Abstract
Background
The administration of antineoplastic agents to pregnant women requires comprehensive evaluation of fetal risks and therapeutic benefits for the mother. In Japan, there has traditionally been a tendency to uniformly classify such cases as ‘contraindicated’ when reproductive toxicity data are insufficient. However, a shift toward a more flexible policy has occurred since the approvals of daratumumab and pembrolizumab in 2017.
Objective
This study quantitatively evaluated the changes in regulatory decisions regarding the administration of antineoplastic agents to pregnant women approved by the Ministry of Health, Labour and Welfare (MHLW) following review by the Pharmaceuticals and Medical Devices Agency (PMDA) of Japan since October 2004 to identify influencing factors and compare decision trends with those of the US Food and Drug Administration (FDA).
Methods
A total of 129 drugs approved by the PMDA and FDA were analyzed. The evaluation included classification of decisions (contraindicated, warning, benefit-based administration), annual trends (Joinpoint regression), and background factors (logistic regression).
Results
Since 2018, the proportion of drugs judged as ‘benefit-based administration’ increased to 82.5%. Significant positive associations were observed between benefit-based administration and the following predictors:more recent PMDA approval year, monoclonal antibody drug modality (vs small molecules), absence or presence offetal death, and absence or presence of reversible alterations. Fetal growth retardation showed a negative association with benefit-based administration; ‘unassessable’ classifications tended to act as a barrier to decision making. The concordance rate with the FDA was only 12.4%, suggesting systematic differences in decision-making approaches.
Conclusion
In recent years, the PMDA has adopted a more flexible approach that prioritizes clinical benefit. Importantly, the presence of assessable information—rather than the absence or presence of risk—was identified as a key factor influencing regulatory decisions. However, this descriptive comparison did not restrict the sample to clinically similar drugs. Therefore, while the observed differences likely reflect regulatory tendencies to some extent, they may also be partially influenced by differences in drug background characteristics.
Supplementary Information
The online version contains supplementary material available at 10.1007/s40801-025-00524-8.
Key Points
| The Japanese regulatory authority (PMDA) has increasingly adopted benefit-based approaches in its decisions on anticancer drug use during pregnancy since 2018. |
| Logistic regression identified the presence of assessable data—not the absence of risk—as the primary driver of regulatory flexibility. |
| Only 12.4% of regulatory decisions aligned with those of the US FDA, highlighting differences in decision-making frameworks. |
Introduction
The administration of antineoplastic agents during pregnancy requires extremely careful consideration due to the necessity of ensuring therapeutic efficacy for the mother while minimizing the impact on the fetus. In particular, when data regarding reproductive and developmental toxicity are limited, the conventional practice is to avoid administration by categorizing the agent as ‘contraindicated.’ In Japan, the Guideline on Nonclinical Evaluation of Antineoplastic Drugs does not mandate the implementation of reproductive and developmental toxicity studies (e.g., fertility studies or pre-/post-implantation development studies) for drugs targeted to patients with advanced cancer. Instead, the Guideline allows the substitution of information regarding reproductive organ toxicity obtained from general toxicity studies [1, 2]. Furthermore, ICH S5(R3) presents a flexible framework for evaluations of toxicity that considers the treatment target and disease severity but does not necessarily require that all reproductive and developmental toxicity studies be conducted [3].
Due to the trend of increasing maternal age at childbirth, the number of women diagnosed with cancer during pregnancy is also increasing [4]. In the case of breast cancer among young women, diagnosis often occurs during pregnancy or immediately after childbirth [5]. These trends suggest there is a need to move away from the conventional uniform contraindication policy toward adoption of a more flexible approach that evaluates individual risks and benefits. In this study, ‘benefit-based administration’ refers to a regulatory judgment that allows the use of a drug during pregnancy when the potential therapeutic benefit to the mother is considered to outweigh the possible risks to the fetus, based on explicit language in official documents. Such regulatory decisions have direct clinical implications: they influence the range of therapeutic options that oncologists can consider and shape the content of counseling provided to pregnant patients facing a cancer diagnosis in Japan. In this context, the anticancer drugs daratumumab and pembrolizumab (which were approved in 2017) are noteworthy because although data regarding reproductive toxicity were limited, the Pharmaceuticals and Medical Devices Agency (PMDA) of Japan permitted benefit-based administration to pregnant women [6, 7]. The cases of daratumumab and pembrolizumab are interpreted as suggesting a shift from the conventional ‘principle of contraindication’ to more flexible judgments based on clinical necessity. We reviewed all available package inserts and related regulatory documents for anticancer drugs approved in Japan between October 2004 and December 2024 and found no other cases where benefit-based language was used under similarly limited reproductive safety data. Therefore, daratumumab and pembrolizumab represent the only publicly documented cases to date reflecting this regulatory shift.
