Statistical Analysis

All analyses were performed in Stata statistical software version 16.1 (StataCorp) with 2-sided tests of hypotheses. We first performed descriptive statistics, reporting frequencies and proportions. We assessed for associations of age, sex, fracture type, year, payer type, and race and ethnicity with neuroimaging use in unadjusted χ² analyses or Fisher exact test when cell size was less than 5. We then used multivariable logistic regression to estimate the association of these factors with odds of neuroimaging use. We a priori elected to include age, sex, fracture type, year, payer type, race and ethnicity, and hospital in our model. The significance level was set α = .05. We repeated our model, restricting the population to infants ages less than 6 months and separately to infants ages 6 to 12 months.

We report odds ratios (ORs) and the estimated probability of neuroimaging for publicly vs privately insured children and at each hospital.^25,26 The estimated probability is based on marginal standardization,²⁰ calculated by treating all individuals in this population as if they were publicly insured; we performed the same calculation for all individuals while treating them as if they were privately insured.²⁵ The calculation across hospitals assumed that all infants in the sample presented to each individual hospital, calculating the mean estimated probability of neuroimaging at each hospital.²⁰ We multiplied the estimated probabilities by 100 and describe these as adjusted percentages for ease of interpretation. We used CIs from the Stata margins command after logistic regression. CIs less than 0 or greater than 100 were clipped to 0.01% and 99.9%, respectively. Data were analyzed from March 2021 through January 2022.