Statistical analysis
Descriptive statistics summarized the proportion of PTB, overall and by subtypes according to mode of conception. To study the effect of confounding by indication of treatment and fertility status, inverse probability of treatment weighting (IPTW) using propensity score was applied.17,19 The weights were derived to obtain estimates representing the average treatment effect for the treated (ATT).17 The underlying propensity score model included maternal age; obesity (i.e., pre-pregnancy BMI > 30 kg/m2, or, if BMI was missing from BORN, we used the OHIP billing code for obesity [ICD-9 278); parity; smoking; pre-pregnancy and gestational diabetes mellitus; chronic hypertension; immigration; rurality; and income quintile; along with the statistical interaction of parity with age, smoking with obesity, history of preterm birth with age, history of preterm birth with diabetes, and income quintile with age (Table S1). The balance between treatment population was evaluated by standardised differences of baseline covariates, using a threshold of > 0.1 signifying important differences.20
The relation between mode of conception and PTB was quantified by absolute rates and risk ratios (RR), derived using modified Poisson regression with a robust error variance, which also accounts for correlated errors among potentially more than one birth occurring in the same woman,21 and adjusted for baselines characteristics selected through ATT weights.17
This study was reviewed for ethical compliance by the Queen’s University Health Sciences & Affiliated Teaching Hospitals Research Ethics Board and received initial clearance on October 29, 2019 (Reference number is 6028050). Data were analyzed using SAS software version 9.1 (SAS Institute, Cary, NC, 2010).