In 2015, the US Food and Drug Administration (FDA) introduced the Pregnancy and Lactation Labeling Rule (PLLR), thus transitioning to a format that clearly presents information regarding administration of drugs during pregnancy in descriptive form [8]. Similarly, new labeling standards have been gradually introduced in Japan since August 2017, leading to a more structured format for package inserts. However, the criteria used to make judgments regarding administration during pregnancy in Japan are still not clearly defined, and no quantitative studies have been conducted to assess changes in regulatory decisions or their validity [9]. In this study, therefore, we chronologically analyzed the transitions in decisions regarding the administration of antineoplastic agents authorized by the MHLW following review by the PMDA since October 2004 to pregnant women. In addition, the factors affecting these decisions were analyzed statistically, and the international positioning of Japan’s regulatory decisions were clarified by comparing them with those of the US FDA.
Materials and Methods
This study targeted newly approved antineoplastic agents categorized as new active ingredient drugs authorized by the MHLW following review by the PMDA between October 2004 and March 2025; these drugs had also been approved by the US FDA. Radiopharmaceuticals and drugs indicated for cancers of the male reproductive organs were excluded.
Information regarding regulatory decisions on drug administration during pregnancy was collected from PMDA approval review reports (including both initial and partial change approvals) and package inserts. Although Japan introduced new labeling standards in 2017, many products continued to use the older format; therefore, all available documents were reviewed in full, regardless of format, and relevant descriptions related to pregnant women were extracted. Each year, anticancer drugs approved by the MHLW following PMDA review were classified into one of three categories—'contraindicated,’ ‘warning,’ or ‘benefit-based administration’—based on predefined criteria. ‘Contraindicated’ referred to explicit prohibitions (e.g., ‘contraindicated in pregnant women’); ‘warning’ referred to cautionary language without absolute prohibition; and ‘benefit-based administration’ referred to statements allowing use during pregnancy when the expected maternal benefit outweighs potential fetal risk. All classifications were based solely on explicit language and excluded inferred interpretations.
For comparison, corresponding information was collected from the US FDA using both the initial and most recent versions of each drug’s label. Although the FDA implemented the Pregnancy and Lactation Labeling Rule (PLLR) in 2015 to standardize labeling formats—such as Sections 8.1 (‘Pregnancy’) and 8.3 (‘Females and Males of Reproductive Potential’)—many labels still follow older formats. Therefore, the entire label was reviewed regardless of format. For drugs labeled under the PLLR, explicit statements about use during pregnancy served as the basis for classification. When such language was absent, risk descriptions were interpreted using the same predefined criteria. Coding was performed by one reviewer and cross-checked by another, with discrepancies resolved through discussion.
For each year, drugs authorized by the MHLW following PMDA review were categorized into three groups based on pregnancy-related decisions: ‘contraindicated,’ ‘warning,’ and ‘benefit-based administration,’ and the proportion of drugs in each group was calculated. In this study, ‘contraindicated’ was defined as a statement in the package insert explicitly prohibiting use during pregnancy (e.g., ‘contraindicated in pregnant women’ or equivalent wording); ‘warning’ as a statement advising caution or discouraging use during pregnancy but not explicitly prohibiting it, often accompanied by descriptions of potential risks without an absolute prohibition; and ‘benefit-based administration’ as a statement permitting use during pregnancy when the anticipated maternal benefit outweighs potential fetal risk, as described in the official package insert or review report. These classifications were based solely on explicit text in the package insert and did not rely on inferred interpretations.
In particular, a Joinpoint regression analysis was conducted to characterize the trend in the proportion of drugs categorized as ‘benefit-based administration’ (benefit-based administration rate). The number of benefit-based administrations was selected as the dependent variable, and calendar year (2004–2024) as the independent variable. The annual percent change (APC) and 95% confidence interval (CI) of the Joinpoint (trend change point) analysis were calculated. Model selection was based on the Weighted Bayesian Information Criterion (WBIC) because it allows for a balance between model fit and complexity, with lower WBIC values indicating better performance. The final model—a linear model with no joinpoints—was selected because it had the lowest WBIC value, indicating that additional joinpoints did not significantly improve model fit. Model diagnostics included examination of residual plots and goodness-of-fit statistics to confirm the appropriateness of the chosen model. To compare changes in benefit-based judgments at two time points, a cross-tabulation of judgment distributions (benefit-based administration vs others) was generated for the two periods (prior to 2017 and from 2018 onward), and statistical significance was evaluated using the continuity-adjusted Chi-square test and Fisher’s exact test.
