Respiratory Syndromic Surveillance with Alert and Analytics System Using Artificial Intelligence

Abstract

The objective of this work is to design and assess a replicable syndromic monitoring system for respiratory infections utilizing artificial intelligence approaches, suitable for primary healthcare environments with constrained resources. An ETL pipeline was established for processing data from 2,777 outpatient records at the Villa Nueva Health Center (April-December 2024). A syndromic classifier based on TF-IDF was created using logistic regression, accompanied by an anomaly detection system that integrates moving average thresholds, CUSUM, and EWMA algorithms. The system's performance was assessed utilizing operational metrics. The syndromic classifier attained an F1-score of 0.999 and an AUC-ROC of 1.00. During the analysis of 39 epidemiological weeks, the system produced 9 alerts (5 red, 4 yellow), resulting in an alert rate of 23.1%, a proxy sensitivity of 60%, and an operational precision of 100%. The alert delay was seven days. Two notable outbreak clusters were identified: weeks 3 to 5 and weeks 28 to 32. Conclusions: The established system illustrates that automated syndromic monitoring is viable in resource-constrained primary care environments, offering a reproducible pipeline, detector, and dashboard solution for early warning of respiratory outbreaks.