EY Healthcare Project: Using Data Analytics and AI to Support Cardiovascular Disease Prevention in Scotland

Analysis of Scotland-wide mortality and hospital discharge data revealed significant regional variation across NHS Health Boards using Power BI, highlighting opportunities for earlier cardiovascular prevention through data-driven insights and AI-enabled support pathways. Several AI models were evaluated during development, including GPT-4o, Claude 4 Opus, Gemini 2.5 Flash and GPT-5, to assess their suitability for safe and structured healthcare communication.

Cardiovascular disease (CVD) remains one of the most significant public health challenges in Scotland, placing sustained pressure on the National Health Service (NHS) and local healthcare systems. Despite improvements in treatment, prevention remains fragmented and heavily reliant on clinician-led intervention. This project explored how data analytics and artificial intelligence could support early cardiovascular disease prevention by identifying national patterns in CVD outcomes and designing a scalable AI-enabled prevention pathway. The project combined healthcare data analysis using Power BI with AI system design to propose a chatbot capable of delivering structured prevention guidance aligned with clinical pathways.

Understanding the Burden of Cardiovascular Disease in Scotland

Cardiovascular disease is consistently one of the leading causes of mortality in Scotland. Between 2017 and 2022, Scotland recorded approximately 44,000 CVD deaths annually, alongside extremely high levels of hospital activity related to cardiovascular conditions. During the same period, hospital discharge data indicates that there were on average 0.77 million CVD-related discharges each year, demonstrating the ongoing burden that cardiovascular disease places on healthcare services.

This produces a discharge-to-death ratio of approximately 17:1, highlighting the chronic nature of cardiovascular disease and the repeated hospital admissions associated with long-term disease management. Regional analysis also revealed variation across NHS boards. For example, NHS Lanarkshire exhibited a lower discharge-to-death ratio of around 11:1, suggesting higher severity or reduced effectiveness of preventative pathways within the region. These findings highlight the need for improved early prevention strategies that can intervene before patients require hospital treatment.

Data Sources and Analytical Approach

The analysis utilised two Scotland-wide datasets covering cardiovascular disease outcomes across NHS regions:

• Mortality data by health board
• Hospital discharge data by health board

Both datasets were combined using regional health board codes, allowing national and regional comparisons to be performed within Power BI. Several analytical steps were undertaken to prepare and explore the data.

Data Preparation

The datasets were cleaned and structured to allow consistent analysis across multiple years. Additional variables were created to enable filtering and comparison across time periods and health boards.

New analytical measures were developed including:

• Year-specific mortality and discharge indicators
• Regional health board filters
• A discharge-to-death ratio metric

These calculated metrics allowed the creation of a national framework for comparing cardiovascular outcomes across Scotland.

Dashboard Development

Power BI dashboards were developed to visualise cardiovascular disease patterns across the country.

The dashboards included visualisations covering:

• Annual mortality trends
• Hospital discharge volumes
• Diagnostic categories
• Age group distributions
• Regional comparisons between NHS boards

This allowed the identification of both national patterns and localised variations in disease burden.

Key Insights from the Data

The exploratory analysis highlighted several important patterns within Scotland’s cardiovascular health data.

Mortality levels were found to remain relatively stable over the period analysed, with annual deaths ranging between approximately 43,000 and 47,000 cases per year. Edited Draft Presentation Case …

Hospital discharge volumes were significantly higher, reflecting the ongoing clinical management required for cardiovascular conditions.

Age distribution analysis revealed that the majority of cardiovascular deaths occur in older populations:

• 62% of deaths occur in individuals aged 75 and above
• 21% occur in individuals aged 65–74
• 15% occur in individuals aged 45–64

These findings highlight the importance of early prevention strategies targeting modifiable risk factors before individuals reach high-risk age groups. Diagnostic analysis further showed that coronary heart disease and heart-related conditions account for the majority of hospital discharges, reinforcing the importance of preventative interventions targeting cardiovascular risk factors.