Binary logistic regression analysis was conducted to evaluate the significance of background factors affecting PMDA decisions regarding administration to pregnant women. The dependent variable was binary: benefit-based administration or other (contraindicated or warning). The explanatory variables included approval year (continuous variable), structural classification of the drug (small-molecule compound, monoclonal antibody, antibody-drug conjugate, bispecific antibody, other), and five representative toxicity findings (embryotoxicity, fetal death, fetal growth retardation, teratogenicity, and reversible alterations), as organized based on the descriptions in ICH S5(R3) [3]. Each of the toxicity findings (i.e., embryotoxicity, fetal death, fetal growth retardation, teratogenicity, and reversible alterations) was classified into one of three categories: ‘present,’ ‘absent,’ or ‘unassessable.’ Here, ‘unassessable’ included cases for which reproductive toxicity data were not reported or nonclinical studies were not conducted. The rationale for selecting ‘unassessable’ as the reference category in logistic regression was to explicitly evaluate the impact of information availability on regulatory judgments. This approach was based on the hypothesis that the absence of assessable data itself may act as a barrier to benefit-based administration decisions, independent of whether toxicity is present or absent. Multicollinearity among explanatory variables was assessed using the variance inflation factor (VIF) to assess potential multicollinearity among the independent variables, and variables with a VIF exceeding 10 were excluded from the final model.
To examine how PMDA’s decision trends are influenced by drug characteristics and temporal changes, a logistic regression model incorporating an interaction term between drug modality (small molecules vs monoclonal antibodies) and approval year was constructed. This interaction term was included to test whether the temporal trend in benefit-based administration differed by drug modality. The dependent variable was whether a drug was categorized as ‘benefit-based administration,’ and the independent variables included drug modality, approval year (as a continuous variable), and their interaction (modality × year).
In addition, to investigate whether the presence of previously approved drugs in the same therapeutic area influenced PMDA decisions, a logistic regression analysis was conducted including a variable for the presence of a ‘preceding drug.’ A preceding drug was defined as one with the same drug modality that had previously been authorized by the MHLW following PMDA review for the same therapeutic area since October 2004, for which a pregnancy-related regulatory judgment was available. This variable was included to test the hypothesis that the availability of prior regulatory precedents within the same therapeutic area and modality could lower uncertainty and thereby increase the likelihood of benefit-based administration judgments. The dependent variable was whether the drug was categorized as ‘benefit-based administration,’ and the independent variables included the presence of a preceding drug, approval year, and drug modality.
As a supplementary analysis to provide an overview of the international positioning of PMDA decisions, judgments regarding administration to pregnant women were compared with those of the US FDA. Judgments by both agencies were matched into the same three categories and cross-tabulated, and concordance was evaluated using Cohen’s kappa coefficient and the chi-square test. Judgments were also reclassified into three additional categories (‘PMDA more conservative,’ ‘FDA more conservative,’ and ‘Consistent’), and simple year-by-year aggregation and comparison of distributional changes were conducted. In this classification, a conservativeness hierarchy was defined as follows: ‘contraindicated’ > ‘warning’ > ‘benefit-based administration.’ A decision by one agency was deemed more conservative if it was higher on this scale than the corresponding judgment by the other agency. This comparison was intended as a descriptive assessment of alignment or divergence between the two agencies’ judgments, with a focus on identifying temporal patterns rather than inferring causality. The categorization of regulatory judgments was based directly on textual descriptions in the official package inserts issued by each agency, rather than on inferred policy frameworks or historical classification systems. It should be noted that this comparison was conducted as a descriptive comparison without statistical adjustment for potential confounding factors such as therapeutic indication, approval timing, or drug background characteristics. Therefore, interpretations should take into account differences in regulatory criteria and institutional contexts between the two agencies. The full list of drugs analyzed, including their structural classifications, regulatory judgments (PMDA and FDA), and reproductive toxicity findings, is provided in Supplementary Table 1 (see electronic supplementary material [ESM]). The dataset in Supplementary Table 1 includes all drugs reviewed, their structural classification, and reproductive toxicity findings extracted from official regulatory documents, enabling independent verification of the analysis.
The significance level was set at 5% for all statistical analyses in this study, and results showing p < 0.05 were considered statistically significant.
Results
As indicated in the Methods section, decisions regarding the administration of antineoplastic agents to pregnant women for drugs approved annually by the PMDA were classified into three categories: contraindicated, warning, and benefit-based administration. Between 2005 and 2017, the proportion of drugs categorized as contraindicated was high, whereas that of benefit-based administration was 26.4%. The complete dataset, including the classification of each drug, regulatory judgments (PMDA and FDA), and related reproductive toxicity findings, is presented in Supplementary Table 1 (see ESM). Figure 1 presents the distribution of classification categories across the study period, while Fig. 2 illustrates the annual trend in the proportion of drugs categorized as benefit-based administration. As no approved drugs meeting the study criteria were identified for October–December 2004, calendar year 2004 was excluded from the analysis. Figure 1 shows the annual distribution of pregnancy-related regulatory decisions, and Fig. 2 depicts year-by-year changes in the proportion of drugs classified as benefit-based administration. From 2018 onward, this proportion rose sharply, with several years reaching 100% (2018–2020 and 2022). Across 2018–2024, the pooled proportion was 82.5%, representing a significant increase compared with 2005–2017 (p < 0.001, continuity-adjusted chi-square test). Joinpoint regression analysis identified a single linear model without a joinpoint as optimal (WBIC = 0.279).
