Association between having a pregnancy complicated by vaginal bleeding and risk of cancer

Setting, design and data sources

This was a registry-based cohort study conducted using individual-level linked data from Danish health and administrative registries, set within Denmark’s universal and tax-funded health care system (29). The Civil Registration System assigns personal identifiers and contains data on date of birth, vital and civil status, and dates of immigration and emigration (29,30). Data on all deliveries are captured by the Medical Birth Registry (MBR) starting in 1973 and includes women’s and pregnancy-related characteristics including but not limited to women’s age at delivery, gestational age, and parity (31,32). Smoking status in pregnancy is collected by the MBR from 1991 and body-mass index (BMI) from 2004. Information on stillbirths at the gestational age of at least 28 full weeks is available in the MBR before April 2004 and at 22 full gestational weeks afterwards; livebirths are recorded regardless of gestational age. Information on highest education level (33), employment status (34) and personal income (35) are stored at Statistics Denmark and available starting in 1980-1981 (36). Hospital encounters are recorded in the Danish National Patient Registry (DNPR) (37,38), including data on medical procedures and discharge diagnoses, coded using the International Classification of Diseases, 8^th revision (ICD-8) in 1977-1993 and 10^th revision (ICD-10) thereafter (37). Hospital admissions are available in the DNPR from 1977, while outpatient clinic and emergency room visits are recorded from 1995 onwards (37,38). Admissions and outpatient specialist visits due to psychiatric conditions are captured as part of the the DNPR (39) and were available for this study from 1995. The Danish National Prescription Registry (40) captures data on prescribed medication dispensings from 1995 onwards and records the Anatomical Therapeutic Classification (ATC) codes of the dispensed drugs and the dates when the prescriptions were redeemed. The Danish Cancer Registry (DCR) captures all primary incident cancer diagnoses (with compulsory reporting from 1987) and contains data on tumour morphology (41).

Study population

Using the MBR, we identified all pregnancies ending in a liveborn or stillborn delivery between 1995 and 2017. We kept one record per delivery of a multifetal pregnancy. Using the DNPR, we obtained and linked data on VB within 20 gestational weeks starting from the last menstrual period date, computed by subtracting gestational age at delivery from the delivery date (Table 1 in the electronic supplementary material). Pregnancies ending in a delivery without a VB record in the DNPR within 20 weeks of gestation were classified as VB-unaffected. The date of delivery served as the start of the follow-up for both VB-affected and VB-unaffected cohorts of pregnancies.

Using the DNPR, we identified pregnancies ending in a miscarriage (spontaneous abortion) or a termination (medical or surgical abortion) in 1995-2017, using relevant diagnostic and procedure codes (Table 1 in the electronic supplementary material). To avoid multiple counting of the same miscarriages or terminations, a woman with repeated event-qualifying diagnostic or procedure code was allowed to re-enter the cohort after 180 days, whereby she was assumed to have a new pregnancy. When multiple dates associated with the pregnancy termination or miscarriage were available (date of procedure, date of hospital admission, date of hospital discharge), we included the earliest. The date of the hospital-based encounter for pregnany termination or miscarriage served as the index date. We excluded from the study pregnancy records of women with cancer or HIV/AIDS diagnosis before the index date, records with missing women’s age at the pregnancy end, and pregnancies ending in childbirth with missing gestational age. The flow-chart for study population construction is presented in Figure 1.

• Download figure
• Open in new tab

Figure 1. Flow chart for the study population ascertainment

Standalone figure

Cancer outcomes

Using the DCR, we ascertained incident diagnoses of cancer. We investigated the long-term risk of any cancer; cancers at specific anatomic sites, focusing on breast, cervical, ovarian and fallopian tube, and uterine cancer, and cancers grouped as hormone-related, haematological, immune-related, smoking or alcohol-related, obesity-related, and cancers of neurological origin, as previously described (42–44)(Table 1 in the electronic supplementary material).

