Value-of-Information Question Selection for Symptom Triage with Missing Patient Evidence

Abstract

The advent of automated digital health platforms has revolutionized the initial stages of patient care, primarily through the deployment of symptom triage systems. However, a pervasive challenge in these automated environments is the prevalence of missing patient evidence, as individuals often fail to report complete or clinically relevant symptom profiles during their initial interaction. This paper explores the application of Value-of-Information theory to optimize sequential question selection in symptom triage, aiming to dynamically acquire the most diagnostic evidence while minimizing patient burden. By conceptualizing the triage process as a sequential decision-making problem under uncertainty, we propose a comprehensive framework that evaluates the expected utility of querying potential symptoms based on the current, incomplete patient state. The research systematically investigates the theoretical underpinnings of information elicitation, contrasting it with traditional missing data imputation techniques that often fail in dynamic clinical contexts. Through extensive simulated interactions using standardized medical datasets, our analysis demonstrates that a Value-of-Information approach significantly enhances diagnostic accuracy compared to static decision trees and random querying baselines. Furthermore, it effectively balances the trade-off between information gain and user fatigue. The findings highlight the critical importance of adaptive questioning mechanisms in overcoming the inherent sparsity of patient-reported data, thereby improving the reliability and safety of algorithmic triage systems in modern healthcare infrastructure.