A New Tool for Pregnancy Research: A Unified Definition for Major Congenital Malformation across ICD Eras

Abstract

Composite major congenital malformation (MCM) outcomes are commonly used to assess teratogenic effects of prenatal medication exposure, but this approach dilutes effect estimates when the risk is confined to a specific MCM. Tree-based scan statistics address this by screening outcomes using a hierarchical tree, enabling detection of specific risks without predefined hypotheses. To apply this method across ICD-9-CM and ICD-10-CM eras, we developed a unified hierarchical outcomes tree for MCM. We selected ICD-9-CM and ICD-10-CM codes classified as congenital anomalies, removing minor malformations, chromosomal anomalies, and single-gene conditions. A multi-level tree was built based on the Multilevel Clinical Classification Software, General Equivalence Mappings, and expert review. We validated the tree using birth cohorts from MarketScan and Medicaid databases (2011–2013; 2016–2018), assessing balance of MCM prevalences within one year of birth via standardized mean differences (SMDs). The final tree included 1,023 codes, organized into 244 clinical MCM groups at the most granular level. We identified 572,107 (2011–2013) and 360,167 infants (2016–2018) in MarketScan and 362,820 and 3,500,589 infants in Medicaid. All SMDs were below 0.1, indicating consistency across coding eras. This hierarchical MCM tree bridges ICD-9-CM and ICD-10-CM, enabling consistent outcome definitions and enhancing detection of specific teratogenic risks.

Plain language summary

Researchers often group major birth defects together in studies to increase the number of cases, but this can hide risks linked to specific defects. Tree-based scan statistics help by looking at outcomes in a detailed, structured way without needing to guess which defects might be affected. To use this method across multiple years, we created and tested a single, unified definition for major birth defects. This structure makes it easier to evaluate birth defects across coding systems and improves the ability to find risks linked to specific defects. Overall, this approach helps detect important risks that might be missed when using broad groupings of birth defects.

Introduction

Almost 98% of medications have undetermined teratogenicity risk at the time of approval,¹ with evidence relying largely on animal data as pregnant women are commonly excluded from clinical trials.² As a result, post-marketing observational studies have become key sources for evaluating medication safety during pregnancy.^1,3 However, despite significant advances in pregnancy research, it can still take decades after a medication’s approval until a reasonably accurate and precise classification of teratogenic risk is available.¹

Research on medications during pregnancy faces multiple challenges including low exposure prevalence, difficulty in accurately measuring teratogenic outcomes, and the rarity of such outcomes.³ Major congenital malformations (MCM) are among the key outcomes used to assess teratogenicity risk due to their surgical, medical, or serious cosmetic importance.⁴ There are nine main classifications of MCM (central nervous system, eye, ear, cardiovascular, orofacial, gastrointestinal, genitourinary, musculoskeletal, and chromosomal defects), which are commonly grouped into 30-50 categories,^5,6 as aggregate occurring in about 3-4% of all deliveries.⁷ To address small case numbers of specific MCMs, pregnancy studies often use composites,^3,8,9 which are routinely dominated by cardiac malformations due to their higher prevalence. Use of composite outcomes, many of which might be unrelated to the targeted drug effect, might introduce noise in outcome ascertainment, mask a true excess risk of a specific MCM, and lead to false negative results. In absence of animal studies or other prior evidence that might guide the specific selection of MCMs for evaluation, researchers are therefore facing a choice between maximizing power with larger MCM aggregates or greater outcome specificity and detectable relative differences in frequency.

An alternative solution is to employ tree-based scan statistics, a data mining method that uses a hierarchical tree structure to screen thousands of outcomes at varying levels of aggregation while adjusting for multiplicity.^10,11 This approach is particularly valuable in medication safety assessments where no prior information on types of specific adverse outcomes is available and has been routinely used by the U.S. Food and Drug Administration (FDA) and Centers for Disease Control and Prevention (CDC) for post-marketing surveillance activities.^12-18 Because tree-based scan statistics assess outcomes frequencies at various levels of aggregation, tree-based scan statistics optimize detection of unknown safety outcomes, simultaneously maximizing case numbers and specificity.

