🧬 DOSAGE: An intelligent decision-support tool designed to help validate antibiotic prescriptions based on patient-specific parameters

License: MIT (code) — see data/SOURCE.md for the dataset's own license (CC BY 4.0)
Authors:

• Ferdous Wahid Anik (Lead Developer & Project Lead)
• Marzia Zaman (Clinical Advisor)
• Tahmina Foyez (Pharmacological Reviewer)
• Khandoker A. Mamun (Supervisor & Technical Advisor)

Contact: mamun@cse.uiu.ac.bd

---

📄 Overview

This is a rule-based clinical decision support tool (FastAPI web app) that validates antibiotic prescriptions against patient-specific parameters — age, weight, renal function (CrCl), and pregnancy status. Given a prescribed dose, it flags overdose, underdose, wrong administration route, or renal adjustment issues, with a 15% tolerance band applied to dosing rules.

It supports two modes:

• Single medication check — validate one prescription via a web form.
• Batch check — upload a CSV of multiple prescriptions and get a validation report back, downloadable as CSV.

The rules and reference ranges the app validates against come from the DOSAGE dataset (see "Data" below) — a knowledge base covering 98 antibiotic generics and over 4,000 dosing scenarios, published in Nature Scientific Data.

---

🚀 Running locally

1. Clone the repo and install dependencies:

   git clone https://github.com/<your-username>/DOSAGE.git
   cd DOSAGE
   pip install -r requirements.txt

2. Download the dataset and add it to the data/ folder — see data/SOURCE.md for the download link and data/README.md for the exact filenames/columns expected.
3. Run the app:

   uvicorn main:app --reload

4. Open http://127.0.0.1:8000 in your browser.

No database or API keys are required — the app reads the dataset directly from the data/ folder at startup.

---

📦 Repository Contents

• main.py — FastAPI backend: dose validation logic, renal/pregnancy checks, single and batch endpoints
• templates/, static/ — front-end (HTML form, CSS, JS) for the web app
• data/ — where the dataset CSVs go locally (not committed; see data/SOURCE.md)
• notebooks/ — Jupyter notebooks with test cases exercising the validation rules
• requirements.txt, render.yaml, Procfile — dependencies and deployment config

---

📊 Data

This tool validates against the DOSAGE dataset: disease-specific dosing, standard dosing, renal-adjusted dosing, and pregnancy risk classifications for 98 antibiotics.

The dataset is published separately and is not bundled in this repo. Download it and see the field-by-field documentation at:

Figshare: https://figshare.com/s/7da6474b776654003959

Citation: Anik, F. W., Zaman, M., Foyez, T., & Mamun, K. A. DOSAGE: A Dataset for Optimized Safe Antibiotic Guidelines and Estimates, Nature Scientific Data, 2025. DOI: https://www.doi.org/10.1038/s41597-025-05761-8

---

🧪 Test Notebooks

The notebooks/ folder contains Jupyter notebooks exercising the validation logic against known test cases (overdose, underdose, renal edge cases, pregnancy flags, etc.).

---

📈 Use Cases

• Integration into CDSS or EHR platforms for real-time prescription audits
• Medical education tools for pharmacological training
• Antimicrobial stewardship analysis
• Reference implementation for AI-assisted medication validation workflows built on the DOSAGE dataset

---

🤝 Contributing

Contributions and feedback are welcome. Please open an issue or submit a pull request.

---

📬 Visit

www.dosage.iriic.uiu.ac.bd