Chronic Disease Burden – SQLite + Python

This repository contains a small but realistic SQLite schema and Python tooling for chronic disease burden tracking. It is modeled after the patient, lab, and risk-score data I worked with during my Machine Learning Internship at Doyen Diagnostics, where I built an end-to-end ML pipeline that improved disease-burden prediction by 15%.

In that internship, I used a semi-supervised DNN with label propagation and a Gaussian kernel to achieve robust health-score modeling from only 78 labeled samples. The model sat downstream of a structured data layer that looked a lot like this one: patients with chronic conditions, longitudinal lab values, and derived risk scores over time. This project isolates that data layer in a standalone SQLite database and shows how patient information can be organized and queried before being exported into an ML pipeline.

Schema at a glance

• patients – Demographics and smoking status, keyed by an external LIS/EMR identifier.
• clinicians – Doctors and their specialties.
• conditions / patient_conditions – Chronic conditions (for example, T2DM, HTN, CKD) associated with each patient.
• visits – Longitudinal visits with dates, reasons, and notes.
• lab_results – Key lab tests (for example, HbA1c, LDL) per visit, with normal ranges.
• risk_scores – Simple rule-based chronic disease burden scores in [0, 1], with LOW / MODERATE / HIGH bands and an explanation string.
• alerts – Flags when overall risk is high or when critical labs are severely out of range.

This mirrors the kind of schema you would expect in a lightweight chronic-disease registry or analytics layer on top of an EHR.

Files

• schema.sql – SQLite schema and indexes for patients, clinicians, conditions, visits, lab_results, risk_scores, and alerts.
• db.py – Small helper for opening SQLite connections with foreign keys enabled.
• init_db.py – Creates chronic_disease.db from schema.sql.
• seed_and_score.py – Inserts sample patients, visits, and lab measurements, then computes a transparent rule-based disease-burden score and raises alerts when thresholds are crossed.
• reports.py – Example read-side queries and reports:
  • Latest risk score per patient.
  • Visit and lab timeline for a given patient.
  • List of open (unacknowledged) alerts.
• requirements.txt – Listed for completeness; the project only relies on the Python standard library (sqlite3).

How to run

   # 1. Initialize the database
   python init_db.py

   # 2. Seed sample data and compute risk scores
   python seed_and_score.py

   # 3. Run example reports
   python reports.py

Contact

For questions or suggestions, please email ani.tubai022@gmail.com or open an issue on the GitHub repository.