Implementation of tree-based scan statistics requires organizing thousands of outcomes into a hierarchical tree. Top levels (closer to the “root” in tree parlance) represent broad categories (e.g., congenital anomalies), which are progressively divided into more specific concepts at lower levels until the “leaf level” that contains the most detailed outcomes, such as individual diagnoses (e.g., anencephaly). Previous studies have used Multi-Level Clinical Classification Software (MLCCS) from the Agency for Healthcare Research and Quality to build five-level trees based on International Classification of Disease, Ninth Revision, Clinical Modification (ICD-9-CM) codes.^12,19-21 However, MLCCS is available for ICD-9 coding but not its successor, ICD-10-CM.¹⁹

In October 2015, the U.S. transitioned from the ICD-9-CM coding system to ICD-10-CM for diagnosis and procedure codes used in clinical settings, expanding from over 14,000 diagnosis codes to over 70,000 codes. This shift introduced challenges in translating clinical concepts across eras. A crosswalk between ICD-9-CM and ICD-10-CM coding systems exists but requires careful review to ensure consistency in clinical meaning of individual codes and in aligning diagnosis groups representing the same clinical construct.

When applying tree-based scan statistics across both eras, it is essential not only to maintain consistency in the meaning of individual clinical diagnostic concepts but also to preserve their hierarchical organization so risks can be detected for specific MCMs or groupings. A unified structure enables continuous medication safety monitoring and thereby increases sample size and statistical power across ICD eras.

Here, we report development of a unified hierarchical structure (an outcome tree) for MCMs across ICD-9-CM and ICD-10-CM eras as a new tool to accommodate ongoing medication safety studies in pregnancy.

Methods

Major congenital malformation outcome tree development

To obtain all relevant MCMs, we selected 444 ICD-9 codes classified under “Congenital anomalies” at level 1 in the 2015 MLCCS¹⁹ and 839 ICD-10 codes under the same category in the 2018 Beta Clinical Classifications Software (Beta-CCS).²² Since these lists exclude header codes which, despite being non-billable, are used frequently on medical claims, we manually added relevant headers and included new codes introduced after 2018 (as of May 1, 2025). We then mapped all ICD-9 codes to ICD-10 codes and vice versa using the 2018 general equivalence mappings (GEMS),²³ merged the bidirectional mappings and removed duplicated pairs. To ensure clinical relevance of outcomes and focus on possible iatrogenic etiologies, we excluded codes related to chromosomal anomalies or single-gene conditions (Table S1), exogenous causes unrelated to medication exposure, or codes referring to personal history. These exclusions were reviewed and curated by a clinical geneticist (SAR).

To facilitate clinical review, we grouped the ICD code mappings into five categories based on the complexity of the matches. Category 1 included simple one-to-one matches, where a single ICD-9 code mapped to a single ICD-10 code and vice versa. Category 2 included one-to-many matches, where one ICD-9 code mapped to several ICD-10 codes, but those ICD-10 codes did not map to any other ICD-9 codes. Category 3 was the reverse; one ICD-10 code mapped to several ICD-9 codes, with no overlap to other ICD-10 codes. Category 4 captured more complex, chained mappings where codes looped back and forth between ICD-9 and ICD-10, creating interconnected chains (Figure 1). Finally, category 5 included codes that could not be mapped, often because they were general or header codes. We addressed these by manually adding each header code to an existing clinical concept that includes their sub-codes (e.g., adding ICD-10 header code Q03 with its sub-codes Q03.0, Q03.1, Q03.8, Q03.9).

Figure 1. Examples of the complex mappings between ICD-9 and ICD-10 codes mapping category 4.

Our outcome tree has a total of five hierarchical levels (levels 1-5) which were developed based on the MLCCS structure. At the broadest concept level (level 1), MLCCS organizes congenital malformations into five main groups: “ Cardiac and circulatory congenital anomalies”, “Digestive congenital anomalies”, “Genitourinary congenital anomalies”, “Nervous system congenital anomalies”, and “Other congenital anomalies”. To make the “other” group more specific, we sub-divided it into five smaller groups including facial anomalies, respiratory system anomalies, musculoskeletal anomalies, skin-related anomalies, and a remaining “other” group. Levels 2 to 4 were based on the MLCCS hierarchy of the mapped ICD-9 codes. Table 1 gives examples of how codes were assigned using these grouping strategies and structured into hierarchical levels.

