Outcomes and Predictive Modeling in Helmet Therapy for Plagiocephaly.

OBJECTIVE: To develop and evaluate a machine learning model to forecast treatment-related changes in cranial vault asymmetry index (CVAI) and cranial index (CI) in infants undergoing cranial orthosis therapy for deformational plagiocephaly (DP) and/or deformational brachycephaly (DB). STUDY DESIGN: This retrospective analysis utilized serial measurements of CVAI and CI for 6,694 infants with DP and/or DB at initiation and throughout cranial orthosis therapy from 4/1/2012 to 10/31/22 across cranial orthosis health services in nine US states. Extreme Gradient Boosting (XGBoost) was employed to predict CVAI and CI at the end of therapy based on initial cranial measurements, patient age and duration of treatment. RESULTS: At initial presentation for helmet therapy, the median [interquartile range (IQR)] age of infants in months was 6 (5-8), with 36.5% (n=2446) having isolated DP, 9.0% (n=600) isolated DB, and 54.5% (n=3648) both DP and DB. Initial median CVAI and CI were 7.6% (IQR: 6.2%-9.4%) and 95% (IQR: 92.6%-98.5%), respectively. The XGBoost models demonstrated high predictive accuracy for final CVAI (explaining 79.4% of the variance [R2 = 0.794] and a mean absolute error [MAE] of 0.756) and CI (86.4% of the variance [R2 = 0.864] with an MAE of 0.936). CONCLUSIONS: A large, retrospective cohort analysis of patients undergoing cranial molding therapy for DP and/or DB enabled development of a machine learning-based model that forecasts treatment-related changes in cranial shape during helmet therapy. Further investigation is needed to assess how this predictive tool may assist in clinical decision making and informed parental counseling regarding cranial orthosis for DP and/or DB.