Prediction of Palatoplasty Timing for Infants With Cleft Lip and Palate Using Machine Learning Algorithm.

This study aims to predict the timing of palatoplasty in infants with cleft lip and palate (CLP) using a machine learning (ML) algorithm. The study included 111 patients with nonsyndromic CLP. Patient subgroups were classified based on birth weight (criterion: -1 SD, low birth weight versus normal birth weight) and cheiloplasty timing (criterion: 5 mo, early cheiloplasty versus late cheiloplasty). Growth trajectories at T2 were compared using T0-weight z -scores between the birth weight subgroups. Changes in the z -scores of weight and height from T1 to T2 were compared between the cheiloplasty timing subgroups. After training the tree-based ML models using cleft type, age, height, and weight at T0, T1, and T2, the palatoplasty timing was predicted with cleft type, weight at T0, and age, height, and weight at T1. The low-birth weight subgroup showed significant catch-up growth during T0-T1 and T0-T2 (all P <0.0001), resulting in no significant difference from the normal birth weight subgroup at T2. Compared with the late cheiloplasty subgroup, the early cheiloplasty subgroup underwent palatoplasty earlier (13.1 versus 14.3 mo; P <0.0001) and showed higher growth rates of weight and height and a greater increase in weight z -scores from T1 to T2 (all P <0.001). The CatBoost algorithm, with a root mean square error of 1.6 months, accurately predicted the palatoplasty timing (mean: actual, 12.8±1.8 mo versus prediction, 12.8±1.0 mo). Use of ML-assisted prediction method may help clinicians decide the timing of personalized palatoplasty in infants with CLP.