Chronic Disease Progressive Model for Diabetes (CDPMD)

Inspiration

The Chronic Disease Progressive Model for Diabetes (CDPMD) was developed in response to the growing global burden of diabetes, which now affects hundreds of millions of people worldwide. Given that diabetes is a major contributor to complications such as retinopathy, nephropathy, and cardiovascular disease, this project was conceived to help bridge the gap between data-driven insights and clinical decision-making. The vision was to build a predictive model that forecasts disease progression while providing clinicians with actionable, evidence-based recommendations aligned with current guidelines from the American Diabetes Association (ADA) and the European Association for the Study of Diabetes (EASD).

What We Learned

During development, we discovered that high-quality, diverse data are essential for building robust predictive models. Adopting FHIR R4 standards ensured that our data were interoperable across various healthcare systems. We combined deep learning techniques with structured clinical rules to strike a balance between predictive performance and clinical interpretability. We also recognized the critical need to mitigate bias—especially for underrepresented populations that face disparities in diabetes outcomes—and implemented fairness audits to improve the model's equity. Finally, we learned that building clinician trust requires transparency, clear explanations, and seamless integration into existing clinical workflows.

How It Was Built

The CDPMD was built using an AI architecture that includes:

1. An AI Engine (DeepSeek) - for modeling disease progression and analyzing clinical patterns, comorbidities, and risk factors.
2. A clinical knowledge graph that encodes ADA/EASD guidelines and evidence-based rules.

Input data were standardized according to FHIR R4, allowing smooth integration with electronic health records (EHRs).

Impact and Future Vision

Preliminary evaluations of the CDPMD indicate that it can significantly reduce the time clinicians spend on manual data analysis and offer reliable predictions of diabetes complications. In the future, we plan to extend the model to address other diabetes types, such as Type 1 Diabetes and gestational diabetes, and to incorporate additional real-world data to further refine its predictions. By continually prioritizing fairness, transparency, and close collaboration with clinicians, we aim to establish CDPMD as an essential tool for enhancing diabetes care and improving patient outcomes.

Built With

• deepseek
• fasthtml
• meldrx
• pydanticai
• python