Trimester of diagnosis affects tumor characteristics and survival in breast cancer during pregnancy: first results from the STURGATE collaboration

Abstract

Breast cancer diagnosed during pregnancy (PrBC) more often shows unfavorable disease characteristics compared to non-pregnant young breast cancer patients. Previous PrBC research suggests marked differences in tumor characteristics and survival between the gestational trimesters of diagnosis. To investigate this, data of two national retrospective PrBC cohorts from Sweden (1992-2018) and the Netherlands (1988-2022) were harmonized and pooled. PrBC patients were matched for age and year of diagnosis to three non-PrBC patients (controls) per country. Frequencies were pooled and compared using non-parametric tests, while fixed effect meta-analysis with stepwise multivariable Cox models was applied for overall survival. PrBC patients diagnosed in the first trimester resembled the non-PrBC controls. In contrast, second and third trimester diagnoses were associated with significantly more aggressive features, including higher rates of triple negative subtype, higher histologic grade, and more advanced stage at presentation compared to controls. Following adjustment for these features and treatment, overall survival remained significantly worse in PrBC patients compared to non-PrBC controls. This association was pronounced in the second and third trimesters but did not reach statistical significance after full adjustments. Altogether, the persistent worse survival for PrBC and the differences between the trimesters suggest trimester-specific biological influences on tumor behavior and survival.

Introduction

Breast cancer occurs in 1:3,000-10,000 pregnancies, corresponding to 17.5 to 39.9 per 100,000 births^1–3. Pregnancy-associated breast cancer (PABC) commonly includes breast cancer diagnosed during pregnancy (PrBC) or within one or two years after delivery (PPBC)⁴.

Compared to age-matched patients diagnosed not near pregnancy, PABC more often shows unfavorable characteristics, such as advanced stage at diagnosis, high histologic grade, and lack of estrogen (ER) and progesterone (PR) receptor expression^5–11. Although these adverse characteristics contribute to the worse prognosis^6,11–15, an independent effect of the pregnancy itself cannot be excluded. Moreover, the underlying causes of these observed differences in tumor biology remain unknown.

Research suggests that when adjusted for stage and tumor biology, PrBC survival is not compromised compared to non-PrBC controls^{7,11,13,16,17}, although contrasting results have also been reported^5,8,12,14. We have in two independent population-based cohorts recently reported that the clinicopathologic profiles vary markedly per trimester. Notably, we observed more unfavorable tumor characteristics and a worse survival for PrBC patients diagnosed in the second and third trimester of pregnancy, also after multivariable adjustment^18–20. Yet, most studies on PrBC do not report on the trimester of diagnosis, nor do they take this into account in multivariable models.

To further evaluate trimester-specific differences, there is a need for trimester-aware large international studies, especially given the rarity of PrBC. This is necessary as factors essential in maintaining a healthy pregnancy, such as rising levels of pregnancy-hormones and a relative immunotolerant state^21–23, differ per trimester and may also differentially influence tumor growth in PrBC.

Both the Netherlands (NL) and Sweden (SE) represent a setting with similar population structure, publicly funded healthcare systems, and specialized breast cancer care, making it well-suited for pooling of data. For that reason, the Stockholm- Utrecht collaboration on pregnancy-associated breast cancer (STURGATE) was set up. In this first study from the STURGATE collaboration, where we merged these two nationwide population-based PrBC cohorts and age-matched controls, we aimed to determine whether women with PrBC have more aggressive tumor characteristics, and in particular, whether such effects differ by trimester of diagnosis. Furthermore, we compared overall survival in PrBC and controls, whether those associations depend on the trimester of diagnosis, and to what extent the effect is mediated by tumor biology, disease stage, and applied treatments.

Results

Cohort characteristics

The pooled cohort included 534 PrBC-patients (NL: 351, SE: 183) and 1,602 matched controls (Table 1). A first trimester diagnosis was more common in the Dutch compared to the Swedish cohort (24.8% vs. 13.7%). Patients with a third-trimester diagnosis comprised 50.6% of the pooled cohort. The majority of PrBC-cases, and their matched controls, were diagnosed after 2007, and were aged 30-39 years, with a mean age of 33.9 years for PrBC-patients and 34.1 years for controls (Table 1).

