Predicting progression of Mycobacterium avium complex pulmonary disease with treatment using explainable artificial...

Area of Science:

• Pulmonology
• Infectious Diseases
• Artificial Intelligence in Medicine

Background:

• Nontuberculous mycobacterial pulmonary disease (NTM-PD) incidence is increasing globally.
• Identifying patients with NTM-PD who will progress to requiring treatment remains a clinical challenge.

Purpose of the Study:

• To develop and validate a predictive model for clinical progression in patients with NTM-PD.
• To identify key predictors of treatment initiation in NTM-PD.

Main Methods:

• Retrospective cohort study of 303 adult patients with NTM-PD not on treatment at enrollment.
• Utilized a point-wise linear neural network to screen 63 clinical, laboratory, microbiologic, physiological, and radiographic variables.
• Developed a logistic regression model incorporating age, treatment history, cavitation, and modified Reiff score.

Main Results:

• The final two-variable model, using four routinely available inputs, achieved an AUC of 0.81.
• Younger age, presence of cavitation, higher modified Reiff scores, and prior treatment history were associated with clinical progression.
• Bayesian network analysis revealed that age and treatment history influenced progression via radiographic disease burden.

Conclusions:

• The modified Reiff score, combined with other routine factors, is a significant predictor of clinical progression in Mycobacterium avium complex pulmonary disease (MAC-PD).
• Explainable AI models were instrumental in demonstrating the importance of the modified Reiff score for predicting MAC-PD progression.