real-world performance data. Each phase includes comprehensive evaluation periods to measure clinical impact, user satisfaction, and technical performance before advancing to more sophisticated capabilities.
AI System Architecture
Our cardiovascular disease prediction model employs a sophisticated multi-layered architecture designed to process diverse data types and deliver actionable insights to healthcare providers and patients.
1
Data Ingestion Layer
Securely collects and processes information from HIE and Sahatna
2
Feature Engineering
Transforms raw data into meaningful predictive variables
3
Prediction Engine
Applies machine learning algorithms to calculate cardiovascular risk
4
Alert System
Triggers notifications based on risk thresholds
5
Clinical Interface
Presents actionable insights to healthcare providers
The prediction target is the likelihood of a cardiovascular event within the next 5 years, calibrated specifically for the Emirati population. Input features include clinical measurements (blood pressure, cholesterol, BMI), demographic data, lifestyle factors (smoking, diet, physical activity), and social determinants of health collected through Sahatna surveys.
The system employs ensemble modeling techniques combining gradient boosting and deep learning approaches to maximize predictive accuracy while maintaining interpretability for clinical users.