ReadmissionGuard

Inspiration

A few weeks ago,I was conducting research on the healthcare industry particularly on prediction and prevention of NCDs(non-communicable diseases) and came across some concerning facts. In the Us, unplanned hospital readmissions are estimated to cost the healthcare system approximately $15–20 billion annually while the average cost of a 30-day all-cause adult hospital readmission is approximately $15,200. The financial burden of hospital readmissions is not one to be taken lightly as it is one of the factors that have pushed over 500 million people into extreme poverty according to the WHO in 2021 from health related costs, though studies indicate that approximately 25% of all hospital readmissions are preventable.

What it does

ReadmissionGuard is a healthcare application that actively works to prevent hospital readmissions by identifying at-risk patients before they require rehospitalization. It achieves this through real-time monitoring and analysis of patient data, providing healthcare providers with actionable insights and intervention recommendations.

Here's how ReadmissionGuard works: When a patient's information enters the system, it processes multiple data points including their medical history, current conditions, medications, previous hospital visits, and ongoing health metrics. This data is stored and interconnected in the Dgraph knowledge graph, creating a comprehensive view of the patient's health journey. The system's AI model, powered by GPT-4 through the Modus framework, analyzes this rich dataset to calculate a readmission risk score. This score is based on patterns identified from successful and unsuccessful patient outcomes, considering factors such as age, chronic conditions, medication adherence, and previous admission history. Based on the calculated risk score, ReadmissionGuard generates personalized intervention recommendations. For example, if a patient shows high readmission risk due to medication management issues, the system might recommend daily nurse check-ins or a medication review. These recommendations are tailored to address specific risk factors identified in the patient's profile. Healthcare providers interact with ReadmissionGuard through an intuitive interface that displays risk assessments, recommended interventions, and patient progress tracking. They can monitor multiple patients simultaneously, with the system highlighting those requiring immediate attention based on their risk scores. The system's effectiveness comes from its continuous learning capability. As new patient data and outcomes are recorded, the AI model refines its predictions, and the knowledge graph expands its understanding of effective interventions. This creates a constantly improving cycle of patient care, where each case contributes to better predictions and recommendations for future patients.

How we built it

ReadmissionGuard was built using a modern technology stack centered around the Modus API framework, which provides the development environment and orchestration layer. The backend is developed in Go, offering robust performance and excellent concurrency handling, while the frontend utilizes AssemblyScript for efficient client-side operations and a responsive user interface.

The core architecture integrates three main components: the Modus framework for application structure and AI model management, GPT-4 for predictive analytics, and Dgraph as the knowledge graph database. These components are configured through the modus.json file, which defines the AI models, data connections, and API endpoints. The system's data flow is orchestrated through well-defined API endpoints that handle patient data processing, risk assessments, and intervention recommendations. The knowledge graph implementation in Dgraph creates a rich network of interconnected patient data, enabling complex queries and pattern recognition, while the AI model analyzes this data to generate accurate risk predictions and personalized interventions.

Challenges we ran into

There were several challenges that were encountered when building ReadmissionGuard such as the scaling of the project. One significant challenge involved integrating the AI model (GPT-4) with real-time patient data while maintaining both accuracy and speed. The system needs to process large amounts of medical data quickly while ensuring the AI predictions remain reliable and clinically relevant.

Another challenge was working with healthcare data in a knowledge graph presents complex data modeling challenges. Designing the right schema in Dgraph to represent intricate medical relationships while maintaining query performance requires was quite demanding.

Accomplishments that we're proud of

One major achievement is the development of an accurate predictive model that identifies at-risk patients before complications arise. Also the successful implementation of the dgraph knowledge graph that maps intricate relationships between patient data points, enabling healthcare providers to understand patterns and act on them quickly.

What we learned

Working with the Modus framework revealed the power of the framework in orchestrating complex systems, particularly how it simplifies the integration of AI models and knowledge graphs into production applications.

What's next for ReadmissionGuard

Built With

• assemblyscript
• dgraph
• go
• gpt-4aimodel
• modusapiframework