Table 1.

Overview of the cohort distribution of the PrBC patients and 1:3 matched non-PrBC controls per country and per trimester of diagnosis

			Pooled		Netherlands		Sweden
			Cases	Controls	Cases	Controls	Cases	Controls
Overall PrBC			534	1602	351	1053	183	549
	Age at diagnosis
		18–29	77 (14%)	240 (15%)	44 (13%)	128 (12%)	33 (18%)	112 (20%)
		30–34	223 (42%)	638 (40%)	151 (43%)	431 (41%)	72 (40%)	207 (38%)
		35–39	183 (34%)	548 (34%)	124 (35%)	378 (36%)	59 (32%)	170 (31%)
		40–49a	51 (10%)	176 (11%)	32 (9%)	116 (11%)	19 (10%)	60 (11%)
		Mean (years)	33.9	34.1	34.0	34.3	33.7	33.8
	P-valueb		0.745		0.745		0.898
	Year of Diagnosis
		1988–1991	6 (1%)	16 (1%)	6 (2%)	16 (2%)	0 (0%)	0 (0%)
		1992–1997	51 (10%)	158 (10%)	29 (8%)	88 (8%)	22 (12%)	70 (13%)
		1998–2002	66 (12%)	187 (12%)	39 (11%)	111 (11%)	27 (15%)	76 (14%)
		2003–2007	98 (18%)	298 (19%)	62 (18%)	189 (18%)	36 (20%)	109 (20%)
		2008–2012	114 (21%)	376 (23%)	77 (22%)	232 (22%)	37 (20%)	144 (26%)
		2013–2018	153 (29%)	428 (27%)	92 (26%)	278 (26%)	61 (34%)	150 (27%)
		2019–2022	46 (9%)	139 (9%)	46 (13%)	139 (13%)	0 (0%)	0 (0%)
	P-valueb		0.952		1.000		0.424
Trimester 1			112 (21%)	N.a.	87 (25%)	N.a.	25 (14%)	N.a.
Trimester 2			152 (28%)	N.a.	93 (26%)	N.a.	59 (32%)	N.a.
Trimester 3			270 (51%)	N.a.	171 (49%)	N.a.	99 (54%)	N.a.

^aAll PrBC cases in Netherlands and Sweden were ≤45.

^bChi-square test.

PrBC breast cancer during pregnancy, N.a. not applicable.

Disease characteristics

PrBC patients had significantly more advanced overall disease stages, in particular lower proportions of stage I disease (16.6% vs. 27.5% in controls) (Fig. 1, Supplementary Table 1). These differences were the biggest if diagnosed in trimester 2 (stage I: 12.9% vs. 24.4%, P = .039), and trimester 3 (stage I: 11.5% vs. 28.3%, P < 0.001). Grade III tumors were significantly more common in PrBC (73.2% vs. 58.5%. Moreover, this proportion was higher in second trimester patients (grade III: 79.5% vs. 57.8%, P < 0.001) and third trimester patients (grade III: 73.6% vs. 59.9%, P < 0.001), while it was similar for first trimester patients (64.9% vs. 56.4%, P = 0.343). Among PrBC, 44.8% had negative hormone receptor (HR) status, versus only 33.4% of controls (P < 0.001), which was more pronounced in trimester 2 (HRneg: 50.0% vs. 31.7%, P < 0.001) and trimester 3 (HRneg: 44.9% vs. 34.4%, P = 0.011). No statistically significant differences were observed for human epidermal growth factor receptor 2 (HER2) status between PrBC patients and controls, both overall (HER2pos: 27.7% vs. 23.6%) and within separate trimesters. In the Swedish cohort, the proportion of HER2-positive tumors was higher in third-trimester patients, and lower in second-trimester patients, although non-significantly (Supplementary Table 2). In a sensitivity analysis restricting the Dutch cohort to most recent years with nearly complete data (when HER2-testing was clinical routine), the HER2-negative proportions were similar between PrBC cases and controls. In contrast, the observed differences in HER2-status by trimester became more pronounced in the Swedish cohort for recent years. A significantly higher proportion of PrBC patients had triple negative breast cancer (TNBC, 34.9% vs. 24.6%), which again was most pronounced for second trimester patients (45.3% vs. 23.1%, P < 0.001) and third trimester patients (34.9% vs. 24.4%, P = 0.001).