Covariables

The potential confounding factors were age, calendar year of pregnancy end, women’s parity, number of previous identifiable pregnancies, socioeconomic factors (education, employment, and year-specific income level divided into quartiles), reproductive factors (pre-index date vaginal bleeding, termination, or miscarriages, placental complications in previous and current deliveries), chronic somatic conditions (obesity, polycystic ovary syndrome, thyroid disorders, chronic kidney disease, chronic liver disease, hyperlipidaemia, diabetes, hypertension, and connective tissue disorder), and history of any psychiatric condition using the entire available lookback period. We additionally adjusted for the use of NSAIDs and steroids for systemic use before the 12 weeks of gestation for VB-affected and VB-unaffected pregnancies and before the index date for terminations and miscarriages using 365 days lookback period for NSAIDs and the entire available lookback period for steroids for systemic use. In the comparisons of pregnancies ending in childbirth, we additionally controlled for placental complications (hypertensive disorders of pregnancy, placenta praevia, and abruptio placenta) not only at the previous deliveries but also at the index pregnancy as a late manifesting proxy of potential placental dysfunction early in pregnancy.

Statistical analyses

We followed each woman from the index date until the earliest of cancer diagnosis, death, emigration, or end of the study on 31 December 2018. We constructed the cumulative incidence curves of any cancer at the end of a 24-year follow-up while treating death and emigration as competing events (Figure 1 in the electronic supplementary material) and calculated cancer incidence rates (IRs) per 10 000 person-years (PYs). We used Cox proportional-hazards regression to compute crude and adjusted hazard ratios (HRs) for the cancer outcomes. In the analyses of all pregnancies, with each subsequent pregnancy of a woman, we accounted for the reproductive history, including prior exposure status, at the previous pregnancies. We computed robust 95% confidence intervals (CIs) taking into account that different non–independent pregnancies of the same woman were analysed (45). Further, we repeated the main analyses following women’s first identifiable pregnancy. In separate sets of analyses, we compared the risk of cancer outcomes following a VB-affected pregnancy vs 1) VB-unaffected pregnancy, 2) termination, or 3) miscarriage.

The HRs were adjusted for potential confounders (women’s age at pregnancy end, calendar period, socioeconomic factors [education, employment, and year-specific income level divided into quartiles], reproductive history [parity, number of pre-index date identifiable pregnancies, history of VB, terminations or miscarriages, hypertensive disorders of pregnancy, placenta praevia, abruptio placentae] and placenta-related complications at the current pregnancy ending in delivery [preeclampsia-eclampsia, placenta praevia, abruptio placentae]; pre-index date thyroid conditions, obesity, polycystic ovary syndrome, chronic obstructive pulmonary disease, chronic liver and kidney diseases, connective tissue disorders and medication use [non-steroid anti-inflammatory drugs and steroid for systemic use]). In analyses of all identifiable pregnancies of a woman, we ascertained the history of all covariables before the index date in a time-varying manner updating the covariables’ status at each subsequent pregnancy.

In a sensitivity analysis, we investigated the association of VB-affected pregnancy with any cancer and hormone-related, haematological, immune-related, smoking or alcohol-related, obesity-related, and cancers of neurological origin additionally adjusted for women’s BMI in analyses of pregnancies ending in a delivery; however, the follow-up was short and accrued few events. We additionally investigated the associations with cancer outcomes restricting the data to women having their last recorded pregnancy at the age of 40 years or later as the age, when most women likely reach the desired family size. We did not impute missing values and included variables with missing data in the statistical models using “missing” as a category.

We used log-minus-log plots to evaluate the proportionality of hazards assumptions, and the curves were close to parallel over the follow-up period for overall cancer. All analyses were conducted on the servers of Statistics Denmark (36) using RStudio (46) and R software for statistical computing (47) versions 3.6.0-4.1.0, including the following packages and their dependencies: tidyverse (48), ggplot2 (49), survival (45), cmprsk (50), prodlim (51), and epiR (52). Reported counts were masked to protect personal data, in compliance with the Statistics Denmark regulations.

Ethics

Patient consent or ethical approval is not required for registry-based research according to Danish legislation. This study was reported to the Danish Data Protection Agency (53) through registration at Aarhus University (record number: AU-2016-051-000001, sequential number 605).

Descriptive characteristics

There were 37 082 VB-affected pregnancies among 35 514 women; 1 363 614 VB-unaffected pregnancies among 783 314 women; 324 328 terminations among 239 638 women and 137 104 miscarriages among 121 353 women in 1995-2017 (Figure 1). Women’s age, parity, civil status, employment, income, obesity, chronic pulmonary, chronic liver and kidney conditions, alcohol abuse, and medication use had similar distributions for VB-affected and VB-unaffected pregnancy cohorts. Of women with VB-affected vs VB-unaffected pregnancies, 29.0% vs 35.0% had higher education, 13.0% vs 4.9% had a previous record of VB diagnosis irrespective of pregnancy outcome, 6.2% vs 2.1% had a history of a previous VB episode within 20 gestational weeks of a pregnancy ending in a delivery, 22.0% vs 12.0% had at least one previous miscarriage; 19.0% vs 15.0% were smokers (Table 1).

