A Shap Interpreter-Based Explainable Decision Support System for COPD Exacerbation Prediction

COPD is a chronic lung disease that can exhibit exacerbation. An exacerbation requires additional therapy. When treating COPD using a pulse oximeter, some SpO2 (blood Oxygen Saturation) levels are to be targeted depending on the prognosis of the disease and the current state of symptoms. GOLD (Global Initiative for COPD) 2020 proposed to evaluate respiratory symptoms severity change for the assessment of exacerbation occurrence and personalized therapy. We proposed a machine learning model to evaluate the change of symptoms severity in the association with SpO2 for exacerbation prediction. Using a Shap interpreter, the contribution of exacerbation predictors, is computed at the individual level. This could guide clinicians to evaluate whether and how much SpO2 contributed to the individual exacerbation probability, helping them make a decision on the initiation of oxygen. This personalized approach contrasts with delivering Oxygen without aligning it with a patient's actual needs. The pre-trained model was integrated with the shap interpreter for the DSS (Decision Support System) design. Afterward, DSS was validated on 56 patients by attaining a concordance rate of 94.9% when comparing the DSS and medically made decisions, about exacerbation events. We report that the proposed DSS is much better than the current clinical assessment, which doesn’t interpret the individual current SpO2 contribution to the exacerbation risk.