Predicting Changes in Cephalic Index Following Spring-mediated Cranioplasty for Nonsyndromic Sagittal Craniosynostosis: A Stepwise and Machine Learning Algorithm Approach.

BACKGROUND

Spring-mediated cranioplasty (SMC) is an increasingly utilized technique to treat patients with nonsyndromic sagittal craniosynostosis, but variables impacting outcomes are incompletely understood. The purpose of this study was to determine variables most predictive of outcomes following SMC, primarily changes in cephalic index (CI).

METHODS

Patients with nonsyndromic sagittal craniosynostosis undergoing SMC at our institution between 2014 and 2021 were included. Cephalic index was measured from patient computed tomography scans, x-rays, or by caliper-based methods. Parietal bone thickness was determined from patient preoperative computed tomography. Stepwise multiple regression analysis, least absolute shrinkage and selection operator, and random forest machine learning methods were used to determine variables most predictive of changes in CI.

RESULTS

One hundred twenty-four patients were included. Stepwise multiple regression analysis identified duration of spring placement ( P =0.007), anterior spring force ( P =0.034), and anterior spring length ( P =0.043) as statistically significant predictors for changes in CI. Least absolute shrinkage and selection operator analysis identified maximum spring force (β=0.035), anterior spring length (β=0.005), posterior spring length (β=0.004), and duration of spring placement (β=0.0008) as the most predictive variables for changes in CI. Random forest machine learning identified variables with greatest increase in mean squared error as maximum spring force (0.0101), anterior spring length (0.0090), and posterior spring length (0.0056).

CONCLUSIONS

Maximum and total spring forces, anterior and posterior spring lengths, and duration of spring placement were the most predictive variables for changes in CI following SMC. Age at surgery and other demographic variables were inferior predictors in these models.