Identifying Gaps in Current Prevention Pathways

Although Scotland already offers several prevention programmes, these initiatives are often delivered separately and lack integration across multiple cardiovascular risk factors.

Existing interventions include:

• Smoking cessation programmes
• Weight management services
• Diabetes prevention programmes
• GP-led lifestyle counselling

However, these services often operate independently, requiring patients to navigate multiple programmes while placing significant time demands on clinicians.

This fragmented approach highlights the absence of a unified system capable of delivering coordinated prevention support at scale.

Designing an AI-Enabled Prevention Chatbot

To address this gap, the project proposed the development of an AI-enabled cardiovascular prevention chatbot designed to support both patients and clinicians.

The system integrates structured clinical pathways with conversational AI to guide users through four major cardiovascular risk prevention pathways:

• Lipid management
• Obesity management
• Glycaemic control
• Smoking cessation Case Study 2_ Ernst Young_Final…

The chatbot would deliver evidence-based guidance aligned with NHS clinical standards while also supporting behaviour change and risk awareness.

Unlike traditional health chatbots, this system integrates rule-based decision logic with a conversational AI interface, ensuring that responses remain safe, structured and clinically appropriate.

Evaluating AI Models for Healthcare Use

Several large language models were evaluated to determine their suitability for healthcare communication scenarios.

The models tested included:

• GPT-4o
• Claude 4 Opus
• Gemini Flash 2.5
• GPT-5

Evaluation scenarios included high-risk medical prompts such as breathing difficulties or long-term smoking behaviour.

The analysis found that GPT-5 demonstrated the most consistent emergency escalation behaviour, correctly directing users to emergency services such as NHS 111 or 999 when appropriate. Case Study 2_ Ernst Young_Final…

This structured and safety-focused behaviour made it the preferred model for the proposed prevention chatbot.

Implementation Considerations

The proposed AI prevention system would be implemented through a phased pilot programme.

Initial deployment could begin within NHS Lanarkshire, where the chatbot would be integrated into:

• GP practice websites
• community health portals
• mobile health applications

Estimated implementation costs include:

• Initial development cost of approximately £50,000
• Ongoing maintenance costs of around £3,600 per year

These costs are relatively small compared with the economic burden of cardiovascular disease, where hospital treatment for major cardiac events can exceed £3,900 to £10,500 per case.

Potential Impact on Healthcare Services

The implementation of an AI-enabled prevention system could deliver several benefits for healthcare services.

First, the chatbot could reduce clinician workload by handling routine prevention guidance and risk screening.

Second, it could improve patient engagement by providing accessible lifestyle and risk management advice outside of clinical appointments.

Third, it would provide consistent delivery of guideline-based prevention advice aligned with NHS recommendations.

By identifying risk factors earlier and supporting behaviour change, the system could ultimately reduce future hospital admissions and improve long-term cardiovascular health outcomes.

Conclusion

This project demonstrates how data analytics and artificial intelligence can be combined to address real-world healthcare challenges.

Through the analysis of national mortality and hospital discharge data, the project identified significant patterns in cardiovascular disease outcomes across Scotland. These insights informed the design of an AI-enabled prevention chatbot capable of delivering structured, evidence-based guidance aligned with NHS clinical pathways.

As healthcare systems continue to face growing demand, data-driven and AI-enabled solutions offer a promising pathway to strengthen prevention strategies, reduce clinical workload and improve long-term population health outcomes.

Contribution

I led the data analysis component of the project, developing Power BI dashboards to explore cardiovascular mortality and hospital discharge trends across Scotland, and contributed to the evaluation of AI models for the proposed CVD prevention chatbot.

Acknowledgement

This work was developed collaboratively with my MSc Data Analytics team members Shivangi, Ibrahim, Amaar, Prashanth and Charan. I would also like to thank our supervisor Kerem Akartunali for his support and guidance during the project.