We conducted a nationwide retrospective cohort study using the Korean National Health Insurance Service (K-NHIS) database, which represents the entire population of South Korea. The database comprises the national records of all covered inpatient and outpatient visits, procedures, and prescriptions from 2004 to 2020 [12]. To improve the causal relationship, we mimicked a targeted trial [13]. We specified our research question, target population, exposure, comparators, outcomes, and timing according to the principles of a randomized controlled trial (Supplementary Table 1).

Our cohort included all live births between January 1, 2005, and December 31, 2019, with a washout period of 2004 and a follow-up period of 2020. The NHIS links all claims data of mothers with the claims data of their offspring. We limited our study to children born to females under 40 years of age who gave birth to their first child, recognizing that maternal age can significantly affect pregnancy outcomes (N = 3,525,002). Of eligible participants, 19,474 were born to mothers with cancer. Due to concerns regarding overdiagnosis issues with thyroid cancer [14], we further excluded children born to mothers with thyroid cancer. Finally, 8031 mothers with cancer were included in the study. Controls were matched with mothers who had cancer at a ratio of 1:3 using a propensity score (PS). This study included 8,031 children born to mothers with cancer and 24,093 controls (Fig. 1).

Flowchart of the study participants. Matching variables were age at delivery, delivery date, income, residential area, history of abortion, history of stillbirth

The requirement for informed consent was waived as this study was conducted using anonymized claims data. This study was approved by the Institutional Review Board of Samsung Medical Center, South Korea (SMC 2021-08-107).

The K-NHIS data comprise individual-level demographics and all records of diagnosis and healthcare utilization (e.g., drug prescriptions and medical procedures) provided during inpatient, outpatient, and emergency department visits. Moreover, NHIS claims for inpatient and outpatient visits, procedures, and prescriptions were coded using the 10th revision of the International Statistical Classification of Diseases (ICD) [15].

A cancer diagnosis was defined as the presence of the same C code more than three times within a year or an inpatient hospitalization with the C code [16]. In Korea, once a person receives a cancer diagnosis, they are registered with the National Cancer Registry with a specific code (V193). This indicates that the individual has been diagnosed with cancer and is receiving special insurance benefits. Once a person receives V193, the code becomes a part of the medical records and subsequent claims. Cancer diagnosis during pregnancy was defined as the date of diagnosis, ranging from the last menstrual period to the date of delivery. The last menstrual period was estimated using a previously validated algorithm to estimate the gestational age in administrative healthcare databases [17]. The type of cancer was identified by the ICD-10 code at the visit based on the cancer definition (Supplementary Table 2). The definition of cancer incidence in offspring was identical to that used for mothers.

We considered a broad range of covariates as potential confounders or proxies for potential confounders: maternal age, income, residential area at delivery, and maternal comorbidities, including history of congestive heart failure, history of abortion and stillbirth, and comorbidities during pregnancy. Data on age and income at the time of the first screening were obtained from an insurance eligibility database. Income level was categorized into percentile groups ( ≤ 30th, > 30th to ≤ 70th, and > 70th percentiles). Residential areas at the time of the first screening examination were classified as either metropolitan or rural. Metropolitan areas were defined as Seoul, six metropolitan cities, and 15 cities with populations > 500,000 officially designated as municipal cities (http://www.mois.go.kr). Maternal comorbidities within the year before birth were defined using ICD-10 codes, including hypertension and diabetes. We also identified adverse events during pregnancy, including hypertensive disorders (ICD-10 codes O14, O11, O15, O13, O16, I10, and O10), gestational diabetes (ICD-10 codes O244), and overt diabetes (ICD-10 codes O240, O241, O242, O243, E10, E11, E12, E13, and E14). Preterm birth was defined as the presence of ICD-10 codes O601, O603, P072, O073, or P590.

For 1:3 matching, a propensity score (PS) was generated using birth date, income, residential area, maternal age at delivery, history of abortion, and history of stillbirth. PS matching was performed to minimize the potential impact of confounders on the exposure outcomes. Matching was performed using a greedy algorithm (caliper = 0.1).

The incidence of cancer in the offspring was tracked from the date of birth until the date of cancer diagnosis, death, or the end of follow-up (December 31, 2020), whichever occurred first. Incidence percent per year was calculated as the number of events per 100 person-years of follow-up. The cumulative incidence of each maternal cancer outcome was evaluated using Kaplan–Meier curves. Hazard ratios (HRs) and the corresponding 95% confidence intervals (CIs) were calculated using Cox proportional hazards models. The proportionality of the hazards was assessed by visual inspection of log-minus log plots and Schoenfeld residuals.

Due to lack of power, exploratory subgroup analysis was performed according to maternal age at delivery (< 30, 30–34, and 35–39 years). In addition, we also performed exploratory stratified analysis by characteristics of maternal cancer including maternal age at cancer diagnosis (9–19, 20–24, 25–29, 30–34 and 35–39), maternal cancer type (non-Hodgkin lymphoma [NHL], colorectal, cervix, breast and ovary), and time between cancer diagnosis and pregnancy (diagnosed during pregnancy, < 1, 1 to < 3, 3 to < 5 and ≥ 5).

All analyses were two-sided, and P-values < 0.05 were considered statistically significant. Statistical analyses were performed using SAS version 9.2 (SAS Institute Inc., Cary, NC, USA) and R software version 3.3.2 (Free Software Foundation Inc., Boston, MA, USA).