OPTIM AID

Inspiration

Managing insulin titration in hospitals is still a largely manual and protocol-driven process. Clinicians juggle dozens of patients, variable nutrition, co-medications, and intermittent glucose monitoring — all under immense time pressure. This leads to suboptimal glycemic control, higher risk of hypo-/hyperglycemia, and increased workload. We wanted to create a solution that reduces mental burden, improves patient safety, and supports clinicians with explainable, real-time recommendations.

What it does

OPTIM AID is a pharmacist and clinician “co-pilot” that delivers real-time, personalised insulin dosing recommendations for hospitalized patients. It integrates: (i) A mechanistic QSP model to simulate glucose–insulin dynamics, (ii) An AI/ML layer trained on inpatient data, to provide dosing guidance that is safe, explainable, and aligned with hospital protocols. The platform features a clinician-friendly dashboard with glucose trends, insulin history, alerts, and a ‘what-if’ simulator for testing dose adjustments before administration.

How we built it

We combine a physiology-based QSP simulation engine with an AI/ML predictive model, trained on structured inpatient datasets. The backend will be developed using R-mrgsolve and Python for simulation + learning integration. The front-end prototype was built in R Shiny platform for rapid visualization. The workflow harmonizes patient data (glucose logs, nutrition, labs, meds) and outputs real-time recommendations accessible via web or tablet.

Challenges we ran into

Balancing complexity vs usability: Designing a backend that captures physiological detail, while keeping the UI simple enough for clinicians. Trust in AI: Addressing the “black box” problem by embedding physiologically grounded QSP models for explainability. Time pressure of the hackathon: Building an end-to-end working demo in a short timeframe.

Accomplishments that we're proud of

Developing a prototype that demonstrate responses and produces dose recommendations. Creating a front-end mockup that clinicians can immediately relate to in their workflow. Designing a hybrid AI–QSP approach that is both innovative and clinically explainable. Building a strong multidisciplinary team across pharmacology, AI, and clinical practice.

What we learned

The importance of co-design with clinicians to ensure adoption in real-world workflows. How hybrid models (mechanistic + AI) can overcome the trust barrier of pure AI. That rapid prototyping withplatforms can accelerate healthcare innovation when paired with strong technical backends. The value of aligning technology with clinical priorities: safety, time efficiency, and interpretability.

What's next for OPTIM AID

Training and Integrating AI/ML and QSP layers to analyze the real time data and provide recommendations. Pilot testing in collaboration with partner hospital TTSH. Regulatory readiness as a clinical decision support tool. Integration with EHR systems for seamless data flow. Expanding beyond insulin to support multi-drug inpatient management (e.g., steroids, GLP-1 analogues). Long term: scaling OPTIM AID to outpatient and chronic care settings, creating a smart, explainable co-pilot for metabolic disease management.

Built With

• r
• rshiny