Table 1.

Examples of the hierarchical tree structure for both ICD-9 and ICD-10 codes

	Level	Level code	Level description	Level 5 code ICD era	ICD code	ICD description
Example of mapping category 1 (1:1 map)	1	14	Congenital anomalies
	2	14.04	Nervous system congenital anomalies
	3	14.04.05	Other nervous system congenital anomalies
	4	14.04.05.00	Other nervous system congenital anomalies
	5	14.04.05.00.01	Anencephaly	ICD-9	740	Anencephalous
				ICD-10	Q00.0	Anencephaly
Example of mapping category 2 (1:n map)	1	14	Congenital anomalies
	2	14.04	Nervous system congenital anomalies
	3	14.04.01	Spina bifida
	4	14.04.01.00	Spina bifida
	5	14.04.01.00.06	Spina bifida lumbar region with hydrocephalus	ICD-9	741.03	Spina bifida lumbar region with hydrocephalus
				ICD-10	Q05.2	Lumbar spina bifida with hydrocephalus
				ICD-10	Q05.3	Sacral spina bifida with hydrocephalus
Example of mapping category 3 (n:1 map)	1	14	Congenital anomalies
	2	14.05	Facial anomalies
	3	14.05.01	Congenital anomalies of the eye
	4	14.05.01.02	Microphthalmos
	5	14.05.01.02.00	Microphthalmos	ICD-9	743.06	Cryptophthalmos
				ICD-9	743.1	Microphthalmos unspecified
				ICD-9	743.11	Simple microphthalmos
				ICD-9	743.12	Microphthalmos associated with other anomalies of eye and adnexa
				ICD-10	Q11.2	Microphthalmos
Example of mapping category 4 (chain map)	1	14	Congenital anomalies
	2	14.04	Nervous system congenital anomalies
	3	14.04.01	Spina bifida
	4	14.04.01.00	Spina bifida
	5	14.04.01.00.07	Spina bifida unspecified region with/without hydrocephalus	ICD-9	741.00	Spina bifida unspecified region with hydrocephalus
				ICD-9	741.90	Spina bifida unspecified region without hydrocephalus
				ICD-10	Q05.4	Unspecified spina bifida with hydrocephalus
				ICD-10	Q07.02	Arnold-Chiari syndrome with hydrocephalus
				ICD-10	Q07.03	Arnold-Chiari syndrome with spina bifida and hydrocephalus
				ICD-10	Q05.9	Spina bifida, unspecified
Example of mapping category 5 (rescuing unmapped codes)	1	14	Congenital anomalies
	2	14.04	Nervous system congenital anomalies
	3	14.04.05	Other nervous system congenital anomalies
	4	14.04.05.00	Other nervous system congenital anomalies
	5	14.04.05.00.11	Congenital hydrocephalus	ICD-9	742.3	Congenital hydrocephalus
				ICD-10	Q03	Congenital hydrocephalus
				ICD-10	Q030	Malformations of aqueduct of Sylvius
				ICD-10	Q031	Atresia of foramina of Magendie and Luschka
				ICD-10	Q03.8	Other congenital hydrocephalus
				ICD-10	Q03.9	Congenital hydrocephalus, unspecified

After establishing the initial tree structure, we removed all minor malformations to support parsimonious models that are confined to clinically highly relevant outcomes. We defined major malformations as defects that require surgery or medical treatment, or those that have serious cosmetic impact,⁴ consistent with the focus of epidemiologic studies by the Metropolitan Atlanta Congenital Defects Program, the National Birth Defects Prevention Study, and the National Birth Defects Prevention Network.^4,6,7,24 If both ICD-9 and ICD-10 codes in a mapped pair indicated minor anomalies, the pair was removed. In cases of discrepancies, decisions were based on the more specific code. For example, if a nonspecific ICD-9 code was mapped to both minor and major ICD-10 codes, only the minor ICD-10 codes were excluded, preserving the mapped pair(s) between the unspecific ICD-9 code and the major malformation ICD-10 code(s). All decisions about code removal and reconciliation of complex or inconsistent mapping were made by a clinical geneticist (SAR). The final hierarchical tree is captured in the Table S2.