Fig. 1. Baseline clinicopathologic disease characteristics of the PrBC and non-PrBC cohorts.

Comparison of various histopathological and clinical characteristics between the PrBC cohort and their respective 1:3 matched non-PrBC controls, subdivided per trimester of diagnosis of the PrBC cases (pooled Netherlands and Sweden). *: P < 0.05 ***: P < 0.001. HR hormone receptor status, HER2 human epidermal growth factor receptor 2, PrBC breast cancer during pregnancy.

Treatment differences per intrinsic subtype

Across subtypes, there were no differences in treatment (chemotherapy, endocrine treatment, HER2-targeted treatment) between PrBC cases and controls. If anything, chemotherapy was more commonly administered in PrBC, which was even significant in TNBC (P = 0.021, Fig. 2). With respect to surgery, we observed more mastectomies in PrBC.

Fig. 2. Treatment per intrinsic subtype for PrBC and non-PrBC cohorts.

Distribution of treatment in the overall PrBC cohort and in the overall matched controls, divided per intrinsic subtype (pooled Netherlands and Sweden). *: P < 0.05. HR hormone receptor status, HER2 human epidermal growth factor receptor 2, PrBC breast cancer during pregnancy, ALND axillary lymph node dissection, SN sentinel node biopsy, BCS breast conserving surgery.

Overall survival

Mean follow-up in the Dutch cohort was 6 years and 10 months (PrBC: 6 years, 4 months, controls: 7 years), and in the Swedish 6 years and 3 months (PrBC: 5 years, 9 months, controls: 6 years, 5 months). In the pooled cohort, 440 deaths were observed (PrBC: 151, controls: 289). The five-year overall survival probability in PrBC was 73.4% (Netherlands) and 76.9% (Sweden) compared to 85.2% (Netherlands) and 86.7% (Sweden) in controls. Correspondingly, the ten-year overall survival probability was 63.7% (Netherlands) and 70.8% (Sweden) in PrBC cases, and for controls 75.9% (Netherlands) and 80.3% (Sweden).

When adjusting for age and year of diagnosis, patients with PrBC had a significantly higher overall mortality compared to controls (model 1 pooled hazard ratio (HR), 1.67 [95% confidence interval (CI): 1.36 to 2.06]), which decreased after further adjusting for histologic grade (model 2 pooled HR, 1.60 [95% CI: 1.29 to 1.98]) and remained similar after adjustment for intrinsic subtype (model 3) (Fig. 3, Supplemental Table 4). In contrast, further adjustment for stage reduced the effect (model 4 pooled HR, 1.31 [95% CI: 1.02 to 1.67]), while adjusting for chemotherapy (model 5) did not change the association. For first trimester PrBC patients, no significant association was observed in any of the models, although the point estimate was somewhat increased in the fully adjusted model 5 (pooled HR, 1.29 [95% CI: 0.71 to 2.33]). Second-trimester PrBC had a significantly higher mortality compared to controls, after adjusting for age and year of diagnosis, histologic grade, and intrinsic subtype (model 3 pooled HR, 1.60 [95% CI: 1.06 to 2.42]). After further adjustment for stage and chemotherapy the point estimate decreased to 1.45, which was non-significant (model 5). Similarly, third trimester PrBC patients had a significantly increased mortality (model 3 pooled HR, 1.92 [95% CI: 1.43 to 2.57]), which also decreased after further adjustments (model 5 pooled HR, 1.24 [95% CI: 0.89 to 1.74]).

Fig. 3. Overall mortality risk per trimester for PrBC versus non-PrBC.

Hazard ratios with 95% confidence intervals comparing overall mortality by PrBC and per trimester to matched non-PrBC controls with stepwise adjustments. Country-specific and pooled estimates. Model 1: Adjusted for matching variables (age and year). Model 2: Adjusted as in Model 1 and grade. Model 3: Adjusted as in Model 2 and intrinsic subtype. Model 4: Adjusted as in Model 3 and TNM stage. Model 5: Adjusted as in Model 4 and chemotherapy. PrBC breast cancer during pregnancy.

