Identifying Significant Cranial Angles for Predicting Normal vs. Syndromic Craniosynostosis Patients: A Stepwise Logistic Regression Approach

Abstract

Motivation: Previous research in identifying significant angles for early detection of syndromic craniosynostosis (SC) was proposed by making a comparison with a 95% confidence interval (CI) of the angular mean from the non-SC patients. Depending on the number of variables and population studied, this method requires one-by-one comparisons, is time-consuming, and is not robust to outliers.

Objective: we proposed the use of a logistic regression model to identify the significant cranial angles that can well discriminate between the syndromic and non-syndromic patients.

Methodology: Twelve angular measurements of 39 patients (17 patients with SC and 22 normal patients aged between 0 to 12 years) who sought treatment at the University Malaya Medical Centre (UMMC) from the year 2012 to 2020 were obtained from the previous study. 13 regression models (12 simple regression and 1 multiple regression) were produced using the simple and multiple stepwise logistic regression. The significant angles were obtained from the best model which was chosen by comparing their p-value and the Akaike Information Criterion (AIC). 

Results: Results from the simple and multiple logistic regression yields TS-Ba-O (P<0.01) and ACF-DS-Ba (P<0.05) as an important factor in discriminating the patient’s condition. The best logistic regression model however suggested two more significant variables; the Na-S-SO (P<0.02) and Na-Apex point DS-Ba (P<0.08). Compared to the previous study, both TS-Ba-O and Na-Apex point DS-Ba were also captured as significant angles using CI methods.

Conclusion: The logistic regression model may serve as a promising method to identify cranial angles associated with abnormality in a patient's cranial growth.