Tree validation - Evaluation of major congenital malformation prevalence across ICD-9 and ICD-10 eras

To evaluate the mapping between the ICD-9 and ICD-10 eras, we compared the prevalence of MCMs at various aggregate group levels using birth cohorts from Merative^™ MarketScan^® Research Databases and U.S. Medicaid databases. The cohorts included infants born in 2011-2013 (ICD-9 era) and 2016-2018 (ICD-10 era) with continuous medical and prescription coverage within 45 days of birth through 1 year. MCMs were identified if an infant had at least one inpatient diagnosis or diagnoses on two separate days in outpatient settings within the first year of life.

MCM groupings with absolute standardized mean differences (SMDs) > 0.1 were manually reviewed to determine whether differences were due to removal of minor malformations or changes in code specificity across ICD eras. To comply with small cell reporting policies and compare prevalences with sufficient precision, we only reported nodes with at least 30 events across both eras and no cell count smaller than 11 per era per database.²⁵

This study was approved by the Institutional Review and Privacy Boards of the University of Florida, the Centers for Medicare and Medicaid Services, and the Harvard Pilgrim Health Care Institute.

Results

The final tree included 1,023 codes (669 from the ICD-10 era), organized into 244 distinct clinical MCM groups at the most granular level, 96 aggregated groups at level 4 and 45 aggregate groupings at level 3 (Figure 2).

Figure 2. Attrition flow chart describing the tree development process.

To validate the tree, we identified 572,107 (2011-2013) and 360,167 infants (2016-2018) in the MarketScan database, and 362,820 and 3,500,589 infants in the Medicaid database for the same period. The substantial difference in the number of eligible infants in the Medicaid database between the two eras was primarily due to more states failing to meet data quality metrics²⁶ - either for Fee-for-Service or Comprehensive Managed Care - during the ICD-9 era.

Two out of 45 aggregate groups at level 3 with fewer than 30 total events (“All other congenital anomalies” and “Gastrointestinal vessel congenital anomalies”) were excluded from analysis. The most common level 3 groups with on average at least 7 events per 1000 infants in MarketScan and Medicaid were atrial septal defect (24 vs. 28), ventricular septal defect (8 vs. 8), congenital anomalies of urinary system (9 vs. 8), other congenital musculoskeletal anomalies (9 vs. 7) and congenital anomalies of respiratory system (7 vs. 8), respectively (Table 2). All level 3 groups had SMD ≤ 0.05 when comparing prevalences from the two ICD eras in both databases, indicating good balance (Table 2). We observed consistent results when the prevalence balance was evaluated at the most granular level 5 with adequate balance across all 244 MCM groups (Table S3).

Table 2.

Comparison of prevalences of major congenital malformations at aggregated levels by the ICD-9 and ICD-10 eras and databases