Discussion

In this pooled analysis of two population-based cohorts, women diagnosed with PrBC exhibited more aggressive tumor biology and more advanced disease stage compared to age-matched controls. Especially, patients diagnosed in the second and third trimesters had higher-grade tumors, HR negativity, and advanced disease. When evaluating the trimester-specific effects of diagnosis, a significantly higher mortality was observed if breast cancer was diagnosed in the second or third trimesters, which remained significant after adjustment for intrinsic subtype, yet became non-significant after further adjustment for disease stage. Importantly, the point estimates remained increased also in the fully adjusted models suggesting insufficient power. This calls for further pooling of national data to increase numbers in the trimester comparisons and validate our findings.

The underlying etiology for the aggressive disease profile in PrBC is likely multifactorial. Tumor biology and growth patterns may be affected by pregnancy-specific factors which differ between the trimesters, such as a relative immunotolerant state and hormone levels^21–24. Estrogen levels rise rapidly and reach high levels during the second and third trimester²², which could influence tumor growth, both through classic hormone-receptor pathways and through the tumor microenvironment^25,26, thereby also hypothetically influencing ER negative breast cancer growth²⁷. Other growth-stimulating factors and hormones also reach high serum levels during pregnancy, and vascular expansion occurs through angiogenesis and capillary growth^28,29. Lastly, besides aggressive tumor biology, potential detection and diagnostic delays may contribute to the more advanced disease stage in PrBC^30–33. Altogether, these potential interactions of rising levels of hormones and growth factors, combined with an immunotolerant state, may hypothetically influence BC development during pregnancy. This warrants further etiologic research, which may also have implications for (aggressive) breast cancer in general.

Although poorer survival in women with PrBC has been reported, several studies have demonstrated no survival disadvantage when differences in tumor characteristics and stage were accounted for through adjustment or matching^{7,11,13,16,17}. Only a few studies, including our own, have assessed trimester-specific effects and found remaining survival differences also after adjustments for disease characteristics, and in particular for second and third trimesters^18–20,34. This suggests a potential additive effect of the pregnancy itself on survival.

In contrast to our study, the largest PrBC study to date by Amant et al. reported no survival differences after adjusting for tumor and treatment factors¹⁶. This study, which pooled two international cohorts, presented a similar prognosis for 662 pregnant and 2081 controls after adjustment for age, stage, grade, hormone receptor status, HER2 status, and histologic type¹⁶. However, the two pooled cohorts individually reported opposed HRs, and one of them was in line with our findings. Compared to our pooled analysis, Amant et al. did not include stage IV patients or patients not receiving chemotherapy, and follow-up was shorter (median 51 months for PrBC). Importantly, their sample included 14.7% third trimester cases, while our population-based study included 50.6% third trimester cases. Such diverging trimester distributions could explain the contrasting findings from our studies, as the total PrBC survival is a weighted average of the trimester-specific effects, and survival is worse in the second and third trimesters. For this reason, reporting on trimester-specific effects is essential as it increases comparability across studies.

The poorer survival in PrBC in our study was not explained by differences in given treatment, as the treatment patterns were similar when accounting for intrinsic subtype differences. Nevertheless, our register-based treatment information lacked details on treatment delays and specific drug doses, limiting the interpretation of the treatment adequacy within this cohort. However, in a previous Swedish study including PABC cases diagnosed between 1992 and 2009, diagnostic and treatment data extracted from medical records did not support diagnostic or treatment delays, nor treatment intensity³⁵. Speculatively, as chemotherapy is only advised from the second trimester, and HER2-targeted agents and immunotherapy are generally not given during pregnancy, this could potentially lead to poorer survival^36,37. Although treatment options are most limited in the first trimester, the observed better prognosis appears primarily driven by more favorable tumor biology rather than treatment timing.

