AI4PEP LLM
From Clinical Narratives to Public Health Signals
The AI4PEP LLM project develops and implements large language model (LLM)-based tools to transform unstructured clinical narratives into structured, actionable information for infectious disease surveillance in Indonesia. The project explores how information already available in clinical records can support earlier detection of disease patterns while complementing existing surveillance systems.
into Public Health Intelligence
Extracting Public Health Signals
from Clinical Narratives
Routine surveillance systems such as EWARS/SKDR rely largely on structured and disease-specific reporting, while valuable clinical information may remain within unstructured medical records. The project develops LLM-based approaches to extract and classify relevant information from clinical narratives to support infectious disease surveillance.
Project Objectives
Develop LLM-Based System
Build and fine-tune NLP models to extract and classify clinical entities from unstructured medical text.
Integrate with Surveillance
Link extracted signals seamlessly with routine epidemiological pipelines like EWARS/SKDR.
Pilot and Evaluate
Conduct real-world validation across healthcare facilities to evaluate accuracy and utility.
Ensure Responsible AI
Implement stringent data privacy, medical ethics, algorithmic fairness, and human oversight frameworks.
AI4PEP LLM Surveillance Platform
Transforming clinical narratives into structured surveillance signals.
- Classify syndromes from clinical notes
- Support SKDR priority diseases
- Explainable AI output with confidence score
- Designed for real-world healthcare settings
Clinical Note
Suggested Classification
- Demam tinggi
- Ruam
- Nyeri sendi
Target SKDR Classes
- Diarrhoea & Jaundice
- Acute Respiratory Infection
- Fever with Rash (Arboviral)
- Vaccine Preventable Diseases
- Zoonotic Diseases
- NTDs and Other Priority Diseases
and syndromes
From Clinical Notes to Public Health Action
Sistem mengubah data klinis tidak terstruktur menjadi informasi yang dapat digunakan untuk surveilans melalui alur sistem terintegrasi. Model memberikan rekomendasi klasifikasi dan informasi pendukung, sementara verifikasi dan keputusan akhir tetap berada pada tenaga kesehatan/pengelola surveilans.
Clinical Notes (EMR)
Free-text data from Puskesmas, hospitals, and laboratories.
Data Processing & Privacy
De-identification, normalization, and secure processing.
LLM / NLP
Interpret symptom patterns using large language models.
SKDR Classification
Map to 24 SKDR syndromes with confidence score and explanation.
Signal Analysis
Spatial & temporal trends and unusual anomaly detection.
Public Health Action
Support early investigation, targeted response, and decision-making.
Testing in Real-World Healthcare Settings
The system is being piloted in selected healthcare facilities in Yogyakarta to evaluate performance, usability, and real-world impact.
Pilot Sites
- RS Akademik UGM (RSA)
- RS Nyi Ageng Serang
Data Sources
- Clinical records (unstructured text)
- Laboratory data
- Integration with EWARS/SKDR
Timeline
Evaluation Focus
- Detection of unusual syndromic signals
- Usability and acceptability
- Implementation feasibility
- Integration with existing workflows
Stronger Surveillance for a Safer Tomorrow
Project Team
Partners & Funders
Latest News
View All News →Building Smarter Surveillance
for Faster Public Health Action.
Combining artificial intelligence, implementation science, and local public-health expertise to strengthen infectious disease preparedness, outbreak detection, investigation, and response.