Level 3 node*	MarketScan Database			Medicaid Database
	Events per 1000 infants: ICD-9 (2011- 2013 birth cohort)	Events per 1000 infants: ICD-10 (2016-2018 birth cohort)	Absolute standardized mean difference (aSMD)	Events per 1000 infants: ICD-9 (2011-2013 birth cohort)	Events per 1000 infants: ICD-10 (2016-2018 birth cohort)	Absolute standardized mean difference (aSMD)
Total eligible infants	572,107	360,167		362,820	3,500,589
Atrial septal defect	12181 (21.29)	10591 (29.41)	0.052	8852 (24.40)	101136 (28.89)	0.028
Ventricular septal defect	4578 (8.00)	3306 (9.18)	0.013	2938 (8.10)	29622 (8.46)	0.004
Congenital anomalies of urinary system	4298 (7.51)	3666 (10.18)	0.028	2406 (6.63)	26870 (7.68)	0.012
Other congenital musculoskeletal anomalies	5527 (9.66)	3251 (9.03)	0.007	3227 (8.89)	24830 (7.09)	0.020
Congenital anomalies of respiratory system	3518 (6.15)	3129 (8.69)	0.030	2410 (6.64)	26581 (7.59)	0.011
Congenital anomalies of genital organs	4903 (8.57)	3657 (10.15)	0.016	2223 (6.13)	22712 (6.49)	0.005
Certain congenital musculoskeletal deformities	3729 (6.52)	3181 (8.83)	0.027	1640 (4.52)	15897 (4.54)	0.000
Congenital anomalies of limbs	2887 (5.05)	1614 (4.48)	0.008	2730 (7.52)	15694 (4.48)	0.039
Pulmonary artery anomalies	1342 (2.35)	1561 (4.33)	0.034	1211 (3.34)	15500 (4.43)	0.018
Congenital anomalies of ear face and neck	1478 (2.58)	1595 (4.43)	0.031	1042 (2.87)	12284 (3.51)	0.011
Other nervous system congenital anomalies	1628 (2.85)	1235 (3.43)	0.010	1252 (3.45)	11860 (3.39)	0.001
Persistent fetal circulation	1336 (2.34)	1077 (2.99)	0.013	962 (2.65)	11306 (3.23)	0.011
Other congenital anomalies	1637 (2.86)	1019 (2.83)	0.001	991 (2.73)	7620 (2.18)	0.011
Microcephalus	429 (0.75)	549 (1.52)	0.023	609 (1.68)	8877 (2.54)	0.019
Pulmonary valve atresia and stenosis	1252 (2.19)	735 (2.04)	0.003	898 (2.48)	6821 (1.95)	0.011
Pyloric stenosis	1019 (1.78)	476 (1.32)	0.012	932 (2.57)	6819 (1.95)	0.013
Cleft palate and/or cleft lip	928 (1.62)	559 (1.55)	0.002	632 (1.74)	6121 (1.75)	0.000
Congenital anomalies of the eye	968 (1.69)	763 (2.12)	0.01	558 (1.54)	5157 (1.47)	0.002
Other circulatory congenital anomalies	1063 (1.86)	782 (2.17)	0.007	697 (1.92)	4666 (1.33)	0.015
Other heart valve congenital anomalies	258 (0.45)	583 (1.62)	0.036	110 (0.30)	6040 (1.73)	0.045
Other lower gastrointestinal congenital anomalies	632 (1.10)	398 (1.11)	0.000	399 (1.10)	4199 (1.20)	0.003
Other congenital anomalies of aorta	476 (0.83)	352 (0.98)	0.005	234 (0.64)	3940 (1.13)	0.016
Coarctation of aorta	530 (0.93)	335 (0.93)	0.000	304 (0.84)	3121 (0.89)	0.002
Congenital anomalies of the integument	141 (0.25)	332 (0.92)	0.028	158 (0.44)	3101 (0.89)	0.018
Endocardial cushion defects	443 (0.77)	279 (0.77)	0.000	228 (0.63)	2083 (0.60)	0.001
Rectal and large intestine atresia/stenosis	362 (0.63)	198 (0.55)	0.003	243 (0.67)	2097 (0.60)	0.003
Other congenital anomalies of the heart	323 (0.56)	244 (0.68)	0.005	201 (0.55)	2118 (0.61)	0.002
Other spinal cord congenital anomalies	364 (0.64)	284 (0.79)	0.006	154 (0.42)	2044 (0.58)	0.007
Tetralogy of Fallot	352 (0.62)	239 (0.66)	0.002	226 (0.62)	1955 (0.56)	0.003
Atresia and stenosis of small intestine	343 (0.60)	165 (0.46)	0.006	244 (0.67)	2003 (0.57)	0.004
Other upper gastrointestinal congenital anomalies	229 (0.40)	282 (0.78)	0.016	145 (0.40)	1623 (0.46)	0.003
Anomalies of intestinal fixation	290 (0.51)	159 (0.44)	0.003	180 (0.50)	1460 (0.42)	0.004
Other anomalies of bulbus cordis and cardiac septal closure	360 (0.63)	192 (0.53)	0.004	171 (0.47)	1303 (0.37)	0.005
Spina bifida	224 (0.39)	149 (0.41)	0.001	150 (0.41)	1497 (0.43)	0.001
Hirshsprungs disease	235 (0.41)	121 (0.34)	0.004	137 (0.38)	1250 (0.36)	0.001
Hypoplastic left heart syndrome	211 (0.37)	119 (0.33)	0.002	126 (0.35)	1260 (0.36)	0.001
Transposition of great vessels	219 (0.38)	112 (0.31)	0.004	102 (0.28)	1223 (0.35)	0.004
Double outlet right ventricle	175 (0.31)	94 (0.26)	0.003	119 (0.33)	1069 (0.31)	0.001
Esophageal atresia/tracheoesophageal fistula	195 (0.34)	112 (0.31)	0.002	79 (0.22)	929 (0.27)	0.003
Aortic valve stenosis	159 (0.28)	154 (0.43)	0.008	92 (0.25)	893 (0.26)	0.000
Common ventricle	142 (0.25)	79 (0.22)	0.002	84 (0.23)	929 (0.27)	0.002
Anomalies of adrenal gland	70 (0.12)	33 (0.09)	0.003	50 (0.14)	617 (0.18)	0.003
Cerebrovascular anomalies	102 (0.18)	37 (0.10)	0.006	29 (0.08)	234 (0.07)	0.002