Fig. 1.
Annual distribution of PMDA judgments on administration of drugs to pregnant women (2005–2024). This figure shows the year-by-year proportions of 129 anticancer drugs classified by the PMDA into three categories: ‘contraindicated,’ ‘warning,’ and ‘benefit-based administration.’ Notably, the proportion of benefit-based administration judgments increased after 2018, reflecting a potential regulatory trend, although year-to-year variations were also observed
Fig. 2.
Annual proportion of drugs judged as suitable for benefit-based administration (2005–2024). Several post-2018 years reached 100% (2018–2020, 2022). The pooled proportion for 2018–2024 was 82.5%. Proportions were calculated per calendar year; variability reflects small denominators in some years. This figure focuses on the ‘benefit-based administration’ category and was used to quantitatively capture its annual trend as a basis for subsequent trend analysis using Joinpoint regression
Between 2005 and 2024, the benefit-based administration rate increased by 11.3% per year (95% CI 6.8–20.6: p < 0.00001). The regression model on a logarithmic scale had a slope of 0.1069 (p = 0.000798) with a small residual sum of squares (Fig. 3).
Fig. 3.
Joinpoint regression analysis of the benefit-based administration proportion. This figure shows the results of statistical analysis of the Joinpoint regression model analyzing the annual trend in the proportion of antineoplastic agents authorized by the Ministry of Health, Labour and Welfare (MHLW) following review by the Pharmaceuticals and Medical Devices Agency (PMDA) and judged suitable for benefit-based administration between 2005 and 2024. The estimated APC was 11.3%, which quantitatively supports an increasing trend toward benefit-based administration decisions after 2018. The y-axis is presented on a logarithmic scale
Binary logistic regression analysis identified PMDA approval year, structural classification, and three reproductive/developmental toxicity findings (fetal death, fetal growth retardation, reversible alterations) as significant predictors (model chi-square = 55.446, df = 15, p < 0.001; Nagelkerke R2 = 0.466; Hosmer–Lemeshow p = 0.392). Later approval year was associated with greater likelihood of benefit-based classification (odds ratio [OR] 1.228, 95% CI 1.104–1.367; p < 0.001). Monoclonal antibodies were more likely than small molecules to be classified as benefit-based (OR 10.511, 95% CI 2.092–52.825; p = 0.004). For fetal death, the odds ratios were 42.688 (Absent, p = 0.027) and 29.451 (Present, p = 0.045). For fetal growth retardation, the odds ratios were 0.002 (Absent, p = 0.002) and 0.005 (Present, p = 0.003). For reversible alterations, the odds ratios were 26.879 (Absent, p = 0.013) and 38.252 (Present, p = 0.008).
Year of PMDA approval was significantly associated with a greater likelihood of a drug being judged as suitable for benefit-based administration (OR 1.228, 95% CI 1.104–1.367; p < 0.001; Table 1). Monoclonal antibodies were more likely than small molecules to receive this classification (OR 10.511, 95% CI 2.092–52.825; p = 0.004). Significant positive associations were observed for fetal death (p = 0.045), fetal growth retardation (p = 0.003), and reversible alterations (p = 0.008), whereas embryotoxicity and teratogenicity showed no significant associations. Compared with unassessable cases, the odds ratios for fetal death were 42.688 when Absent (p = 0.027) and 29.451 when Present (p = 0.045); for fetal growth retardation, 0.002 when Absent (p = 0.002) and 0.005 when Present (p = 0.003); and for reversible alterations, 26.879 when Absent (p = 0.013) and 38.252 when Present (p = 0.008).
Table 1.
