Maternal Risk Factors vary between Subpopulations of Children with Autism Spectrum Disorder

Abstract

Previous work identified three subgroups of children with ASD based upon co-occurring conditions (COCs) during the first five years of life. This work examines prenatal risk factors, given by maternal medical claims, for each of the three subgroups: children with a High-Prevalence of COCs, children with mainly developmental delay and seizures (DD/Seizure COCs), and children with a Low-Prevalence of COCs. While some risk factors are shared by all three subgroups, the majority of the factors identified for each subgroup were unique; infections, anti-inflammatory and other complex medications were associated with the High-Prevalence COCs group; immune deregulatory conditions such as asthma and joint disorders were associated with the DD/Seizure COCs group; and overall pregnancy complications were associated with the Low-Prevalence COCs group. Thus, we have found that the previously identified subgroups of children with ASD have distinct associated prenatal risk factors. As such, this work supports subgrouping children with ASD based upon COCs which may provide a framework for elucidating some of the heterogeneity associated with ASD.

Lay Summary

Children diagnosed with Autism Spectrum Disorder (ASD) are commonly diagnosed with co-occurring conditions (COCs) as well. Medical events that occur during a woman’s pregnancy can affect the outcome of ASD but it is unclear how these events affect COCs. Medical events during pregnancy, identified using insurance claims, were found to vary for the three subgroups of children diagnosed with ASD and grouped by COCs. This supports subgrouping children with ASD based upon their COCs.

1. Introduction

The current diagnostic definition for Autism Spectrum Disorder (ASD) is involves two domains: (1) social-communication deficits, and (2) restricted and/or repetitive behaviors and/or interests (Anon 2013). While this definition encompasses the core symptoms of ASD, individuals with ASD are also commonly afflicted by a variety of co-occurring conditions (COCs), which are health conditions separate from the core symptoms which define ASD (Levy et al. 2010; Matson and Goldin 2013; Mosner et al. 2019; Rosen et al. 2018).

A recent study has shown that 95% of children with ASD have at least one COC (Soke et al. 2018) though it is common for a child with ASD to have multiple COCs (Aldinger et al. 2015) with the prevalence of COCs increasing with age (Kohane et al. 2012). Common COCs include gastrointestinal (GI) disorders (Bauman 2010), epilepsy or seizure disorders (El Achkar and Spence 2015; Thomas et al. 2017; Viscidi et al. 2013), psychiatric disorders (Mosner et al. 2019; Reaven and Wainer 2015; Rosen et al. 2018), developmental disabilities (Rubenstein et al. 2018), and sleep issues (Mazurek et al. 2019). Some studies have found patterns of COCs which cluster together such as GI issues with sleep disorders (Aldinger et al. 2015; Neumeyer et al. 2019), sleep disorders with epilepsy (Accardo and Malow 2015), and GI issues with impaired immune response (Rose et al. 2018). These common associations between COCs for ASD may prove useful in identifying biological ASD subtypes (Aldinger et al. 2015; Bauman 2010; Tye et al. 2019). Children with ASD have been shown to be negatively affected by the presence of COCs (Rosen et al. 2018), as COCs have been linked to adverse childhood experiences (Kerns et al. 2017), contribute to a greater need of psychological services (Posserud et al. 2018), have reduced positive outcomes later in life (McCauley, Elias, and Lord 2020), and affect the stability of an ASD diagnosis (Close et al. 2012; Wu et al. 2017). Specific COCs influence the age at which children are diagnosed with ASD (Soke et al. 2018). ASD has been associated with a high economic burden (Croen et al. 2006; Lavelle et al. 2014; Rogge and Janssen 2019) which increases in the presence of COCs (Buescher et al. 2014; Horlin et al. 2014; Lin 2014).

Current research has identified maternal risk factors for ASD in conjunction with certain COCs. Recurrent maternal GI issues, hypertension during pregnancy, and gestational diabetes were found to significantly increase the risk of ASD and GI disorders (Chandradasa et al. 2017). Maternal psychiatric disorders have been found to be directly and in-directly associated with a diagnosis of ASD and epilepsy (Ekinci et al. 2010; Jokiranta et al. 2014). Preterm birth, poor fetal growth, maternal obesity, and pregestational diabetes have shown a significant increased risk for ASD and intellectual disability (ID) (Joseph et al. 2017; Langridge et al. 2013; Li et al. 2016; Schieve et al. 2015). Anemia diagnosed during pregnancy was also found to increase the risk for ASD with ID, as well as ASD with ADHD (Wiegersma et al. 2019). Additionally, some evidence showed that maternal reported infection significantly increased the risk for diagnoses of ASD with ADHD (Hall et al. 2021) and children with this diagnostic combination were also found to be in increased risk of developing anxiety and mood disorders (Gordon-Lipkin et al. 2018). Lastly, for women who have never been married, were less educated, and who experienced bleeding during pregnancy these events were found to be significant predictors for diagnoses of ASD with developmental disorders (Zauche, Darcy Mahoney, and Higgins 2017).

In our previous work, we used medical claims of children with an ASD diagnosis to show three distinct patterns of COCs: children with a large number of different COCs, children with developmental delays (DD) or seizures as COCs, and children with a number of COCs only slightly more prevalent than children without ASD (Vargason, Frye, et al. 2019). To further investigate the characteristics of these clusters as well as validate this clustering approach we investigate whether there are distinct prenatal risk factors associated with these clusters of children. This retrospective study aims to associate maternal risk factors of having a child with ASD with the identified COCs clusters. Factors are represented by diagnostic, pharmacy, and procedural claims made during pregnancy from a large national database.

2. Materials and Methods

2.1. Cohort/Cluster Identification

All data used for this study were taken from the OptumLabs^® Data Warehouse (OLDW), a de-identified claims database with family identifiers, socioeconomic status information, medical and pharmacy claims, laboratory results, and enrollment records for commercial and Medicare Advantage (MA) enrollees (OptumLabs 2020). This database contains longitudinal health information on enrollees and patients, representing a diverse mixture of ages, ethnicities, and geographical regions across the United States. Children, born between the years 2000 and 2010, were previously identified from the OLDW and were investigated for a diagnosis of ASD using their medical claims up until 5 years of age (Vargason, McGuinness, and Hahn 2019). A child was considered to have a diagnosis of ASD if they had at least two separate ICD-9 diagnostic claims of autistic disorder (ICD-9 299.0), Asperger syndrome (ICD-9 299.8), or unspecified pervasive developmental disorder (ICD-9 299.9), and no diagnostic claims for childhood disintegrative disorder (ICD-9 299.1) or Rett syndrome (ICD-9 330.8) (Grivas, Frye, and Hahn 2021; Vargason, Frye, et al. 2019; Vargason, McGuinness, et al. 2019).