^*Level 3 nodes are sorted in descending order based on the total number of events across ICD-9/ICD-10 codes and databases.

Discussion

We created a hierarchical MCM tree that can be used to screen for safety signals of medications, biologics and other exposures across ICD eras, using tree-based scan statistics. The tree captures 244 distinct MCMs at the most granular level and offers several higher-level groupings of MCMs that might share similar etiology and hence might present appropriate composites that can enhance statistical power when evaluating rare events. MCMs were curated to include only major anomalies to focus on the most clinically relevant outcomes and optimize statistical power when employing tree-based scan statistics across all leaves and branches. We removed genetic syndromes (chromosomal anomalies and single-gene conditions) to remove noise in inferential analyses because these conditions are present at conception and thus could not possibly be due to exposures to medications or biologics during pregnancy. The trend analyses showed that prevalence estimates at all grouping levels were comparable across ICD cohorts in both MarketScan and Medicaid databases with all SMDs below 0.1, reducing risk of measurement biases when evaluating exposures over longer study periods. While optimized to use as a tree, investigators may also want to select one or several MCM groups when evaluating hypotheses about specific teratogenic pathways, which is feasible given validated crosswalks between ICD eras on all group levels.

Due to the rarity of specific MCMs and the uncertainty around which specific MCM to target for evaluation when teratogenic pathways are unclear, investigators commonly use composite outcomes to increase proportion case numbers and statistical power.³ However, counts of cases are driven by more frequent MCMs, e.g., septal defects among cardiac anomalies or cardiac anomalies when evaluating all MCMs.^6,27 Thus, if the teratogenic effect lies within rarer MCMs, elevated risks might be masked when evaluating large composites.^3,28 Using a tree structure with multiplicity adjustment is a potential solution, maximizing power by testing MCM groups at various levels of prevalence and specificity.

Although tree-based scan statistics can improve statistical power to detect rare MCMs, small sample sizes of medication exposure during pregnancy remain a major challenge. According to the Slone Epidemiology Center Birth Defects Study over a 33-year period from 1997-2003, all medications other than the 20 most common had exposure prevalences less than 0.5% during the first trimester.²⁹ Similarly, an analysis of 120 medical products with pregnancy-related labeling changes among 7.1 million livebirth pregnancies within 13 Sentinel Data Partners in the US (2008 to 2023) found that 80% had less than 0.1% use during the first trimester.³⁰ A previous simulation study showed that Poisson tree-based scan statistics require 4,000 exposed pregnancies to detect a two-fold increased risk for an outcome occurring at 8 events per 1,000 births with 85% power, while Bernoulli tree-based scan statistics require twice that sample size.¹⁶ Given these constraints, our harmonized tree allows researchers to pool data across ICD eras, further improving statistical power for tree-based scan analyses.

