WellDoc - AI Risk Prediction Engine

A professional healthcare dashboard prototype for predicting chronic care patient deterioration risk using AI.

🏥 Overview

WellDoc is an AI-driven Risk Prediction Engine that forecasts whether a chronic care patient is at risk of deterioration in the next 90 days. The dashboard provides clinician-friendly, explainable predictions with actionable insights.

✨ Features

🎯 Core Functionality

• Risk Prediction: Binary classification with probability scores (0-100%)
• Dashboard Scores: Individual assessments for Vitals, Medication, Lifestyle, and Labs
• Explainable AI: SHAP-based feature importance and risk factor analysis
• Real-time Monitoring: Patient cohort overview with trend analysis

📊 Dashboard Views

1. Overview Dashboard: Key metrics, risk distribution, and recent alerts
2. Cohort View: Complete patient list with filtering and sorting
3. Patient Detail: Individual patient deep-dive with charts and recommendations
4. Analytics: Model performance metrics and clinical outcomes

🎨 Design Features

• Dark Theme: Professional healthcare-focused design
• Modern UI: Built with shadcn/ui components and Tailwind CSS
• Responsive: Works on desktop and tablet devices
• Professional: Clean, medical-grade interface without emojis

🛠 Tech Stack

• Frontend: Next.js 15 + TypeScript + Tailwind CSS
• UI Components: shadcn/ui + Radix UI
• Charts: Recharts for data visualization
• Icons: Lucide React
• Data: Synthetic demo data with realistic medical scenarios

🚀 Getting Started

Prerequisites

• Node.js 18+
• npm or yarn

Installation

1. Clone and Install

   cd welldoc
   npm install

2. Run Development Server

   npm run dev

3. Open Browser Navigate to http://localhost:3000

📱 Demo Navigation

Dashboard Overview (/)

• Key metrics: Total patients, high-risk count, average risk score
• Risk distribution chart
• Recent alerts and high-risk patients
• Quick action buttons

Cohort View (/cohort)

• Complete patient list with search and filtering
• Risk level distribution summary
• Sort by name, risk score, or last updated
• Direct links to patient details

Patient Details (/patients/[id])

• Overall risk score with trend indicators
• Dashboard scores breakdown (Vitals, Medication, Lifestyle, Labs)
• Interactive charts: Vital signs, lab results, medication adherence
• Risk factors analysis with impact percentages
• Clinical recommendations

Analytics (/analytics)

• Model performance metrics (AUROC, AUPRC, Precision, Recall)
• Feature importance analysis
• Prediction accuracy trends
• Intervention outcomes and patient timeline
• Clinical impact summary

🧪 Demo Data

The application includes 5 synthetic patients with different risk profiles:

1. John Mitchell (High Risk - 85%): Declining vitals, poor medication adherence
2. Sarah Chen (Low Risk - 15%): Well-controlled pre-diabetes, good lifestyle
3. Robert Williams (Medium Risk - 55%): Heart failure with mixed indicators
4. Maria Rodriguez (High Risk - 78%): Diabetes with kidney complications
5. David Thompson (Medium Risk - 42%): Metabolic syndrome, improving trends

🏗 Architecture

Data Flow

Synthetic Patient Data → Feature Engineering → Binary Classification Model
                                           ↓
Risk Probability → Dashboard Scores → SHAP Analysis → Clinical Insights

Score Derivation

• Risk Score: Direct model probability (0-100%)
• Category Scores: Aggregated SHAP values for feature groups
• Trends: Time-series analysis (improving/stable/declining)
• Recommendations: Rule-based clinical guidelines

📈 Model Concept

While this is a frontend prototype with synthetic data, the underlying ML approach:

• Model: XGBoost/LightGBM for binary classification
• Features: Vitals trends, medication adherence, lab results, lifestyle factors
• Target: 90-day deterioration risk (binary)
• Explainability: SHAP values for global and local explanations
• Metrics: AUROC >0.80, AUPRC >0.70, calibrated probabilities

🎯 Prototype Goals

This 1-day prototype demonstrates:

• ✅ End-to-end dashboard for risk prediction
• ✅ Professional healthcare UI/UX
• ✅ Multiple visualization types
• ✅ Clinician-friendly explanations
• ✅ Scalable component architecture
• ✅ Realistic medical scenarios

🔮 Future Enhancements

• Real ML Pipeline: Integrate actual model training and inference
• Live Data: Connect to EMR systems and real patient data
• Advanced Charts: Time-series forecasting and trend prediction
• Mobile App: React Native companion for on-the-go monitoring
• Alerts System: Real-time notifications and escalation workflows

📄 License

This is a prototype developed for demonstration purposes.

---

Built with ❤️ for better patient outcomes