Previous work utilized this child cohort as a whole to investigate COCs diagnoses made within the first 5 years of life (Vargason, Frye, et al. 2019). Mothers of these children were previously identified within the OLDW using delivery claims and family identifiers, and children without an identified mother were excluded (Grivas et al. 2021). Thus, the cohort used in this study consisted of 1,258 children diagnosed with ASD (ASD cohort) and 122,559 children representing the population with no ASD diagnosis (POP cohort) with previously identified mothers and available COCs information, see Table 1. Approval from the Institutional Review Board was not needed as this retrospective study used de-identified data.

Table 1.

Child Cohort Progression

Child Cohort Identification	Number of Children
Original child cohort (Vargason, McGuiness, et al., 2019)	283,644
Children with COC information on file (Vargason, Frye, et al. 2019)	282,971
Children with identified mothers (Grivas, et al., 2021)	< 123,828
Children with identified mothers and COC information (current work)	123,817

Seven broad categories of COCs were used in our previous work: auditory disorders, DD, GI symptoms, immune-related conditions, psychiatric disorders, seizure disorders, and sleep disorders (Vargason, Frye, et al. 2019). All diagnostic definitions included for these categories of COCs, as well as ASD, are fully outlined in the previous work. The diagnoses of DD made were distinct from ASD however this study could not discern which diagnoses were made first. Analysis on temporal patterns of COCs showed that the ASD cohort could be grouped into the following three clusters (see Figure 1, identify ASD clusters): the first (High-Prevalence COCs) cluster had a majority of auditory, GI, immune, psychiatric, and sleep disorders and a significantly higher prevalence of COCs compared to the ASD cohort as a whole. The second cluster (DD/Seizure COCs) had the highest prevalence of DD and seizure disorders and had an overall similar pattern of COCs to the entire ASD cohort. The third cluster (Low-Prevalence COCs) had the lowest prevalence of COCs in all categories and most closely resembled the COCs patterns in the POP cohort.

Figure 1.

Flowchart of ASD cluster identification using 7 categories of COCs (top) and identification of prenatal risk factors from the 3 resulting ASD clusters (bottom).

Clustering of the children with an ASD diagnosis associated with this study identified 378 (30.0%) children belonging to the High-Prevalence COCs cluster, 304 (24.2%) children belonging to the DD/Seizure COCs cluster, and 576 (45.8%) children belonging to the Low-Prevalence COCs

cluster. Mothers of these children were previously identified and details of the procedure linking the mothers to the children can be found in (Grivas et al. 2021); maternal medical claims made during a mother’s pregnancy (or child’s gestation) were associated with the cluster or cohort affiliation for each child (Figure 1, identify prenatal risk factors). Mothers of multiple children have unique maternal events with the exception of multiple births, whose events are duplicated; this study identified 5,721 (4.6%) of children belonging to multiple births (twins/triplets/quadruplets), 32,046 (25.9%) children with siblings, and the remaining 86,050 (69.5%) with no identified sibling. Clustering of COCs identified within the POP cohort was not performed due to the smaller prevalence of COCs associated with this group. A result of this is that matching POP COCs to the ASD clusters would result in inadequate sample sizes for comparison.

2.2. Medical Claims Identification

Maternal medical claims were identified using the International Classification of Diseases, version 9 (ICD-9), National Drug Code (NDC), and Current Procedural Terminology (CPT) codes which were grouped into numerous diagnostic, pharmacy, and procedural variables, respectively (Grivas et al. 2021). For this study, variables were discarded if they contained fewer than 11 claims in order to stay compliant with OptumLabs Data Policy. Identified claims were used to determine women’s gestational ages and associated trimesters. The following sociodemographic variables were also included: race, home ownership, education level, income level, age group, and “Previous ASD” (a binary indicator for women who have had previous children with ASD identified by this study). It has been well established that women who have a child diagnosed with ASD are at significantly higher risk of having subsequent children with ASD (Hollowood et al. 2018; Ozonoff et al. 2011).

While maternal medical claims are unique for each birth, maternal sociodemographic information was assumed to remain the same as it was not possible to extract time-dependent sociodemographic information about individuals from the data set (i.e., siblings have the same race, home ownership status, education level, and income level on file as their mother). To avoid skewing sociodemographic trends from sibling bias, sociodemographic information representing only unique mothers was analyzed and was defined as follows: women belonging to the POP cohort have no children with ASD and women with children with an ASD diagnosis are placed into each cluster based on the cluster affiliation of their first child with ASD. Only 12 women had multiple children with ASD belonging to different clusters and their inclusion did not affect the results. The age category associated with each woman refers to the age of their first child (POP cohort) or first child with ASD (ASD clusters). These definitions are only reflected in the sociodemographic information and do not pertain to the analysis of risk factors or medical claims.

2.3. Statistical Analysis

A chi-square analysis was used to determine statistically significant differences between all pair-wise proportions of sociodemographic variables. Pair-wise statistically significant differences in variance and mean were calculated for age and for number of medical claims. Difference in variance was calculated using an F-test at a 5% significance level. Differences in mean were calculated using a Two-Sample T-test for equal variance or a Welch’s Two-Sample T-test for unequal variance, at a 5% significance level. The number of diagnostic, pharmacy, and procedural claims, as well as all three combined, were summed throughout the entire pregnancy as well as each trimester. This data were assumed to be normally distributed due to large sample sizes. To discern statistically significant differences between the number of medical claims made between the POP cohort and/or COC clusters, the F- and T-test values reported were calculated by averaging p-values on random samples of 300 from the cohort/cluster, repeated 1,000 times.

Logistic regression analysis was used to estimate the relationship between maternal medical claims (made during pregnancy) and having a child with ASD belong to the High-Prevalence, DD/Seizure, or Low-Prevalence COCs cluster, quantified through adjusted odds ratios (ORs) using a 95% confidence interval (Hosmer, Lemeshow, and Sturdivant 2013; Wright 1995), see Figure 1, prenatal risk factor identification. Highly correlated pairs of claims (r > 0.7) were identified using Pearson correlation analysis; unadjusted logistic regression analysis on these pairs identified the claim of lower significance which was removed from the study. The number of variables included in each logistic regression analysis differs due to the different number of mothers and associated claims contained within each cluster (for an extensive list of all variables investigated and their associated number of claims, please see supplemental material Table S1).

Following the protocol outlined in our previous work, the logistic regression analysis consisted of two logistic regression models containing common variables between the POP cohort and ASD COCs cluster, however one model includes the Previous ASD variable and the other does not as the Previous ASD variable was found to be highly skewed towards the ASD cohort due to bias from multiple gestations (Grivas et al. 2021). A final logistic regression model was made using only variables found to be significant in the two previously described models in order to determine falsely significant variables as well as to adjust for multiple comparisons by greatly reducing the number of variables included. Due to computational restrictions, all logistic regression analyses were made using 10% of the POP cohort data, stratified based on sociodemographic variables. These analyses were used to compare each ASD COCs cluster to the POP cohort and were repeated for variables segmented by each trimester. Due to the nature of the coding schemes used in insurance claims, some variable descriptions are vague as they encompass a variety of further defined subcodes. For example, the ICD-9 code 649 “Other Conditions Complicating Pregnancy” can be segmented into the following subcodes: 649.0 “Tobacco Use Disorder Complicating Pregnancy”, 649.1 “Obesity Complicating Pregnancy”, 649.2 “Bariatric Surgery Complicating Pregnancy”, etc. The choice of using an ICD-9 whole- or decimal-value claim is made at the discretion of the medical personnel. Subcodes of statistically significant diagnostic or procedural variables that are vaguely defined were evaluated in separate logistic regression analyses to better elaborate on the variable definitions. This analysis is represented in Figure 1, prenatal risk factor identification, by the purple arrows where each COC cluster was combined with the POP cohort (“+” symbol) for comparative analysis described above.