In addition, for chemotherapy, it is known that physiological alterations during pregnancy result in pharmacokinetic changes such as increased plasma volume and clearance, leading to decreased maximum plasma concentrations^38,39. This raises the question whether we are underdosing chemotherapy in pregnant patients, which may potentially contribute to the worse prognosis in PrBC patients. Key strengths of our study include the large, population-based cohorts with high-quality clinical information from cancer registries in the Netherlands and Sweden.

The well-harmonized matching procedures ensured comparability to relevant control populations in both countries, a feature that many previous studies lack. Importantly, our well-developed study protocol and federated analysis approach allow additional population-based cohorts to join this collaboration, thereby increasing the number of trimesters further. Age and year-matching enabled direct comparisons of baseline characteristics between PrBC and controls.

Our study utilized multiple imputation to increase the analytical samples and decrease selection bias due to missing data. Despite the rich data, we could not assess the mediating effect of several known prognostic indicators, such as Ki67-levels (although grade is a surrogate), precise ER expression levels (only positive/negative), levels of tumor-infiltrating lymphocytes, and presence of lymphovascular invasion, as this information was missing for the vast majority of patients in both cohorts. However, our point estimates were thoroughly adjusted for common patient, tumor, and treatment characteristics. Nevertheless, residual confounding within our broader ER levels or stage levels cannot be excluded.

A further limitation was the lower proportion of first-trimester cases in Sweden, due to the inclusion criteria in the Medical Birth Register (MBR), which reflects that it is prohibited to register pregnancy terminations according to Swedish law. Comparing characteristics between the Swedish and Dutch first-trimester cases, however, did not indicate a major selection bias. In conclusion, women diagnosed with breast cancer in the second and third trimesters exhibit more aggressive disease characteristics, differing substantially from the first trimester cases and non-PrBC controls. The aggressive disease characteristics and potentially suboptimal treatment (underdosing of chemotherapy, delay of targeted therapy) translate into women with PrBC having a worse prognosis. However, an independent effect of pregnancy on survival cannot be ruled out. This highlights the importance of considering trimester distributions in PrBC survival studies, and stresses the need to explore differences in tumor growth promoting mechanisms across trimesters.

Methods

Study populations

In this population-based study, women aged 18–50 years diagnosed with invasive breast cancer (International Classification of Diseases version 10: C50) during pregnancy in the Netherlands from 1988 to 2022 and in Sweden from 1992 to 2018 were included (Table 2).

Table 2.

Overview of the harmonized study protocols in both the Netherlands and Sweden

	Netherlands	Sweden
Study Design	Nationwide Matched Cohort Study	Nationwide Matched Cohort Study
Inclusion Criteria	All women diagnosed with invasive breast cancer (ICD-10: C50) with age- and year of diagnosis limits of: -Ages: 18–50 -Years: 1988–2022	All women diagnosed with invasive breast cancer (ICD-10: C50) with age- and year of diagnosis limits of: -Ages: 18–50 -Years: 1992–2018
Exclusion Criteria	None	None
PrBC Definition	Diagnosis of invasive breast cancer (ICD10: C50) during pregnancy, as ascertained from clinical data from pathology reports.	Diagnosis of invasive breast cancer (ICD10: C50) within nine months before a recorded birth in the Medical Birth Register (MBR).
Data Source	PrBC Cases -Nationwide Pathology Registry (PALGA) -Nationwide Cancer Registry (NCR) -Non-PrBC Controls -Nationwide Cancer Registry (NCR)	PrBC Cases and non-PrBC Controls -Swedish National Cancer Register -Swedish Quality Register for Breast Cancer (NKBC) -Medical Birth Register (MBR), including live and stillbirths with gestational age of ≥22 weeks ( ≥28 weeks before July 2008).
Trimester Definitions	Trimester 1: Weeks 1–13+6 Trimester 2: Weeks 14+0 - 26+6 Trimester 3: ≥27+0	Trimester 1: Weeks 1 - 13+6 Trimester 2: Weeks 14+0 - 26+6 Trimester 3: ≥27+0
Matching to Controls	1:3 matching without replacement on: -Age at Diagnosis -(+/- 1 year) -Year of Diagnosis -( +/- 1 year)	1:3 matching without replacement, except for two controls, on: -Age at Diagnosis -( +/- 1.5 year) -Year of Diagnosis -( +/- 2 year) -Quality Registry -( <2008 and ≥2008)

PrBC: breast cancer during pregnancy; ICD: International Classification of Diseases.