Pregnancies ending in a termination were more likely to be among women who were younger (<24 years), unmarried, nulliparous and unemployed, had a higher prevalence of psychiatric conditions history, a lower level of the highest attained education and were in the first income quartile compared with pregnancies ending in a delivery or miscarriage (Table 1). Pregnancies ending in a miscarriage were more likely than VB-affected pregnancies to be of women older than 35 years and 15.0% had a previous miscarriage (Table 1).

Cumulative incidences and incidence rates of the cancer outcomes

The median (25^th-75^th percentile) follow-up was 12.6 (6.9-18.2) years. At the end of the follow-up, there were 1 320 cancer events following VB-affected pregnancy, 40 420 events following VB-unaffected pregnancy, 10 300 events following a termination and 4 790 events following a miscarriage (Table 2). At the end of the 24-year follow-up, the cumulative incidence of any cancer was 7.4% following VB-affected pregnancy and 7.5% following VB-unaffected pregnancy, 7.3% following termination and 7.9% following a miscarriage (Figure 1 in the electronic supplementary material). The cumulative incidence of groups of site-specific cancers was also similar between the investigated cohorts. Cancers classified as hormone-related were the most prevalent group with cumulative incidences at the end of follow-up of 3.2%, 3.0%, 2.6%, and 3.0% following VB-affected pregnancy, VB-unaffected pregnancy, termination, and miscarriage, respectively.

The incidence rate per 10 000 PYs of any cancer was 25.36 (95% CI: 24.02-26.76) for VB-affected pregnancies and 24.04 (95% CI: (23.81-24.28) for VB-affected pregnancies (Table 2). The incidence rates per 10 000 PYs of hormone-related cancers were 9.83 (9.01-10.71) for VB-unaffected pregnancies, 9.19 (9.05-9.34) for VB-affected pregnancies, 8.20 (7.94-8.48) for terminations, and 10.56 (10.09-11.05) for miscarriages. Similarly, the incidence rates per 10 000 PYs of smoking or alcohol-related cancers and obesity-related cancers were slightly higher following VB-affected than VB-unaffected pregnancy. The incidence rates of premenopausal breast cancer, cervical cancer, ovary and fallopian tube cancer, uterine cancer, and cancers at other sites were similar for all investigated cohorts (Table 2 and Table 2 in the electronic supplementary material).

Hazard ratios for the cancer outcomes

The HR for any cancer in the VB-affected pregnancy cohort was 1.03 (95% CI: 0.97-1.08) vs VB-unaffected pregnancy cohort, 1.03 (95% CI: 0.97-1.09) vs termination cohort, and 0.90 (95% CI: 0.84-0.95) vs miscarriage cohort. In the analyses including the first pregnancy of a woman, the adjusted models yielded nearly identical results as analyses including all identifiable pregnancies of a woman (Table 3).

There was no strong evidence for an association between VB-affected pregnancy and subsequent risk of hormone-related, haematological, immune-related, obesity-related, cancers of neurologic origin, or other cancers across all contrasts (Table 3). When investigating the risk of smoking or alcohol-related cancers following VB-affected vs VB-unaffected pregnancy the HR was 0.98 (95% CI: 0.79-1.22).

However, when following a first identifiable VB-affected vs VB-unaffected pregnancy of a woman, we observed 1.3-fold increased hazard of smoking or alcohol-related cancers (HR: 1.29, 95% CI: 0.78-2.13) (Table 3). For contrasts of VB-affected pregnancy ending in childbirth vs pregnancy ending in termination or miscarriage, the HRs for smoking or alcohol-related cancers were consistent with slightly reduced risk to no association (HR: 0.84, 95% CI: 0.65-1.09 and HR: 0.77, 95% CI: 0.57-1.04, respectively) in analyses of all identifiable pregnancies. The latter associations could not be adjusted for women’s smoking status, since this information was only available for pregnancies ending in a delivery.