Results of logistic regression analysis of factors associated with PMDA benefit-based administration judgments for pregnant women
| Variable | Odds ratio | 95% CI | p value |
|---|---|---|---|
| PMDA approval year | 1.228 | 1.104–1.367 | <0.001 |
| Structural classification | |||
| Monoclonal antibody | 10.511 | 2.092–52.825 | 0.004 |
| Antibody-drug conjugate | 10.590 | 0.950–118.079 | 0.055 |
| Bispecific antibody | 13.089 | 0.718–238.618 | 0.083 |
| Other | 0.595 | 0.072–4.922 | 0.630 |
| Embryotoxicity | |||
| Absent | 0.937 | 0.126–6.969 | 0.950 |
| Present | 0.311 | 0.044–2.212 | 0.243 |
| Fetal death | |||
| Absent | 42.688 | 1.528–1192.899 | 0.027 |
| Present | 29.451 | 1.074–807.504 | 0.045 |
| Fetal growth retardation | |||
| Absent | 0.002 | 0.000–0.104 | 0.002 |
| Present | 0.005 | 0.000–0.159 | 0.003 |
| Teratogenicity | |||
| Absent | 5.310 | 0.238–118.670 | 0.292 |
| Present | 4.087 | 0.239–69.904 | 0.331 |
| Reversible alterations | |||
| Absent | 26.879 | 1.981–364.671 | 0.013 |
| Present | 38.252 | 2.596–563.558 | 0.008 |
| Constant | 0.000 | <0.001 |
Reference categories for the categorical variables were as follows: for structural classification, ‘small-molecule drug’; for embryotoxicity, fetal death, fetal growth retardation, teratogenicity, and reversible alterations, ‘unassessable’ in all cases
CI confidence interval, PMDA Pharmaceuticals and Medical Devices Agency
A logistic regression model with an interaction term between drug modality and approval year found both factors to be independent predictors (p = 0.007 and p < 0.001, respectively), with no significant interaction (p = 0.264), indicating that the upward trend applied to both modalities (Table 2). The presence of a ‘preceding drug’ approved during the study period (since October 2004) in the same therapeutic area was not statistically significant (OR 0.579, 95% CI 0.219–1.532; p = 0.272), indicating that prior approvals did not substantially influence benefit-based judgments (Table 3).
Table 2.
Logistic regression analysis of benefit-based administration: interaction between approval year and drug modality (small molecules vs monoclonal antibodies)
| Variable | B | SE | Odds ratio (95% CI) | p value |
|---|---|---|---|---|
| Approval year (continuous) | 0.211 | 0.056 | 1.235 (1.107–1.378) | <0.001* |
| Drug modality (monoclonal antibody = 1) | 251.367 | 223.808 | — | 0.261 |
| Approval year × drug modality | –0.125 | 0.112 | 0.883 (0.711–1.098) | 0.264 |
| Constant | –424.197 | 112.961 | — | <0.001 |
This analysis included 108 drugs for which information on structural classification and approval year was available. Only small molecules (n = 82) and monoclonal antibodies (n = 26) were included in the model to evaluate the interaction between approval year and drug modality. Other drug modalities (e.g., antibody-drug conjugates, bispecific antibodies, and others) were excluded. The coefficient for the monoclonal antibody group was unstable due to high correlation with the interaction term and thus its odds ratio is not interpretable
B regression coefficient, CI confidence interval, SE standard error
aModel omnibus test: χ2 = 26.918, df = 3, p < 0.001; Nagelkerke R2 = 0.294; Hosmer–Lemeshow p = 0.387
Table 3.
Logistic regression analysis of benefit-based administration by presence of prior approved drugs, drug modality, and approval year
| Variable | B | SE | Odds ratio (95% CI) | p value |
|---|---|---|---|---|
| Prior approved drug (yes = 1) | –0.546 | 0.498 | 0.579 (0.219–1.532) | 0.272 |
| Approval year | 0.203 | 0.048 | 1.226 (1.118–1.344) | <0.001* |
| Drug modality (reference: small-molecule compound) | 0.076a | |||
| Monoclonal antibody | 1.333 | 0.550 | 3.794 (1.275–11.288) | 0.015* |
| Antibody-drug conjugate | –1.230 | 1.039 | 0.292 (0.039–2.206) | 0.236 |
| Bispecific antibody | –0.292 | 0.954 | 0.747 (0.115–4.865) | 0.760 |
| Others | 0.210 | 0.829 | 1.234 (0.240–6.343) | 0.800 |
| Constant | –410.039 | 97.593 | — | <0.001 |
The dependent variable was benefit-based administration, coded as 1; non-benefit-based administration (e.g., contraindicated or warning) was coded as 0. ‘Prior approved drug’ was coded as 1 if a drug with the same modality had been previously approved within the same therapeutic area (e.g., monoclonal antibodies approved for lung cancer) since October 2004. Approvals in different therapeutic areas (e.g., colorectal cancer) were not considered as prior drugs in this context. Drug modality categories were coded as follows: 1 = small-molecule compound (reference), 2 = monoclonal antibody, 3 = antibody-drug conjugate, 4 = bispecific antibody, 5 = others
B regression coefficient, SE standard error, OR odds ratio, CI confidence interval
*p < 0.05
aWald test for overall significance of drug modality (df = 4, p = 0.076)
In comparison with FDA decisions (Table 4), only 12.4% of drugs showed concordance (Cohen’s κ = 0.026), and in approximately half of cases, the FDA adopted a more conservative stance, particularly after 2018.
Table 4.
