JMIR Publications

ABSTRACT

Background:

Acute Exacerbations of Chronic Obstructive Pulmonary Disease (AECOPD) are major determinants of disease progression, hospitalization, mortality, and healthcare costs of patients affected by COPD. Accurate prediction of exacerbations is therefore essential to enable timely intervention and personalized disease management. Artificial intelligence (AI) has emerged as a promising approach for AECOPD prediction using either thoracic imaging or home-based monitoring. While imaging provides detailed structural information on lung pathology, home-based monitoring captures longitudinal physiological, behavioural, and environmental changes that may precede an exacerbation.

Objective:

The aim of this systematic review is to compare AI models based on imaging and home-based or sensor-based monitoring for the prediction of AECOPD, evaluating their predictive performance, methodological quality, clinical utility, feasibility, and potential for real-world implementation.

Methods:

A systematic review was conducted according to the PRISMA guidelines. Studies using AI to predict AECOPD from thoracic imaging or home-based monitoring data were identified through a systematic literature search. Data on study characteristics, prediction tasks, AI methodologies, predictive performance, and validation strategies were extracted, and methodological quality was assessed using the Prediction Model Risk of Bias Assessment Tool.

Results:

Twentytwo studies met the inclusion criteria, including six imaging-based and sixteen home-based studies. Imaging models primarily combined Computed Tomography-derived biomarkers with clinical variables to predict AECOPD, reporting AUC values between 0.80 and 0.97. Home-based approaches integrated longitudinal physiological, behavioural, environmental, and patient-reported data to predict AECOPD, with greater variability in performance; despite this, several studies reported AUC values above 0.95. However, across both approaches, high predictive performance was frequently linked to methodological limitations, including inadequate external validation, heterogeneous outcome definitions, limited sample sizes, and moderate-to-high risk of bias. Imaging studies were methodologically more homogeneous, whereas home-based studies showed variability in monitoring technologies, data acquisition protocols, prediction horizons, and predictive methods.

Conclusions:

Imaging-based and home-based AI models should be considered as complementary rather than competing approaches to AECOPD prediction. Imaging-based models are best suited for risk stratification by characterizing lung structural abnormalities, whereas home-based monitoring enables continuous assessment of physiological changes associated with imminent clinical deterioration. Future research should prioritize multimodal AI models integrating structural imaging biomarkers with longitudinal health data, supported by standardized methodologies, rigorous external validation, and prospective evaluation in real-world clinical settings to facilitate translation into routine COPD care.