In the Netherlands, cases were identified through the Dutch nationwide network and registry of histo- and cytopathology (PALGA), and then individually linked to the Dutch Cancer Registry (NCR) for data on clinicopathologic characteristics, treatment and survival. Non-pregnant controls were identified through the NCR. For bilateral or multiple primary tumors, the tumor with the highest grade, or (when equal) the largest size, was included.

In Sweden, women were identified in the National Cancer Register, recording all invasive tumors diagnosed in Swedish residents. The first breast cancer diagnosis for each woman was included, and for synchronous bilateral disease within 90 days, we chose the case with the largest tumor size and highest nodal stage. Data on clinicopathological characteristics and treatment were individually linked from the Swedish Quality Registers for Breast Cancer, including the historical registers for years 1992-2007 and the National Quality Register for Breast Cancer for 2008-2018.

In each country, we used a matched cohort design, where both countries implemented a joint study protocol to ensure the design, matching procedures, and variable definitions were harmonized (Table 2).

Identification of PrBC cases

In the Netherlands, pathology reports were manually screened for predefined keywords (e.g., ‘pregnancy’, ‘gestation’, ‘amenorrhea duration’) to identify PrBC cases. In addition, patients with a record of a placenta or endometrial curettage close to their breast cancer diagnosis were screened for PrBC.

In Sweden, information on childbirths was individually linked from MBR, which records all live and stillbirths ≥22 weeks of gestation since 1973 ( ≥28 weeks before July 2008). If the date of delivery from MBR was incomplete, birthdate information from the Swedish Multigeneration Register was used. Pregnancy duration was retrieved based on the known delivery date and gestational age reported at delivery, and for some cases in the Netherlands, from the mentioned gestational age at diagnosis in the pathology report.

In both countries, PrBC was subdivided by trimester 1 (week 1-13⁺⁶), trimester 2 (week 14⁺⁰-26⁺⁶) and trimester 3 (week 27⁺⁰ or later) (Table 2).

Matching of controls

In both countries, controls were women diagnosed with breast cancer not near pregnancy. Controls were randomly selected without replacement and matched 3:1 to PrBC patients. In the Netherlands, the matching criteria were age at diagnosis (+/- 1 year) and year of diagnosis (+/- 1 year). Due to fewer eligible controls in certain age groups in Sweden, the matching criteria were adapted as age (+/- 2 years), year (+/- 1.5 years), and quality register used (1992-2007/2008-2018; due to differing variable content). For two cases with few eligible controls, one control was used twice.

Tumor characteristics and treatments

For ER and PR status, both countries use a threshold of 10% nuclear expression for positivity. Hormone receptor status was defined as positive if either ER and/or PR was considered positive. Furthermore, HER2 status was assessed using the guidelines at the time of testing, meaning 3+ immunohistochemistry scores or an amplified result on in situ hybridization for 2+ cases. Intrinsic subtypes were defined by combining HR status and HER2 status, resulting in four intrinsic subtypes: HR+/HER2+ (triple positive), HR+/HER2- (luminal), HR-/HER2+ (non-luminal/HER2-driven), and HR-/HER2- (triple negative).

Overall disease stage was based on the reported pathological TN stage and clinical M stage for those primary operated, and on the reported clinical TNM for neoadjuvantly treated patients and those not operated.

Surgical type was categorized as mastectomy with either axillary lymph node dissection (ALND) or sentinel lymph node biopsy (SN), breast conserving with ALND or SN, or no surgery/unknown. Systemic treatments were chemotherapy (yes/no), endocrine treatment (yes/no), and HER2-targeted treatment (yes/no). Radiotherapy was categorized as yes/no. Treatment analyses were stratified by intrinsic subtype.

Statistical Methods

The analysis followed a two-stage approach: firstly, country-specific analyses were performed; and secondly, aggregated data and parameters were pooled to derive joint estimates, ensuring that no individual-level data were exchanged between countries.

