Skip to content

replit.md

Latest commit

 

History

History
93 lines (70 loc) · 3.89 KB

replit.md

File metadata and controls

93 lines (70 loc) · 3.89 KB

Clinical Decision Support: Preventive Diabetes Risk Assessment

A full-stack web application for preventive diabetes risk assessment that combines a Python-based logistic regression model with a React frontend. The system surfaces early risk signals from routine patient data and communicates findings differently for clinicians and patients.

Features

Core Functionality

  • Risk Assessment Form: Input patient data including age, gender, hypertension, heart disease, smoking history, BMI, HbA1c level, and blood glucose level
  • Dual-View Results:
    • Clinician View: Displays precise risk percentage, top contributing factors with impact analysis, model confidence, and recommended follow-up actions
    • Patient View: Shows simplified risk category (LOW/MODERATE/HIGH), factors in plain language, and preventive lifestyle advice
  • Assessment History: View all past risk assessments with timestamps and risk categories
  • Data Visualization: Interactive bar charts showing risk factor contributions (clinician view)

Technical Architecture

Frontend (client/)

  • React with TypeScript
  • TanStack Query for server state management
  • React Hook Form with Zod validation
  • Recharts for data visualization
  • Tailwind CSS with dark mode support
  • Framer Motion for animations

Backend (server/)

  • Express.js REST API
  • PostgreSQL database (via Drizzle ORM)
  • Python integration for ML predictions
  • Route validation with Zod schemas

Machine Learning (analyze.py)

  • Logistic Regression model using scikit-learn
  • Feature engineering and preprocessing
  • StandardScaler for feature normalization
  • Interpretable risk scoring and factor analysis
  • Dynamic advice generation based on risk level

Data Model

Assessment Table

  • Patient demographics: gender, age
  • Medical history: hypertension, heart disease, smoking history
  • Clinical measurements: BMI, HbA1c level, blood glucose level
  • Model outputs: risk score (0-100%), risk category, contributing factors
  • Timestamp for tracking

API Endpoints

  • POST /api/assessments - Submit patient data for risk assessment
  • GET /api/assessments - Retrieve assessment history

Risk Categorization

  • LOW (<20%): Monitor annually, no immediate intervention
  • MODERATE (20-50%): Lifestyle counseling, repeat HbA1c in 6 months
  • HIGH (>50%): Refer for diagnostic testing, consider intervention

Development

Dependencies

  • Node.js: Express, TypeScript, Drizzle ORM
  • Python: pandas, numpy, scikit-learn
  • React: React 18, React Query, React Hook Form

Database

PostgreSQL with Drizzle ORM. Schema managed via npm run db:push.

Running the Application

The workflow "Start application" runs npm run dev which starts both the Express backend and Vite frontend on port 5000.

Clinical Decision Support Workflow

  1. Data Input: Clinician enters patient data via the assessment form
  2. Processing: Backend invokes Python script with patient data
  3. Model Inference: Logistic regression model calculates risk probability and identifies key factors
  4. Result Storage: Assessment saved to database with predictions
  5. Dual Display: Results shown in both clinician and patient-friendly formats
  6. Follow-up: System suggests appropriate next steps based on risk level

Safety & Ethics

  • Not a Diagnostic Tool: Frames everything as decision support, not medical diagnosis
  • Interpretability: Uses logistic regression for transparent, explainable predictions
  • Uncertainty Communication: Includes confidence measures and suggests clinical judgment
  • Bias Awareness: Model trained on balanced data with attention to demographic factors

Future Enhancements

  • Longitudinal patient tracking
  • Counterfactual reasoning ("what would reduce risk most?")
  • Cohort discovery and population-level insights
  • Integration with electronic health records (EHR)
  • Advanced bias detection and fairness metrics