Concordance between PMDA and US FDA judgments on the use of antineoplastic drugs during pregnancy and annual trends in judgment tendencies
| Item | Content |
|---|---|
| Total number of drugs | 129 |
| Number of concordant judgments (percentage) | 16 (12.4%) |
| Cohen’s kappa coefficient | 0.026 |
| Number of drugs judged more conservatively by PMDA | 48 (37.2%) |
| Number of drugs judged more conservatively by FDA | 65 (50.4%) |
| Yearly trend in conservative judgments | |
| Before 2017 | PMDA was more conservative (47 drugs) |
| From 2018 onward | FDA was more conservative (47 drugs) |
| Statistical significance of trend change | p < 0.001 (chi-square test) |
This analysis included 126 drugs for which structural classification information was clearly available. A logistic regression model was employed to examine whether a drug was categorized as ‘benefit-based administration,’ with particular focus on the interaction between approval year and drug modality. Conservativeness was defined according to the following hierarchy: ‘contraindicated’ > ‘warning’ > ‘benefit-based administration’
FDA US Food and Drug Administration, PMDA Pharmaceuticals and Medical Devices Agency
Discussion
This study demonstrated that the proportion of drugs the PMDA judged to be suitable for benefit-based administration to pregnant women has consistently increased since 2005. The optimal model in the Joinpoint regression analysis was linear (Fig. 3), suggesting a stable and continuous trend. As the residual sum of squares was small and the trend could be explained without a complex model, the increase was deemed statistically stable.
As shown in Figs. 1 and 2, the proportion has increased markedly since 2018, with quantitative analysis confirming statistical significance. The Joinpoint regression model also demonstrated an annual increase of 11.3%, quantitatively supporting the trend. This indicates that in situations such as advanced cancer in pregnant women, regulatory decisions have shifted from uniform contraindication based on nonclinical data to flexible judgments considering clinical necessity.
Such flexibility is exemplified by daratumumab and pembrolizumab, approved in 2017 [6, 7], which were judged suitable for benefit-based administration despite limited reproductive toxicity data—a turning point toward individualized regulatory approaches. These decisions were likely influenced by the severity and urgency of the indications—multiple myeloma and urothelial carcinoma—where treatment during pregnancy may be justified due to their life-threatening nature and narrow therapeutic windows. Although other life-threatening cancers, such as advanced melanoma, also share these characteristics, the regulatory documents for urothelial carcinoma appear to mark a turning point toward more flexible, benefit-based decisions. Furthermore, the regulatory documents explicitly stated that administration during pregnancy may be considered when the expected maternal benefit outweighs the potential fetal risk. This suggests that risk–benefit balancing, rather than rigid contraindication, was a central factor in these cases.
A similar regulatory approach is observed in certain non-oncologic drugs. For instance, agents such as ramelteon or nifedipine are permitted for use during pregnancy when the expected maternal benefit outweighs the potential fetal risk, despite known fetal toxicity in animal studies. These examples collectively illustrate a broader trend toward individualized, context-sensitive regulatory judgments.
However, this study did not directly examine changes in official regulatory guidelines or internal decision-making protocols. As such, the extent to which this shift reflects policy-level change versus case-by-case clinical assessment remains unclear. Future research analyzing internal review documents or advisory panel statements may help clarify the basis for this apparent regulatory flexibility.
Analysis of background factors revealed that more recent PMDA approval year was associated with higher likelihood of benefit-based administration, particularly for monoclonal antibodies. Their limited placental transferability, consistent pharmacokinetics, and predictable actions may influence safety evaluations and decisions [10, 11].
Among reproductive and developmental toxicity findings, fetal death, fetal growth retardation, and reversible alterations showed statistically significant associations (Table 1). Regulatory authorities may value whether risk is assessable over its magnitude. Even with certain risks present, adequate supporting data can allow for benefit-based determinations. Conversely, absence of information—‘unassessable’—may be the greatest barrier. Here, ‘unassessable’ refers to cases where developmental and reproductive toxicity (DART) data were unavailable or unperformed.
Notably, however, fetal growth retardation showed a negative association with benefit-based administration regardless of assessability. This finding implies that certain toxicity types—such as growth-related outcomes—may be inherently weighted more heavily in regulatory evaluations, possibly due to their long-term and less reversible nature. Thus, while data availability facilitates benefit-based decisions in general, the qualitative nature of the findings still influences regulatory discretion.
Fetal death and reversible alterations showed significant positive associations, while fetal growth retardation was linked to a significantly lower likelihood of benefit-based administration regardless of categorization, suggesting greater caution for this finding.
