Explainable Knowledge Graph Framework for Tuberculosis Governance Using National Health Insurance Claims at Population Scale
TBGraph is an end-to-end, explainable knowledge graph framework that transforms population-scale TB insurance claims into governance-ready equity insights. It exposes disparities in tuberculosis care access between subsidized (PBI) and non-PBI patients across hospital tiers and 500+ Indonesian cities/regencies, with full GNN explainability.
┌─────────────────────────────────────────────────────────────────────────────┐
│ TBGraph Processing Pipeline │
└─────────────────────────────────────────────────────────────────────────────┘
┌──────────────────┐
│ BPJS Kesehatan │
│ Claims Data │ (500+ cities/regencies · PBI & non-PBI patients)
└────────┬─────────┘
│
▼
┌──────────────────┐
│ Stage 1 · Ontolo │
│ gy Design │ RDFLib · OWL/SKOS schema · tbgraph_ontology.ttl
└────────┬─────────┘
│
▼
┌──────────────────┐
│ Stage 2 · KG │
│ Construction │ NetworkX · RDFLib · entity linking · triple store
└────────┬─────────┘
│
▼
┌──────────────────┐
│ Stage 3 · GNN │
│ Equity Reason- │ PyTorch Geometric · Graph Attention Network (GAT)
│ ing │ PBI vs. non-PBI · hospital tier · regional equity
└────────┬─────────┘
│
▼
┌──────────────────┐
│ Stage 4 · Expla- │
│ inability │ GNNExplainer · subgraph attribution · feature masks
└────────┬─────────┘
│
▼
┌──────────────────┐
│ Policy Report │
│ & Region Stats │ tbgraph_policy_report.json · tbgraph_region_stats.csv
└──────────────────┘
The real BPJS Kesehatan TB claims data used in this research is restricted and cannot be redistributed. Access is subject to formal data-sharing agreements with BPJS Kesehatan and the Indonesian Ministry of Health.
A synthetic data generator is included in the notebook. Navigate to Stage 1A (TBGraph_Pipeline.ipynb) to generate a synthetic cohort that mirrors the statistical structure of the original dataset — including patient insurance category (PBI/non-PBI), hospital tier distributions, regional codes, treatment outcomes, and visit frequency. All downstream pipeline stages run identically on synthetic data.
If you use TBGraph in your research, please cite:
@inproceedings{anonymous2026tbgraph,
title = {TBGraph: An Explainable Knowledge Graph Framework for Tuberculosis
Governance Using National Health Insurance Claims at Population Scale},
author = {Hafidz, I, Rakmawati, N.A, Rusdiansyah, A.},
booktitle = {AI4GOOD Workshop at ICML 2026},
year = {2026},
note = {Workshop paper}
}This project is licensed under the MIT License.
- BPJS Kesehatan for facilitating access to anonymized national health insurance claims data under formal data governance agreement.
- Institut Teknologi Sepuluh Nopember (ITS) Surabaya for institutional support and research infrastructure.
- World Health Organization Global Tuberculosis Report 2024, which informed the equity framing and epidemiological context of this work.