| Introduction: Children born very low birth weight (VLBW) have an increased risk of impaired preschool motor coordination, which may have negative effects on the child's mental and physical health. Physical and occupational therapy services are suggested to attenuate the negative effects of poor preschool coordination. We estimated the effect of physical and occupational therapy services delivered in early childhood on preschool motor coordination among VLBW children. To control for confounding, we implemented propensity score (PS) methods estimated using traditional logistic regression (LR) and tree based methods. Methods: Using the Early Childhood Longitudinal Study Birth Cohort (ECLS-B) we estimated the effect of therapy on: skipping eight consecutive steps, hopping five times, standing on one leg for ten seconds, walking backwards six steps on a line, jumping distance, and change in jumping distance from preschool to kindergarten. We estimated the PS using random forest classification, bagging, and a single tree using the R statistical program and with LR in SAS 9.2. Using linear regression, we modeled the estimated effect of therapy on the distance that the child jumped. We weighted the adjusted models using inverse probability of treatment weights estimated from all four methods. We modeled all other end points as stated using LR. Results: Approximately 500 children were VLBW. RF and Bagging produced the best covariate balance between treatment groups (MSD 0.07, 0.03). The single classification tree produced the worst covariate balance (MDS 0.18). When estimating the PS with RF, treated VLBW children were 2.39 times as likely to successfully skipping eight steps (OR: 2.39, 95% CI: 0.75, 7.51) compared to the untreated group. Treated children jumped an additional 1.79 inches (95% CI: -2.21-5.79) further and were also 52% (OR: 1.52, 95% CI: 0.51, 4.54) more likely to successfully complete the backwards walking task. There was little effect of therapy on other endpoints. Effect estimates were similar among models weighted with RF, bagging, and LR. Conclusion: Providing therapy to VLBW children, may improve the child's school age motor coordination. RF is a useful method to improve covariate balance when estimating the PS and to potentially reduce bias in observational studies. |
