A systematic literature review of explainable risk assessment models for bronchial asthma

Abstract

Background: The reduced quality and risks to life brought on by bronchial asthma (BA) have heightened the need for trustworthy risk assessment solutions with deliberate interpretability and transparency. Improper management of BA, such as ignoring symptoms, improper inhaler technique, or recent admissions to the intensive care unit (ICU), puts a patient at a higher risk of future asthma exacerbations, complications, or even death. This paper details a systematic literature review on recent literature to identify and analyse current explainable artificial intelligence (XAI) risk assessment models used in BA or the assessment of risk in healthcare using XAI.

Methods: A systematic review of English literatures was conducted through Science Direct, Association for Computing Machinery (ACM) Digital Library, Springer, PubMed, and Scopus between January 1, 2019 and October 26, 2023. All studies that incorporated XAI or risk assessment models for BA or health were included for this review. A combination and permutation of the following search terms was used: “explainable artificial intelligence”, “risk assessment”, “risk assessment model”, “asthma”, and “health”.

Results: A total of 43 literatures were included after screening through 689 literatures combined from the specified sources, with duplicates and materials not meeting the inclusion criteria removed. Among them, five of the literatures conducted research on asthma, while seven conducted research on lung-related diseases using explainable machine learning (ML) or deep learning (DL) techniques. The model that had better performance when compared to the other models in the 12 most relevant literature out of the 43 was extreme gradient boosting (XGBoost), with it having better performance two out of the three times it was compared to other models. The most common output was risk prediction with 36 literatures, followed by diagnosis with seven literatures and classification with one.

Conclusions: XAI has been used within the domain of asthma for diagnosis or prediction of future hospital visits; however, there is a scarcity for studies on explainable predictive models for asthma exacerbation risks. Research on XAI within this domain has the potential to contribute towards explainability in asthma risk prediction.