As compared with having VB-unaffected pregnancy, there was no evidence of an increased risk of premenopausal breast cancer (HR: 1.01, 95% CI: 0.92-1.11), cervical cancer (0.95, 0.78-1.16), ovary and fallopian tube cancer (1.20, 0.84-1.71), and uterine cancer (0.85, 0.49-1.42) following VB-affected pregnancy (Table 4). Similarly, we found no association with other investigated site-specific cancers across comparisons with VB-unaffected pregnancy cohort and cohorts of pregnancies ending in a termination or miscarriage (Table 3 in the electronic supplementary material).

Sensitivity analysis of pregnancies ending in delivery from 2004 onwards with additional adjustment for woman’s BMI in pregnancy showed results generally consistent with the findings from the main analyses. The HR for smoking or alcohol-related cancers following a first identifiable VB-affected vs VB-unaffected pregnancy of a woman shifted below the null value (HR: 0.50, 95% CI: 0.19-1.33); however, the estimate should be interpreted with caution since it was based on few cancer events and, thus, was imprecise (Table 4 in the electronic supplementary material). HRs for other cancers at specific sites are available in Table 5 in the electronic supplementary material. There was no increased risk of any cancer in analyses of women having their last identifiable pregnancy at the age of 40 years or later (Table 6 in the electronic supplementary material).

Principal findings

In this registry-based cohort study, we found no evidence of an association between VB in pregnancy carried to delivery and subsequent risk of cancer. When compared with having VB-unaffected pregnancy, there was no increased risk of any cancer and specifically breast, cervical, ovary and fallopian tube, and uterine cancer in women having VB-affected pregnancy. Analysis of a first identifiable VB-affected vs VB-unaffected pregnancy of a woman showed a 1.3-fold increased hazard of smoking or alcohol-related cancers; however, this association attenuated towards the null value in a sensitivity analysis. Similarly, we found no evidence of an association with any cancer or site-specific cancers in women when contrasting having VB-affected pregnancy with termination or miscarriage.

Comparison with other studies

This study is in line with a previous Danish study reporting no association between VB in pregnancy irrespective of pregnancy end and occult cancer (54). The results of this study also indirectly support the findings of previous research showing that the risk of breast, ovarian and endometrial cancer was not increased in women with a gestation duration of at least 33 weeks (1,2,27).

Strengths and limitations

Due to nearly complete administrative follow-up for every woman, selection bias at the inclusion of pregnancies in the study is negligible, but potential selection bias from unidentified early miscarriages cannot be ruled out. In the analyses, we controlled for multiple potential confounding factors. Data on proxies of the lifestyle such as smoking and BMI were available only for pregnancies ending in childbirth; moreover, data on BMI was only available for a data subset starting from 2004 and data on in-vitro fertilization or subfertility status were unavailable. The 1.3-fold increased hazard of smoking or alcohol-related cancers in the contrast of a first identifiable VB-affected vs VB-unaffected pregnancy of a woman can potentially be explained by residual confounding by lifestyle factors such as diet, physical activity, and BMI. At the same time, no lifestyle factor of women with VB in pregnancy is likely to dilute the associations and, thus, is unlikely to explain the null associations in analyses contrasting VB-affected pregnancy ending in delivery with having a termination or miscarriage.

Records of VB diagnosis in pregnancy due to threatened miscarriage were not validated in the DNPR. The validity of recording miscarriage diagnoses and procedures for pregnancy termination in the DNPR is high (positive predictive value >90%) (37,55). We could potentially miss recurrent miscarriages within 180 days of each other. However, owing to high specificity of miscarriage diagnoses and few false positives, the misclassification of miscarriages is not a major concern. A diagnostic work-up for threatened miscarriage includes an ultrasound for establishing pregnancy viability, foetal heartbeat, and to eliminate other causes of VB such as ectopic pregnancy (8,56). The prevalence of VB in this study is low (<3%) compared with estimates reported elsewhere (up to 30%) (8–12). We argue that identified VB-affected pregnancies in this study had high specificity, however, a sizeable proportion of VB episodes not resulting in a hospital-based encounter were missed. This may cause bias only if captured episodes of VB systematically differ from all such episodes.

Owing to compulsory reporting of incident primary tumours, the accuracy of cancer diagnoses in the DCR and its coverage is very high, with nearly 90% of tumours having a morphological verification (41). Thus, misclassification of the outcome is unlikely in this study and is not expected to explain the null findings. We did not censor women who would no longer be at risk of endometrial, cervical, and uterine cancers due to hysterectomy (57,58) and the risks of these cancer outcomes may be overestimated.