Toward unsupervised outbreak detection through visual perception of new patterns.

Area of Science:

• Epidemiology
• Public Health Surveillance
• Data Visualization

Background:

• Conventional outbreak detection relies on predefined syndromes (e.g., influenza).
• Emerging diseases lack prior clinical information, hindering traditional statistical detection.
• This paper addresses the need for methods to detect novel outbreaks without a priori knowledge.

Purpose of the Study:

• To present a novel visual method for facilitating outbreak detection.
• To enable the identification of emerging diseases with unknown clinical presentations.

Main Methods:

• Utilizes a visual representation of symptoms and diseases coded using the International Classification of Primary Care 2nd version (ICPC-2).
• Transforms surveillance data into color-coded cells (white to red) indicating sign frequency.
• Employs a graphic reference frame mimicking body anatomy for pattern analysis over time.

Main Results:

• Demonstrated through retrospective analysis of General Practitioner (GP) and Hospital Emergency Department (HED) data.
• Successfully visualized the 2003 heat wave's health impact using GP data, which conventional systems missed.
• Identified flu epidemics visually using HED data, complementing standard statistical methods.

Conclusions:

• Leverages human visual pattern recognition for detecting unexpected health events.
• Requires convenient image representation of epidemiological surveillance and trained "epidemiology watchers".
• Proposes a system where "epidemiology watchers" signal alerts from "image walls" for validation by field epidemiologists.