Send message to Authors

Abstract

In the context of healthcare digitalisation in Central Eurasia, pre hospital Patient Decision Support Systems (PDSS) are gaining importance as tools for guiding patients before they seek in person care. Existing triage protocols (e.g., NHS Pathways or Schmitt–Thompson algorithms) are designed for interactive settings that rely on sequential clarification questions. Applying such protocols to static, unstructured layperson narratives presents a methodological challenge. Due to limited clinical context, both human annotators and artificial intelligence (AI) systems tend to overestimate risk (up triage) following a “safety first” heuristic, which may lead to biased interpretations and reduced practical usefulness.

This study addresses the first stage of developing a system of Medical Action Recommendation from Layperson Narratives (MARLaN) by focusing on dataset construction and annotation design. We propose a four level classification framework for pre triage of layperson texts that integrates international approaches to legitimising “self care” outcomes with national medical triage standards of the Republic of Kazakhstan (Ministry of Healthcare Orders No. 27 and No. 225/2020). A key feature of the scheme is the inclusion of an “Insufficient Data” category (Grey Zone), which captures cases where available information is inadequate for safe decision-making and reduces forced or speculative judgments.

The dataset is developed in collaboration with medical professionals, who annotate layperson-style health queries according to a structured codebook. This resource enables analysis of how individuals without medical training describe symptoms and how such narratives can be systematically interpreted within a local healthcare context. When fully implemented, the proposed approach (MARLaN) may contribute to safer patient routing by reducing the risk of missing cases requiring immediate medical attention, while also helping to alleviate unnecessary healthcare visits through appropriate use of self-care recommendations, thereby allowing healthcare systems to focus resources on patients with genuine need.

From a regional perspective, the study contributes to ongoing discussions on healthcare accessibility, digital mediation, and patient behavior in Central Eurasia, and provides a practical foundation for future development of AI-assisted patient routing tools adapted to local institutional frameworks.