Lastly, logistic regression analyses were also performed using only the ASD clusters for completeness. For this analysis, the DD/Seizures COCs cluster was used as a reference in comparison to High-Prevalence COCs (HP-DD/S) and Low-Prevalence COCs clusters (DD/S-LP), and the Low-Prevalence COCs cluster was used as a reference in comparison to the High-Prevalence COCs cluster (HP-LP). Thus, variables that were significant associated with the High-Prevalence COCs cluster had ORs > 1 within the HP-DD/S and HP-LP analyses. Variables significant associated with the DD/Seizures COCs cluster had ORs < 1 within the HP-DD/S and DD/S-LP analyses. Lastly, variables significantly associated with the Low-Prevalence COCs cluster had ORS > 1 within the DD/S-LP analysis and ORs < 1 within the HP-LP analysis. This analysis is also represented in Figure 1, prenatal risk factor identification, by the red arrows where only the ASD clusters were combined (represented by the “+” symbol) for comparative analysis as described above.

All statistically significant results presented within the remainder of this work will be referenced as ‘significant’ for brevity of presentation.

3. Results

3.1. Cluster/Cohort Results

Sociodemographic information for each woman cohort/cluster can be found in Table 2. Based on the cohort definitions outlined for the sociodemographic information, it was unsurprising that women with children affiliated with any ASD cluster had a significantly higher percentage of subsequent children with an ASD diagnosis (Previous ASD) in comparison to the POP cohort. Interestingly, the Low-Prevalence COCs cluster showed a significantly higher percentage of subsequent children with ASD (Previous ASD) in comparison to both the High-Prevalence and DD/Seizure COCs clusters. Compared to the POP cohort, women belonging to the ASD High-Prevalence COCs cluster were found to have a significantly higher education, income (between $125,000-$199,999), and were of older age (greater than 30 years) at the time of their first child. Women in this cluster were also found to be significantly more educated when compared to women in the Low-Prevalence COCs cluster. Women belonging to the Low-Prevalence COCs cluster were of significantly higher incomes and older age (greater than 40 years) than women in the POP cohort. While all cohorts had a majority of women identifying as white, the ASD Low-Prevalence COCs cluster had a significantly higher percentage of white identifying women than the High-Prevalence COCs cluster, and both the High-Prevalence and DD/Seizure COCs clusters had a significantly higher percentage of women identifying as Asian than the POP cohort. Notably, the High-Prevalence and DD/Seizure COCs clusters found no significant differences between any sociodemographic variable. Statistical tests on women’s age within each cohort/cluster showed an agreement that the High-Prevalence and Low-Prevalence COCs cluster had significantly greater mean age overall, as well as between 30 and 39 years, see Table 3, though the differences were modest. No significant differences were found for greater than 40 years of age due to small sample sizes.

Table 2.

Sociodemographic Information on Women Cohorts

		ASD			POP
		HP†	D/S†	LP†	P†
	Total	367	>293	546	102,836	Significant Differences
		(%)	(%)	(%)	(%)	HP,P	D/S,P	LP,P	HP,D/S	HP,LP	D/S,LP
Race	White	69.2	70.8	75.6	73.4					*
	Asian	12.3	11.7	8.6	7.7	**	**
	Black	6.8	9.1	7.1	8.0
	Hispanic	11.7	8.4	8.6	10.8
Home Ownership	Probable Renter or Unknown	9.8	10.1	11.0	10.4
	Probable Owner	90.2	89.9	89.0	89.6
Education Level	High School Diploma or Less	10.6	12.8	15.2	18.9	***	**	*		*
	Less than Bachelors Degree	52.3	57.0	55.1	52.8
	Bachelors Degree Plus	37.1	30.2	29.7	28.2	***				*
Income Range	< $40,000	5.4	6.0	7.0	7.6
	$40,000–$74,999	17.2	13.4	18.7	19.2		*
	$75,000–$124,999	26.7	33.2	27.1	31.0
	$125,000–$199,999	29.4	24.5	24.5	23.1	**
	> $200,000	21.3	22.8	22.7	19.1			*
Age Range	< 30	23.2	28.2	25.8	32.0	***		**
	30–39	67.8	66.1	65.0	62.2	*
	40–49	9.0	5.7	9.2	5.8	**		***
Previous ASD	Yes	14.2	17.4	25.5	0.0	***	***	***		***	**
	No	85.8	82.6	74.5	100.0	***	***	***		***	**

^†Abbreviations: HP, High-Prevalence COCs cluster; D/S, DD/Seizure COCs cluster; LP, Low-Prevalence COCs cluster; P, POP Cohort

^*p-value < 0.05;

^**p-value < 0.01;

^***p-value < 0.001

Table 3.

Basic Statistics on Age for Women Cohort/Clusters

	Mean (Standard Deviation)
	POP†		HP†	DD/S†		LP†
All Ages	31.9 (4.8)		33.3 (4.7)	32.4 (4.6)		32.8 (4.9)
< 30	26.5 (2.3)		27.0 (2.0)	26.9 (2.2)		26.5 (2.4)
30–39	33.7 (2.7)		34.3 (2.7)	33.9 (2.5)		34.1 (2.7)
40–49	41.4 (1.6)		41.8 (1.9)	41.7 (1.7)		41.5 (1.7)
		Significant Difference in Mean Age
	HP, POP	D/S, POP	LP, POP	HP, D/S	HP, LP	D/S, LP
All Ages	***		***	*
< 30	*
30–39	***		*
40–49

^†Abbreviations: POP, POP Cohort; HP, High-Prevalence COCs Cluster; DD/S, DD/S COCs Cluster; HP, High-Prevalence COCs Cluster.

^*p-value < 0.05

^**p-value < 0.01

^***p-value < 0.001.

3.2. Maternal Medical Claim Trends

Correlation analysis for all variables during the entire pregnancy revealed the same six pairs of highly correlated variables (r > 0.7) as were found in our previous work (Grivas et al. 2021), regardless of ASD cluster. These highly correlated variable-pairs correspond to a cardiovascular procedure, diabetic materials, thyroid disorder, and multiple pairs related to receiving a vaccination. Variables with the higher p-value (evaluated by an unadjusted logistic regression) remained in the analysis and are shown in bold in the supplemental material, Table S2. Basic statistics on the number of total medical claims made within each cohort/cluster are shown in Table 4. No significant difference in variance was found between any pair-wise comparison (not shown). The High-Prevalence and DD/Seizure COCs clusters had significantly higher mean number of claims than the POP cohort (though the difference was minor) and no significant difference in mean was found between the number of claims from the ASD Low-Prevalence COCs cluster and the POP cohort. There was also no significant difference between any mean number of claims between any ASD clusters.