Supplementary examination of toxicity interactions showed no statistically significant effects (p > 0.05), indicating that PMDA evaluates each finding independently. These tendencies may relate to drug structural characteristics and clinical risks during pregnancy. In early pregnancy (organogenesis), teratogenicity and miscarriage risks are higher; van Gerwen et al. [12] reported 21.7% congenital abnormalities when chemotherapy was administered before 12 weeks (OR 9.24, 95% CI 3.13–27.30). Later trimesters see reduced teratogenicity but ongoing perinatal risks [13, 14].
Although this study is observational and descriptive in nature, regression analyses were employed to explore potential factors associated with regulatory judgments. The intention was not to establish definitive causal relationships, but rather to identify patterns that may inform hypothesis generation. Recognizing the limitations of observational data, these findings are interpreted as suggestive rather than conclusive. Further research involving qualitative assessments or internal decision records may be required to validate these associations.
Comparison of PMDA and US FDA decisions (Table 2) showed only 12.4% concordance, with Cohen’s kappa 0.026, indicating no systematic agreement. This analysis did not adjust for differences in indication, approval timing, or regulatory framework; thus, while suggesting divergent trends, it does not compare the validity of standards.
Since 2018, PMDA has shown increasing flexibility despite limited nonclinical data, while FDA has maintained a conservative stance. Differences in package insert structure and language between the two agencies may influence interpretation and represent a limitation of direct comparison. Conservativeness was determined based on the predefined hierarchy (contraindicated > warning > benefit-based administration) as outlined in the Methods section.
The PLLR introduced by the FDA in 2015 replaced pregnancy risk categories with a narrative format (Sections 8.1–8.3) [8], aiming to improve information transparency, though many labels still rely on animal data [15]. In Japan, a 2017 MHLW notification promoted labeling reform, but standardization of judgment criteria for pregnant women remains limited, and responses vary by drug. The PMDA’s flexible stance overlaps with this reform, but causality is unclear.
This retrospective, descriptive study used publicly available information and did not assess actual clinical administration or outcomes. Decisions in labeling reflect pharmaceutical company applications and PMDA consultations, not necessarily PMDA’s independent judgment. Clinical context was not matched between compared drugs, so differences in labeling may reflect drug background rather than stance. In addition, the classification of drugs into biologics, small molecule compounds, and others may oversimplify pharmacological diversity. In particular, the “small molecule” category encompasses mechanistically diverse agents such as cytotoxic chemotherapies and targeted therapies. We did not further subdivide this group due to the limited number of products, which would have compromised statistical interpretability. This is acknowledged as a limitation, and future studies with larger datasets may allow more granular classification. Future studies should refine comparisons before causal analyses.
Although our analysis included structural classifications such as small molecules and monoclonal antibodies, we did not further subdivide small molecules (e.g., cytotoxic vs targeted agents) due to the limited number of drugs and the descriptive nature of this study. We recognize this as a limitation, and future studies with larger datasets may allow more granular classification. Nevertheless, the current classification still enables identification of broad regulatory trends by contrasting biologics and non-biologics.
To clarify how drugs judged suitable for benefit-based administration are used in practice and affect outcomes, further research using field surveys and real-world data is needed. International decision-making criteria ensuring consistency while allowing flexibility would be desirable. Regulatory authorities should explicitly state decision bases and data limitations. Enhancing transparency in the review process and information disclosure will also be important.
Conclusion
This study quantitatively demonstrated that regulatory decisions by the PMDA regarding the administration of antineoplastic agents to pregnant women have tended to shift toward a more benefit-oriented and flexible approach since 2017. In particular, approval year, drug modality, and the availability of DART information were found to significantly influence decision making. The findings also suggest potential discrepancies between the PMDA and FDA judgments, as well as the critical role of having assessable information in regulatory decisions.
Going forward, it will be essential to enhance the transparency and, where appropriate, the international harmonization of regulatory decisions concerning drug use during pregnancy by evaluating real-world treatment outcomes and safety profiles. However, given that this study was descriptive in nature and did not control for clinical comparability between drugs, any conclusions regarding harmonization should be interpreted with caution. These findings underscore the importance of ensuring that regulatory decisions regarding medication use in pregnant women are grounded in scientific evidence and policy coherence.
Supplementary Information
Below is the link to the electronic supplementary material.
Acknowledgements
The authors thank FORTE Science Communications (https://www.forte-science.co.jp/) for English language editing.
Funding
No funding was received for this work.
Declarations
Conflict of interest
The authors declare that they have no financial or non-financial competing interests, including personal, professional, or institutional relationships, that could be perceived as influencing the work reported in this manuscript.
Ethics approval
Not applicable.
Consent to participate
Not applicable.
Consent for publication
Not applicable.
Availability of data and material
All data used in this study were obtained from publicly available regulatory documents issued by the PMDA and FDA. While summary findings are presented in the manuscript, detailed drug-level data including structural classification, regulatory judgments (PMDA and FDA), and reproductive toxicity findings are provided in Supplementary Table 1 (see ESM), enabling independent verification of the study results. The analysis can be independently verified by consulting the same public regulatory sources.