Frequencies of tumor factors and treatments were compared using Pearson’s χ² test and, if any cell count <5, using Fisher’s Exact test per country separately and for pooled numbers of both countries.

Overall survival was defined as all-cause death within ten years from diagnosis, with survival times defined from the date of diagnosis until the date of death, date of emigration, end of study (Netherlands: January 2023, Sweden: December 2018), or ten years of follow-up, whichever came first. To estimate the association between PrBC status and overall mortality rates, Cox proportional hazards regression models were applied per country. Hazard ratios with 95% CIs were estimated with stepwise adjustment: no covariate adjustment beyond the matching variables (model 1), additional stepwise adjustment for grade (model 2), intrinsic subtype (model 3), stage (model 4), and chemotherapy (model 5). We applied fixed effect meta-analysis to pool the two country-specific estimates into a weighted average pooled HR with 95% CI. In the survival analysis, we performed multiple imputation with chained equations (MICE) for variables with missing values.

Statistical software included Stata version 18/BE for the Swedish analysis, and in the Netherlands, R version 4.3.1 (R Foundation for Statistical Computing, Vienna, Austria), with R packages mice (v.3.17.0), survival (v3.8-3), survminer (v0.5.0), gtsummary (v2.1.0), stats (v4.4.2), and ggplot2 (v3.5.1).

Multiple Imputation

We applied MICE to accommodate and adjust for missing information in important covariates⁴⁰. MICE is a two-step procedure involving i) the imputation step generating imputed datasets from an imputation model, and ii) the analysis step where the analysis model is applied to each imputed dataset and the resulting estimated parameters from each imputed dataset are pooled together to a combined estimate for each parameter of interest using Rubin’s rules.

Table 3 lists the included variables in the imputation model (step 1) and the analytical model (step 2) per country, as well as their percentage (%) missing. Variables with missing data were imputed, using both imputed variables and complete variables in the imputation model. All variables in the analysis model were included in the imputation model. Some variables were auxiliary, meaning they were only included in the imputation model and not in the analysis model.

Table 3.

Overview of variables included (marked X) in the imputation model and analytical model for the Netherlands and Sweden

	The Netherlands			Sweden
	Missing among N = 1404 (proportion missing)	Imputation model (N = 1404)	Analysis model (N = 1404)	Missing among N = 5490b (proportion missing)	Imputation model (N = 5490)	Analysis model (N = 732)
Age at diagnosis	Complete	X	X (conditioning on matched sets)	Complete	X	X (conditioning on matched sets)
Year of diagnosis	Complete	X	X (conditioning on matched sets)	Complete	X	X (conditioning on matched sets)
PrBC trimesters, controls	Complete	X	X	Complete	X	X
Parity prior to index pregnancy	N.a.	N.a.		Complete	X
Country of birth	N.a.	N.a.		Complete	X
T stage	N.a.			410/5490 (7.5%)	X
N stage	N.a.			411/5490 (7.5%)	X
Overall stage	4/1404 (0.3%)	X	X	740/5490 (13.5%)		X
Grade	247/1404 (17.6%)	X	X	2330/5490 (42.4%)	X	X
Histologic type	Complete	X
ER	290/1404 (20.7%)	X		1910/5490 (34.8%)	X
PR	301/1404 (21.4%)	X		1958/5490 (35.7%)	X
HER2	388/1404 (27.6%)	X		1400/5490 (25.5%)	X
Intrinsic subtype a	406/1404 (28.9%)		X	2522/5490 (45.9%)		X
Chemotherapy	Complete	X	X	1561/5490 (28.5%)	X	X
Surgery type (SE) Surgery yes/no (NL)	Complete	X		599/5490 (10.9%)	X
Radiotherapy	N.a.			2095/5490 (28.2%)	X
Endocrine therapy	Complete	X		1953/5490 (35.6%)	X
Death indicator any cause, Survival time / Cumulative Hazard (Nelson Aalen) any cause	Complete	X	X	Complete	X	X

^aIntrinsic subtype based on ER, PR, and HER2.

^bIn Sweden, 732 PrBC and controls, and 4758 post-partum cases and controls.