Table 4.

Basic Statistics on Number of Medical Claims made within each Cohort/Cluster

		Mean (Standard Deviation)
		ASD Cluster		POP Cohort
High-Prevalence (HP)		27.8 (8.9)		25.8 (8.1)
DD/Seizures (D/S)		28.7 (9.0)		25.6 (8.0)
Low-Prevalence (LP)		27.0 (9.3)		26.1 (8.3)
		Significant Difference in Means
HP, POP	D/S, POP	LP, POP	HP, D/S	HP, LP	D/S, LP
*	***

^*p-value < 0.05

^**p-value < 0.01

^***p-value < 0.001

The total number of diagnostic, pharmacy, and procedural claims for both cohorts can be found in Table 5. All ASD clusters had a similar average number of medical claims as the ASD cohort as a whole; all women had a majority of procedural claims (13–14), followed by diagnostic claims (10–12), and fewer pharmacy claims (1–2). Compared to the POP cohort, the High-Prevalence and DD/Seizure COCs clusters had a significantly higher number of diagnostic and pharmacy claims on average, while only the DD/Seizure COCs cluster had a significantly higher number of procedural claims. Additionally, the DD/Seizure COCs cluster had a higher mean number of pharmacy claims than the Low-Prevalence COCs cluster. Although these differences were statistically significant, they are small (1–2).

Table 5.

Basic Statistics on Number of Diagnostic, Pharmacy, or Procedural Claims made within each Cohort/Cluster

			Mean (Standard Deviation)
			ASD Cluster		POP Cohort
Diagnostic Claims		High-Prevalence (HP)	11.6 (4.0)		10.8 (3.9)
		DD/Seizure (D/S)	12.1 (4.2)		10.7 (3.9)
		Low-Prevalence (LP)	11.5 (4.4)		11.1 (4.0)
Pharmacy Claims		High-Prevalence (HP)	2.4 (2.1)		2.0 (1.7)
		DD/Seizure (D/S)	2.6 (2.1)		1.9 (1.7)
		Low-Prevalence (LP)	2.2 (1.9)		2.0 (1.7)
Procedural Claims		High-Prevalence (HP)	13.8 (4.3)		13.1 (4.1)
		DD/Seizure (D/S)	14.0 (4.4)		12.9 (4.1)
		Low-Prevalence (LP)	13.3 (4.6)		13.1 (4.1)
		Significant Difference in Mean Number of Claims
	HP, POP	D/S, POP	LP, POP	HP, D/S	HP, LP	D/S, LP
Diagnostic	*	***
Pharmacy	*	**				*
Procedural		**

^*p-value < 0.05

^**p-value < 0.01

^***p-value < 0.001

3.3. Logistic Regression Results

3.3.1. ASD Clusters vs. POP Cohort

Significant variables determined in the final logistic regression analysis are shown in Table 6 and are grouped based on common themes and factors that are found in common between the analyses are shown in bold; full results can be found in the supplemental material Table S3. The Low-Prevalence COCs cluster had the greatest overall number of significant variables (16), followed by the DD/Seizure COCs (15), and the High-Prevalence COCs (14). However, the High-Prevalence and DD/Seizure COCs cluster had more variables associated with an increased risk of ASD (13) compared to the Low-Prevalence COCs (10) while the latter had more variables associated with a decreased risk (5) compared to the DD/Seizure COCs (4) and High-Prevalence COCs (1) clusters. Aside from sociodemographic variables, the three analyses found few overlapping significant variables (see bolded variables, Table 6). Depressive disorder was associated with an increased risk of belonging to both the High- and Low-Prevalence COCs clusters while vaccinations were significantly associated with a decreased risk. A prescription for nutritional vitamins and cesarean related procedures were both associated with an increased risk of belonging to the DD/Seizures and Low-Prevalence COCs clusters. However, Cesarean or Extractor Delivery associated with the Low-Prevalence COCs cluster was comprised of two different significant diagnostic subcodes (ICD-9 669.7 and 669.5, respectively), both found with a significant increased risk of belonging to the Low-Prevalence COCs cluster (supplemental material Table S4).

Table 6.

Logistic Regression Analysis on POP vs ASD Cohorts Significant Variables

	Adjusted Odds Ratio (95% CI) P-value
	High-Prevalence COCs Cluster	DD/Seizure COCs Cluster	Low-Prevalence COCs Cluster
Sociodemographic Factors	Previous ASD 12.4 (3.93, 33.4) *** Education – Bachelors Degree Plus 2.22 (1.46, 3.44) *** Age – 40–49 1.74 (1.13, 2.63) * Education – Less than Bachelors Degree 1.73 (1.21, 2.54) ** Race – Asian 1.57 (1.11, 2.17) **	Previous ASD 10.5 (2.85, 30.9) *** Race – Asian 1.75 (1.18, 2.52) **	Previous ASD 19.4 (8.88, 42.0) *** Age – 40–49 1.60 (1.13, 2.25) **
Mental Disorders	Depressive Disorder 2.73 (1.48, 4.64) ***	Antidepressants Prescription 2.05 (1.40, 2.94) *** Procedure Services Psychiatry 1.64 (1.13, 2.33) **	Adjustment Reaction Disorder 16.3 (6.87, 37.6) *** Episodic Mood Disorder 4.56 (2.14, 8.97) *** Depressive Disorder 2.24 (1.30, 3.66) **
Pregnancy Related Factors	Anesthesia Procedures Lower Abdomen 2.26 (1.31, 3.66) ** Problems with Amniotic Cavity 1.59 (1.21, 2.05) *** Umbilical Cord Complications 1.34 (1.06, 1.68) *	Cesarean Procedure 1.64 (1.15, 2.40) ** Nutritional Vitamins Prescription 1.40 (1.10, 1.79) ** Fetal Distress or Poor Fetal Growth 1.36 (1.08–1.72) **	Multiple Gestation 1.80 (1.32, 2.41) *** Cesarean or Extractor Delivery 1.54 (1.25, 1.89) *** Hypertension Complicating Pregnancy 1.37 (1.08, 1.72) ** Nutritional Vitamins Prescription 1.26 (1.05, 1.50) *
Non-Pregnancy Related Factors	Screening for Blood Disorders 2.18 (1.39, 3.32) *** Analgesic NSAID Prescription 2.20 (1.20, 3.72) ** Respiratory Corticosteroids 2.04 (1.28, 3.10) ** Urinary Tract Infection 1.54 (1.18, 2.00) **	Diseases of Endocardium 2.53 (1.39, 4.27) ** Asthma 2.26 (1.48, 3.34) *** Joint Disorders 1.84 (1.06, 3.00) * Services Hospital Inpatient 1.57 (1.14, 2.13) ** Services Consultation 1.53 (1.19, 1.96) *** Abdominal Pain 1.51 (1.06, 3.00) **	Procedures Pulmonary 1.87 (1.24, 2.73) ** Bacterial Infection 1.70 (1.05, 2.63) *
Protective Factors	Vaccinations 0.55 (0.37, 0.79) **	Other Conditions Complicating Pregnancy 0.44 (0.24, 0.73) ** Evaluations Physical Medicine and Rehabilitation 0.62 (0.39, 0.93) *	Vaccinations 0.51 (0.36, 0.70) *** Vaginal Delivery Procedures 0.61 (0.43, 0.89) ** Procedures Immunology 0.76 (0.62, 0.94) ** Other Conditions Classified Elsewhere Complicating Pregnancy 0.78 (0.65, 0.93) ** Procedures Other Pathology and Laboratory 0.78 (0.64, 0.95) *

Note: common variables between COCs clusters are bolded.

