Two-Stage Transformer Predicts AECOPD Risk and Timing.

Acute exacerbation of chronic obstructive pulmonary disease (AECOPD) requires timely prediction due to its rapid worsening. Existing machine learning methods often rely on episodically collected clinical data, leading to delays that limit their utility in home monitoring. This research leverages continuous data from home ventilators, which record respiratory status, to overcome these limitations. The paper proposes a novel two-stage framework that directly processes raw pressure and flow waveforms from the most recent seven days of home ventilator use. The first stage employs a classification model to identify patients at high risk of a severe exacerbation. The second stage then uses a regression model to estimate the number of days remaining before the predicted event occurs. Experimental results demonstrate that this two-stage model significantly outperforms traditional baseline models. The Stage 1 classifier achieved an F1 score of 0.91, while the Stage 2 regression model yielded an RMSE of 1.00 days and an R^2 of 0.76. This framework provides clinicians with both an early warning of an impending severe exacerbation and a precise, actionable lead time, enhancing patient care in home settings.