MedOptix Forecasting

Python SARIMAX Power BI Streamlit Healthcare AI

Overview

MedOptix Forecasting is a predictive capacity management suite developed as part of The HealSight Initiative to address critical inefficiencies in hospital operations.

By transitioning from reactive responses to proactive AI-driven planning, the system forecasts patient inflows 7–30 days in advance and increased bed utilization efficiency to 88% and reduced overtime costs significantly.

Project Objectives

Forecasting: Predict patient inflows 7–30 days in advance with >85% accuracy using SARIMAX.
Real-Time BI: Visualize real-time bottlenecks across ED, ICU, MED, and SURG wards.
Resource Optimization: Optimize bed allocation to reduce overflow incidents and overtime costs.
Scenario Planning: Stress-test capacity against hypothetical surges.

View Source Code Live Demo App

Project Report: MedOptix Forecasting

Author: Francis Afful Gyan | Business Analyst & Data Scientist

1. Impact & Results

Bed Efficiency

88%

Overtime Savings

€35k/mo

Forecast Accuracy

88.4%

Wait Time Impact

-25%

2. Business Problem: The Reactive Trap

The hospital network was operating purely reactively, making staffing decisions only after surges occurred. This led to a 20% inefficiency in bed allocation (beds empty in one ward, overflow in another) and a financial strain of €125,000/month spent on reactive overtime.

3. Solution Architecture & Tools

Implemented a "Hybrid Intelligence" stack combining deep analytics with user-friendly interfaces.

Data Processing: Python (Pandas, NumPy) for cleaning 19,000+ records.
Modeling: Statsmodels (SARIMAX) for time-series forecasting.
Frontend: Streamlit (Python) for the interactive scenario planning tool.
BI Layer: Microsoft Power BI (Embedded) for real-time executive dashboards.
Deployment: Docker & Docker Hub (Containerization) on Streamlit Cloud.

Why SARIMAX?

Selected SARIMAX over Prophet or standard ARIMA because of Explainability. It explicitly models "External Drivers" (like Staffing Index), allowing administrators to understand why a surge is coming, not just that it is coming.

4. Key Data Insights

A. The "Staffing Index" Correlation: When the Staffing Index falls below 0.95, the discharge rate slows significantly, creating an "artificial" demand surge the following day.

B. The Overflow Trigger: I identified a "Tipping Point" at 92% occupancy. Once exceeded, the rate of Overflow Incidents doubles. The dashboard now flags risks before this threshold is hit.

C. Demographic Wait Times: Patients aged 40–59 experience the longest wait times (~60 mins), revealing a "priority gap" in triage protocols compared to seniors and pediatric cases.

5. Model Performance Metrics

The SARIMAX model was validated against historical data with high precision:

R² Score

88.4%

Variance Explained

RMSE

0.553

Avg Error (Patients)

MAE

0.463

Absolute Error

6. Key Deliverables

AI Forecast Tool: Allows managers to input current stats and "Stress Test" capacity (e.g., "What if 10% of beds close?").
Executive Dashboard: Real-time tracking of KPIs with an automated traffic-light risk system (🟢 🟡 🔴).
Containerized Deployment: Packaged in Docker to ensure identical performance across all environments.

7. Recommendations & Next Steps

Operational Adoption

Action: Mandate the use of the Power BI Risk Card during morning operational standups.

Proactive Staffing

Action: Trigger overtime shifts only when the Forecast Tool predicts "High Risk" capacity for the next 48 hours, eliminating reactive spending.

Strategic Roadmap

Action: Review triage protocols for the 40–59 age demographic and move from batch CSV updates to real-time IoT streaming using Azure Event Hubs.

8. Conclusion

MedOptix Analytics has successfully demonstrated that data-driven decision-making can solve physical capacity problems. By predicting demand before it happens, we have moved the hospital network from "Putting out fires" to "Preventing them."