^*p-value < 0.05

^**p-value < 0.01

^***p-value < 0.001

Results of the logistic regression analysis segmented by trimester can be found in the supplemental material Tables S5–S6 and a summary in Table 7. More significant variables were found in the third trimester (13–15), regardless of ASD cluster, compared to the first (7–10) and second (6–9) trimesters. All clusters found a majority of variables significantly associated with an increased risk, regardless of trimester or medical category (especially pharmacy variables which found only an increased association). Overall, very few variables (1–2) were found in common with the exception of selected pharmacy and procedural variables between the High-Prevalence and DD/Seizure COCs during the third trimester.

Table 7.

Summary of Significant Variables found in Logistic Regression Results POP vs. ASD Clusters segmented by Trimester

	First Trimester			Second Trimester			Third Trimester
	HP†	DD/S†	LP†	HP	DD/S	LP	HP	DD/S	LP
Diagnostic Variables
Total (Increasing OR†)	4 (2)	4 (3)	5 (3)	3 (3)	1 (1)	3 (2)	5 (4)	3 (2)	9 (7)
Common	1	1	0	1	0	1	1	0	1
Pharmacy Variables
Total (Increasing OR)	2 (2)	2 (2)	1 (1)	1 (1)	1 (1)	1 (1)	3 (3)	3 (3)	1 (1)
Common	1	1	0	0	0	0	2	3	1
Procedural Variables
Total (Increasing OR)	1 (1)	1 (1)	4 (2)	2 (2)	4 (3)	5 (3)	5 (4)	8 (7)	5 (3)
Common	0	1	1	0	1	1	4	3	1

^†Abbreviations: HP, High-Prevalence COCs cluster; DD/S, DD/Seizure COCs cluster; LP, Low-Prevalence COCs cluster; OR, odds ratio.

3.3.2. ASD Clusters Only

Significant variables from the logistic regression analyses only containing ASD clusters are shown in Table 8 (see supplemental material Table S7 for full models) and factors found in common between analyses are shown in bold. From the HP-DD/S analysis, ORs > 1 and ORs < 1 represent variables associated with the High-Prevalence and DD/Seizures COCs clusters, respectively. From the DD/S-LP analysis, ORs > 1 and ORs < 1 represent variables associated with the Low-Prevalence and DD/Seizures COCs clusters, respectively. Lastly, from the HP-LP analysis, ORs > 1 and ORs < 1 represent variables associated with High- and Low-Prevalence COCs clusters, respectively. While no specific trend in variables was found that influenced COC prevalence, there were a greater number of variables associated with an increased risk of COCs within the DD/S-LP and HP-LP analyses (6) than the HP-DD/S analysis (3). The HP-DD/S analysis found more variables associated with a decreased risk of COCs (4), followed by the HP-LP analysis (3) then the DD/S-LP analysis (2), however the differences are minimal. All three analyses found few (1–2) variables associated with sociodemographic and pharmacy variables and more associated with diagnostic and procedural variables (2–4) except the HP-LP analysis which found a larger amount of significant diagnostic variables (5) compared to pharmacy variables (1). Lastly, significant variables in common between analyses are bolded in Table 8 and they are associated with the High-Prevalence COCs cluster (high education and administration of highly complex drugs) and with the DD/Seizures COCs cluster (asthma and anesthesia obstetric procedures).

Table 8.

Highly Significant Adjusted Logistic Regression Results between ASD Clusters

	Odds Ratio (95% CI) P-value
	HP†-DD/S†	DD/S-LP†	HP-LP
Sociodemographic Variables
Education – Bachelors Degree Plus	2.64 (1.34, 5.25) **		2.49 (1.43, 4.40) **
Race – Hispanic			1.73 (1.09, 2.73) *
Diagnostic Variables (ICD-9 Code)
Infectious/Parasitic Conditions Complicating Pregnancy (647)			2.51 (1.11, 5.89) *
Hypertension Complicating Pregnancy (642)		1.69 (1.09, 2.67) *
Urinary Tract Infection (599.0)			1.58 (1.09, 2.29) *
Other Conditions Classified Complicating Pregnancy (648)			1.34 (1.01, 1.77) *
Joint Disorders (719)		0.25 (0.11, 0.55) ***
Asthma (493)	0.27 (0.11, 0.61) **	0.30 (0.15, 0.60) ***
Obstructed Labor (648)			0.50 (0.29, 0.85) *
Other Complications of Labor (669)			0.60 (0.41, 0.87) **
Other Problems Affecting Mother (656)	0.69 (0.49, 0.96) *
Pharmacy Variables
Gastrointestinal Prokinetic Agents		0.34 (0.13, 0.84) *
Antidepressants		0.45 (0.26, 0.75) **
Anti-inflammatory Glucocorticoids			0.56 (0.34, 0.93) *
Procedural Variables
Procedures Pulmonary		2.31 (1.05, 5.42) *
Studies Cytogenetic	2.71 (1.46, 5.26) **
Administrations Highly Complex Drug or Biological Agent	1.78 (1.06, 3.05) *		1.83 (1.17, 2.87) **
Anesthesia Procedures Obstetric	0.52 (0.32, 0.83) **	0.43 (0.28, 0.66) ***
Services Hospital Inpatient		0.52 (0.34, 0.80) **
Services Emergency Department	0.58 (0.36, 0.92) *

Note: Common variables between COCs clusters are bolded.

^†Abbreviations: HP, High-Prevalence COCs cluster; DD/S, DD/Seizure COCs cluster; LP, Low-Prevalence COCs cluster.

^*p-value < 0.05

^**p-value < 0.01

^***p-value < 0.001.

