Statistical Analysis
The exposure was AYA cancer, and the outcomes were preterm birth and SMM. Descriptive characteristics were calculated using frequency and percentages and compared by exposure group Student’s t-test or Chi-square test of proportions, as appropriate. Distributions of continuous variables were assessed for normality and reported as mean±SD. Due to skewed distribution, the Maternal Comorbidity Index was log-transformed for analysis. Analysis was stratified by singleton versus multiple live birth, because of known effect modification by pregnancy plurality.33
To compare differences in maternal characteristics between births for exposed and unexposed females, we used log-binomial regression models to estimate risk ratios (RR) and 95% confidence intervals (CI) for the outcomes of interest with adjustment for potential confounders (Model 1). We tested for effect modification based on p-values from cross-product terms in regression models. Maternal comorbidities were hypothesized mediators, or causal intermediates, of the relationship between AYA cancer and adverse perinatal outcomes (Model 2). Mediation analysis was undertaken to obtain estimates accounting for the mediator and to estimate the proportion mediated by comorbidities, accounting for potential interactions and non-linear relationships.34
Mediation analysis via a counterfactual framework was conducted using PROC CAUSALMED in SAS35, and GLM regression models were analyzed in R. Subgroup analyses were performed: 1) excluding gynecological cancers because of known associations between these and preterm birth, and 2) restricting the study period from 2010 to 2019 to considering temporal advances in cancer and perinatal care. All tests of significance were two-tailed, and alpha was 0.05.