Development and validation of machine learning for early mortality in systemic sclerosis

Area of Science:

• Rheumatology
• Medical Informatics
• Prognostic Research

Background:

• Clinical predictors of systemic sclerosis (SSc) mortality vary significantly across populations and healthcare settings.
• A need exists for a simple, precise predictive tool for early mortality in SSc patients to guide general practitioners.
• Early identification of high-risk SSc patients is crucial for timely intervention and improved outcomes.

Purpose of the Study:

• To develop and validate a simple predictive model for early mortality in systemic sclerosis (SSc) patients.
• To utilize machine learning algorithms for accurate SSc mortality prediction.
• To create a practical referral tool for general practitioners managing SSc.

Main Methods:

• A historical cohort study design was employed, analyzing data from 528 adult SSc patients between 2013 and 2020.
• Various machine learning algorithms, including deep learning and ensemble methods, were utilized for mortality classification.
• Model performance was assessed using area under the receiver operating characteristic curve (auROC) and confusion matrix metrics.

Main Results:

• Two models demonstrated high predictive performance, with Model 1 (mRSS and WHO-FC ≥ II) outperforming Model 2 (mRSS and WHO-FC ≥ III).
• Internal validation showed good accuracy and auROC values for both models.
• High specificity was observed in both models, indicating reliable identification of non-mortality cases.

Conclusions:

• A simplified machine learning model incorporating the modified Rodnan skin score and WHO functional class can effectively predict early mortality in SSc.
• This predictive model can serve as a valuable tool for guiding early specialist referrals and informing patient management strategies.
• Further external validation across diverse SSc clinics is recommended to enhance generalizability.