Diagnostic variables identified within the ASD only analysis (Table 8) were further evaluated using their ICD-9 subcodes (see supplemental material Table S8). The variable Other Conditions Classified Complicating Pregnancy (ICD-9 648) from the HP-LP analysis showed a significant association with the High-Prevalence COCs cluster (OR > 1), however, the subcode analysis only found a significant association with Other Current Conditions Classified Elsewhere complicating pregnancy (ICD-9 648.9), meaning that this association was coded elsewhere within the ICD-9. Similarly, Obstructed Labor (ICD-9 648) and Other Complications of Labor (ICD-9 669) subcodes either did not have enough claims to be included in the subcode analysis or did not find any significant claims associated with their codes. From the HP-DD/S analysis, the diagnostic variable Other Problems Affecting Mother (ICD-9 656) did not find any significant association with any subcodes however, the subcodes Fetal Distress (656.3), Poor Fetal Growth (656.5), and Unspecified Fetal and Placental Problems (656.9) were found with a significant increased risk associated with the DD/Seizures COCs cluster when compared to the POP Cohort (supplemental material Table S4).

4. Discussion

While current research has investigated maternal risk factors of ASD co-occurring with other disorders (Ressel et al. 2020; Rosen et al. 2018; Schieve et al. 2015), there is limited research on how these risk factors vary for ASD subgroups based on COCs patterns. Thus, our findings will focus on risk factors for ASD as a whole and when applicable, COC-based subgroups. This discussion section will be outlined as follows: (1) discussion of findings on maternal medical claims associated with COCs clusters, (2) significant risk factors associated with each COCs cluster, identified through the logistic regression where each ASD cluster was compared to the POP cohort (Table 6), and (3) overall trends for risk factors associated with different ASD subgroups based on COCs and the impact this may have on our current understanding of ASD. Many of our current findings coincide with those identified in our previous study, risk factors associated with the ASD group as a whole, of which those findings will be briefly summarized (Grivas et al. 2021).

4.1. ASD COCs Cluster Results

Clustering results found a majority of children (greater than 40%) identified belong to the Low-Prevalence COCs cluster, similar to what was found in previous work (Vargason, Frye, et al. 2019). However, this study found a larger percentage of children belonging to the High-Prevalence COCs cluster (30.0% compared to 23.7% found in previous work) and a smaller percentage belonging to DD/Seizures COCs cluster (24.2% compared to 26.5% found in previous work). These small discrepancies come from the differences in child cohorts identified, as outlined in Table 1, and do not significantly impact the results found in this study.

While there were some significant differences in sociodemographic prevalence between the COCs clusters and the POP cohort, there were minimal significant differences in socioeconomic status between the COCs clusters. No significant differences were found for any income level between COCs clusters and for any education level between the DD/Seizures and Low-Prevalence COCs clusters. Aside from women who identified as Asian, which is further discussed below, no trends in ethnicity were found to be correlated with any COCs cluster in our results presented here nor in our previous paper (Vargason, Frye, et al. 2019). Thus, the sociodemographic status associated with each COCs cluster do not significantly impact the results found in this study.

The High-Prevalence COCs cluster found a significantly greater mean number of medical claims than the POP cohort, both overall (Table 4) as well as diagnostic and pharmacy category claims (Table 5). The DD/Seizures COCs cluster also had significantly greater mean number of medical claims than the POP cohort, both overall and regardless of medical category (diagnostic, pharmacy, and procedural). Lastly, the Low-Prevalence COCs cluster found no significant difference in mean number of medical claims overall or regardless of medical category compared to the POP cohort. These findings are synonymous to those of previous work which showed that the Low-Prevalence COCs cluster had most similar COCs patterns to that of the POP cohort, while the High-Prevalence and DD/Seizures COCs clusters COCs patterns showed greater prevalence of additional medical conditions (Vargason, Frye, et al. 2019).

This work also found that a greater number of significant factors occurred in the third trimester, regardless of ASD cluster, compared to the first and second. While this may suggest that maternal medical events that occur in the third trimester are most influential on a diagnosis of ASD and associated COCs, this finding may also be a result of sample bias. Women are more likely to have doctor appointments during their third trimester, as they are closely monitored for giving birth. Thus, this increase in frequency of visits may lead to an increase in medical claims.

4.2. Significant Factors Associated with each COCs Cluster

4.2.1. Sociodemographic Factors

Having a previous child with ASD (Previous ASD, Table 6) showed one order of magnitude increased risk regardless of COCs cluster affiliation. Having a previous child diagnosed with ASD was also the greatest risk factor for the ASD cohort overall and has shown an elevated recurrence risk of ASD when compared to the general population (Grivas et al. 2021). The High-Prevalence and DD/Seizure COCs clusters found a greater than 70% increased risk associated with being of Asian Race (Table 6), however, there was a significantly greater prevalence of women identifying as Asian (Table 2), greater than those of the current ASD surveillance estimates which show similar prevalence rates for ASD among different races and ethnicities (Maenner et al. 2021), thus this finding was most likely due to sample bias. Being of older age (greater than 40 years) was associated with a significantly increased risk (greater than 60%) for the High- and Low-Prevalence COCs clusters. Being of older age is a known risk factor for ASD especially for women over the age of 35 with their first pregnancy (Grivas et al. 2021).

Novel to this study, the High-Prevalence COCs clusters found a significantly increased risk associated with higher education (any college degree or graduate education, Table 6). Higher education is confounded by being of older age, where women who pursue a higher education may have children later in life, and as such other studies have found mixed associations between ASD and maternal education (Croen et al. 2007; DiGuiseppi et al. 2016; Pinborough-Zimmerman et al. 2011). Maternal education and age were not found to be highly correlated (Pearson correlation coefficient 0.2) and this study could not determine potential confounding factors. However, some studies found that the prevalence of higher education did change depending on the ASD subgroup, especially ASD without the COC of intellectual disability (Bhasin and Schendel 2007; Kalkbrenner et al. 2019).

4.2.2. Mental Disorders

All COCs clusters found significantly increased risk associated with mental illness. A diagnosis of depressive disorder was found in common between the High- and Low-Prevalence COCs clusters and was associated with more than twice increased risk for belonging to those clusters compared to the POP cohort. The Low-Prevalence COCs cluster also found an increased risk associated with other mental disorders of Adjustment Reaction and Episodic Mood, ORs 16.3 (6.87, 37.6) and 4.56 (2.13, 8.97), respectively. The DD/Seizure COCs cluster found an increased risk associated with a prescription for antidepressants, OR 2.05 (1.40, 2.94) and having a psychiatry service procedure, OR 1.64 (1.13, 2.33), both of which were significant for the ASD cohort as a whole (Grivas et al. 2021). All of these findings support that the association between maternal mental disorders and a diagnosis of ASD is most likely due to a combination of mental illness, treatment, and genetics (Bastaki, Alwan, and Zahir 2020; Grivas et al. 2021). Interestingly, each COCs cluster had an overall unique pattern of associated mental disorders. One previous study found that familial history of psychiatric problems was able to predict severity of psychiatric symptom category in children (Gadow, DeVincent, and Schneider 2008), suggesting that beyond a common theme of mental illness, subgroups of ASD may be associated with unique mental disorders.