While we removed chromosomal anomalies and single-gene conditions that arise at conception from the outcome tree, we did not remove infants that experienced these codes from our trend evaluation as we were looking to generate more inclusive rates of MCM. We typically do remove these infants when we perform pharmacoepidemiology studies as these infants may present additional structural anomalies that are due to the underlying genetic etiology. For example, atrioventricular septal defects are commonly associated with trisomy 21.³¹ To support appropriate cohort definition in future studies, we provide a comprehensive list of chromosomal anomalies and single-gene disorders that researchers may use to identify and exclude affected deliveries when evaluating medication-related risks.

There are several limitations to our study. First, it is not always possible to establish exact mappings between ICD-9 and ICD-10 codes due to inherent differences in definitions between the two coding systems. In some instances, we accepted approximate mappings to preserve the number of outcomes represented on the tree. Additionally, because of complex mapping relationships (e.g., one-to-many, many-to-one, or many-to-many), codes within a particular granular clinical concept (level 5 groups) should be interpreted as approximate. Second, the mapping process, including code review and decision on whether a code represents a major or minor MCM or how different codes were grouped, was based on the best judgment of a clinical reviewer with 28 years of experience working on epidemiologic studies of congenital malformations. As such, the tree should be used and interpreted with this subjectivity in mind.

Third, the database sizes varied across years, particularly within the Medicaid population. This variation is unavoidable due to differences in data quality across states and within Fee-for-Service and Comprehensive Managed Care plans, which vary by state and year.^26,32-34 Specifically, with the introduction of managed care in Medicaid, data quality dropped initially because encounter detail was not submitted by managed care plans for reimbursement. Hence, because we required infants to have at least one year of continuous enrollment, most eligible infants came from states that met data quality standards for two consecutive years. As managed care encounter capture improved over the last decade, more children met enrollment criteria, contributing to substantial differences in cohort sizes within the Medicaid database. However, it is unlikely that exclusion of state Medicaid programs, especially in the earlier study period, yielded different MCM prevalences.

Lastly, the tree was developed based on MLCCS, which organizes MCMs based on organ systems. It is possible that MCMs might be better organized based on specific teratological or genetic pathways that might result in clusters of anomalies across several organ systems, such as the VATER syndrome. The hierarchical structure of the MCM tree is designed to support evaluation at varying levels of specificity, but this system is not intended to imply that all conditions grouped together share a common etiology. Although we attempted to cluster MCMs with similar etiologies wherever possible, it is unavoidable that some groups, particularly the “Other” categories, include conditions with heterogeneous causes. These “Other” groups often arise from the need to reconcile non-specific ICD codes across the two coding eras. Therefore, when a signal is observed in an “Other” category, it is important to review the specific ICD-9 and ICD-10 codes contributing to that grouping and consider whether any biologically plausible mechanism could explain the observed association. Future research on defining such clusters might create more targeted groupings for safety studies and could be incorporated into the current tree structure.

To conclude, we successfully developed and evaluated an MCM tree that spans ICD-9-CM and ICD-10-CM coding eras. Overall, there was good agreement between ICD-9-CM and ICD-10-CM-based groups at various levels. This tree structure facilitates use of tree-based scan statistics, allowing for the detection of specific MCM signals that may be obscured in composite outcome evaluation. By allowing both broad and granular outcome assessments, our approach enhances the ability to identify potential teratogenic risks with greater specificity and clinical relevance.

Take home messages.

• Teratogenicity studies may commonly assess major congenital malformations (MCM) as composites to maximize case numbers, but this approach may dilute effect estimates when the risk is confined to a specific MCM.

• We structured MCM outcomes into a hierarchical tree with varying clinical granularity using ICD-9 and ICD-10 codes to facilitate assessments across coding eras.

• By removing minor MCMs, chromosomal anomalies and single-gene conditions, we optimized the tree to focus on clinically highly relevant outcomes in assessments of teratogenic medications and other prenatal exposures.

• The optimized tree supports tree-based scan statistics, which enhances the ability to detect specific teratogenic risks that may be missed when using composite outcomes.

• By enabling both broad and granular outcome evaluation, our optimized tree enhances the ability to identify potential teratogenic risks with greater specificity.