PrBC breast cancer during pregnancy, ER estrogen receptor, PR progesterone receptor, HER2 human epidermal growth factor receptor 2, SE Sweden, NL The Netherlands, N.a. not applicable.

The outcome was included in the imputation model through the event indicator for overall survival (OS) and the Nelson-Aalen estimator for OS, as suggested by Falcaro et al.⁴¹. In both countries, patients who received HER2-targeted therapy were deterministically predicted as HER2-positive, if HER2 status was unknown. In Sweden, patients with missing values for M stage were deterministically set to M0, as this follows coding practices in the Swedish cancer registers.

In the first imputation step, we set up an imputation model in both countries separately. For the Netherlands, we used the mice package in R (v.3.17.0) to generate a total of m = 30 imputed datasets, with 20 iterations (seed 500). The number of imputed datasets m was chosen to be larger than the highest percentage of missing values among the variables included in the imputation model (HER2 = 27.6% missing), according to the rule of thumb by White and colleagues⁴⁰. In the chained equations, logistic regression models were applied to binary outcome variables (ER, PR, and HER2), while multinomial regression was used for categorical outcome variables (grade and disease stage). The imputation model included the variables listed in Table 3. For Sweden, we used the Stata mi package, including the command mi impute (Stata/BE 17.0). A total of m = 61 imputed datasets, with 50 iterations, were generated. Also here, the number of imputed datasets m was chosen based on the highest percentage of missing values among the variables in the model (grade = 42.4%) before the deterministic prediction. In the chained equations, logistic regression models were applied to binary outcome variables (ER, PR, HER2, chemotherapy, radiotherapy, endocrine therapy), while multinomial regression was used for categorical outcome variables (grade, tumor size T, nodal involvement N, surgery type). The imputation model included the variables listed in Table 3. To increase power and stability in the imputation model step in Sweden, where the available data were limited in the pregnant cohort, the imputation analysis was performed on a cohort with pregnant and postpartum cases (within 2 years of delivery), and their respective controls, similar to before¹⁸. The analysis models were performed only on the pregnant cohort. This way, information on (missing) values for imputed variables were shared across pregnant and postpartum cases and their respective controls.

In the second step, we continued with the analysis model. To estimate pooled hazard ratios for OS, we fitted Cox models and applied Rubin’s rules to the imputed datasets⁴⁰. We combined the ER, PR, and HER2 into an intrinsic subtype variable, which was then included in the analysis model. The analysis model included the variables listed in Table 3.

Ethical approval

The Dutch study was approved by scientific and privacy committees of both PALGA and IKNL and the local ethical research committee (No. 19/366), whereas the Swedish study was approved by the Swedish Ethical Review Authority (DNR: 2022-02992-02) and the Ethical Review Board in Stockholm (DNR: 2010-1950-31/4; 2011-599-32, 2018/1293-32). All research was conducted in accordance with the Declaration of Helsinki.

Author contributions

Conception and design: C.B., Cv.D., A.J., L.G., I.F. Collection and assembly of data: C.B., L.G., A.J. Data analysis and interpretation: C.B., Cv.D., Evd.W., Pv.D., A.J., L.G., I.F. Manuscript writing: C.B., Cv.D., A.J., I.F. Final approval of manuscript: All authors.

Data availability

All relevant aggregated data supporting the findings of this study are available within this paper and can be found in either the tables or the supplemental materials. The individual patient data that support the findings in this study are not openly available due to privacy restrictions of the respective registries in the Netherlands and Sweden. Aggregated data is available upon reasonable request from the corresponding author.

Competing interests

PvD is a member of the advisory boards of Visiopharm, Paige, and Sectra. CvD received institutional research funding from Visiopharm, Paige, Ellogon.AI, and Gilead Sciences, and (speaker) honoraria from Benecke, Congress Care and IQVIA. CB has received institutional research funding from Gilead Sciences and (speaker) honoraria from Congress Care and Congress & Meeting Services Holland. IF has received institutional funding from MSD as part of a Master Collaboration Agreement between MSD and Karolinska Institutet. All other authors report no conflict of interest.

Supplementary information

The online version contains supplementary material available at 10.1038/s41523-026-00918-z.