4.2.3. Pregnancy-Related Factors

All three COCs clusters found a similar number of significantly increased risk factors associated with pregnancy-related medical claims (Table 6). Many of these claims represent unplanned pregnancy complications such as problems with amniotic cavity, umbilical cord complications, fetal distress or poor fetal growth, and hypertension complicating pregnancy. There is a higher prevalence of pregnancy problems for mothers of children with ASD (Visser et al. 2013) and many studies have shown that enduring pregnancy complications or adverse pregnancy conditions are associated with an increased risk of ASD (Gardener, Spiegelman, and Buka 2009; Getahun et al. 2017; Lyall et al. 2017; Nadeem et al. 2020; Zhang et al. 2010). Though our study found mostly unique associations of pregnancy complications, it is clear that these risks are equally prevalent among ASD subgroups (Visser et al. 2013).

A prescription for nutritional vitamins was the only common pregnancy-related claim between the DD/Seizure and Low-Prevalence COCs clusters, ORs 1.40 (1.10, 1.79) and 1.26 (1.05, 1.50), respectively. However, nutritional supplements have previously shown overall positive effects on child development (Bastaki et al. 2020). A prescription for prenatal vitamins may not accurately represent the relationship between nutritional supplements and ASD (Grivas et al. 2021) as prenatal vitamins are available over the counter and are thus not always reflected in insurance prescription claims. Prescriptions for prenatal vitamins, especially folate, may differ from those available over the counter as prescriptions contain much higher doses of folate. Although some studies have demonstrated that high doses of prenatal folate supplementation reduces the risk of ASD (Frye, Slattery, and Quadros 2017; Schmidt et al. 2019), other studies have found higher blood folic acid concentrations were associated with an increased risk of ASD (Egorova et al. 2020; Raghavan et al. 2018). However, the latter is more likely due to the type of folate within the supplementation (oxidized vs reduced) and individual variations in folate metabolism (Frye et al. 2017). For example, abnormalities in the folate pathway, such as the folate receptor alpha autoantibody, may be seen in mothers of children with ASD (Quadros et al. 2018). Such abnormalities may increase blood levels of folic acid because this oxidized synthetic form of folate cannot be readily metabolized into the folate cycle when certain abnormalities in the folate pathway exist (Frye, Slattery, Quadros, 2017) and during the time period of this study, prescription prenatal vitamins contained folic acid. In such cases, reduced forms of folate are preferred but most prenatal multivitamins do not contain these preferred forms of folate. Unmetabolized folate has also been associated with an increased risk of other adverse effects in children such as food allergies (McGowan et al. 2020). Additionally, it was out of the scope of this study to discern the type of vitamin supplementation associated with these prescriptions and as such we could not determine if this relationship was due to a vitamin or nutritional deficiency. Further analysis was done to analyze other correlated medical claims with a claim for nutritional vitamins in order to identify potential confounding health conditions, however, the results were inconclusive.

Multiple gestation was found to have a significant association with the Low-Prevalence COCs cluster, however, this study was unable to determine if the twins identified were monozygotic and dizygotic and thus this association may be due to the higher rate of ASD that occurs with multiple gestations. A cesarean procedure was found to have a greater than 50% increased risk associated with the DD/Seizure and the Low-Prevalence COCs clusters, however, cesarean delivery may be confounded by anesthesia administration (Grivas et al. 2021). Anesthesia procedures were found to have a significant association with the High-Prevalence COCs cluster, OR 2.26 (1.31, 3.66). While some studies have found significantly increased risk with the use of anesthesia (Al-maqati et al. 2021; Chien et al. 2015; Huberman Samuel et al. 2019) there are many confounders to be considered such as fetal distress and, as mentioned prior, cesarean delivery (Sagi-Dain et al. 2020). Unfortunately, this study was unable to determine if cesarean delivery was elective or non-elective. Additionally, instrumental delivery has been shown to be a significant factor associated with ASD (El-Baz, Ismael, and El-Din 2011).

4.2.4. Non-pregnancy Related Factors

Non-pregnancy related factors are medical events that occur during pregnancy but are unrelated to the pregnancy. The High-Prevalence and DD/Seizures COCs clusters found a similar amount of non-pregnancy related risk factors (4–5) while the Low-Prevalence COCs cluster found fewer (2) risk factors. Interestingly, there were no shared non-pregnancy related risk factors among any COCs clusters. Other studies, including our previous work, found that various procedures, both hospital or out-patient, unrelated to the pregnancy increased the risk of ASD in general (Grivas et al. 2021; Zhang et al. 2010).

Prescribed maternal medications of non-steroidal (Analgesic NSAIDs) and steroidal (corticosteroids) anti-inflammatory drugs both found a greater than two times increased risk associated with the High-Prevalence COCs cluster (Table 6). Many studies have investigated an immune-mediated subtype of ASD, in which corticosteroids are a common treatment for children with ASD and inflammatory related COCs or symptoms (McDougle et al. 2015; Thom et al. 2019). A prescription for glucocorticoids during the first trimester was associated with an increased risk for ASD overall (Grivas et al. 2021) however, research is limited on the effect of maternal corticosteroid usage and its relation to ASD. Though postnatal exposure to corticosteroids has been associated with an increased risk of children being diagnosed with ASD (Davidovitch et al. 2020). In addition, research supports the association between maternal inflammatory environment and an increased risk for ASD (Bilbo et al. 2018; Meltzer and Van de Water 2017). The remaining factors associated with the High-Prevalence COCs cluster, i.e. screening for blood disorders and urinary tract infection, were also found to increase the risk of ASD in general (Grivas et al. 2021). Screening for blood disorders only had one diagnostic subcode with a sufficient amount of medical claims to include in the sub-code analysis, Screening for Iron Deficiency Anemia, however, it was not statistically significant (supplemental material table S4). Maternal anemia has been associated with an increased risk for ASD, especially when diagnosed in early pregnancy (Wiegersma et al. 2019).

A diagnosis of asthma showed a significantly increased risk in association to the DD/Seizures COCs cluster with an OR of 2.26 (1.48, 3.34). Maternal asthma has shown an increased risk in ASD (Gong et al. 2019). Joint disorders, such as arthritis, were also found with a significantly increased risk for the DD/Seizures COCs cluster, OR 1.84 (1.06, 3.00). Maternal rheumatoid arthritis has shown increased risk for ASD (Rom et al. 2018) although a diagnosis of rheumatoid arthritis during pregnancy was not found to be significant in this study. Both asthma and arthritis are associated with an overactive immune system and maternal autoimmune conditions in general have been highly researched with regards to their effect of increased risk of ASD in children (Chen et al. 2016; Hughes et al. 2018; Keil et al. 2010; Modabbernia, Velthorst, and Reichenberg 2017). The DD/Seizures COCs cluster found Diseases of the Endocardium to be the highest non-pregnancy risk factor, with an OR of 2.53 (1.39, 4.27). Diseases of the endocardium, such as endocarditis, have shown worse maternal and fetal outcomes (Dagher et al. 2021), however, little research has been done on their effect on ASD. The remaining non-pregnancy related factors associated with the DD/Seizures COCs cluster are also factors for the ASD cohort overall (Grivas et al. 2021).