Code availability
Not applicable.
Authors’ contributions
HK contributed substantially to the conception and design of the study, performed the research, and analyzed the data. KM contributed to the drafting and critical revision of the manuscript, verified the coding results for classification consistency, and also participated in data analysis. Both authors approved the final version of the manuscript and agree to be accountable for all aspects of the work in ensuring its accuracy and integrity.
References
- 1.International Conference on Harmonisation of Technical Requirements for Registration of Pharmaceuticals for Human Use (ICH). Nonclinical Evaluation for Anticancer Pharmaceuticals S9. Tokyo: Pharmaceuticals and Medical Devices Agency (PMDA); 2009. https://www.pmda.go.jp/files/000156321.pdf.
- 2.International Conference on Harmonisation of Technical Requirements for Registration of Pharmaceuticals for Human Use (ICH). Questions and Answers: Nonclinical Evaluation for Anticancer Pharmaceuticals S9 Q&A. Tokyo: Pharmaceuticals and Medical Devices Agency (PMDA); 2018. https://www.pmda.go.jp/files/000231392.pdf.
- 3.International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use (ICH). Detection of Toxicity to Reproduction for Human Pharmaceuticals S5(R3). Tokyo: Pharmaceuticals and Medical Devices Agency (PMDA); 2020. https://www.pmda.go.jp/files/000156671.pdf.
- 4.Boudy AS, Grausz N, Selleret L, et al. Use of tyrosine kinase inhibitors during pregnancy for oncogenic-driven advanced non-small cell lung carcinoma. Lung Cancer. 2021;161:68–75. 10.1016/j.lungcan.2021.09.001. [DOI] [PubMed] [Google Scholar]
- 5.Favero D, Lapuchesky LS, Poggio F, et al. Choosing the appropriate pharmacotherapy for breast cancer during pregnancy: what needs to be considered? Expert Opin Pharmacother. 2023;24(18):1975–84. 10.1080/14656566.2023.2293167. [DOI] [PubMed] [Google Scholar]
- 6.Pharmaceuticals and Medical Devices Agency (PMDA). Darzalex Intravenous Infusion 100 mg, 400 mg (Daratumumab [Genetical Recombination]) Review Report. Tokyo: PMDA; 2017 Aug 30. https://www.pmda.go.jp/files/000234203.pdf
- 7.Pharmaceuticals and Medical Devices Agency (PMDA). Keytruda (pembrolizumab [genetical recombination]) Review Report. Tokyo: PMDA; 2017 Sep 22. Japanese. https://www.pmda.go.jp/files/000226881.pdf.
- 8.US Food and Drug Administration. Content and format of labeling for human prescription drug and biological products; requirements for pregnancy and lactation labeling—final rule. Fed Regist. 2014;79(233):72064–103. [PubMed] [Google Scholar]
- 9.Ministry of Health, Labour and Welfare (MHLW), Pharmaceutical Evaluation Division. Guidance on the format of package inserts for prescription drugs. Tokyo: MHLW; 2020. Japanese. https://www.pmda.go.jp/files/000218446.pdf.
- 10.Barrow P. Review of embryo-fetal developmental toxicity studies performed for pharmaceuticals approved by FDA in 2022 and 2023. Reprod Toxicol. 2024;130:108727. 10.1016/j.reprotox.2024.108727. [DOI] [PubMed] [Google Scholar]
- 11.Barrow P. Review of embryo-fetal developmental toxicity studies performed for pharmaceuticals approved by FDA in 2020 and 2021. Reprod Toxicol. 2022;112:100–8. 10.1016/j.reprotox.2022.06.012. [DOI] [PubMed] [Google Scholar]
- 12.van Gerwen M, Maggen C, Cardonick E, et al. Association of chemotherapy timing in pregnancy with congenital malformation. JAMA Netw Open. 2021;4(6):e2113180. 10.1001/jamanetworkopen.2021.13180. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Maggen C, van Gerwen M, Van Calsteren K, et al. Management of cancer during pregnancy and current evidence of obstetric, neonatal and pediatric outcome: a review article. Int J Gynecol Cancer. 2019;29(2):404–16. 10.1136/ijgc-2018-000061. [DOI] [PubMed] [Google Scholar]
- 14.Triarico S, Rivetti S, Capozza MA, et al. Transplacental passage and fetal effects of antineoplastic treatment during pregnancy. Cancers (Basel). 2022;14(13):3103. 10.3390/cancers14133103. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Byrne JJ, Saucedo AM, Spong CY. Evaluation of drug labels following the 2015 pregnancy and lactation labeling rule. JAMA Netw Open. 2020;3(8):e2015094. 10.1001/jamanetworkopen.2020.15094. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.