Abortion and Female Cancer Risk

We first identified the exposure cohort of women of child-bearing age, aged 20 to 45 years, who had abortion records with an induced or spontaneous abortion diagnosis based on ICD-9-CM diagnosis codes 634 to 637 in an outpatient or inpatient record from the NHIRD between 1 January 2004 and 31 December 2007 (n = 278,850). The first date of the abortion record was defined as the index date. For the non-abortion comparison cohort, we included all women aged 20 to 45 years from the NHIRD in 2004 to 2007, then excluded those who had abortion records during the entire study period until the study end date (31 December 2007) (n = 5,001,653), in order to clearly investigate the relationship between exposure of abortion and subsequent female cancer events from the comparison of 2 cohorts for at least a 10-year follow-up period. Given that the non-abortion cohort lacked specific event dates, we randomly assigned index dates based on the dynamic frequency distribution of the first date of cohort identification date to the abortion date from the abortion cohort. For both cohorts, women with any cancer diagnosis or death record prior to the index date, or with any values (e.g., birthday or sex) missing from the databases, were excluded. Finally, the abortion cohort (n = 269,050) and non-abortion cohort (n = 4,715,170) were included for analysis.
To compare the potential female cancer risk between comparable abortion and non-abortion cohorts, the propensity score caliper matching method with 1-to-3 match was used to generate adequate comparison groups based on propensity score in order to reduce the confounding bias from basic characteristics. Propensity score was generated using a logistic regression model including baseline age categories, average monthly payroll groups, fertility, diabetes mellitus, polycystic ovarian syndrome, endometrial hyperplasia, endometriosis, hormone-related drugs, and Charlson comorbidity index (CCI) categories [16 (link),17 (link)]. In addition, standardized differences were calculated in covariates between matched cohorts, and all differences less than 10% indicated acceptable matching [16 (link),17 (link)]. The final matched cohort included 269,050 women in the abortion group and 807,150 in the non-abortion group. Figure 1 presents the flow chart of inclusion and exclusion criteria in the study population.