Artificial Intelligence for Pandemic, Epidemic, Preparedness & Response (AI4PEP)

In response to rising global dengue cases in May 2024, the ExplainDengue platform was built to anticipate outbreaks by leveraging the climate-sensitive nature of the disease.

While routine surveillance often catches outbreaks late, this system uses a graph neural network to forecast emerging risks.

It integrates dengue case data from Indonesia’s EWARS/SKDR system for Yogyakarta with location details, time-series data, climate variables from OpenWeather and NASA POWER, and a satellite-derived NASA vegetation index.

ExplainDengue connects human, environmental, and climate data for various levels of health offices. Future plans include adding syndromic-surveillance signals from healthcare facility records.

In 2025, the AI4PEP team held a workshop with district health offices in Sleman and Gunungkidul to trial the platform and establish locally grounded alert thresholds.

By utilizing case data, outbreak reports, and SKDR alert histories, the team and local stakeholders mapped out when surveillance signals should trigger public health action.

Moving forward, the team is refining the platform with local partners to expand it into a validated early-warning system across Yogyakarta, modeling responsible AI-enabled surveillance for Indonesia.

Indonesia’s routine surveillance system, EWARS/SKDR, relies on structured and disease-specific reports.

This creates two significant blind spots: detailed clinical data remains hidden in unstructured free-text notes, and the system lacks the agility to detect novel or unexpected syndromes that are essential for pandemic preparedness.

Additionally, manual reporting places a significant burden on frontline health workers.

We are building a system that uses large language models and natural language processing to evaluate unstructured data from Puskesmas, laboratories, and hospitals.

By interpreting symptom patterns expressed through local terms and abbreviations before a diagnosis is finalized, the system groups them into syndromic categories such as acute febrile, respiratory, and gastrointestinal syndromes.

These categories can then be analyzed to identify unusual spatial and temporal clustering.

The system also reduces health workers’ reporting burdens by extracting relevant information from existing records without requiring additional reporting forms.

This tool complements EWARS/SKDR and integrates with ExplainDengue. Within the AI4PEP framework, facility-level syndromic signals are merged with predictive analytics.

This provides public health teams with both a forecast of known disease risks and a broad, symptom-based lookout for unexpected health threats.

Created within AI4PEP’s ethics and governance stream, the system incorporates strong privacy safeguards, alignment with national governance, and strict human oversight.

The AI identifies and explains candidate signals, while public health personnel remain responsible for interpretation and decision-making, embedding transparency, fairness, and accountability.

As the system matures, AI4PEP aims to demonstrate how responsible and explainable AI can enhance frontline surveillance—helping Indonesian health offices detect outbreaks earlier, respond faster, and prepare for previously unseen diseases.

AI-driven outbreak detection requires public trust because it relies on sensitive data and may significantly affect individuals and communities.

Consequently, AI4PEP integrates ethics and governance from the start by developing consensus-based ethical guidelines for the responsible use of AI in Indonesian infectious disease surveillance.

AI surveillance raises crucial questions concerning data protection, equity, model transparency, and accountability in high-stakes decision-making.

To maintain public trust, global standards must be adapted so that they are meaningful, applicable, and feasible within the Indonesian context.

AI4PEP’s ethical initiatives run parallel to technical progress. Rather than serving only as a final report, the guideline functions as an active design and governance framework.

It establishes upfront requirements for privacy, access control, transparency, alert proportionality, equity, and human oversight.

The resulting framework addresses privacy, fairness, transparency, accountability, human control, public benefit, vulnerability inclusion, and redress mechanisms.

In line with AI4PEP’s focus on equitable systems led by the Global South, the study prioritizes gender equality and vulnerability safeguards.

These safeguards are intended to protect populations most at risk of harm, exclusion, or unequal treatment.

This collaboration creates a tailored and consensus-driven guideline that shapes the technical, operational, and institutional readiness required for responsible and trusted AI surveillance in Indonesia.