Statistical Analysis
Patient-reported surveys were scored according to the scoring guides of the individual test developers, including methods for handling any missing data. The CCSS-NCQ was scored by calculating a total raw score for each of the 4 domains, then converted to a Z-score using a healthy comparison group as reference, with higher Z-scores corresponding to worse neurocognitive scores.19 Responses for the Neuro-QoL were summed as total raw scores and converted to standardized T-scores with a mean score of 50 and a standard deviation of 10 using the general U.S. population as reference, with lower T-scores suggesting lower cognitive quality of life.21 The PROMIS Sleep Short Form 4a was summed as a total raw score and then translated into a T-score with a mean of 50 and a standard deviation of 10 based on the general population, with higher T-scores signifying worse sleep quality.22 Hearing issues were defined as self-reported moderate/severe hearing trouble or deafness, current use of hearing aid, or moderate/severe tinnitus.
Descriptive statistics including frequency distributions and medians and IQR (interquartile range) were calculated for demographic and treatment variables. Primary outcome variables, including individual CCSS-NCQ domain and Neuro-QoL scores, were further dichotomized into 2 groups based on perceived clinical impairment. Consistent with previous studies, impairment was defined as Z-score >1.28 for the CCSS-NCQ (corresponding to the worst 10th percentile of scores based on healthy control age-adjusted norms)19 and T-score <40 for the Neuro-QoL (corresponding to 1 SD below the standardized mean).24Scores were compared using chi-squared test for categorical variables and analysis of variance or t-tests as appropriate for continuous variables. To assess the potential influence of non-responders on the reported group averages, we also applied inverse probability sampling weights accounting for sex, age at HCT, and current age of the entire eligible population. Finally, the strength of the associations between certain clinical features and impairment on the CCSS-NCQ or Neuro-QoL were examined using multivariable logistic regression and reported as odds ratios (OR) with 95% confidence intervals (95% CI). The models for risk of neurocognitive impairment included dichotomized co-variates, defined as presence or absence of age <10 years at HCT, hearing issues, stroke/seizures, and sleep impairment adjusted for each other, as well as sex and current age. Other potential co-variates were examined but ultimately not included in the multivariable models. All analyses were completed using Stata (Version 16, StataCorp, College Station, TX).