Lastly, the Low-Prevalence COCs cluster found a diagnosis for bacterial infection was associated with a 70% increased risk (1.05, 2.63). Maternal bacterial infections have been associated with an increased risk of ASD (Zerbo et al. 2015), especially if they require hospitalization (Jiang et al. 2016). Pulmonary procedures, the remaining risk factor associated with this cluster, was also associated with an increased risk of ASD in general (Grivas et al. 2021).

4.2.5. Protective Factors

Protective factors describe medical events that showed a significantly decreased association with any ASD COCs cluster. The High-Prevalence and DD/Seizures COCs clusters had fewer (1–2) protective factors compared to the Low-Prevalence COCs cluster (5). All protective factors were associated with standard obstetrical procedures such as vaginal delivery, physical rehabilitation, standard blood tests (Procedures Immunology, and Procedures Other Pathology and Laboratory), vaccinations, or other noted pregnancy conditions (Other Conditions Complicating Pregnancy and Other Conditions Classified Elsewhere Complicating Pregnancy). The latter two were further evaluated based on their diagnostic subcodes (supplemental material Table S4); the only subcode that showed a significantly decreased association with Other Conditions Classified Elsewhere Complicating Pregnancy was undefined. Although there were not enough claims to further evaluate the subcodes associated with Other Conditions Complicating Pregnancy, this variable was previously significantly associated with Uterine Size Date Discrepancy when looking at the ASD cohort as a whole (Grivas et al. 2021). Aggressive obstetrical care has been previously shown to reduce adverse outcomes in pregnancy including ASD (Cheng et al. 2019; Grivas et al. 2021).

The only protective factor that was found in common between analyses (shown in bold) was a vaccination procedure, which showed a substantial decrease in High-Prevalence and Low-Prevalence COCs cluster association by almost half, although it was not found a factor for the DD/Seizures COC cluster. Vaccinations were also found to be a significant protective factor for the ASD cohort as a whole (Grivas et al. 2021). However, this study cannot discern the type of vaccination administered and some vaccinations, e.g., the flu vaccine, can be administered outside of a medical setting (such as a local pharmacy) and would not always appear in insurance claims data.

4.2.6. ASD Clusters Only Analysis

Many of the statistically significant variables associated with the COC clusters from the ASD vs. POP cohort analysis (Table 6) are also found to share the same cluster association from the ASD clusters only analysis (Table 8). For example, as outlined in Table 8, the High-Prevalence COCs cluster found a significant association with higher education, urinary tract infection, and complex drug administration. The DD/Seizures COCs cluster found a significant association with joint disorders, asthma, a prescription for antidepressants, obstetrical procedures using anesthesia, various hospital procedures, and other problems affecting the mother. Lastly, the Low-Prevalence COCs cluster found significant associations with hypertension, pulmonary procedures, and other complications of labor. All of these factors have been addressed in their relation to ASD above. All significant associations outlined in Table 8 were unique for each COCs cluster. The only variables that appear more than once in this table are either associated with the High-Prevalence COCs cluster (higher education and administration of complex drugs) or the DD/Seizures COCs cluster (asthma and obstetrical procedures using anesthesia). Additionally, many of the variables found to be significantly associated with the DD/Seizures COCs cluster, as outlined above, were also found to have a significant association with the ASD cohort as a whole (Grivas et al. 2021).

4.3. General Trends of ASD Subgroups based on COCs

While the results of the logistic regression share overarching themes, interestingly, the individual factors associated with each COCs cluster were largely unique. This trend continued even when segmented by trimester (supplemental material Table S5) and with the ASD-only analysis (Table 8). In addition to unique associations, many factors appeared to be robust, their cluster affiliation shown within both the ASD vs. POP cohort and ASD clusters only analyses (Tables 6 and 8, respectively).

Similarly, the DD/Seizure COCs cluster identified the most significant factors in common with those identified by the ASD cohort as a whole (Grivas et al. 2021). This supports the finding that the patterns associated with the DD/Seizures COCs cluster are most similar to those of the entire ASD cohort (Vargason, Frye, et al. 2019). All COCs clusters were associated with factors that have been shown with an increased risk of ASD in general. In addition to these factors, each COCs cluster shows a unique pattern of risk factors that suggest potential prenatal risk-factor profiles as outlined below.

The High-Prevalence COCs cluster found significant increased association with urinary tract infections (Tables 6 and 8), infectious or parasitic conditions (Table 8), administration of complex drugs in-hospital (Table 8), and prescribed anti-inflammatories (Table 6). These findings suggest that the High-Prevalence COCs cluster may be associated with infectious or inflammatory events during pregnancy. The DD/Seizures COCs cluster found more factors associated with maternal immune disruption through factors such as asthma and arthritic-like (joint disorder) conditions (Tables 6 and 8). While the Low-Prevalence COCs cluster mostly found factors associated with pregnancy-related issues (Tables 6 and 8).

These potential risk factor profiles require further investigation, especially to account for associations with (different) pregnancy related complications as well as factors that influence ASD in general. However, this suggests that unique patterns of risk factors are associated with different ASD COCs patterns and may identify subgroups of the disorder (Gadow et al. 2008; Langridge et al. 2013). Other research has also found differentiating risk factors for ASD subgroups based on varying COCs (Langridge et al. 2013). Identifying these unique ASD phenotypes may help further understand the shared developmental pathways for both ASD as well as the COCs (Muhle et al. 2018; Oerlemans et al. 2016; Ousley and Cermak 2014). In general, children with ASD will greatly benefit from individualized monitoring, diagnosis, and treatment based on their ASD phenotype (McDonald et al. 2020).

4.4. Strengths and Limitations

One advantage of this study is that it uses a large, heterogeneous data set of privately insured individuals across the United States, utilizing the same provider. Medical claims documented through insurance records provide a better understanding of the medical events that may have occurred during pregnancy, especially in comparison to participation-based surveys. However, the data provided for this study does not include any information on individuals enrolled in Medicaid and the use of insurance claims may not convey a complete understanding of environmental factors such as those that are not recorded within insurance claims (i.e. over-the-counter medications). Furthermore, the data from this study assume that these individuals utilized their medical coverage and are thus not limited by any barriers preventing medical attention. Additionally, this study focuses on events that occurred during a woman’s pregnancy, it was not able to discern any mitigating factors that may have occurred before or after this time period.

All pharmacy claims made within this study were represented by filled prescriptions. This study could not determine the type of prenatal vitamins prescribed, nor could it determine the type of vaccination administered. Lastly, some interpretations of factor associations are limited due to confounding factors that could not be further investigated by this study (such as higher education with older age or prenatal vitamins).

5. Conclusion

The complex heterogeneity of ASD will benefit from understanding the biological impact of ASD subgroups, especially based on COCs. The presence of COCs both impacts the severity of an ASD diagnosis and negatively affects children overall (Meguid et al. 2018; Rosen et al. 2018). This study showed that ASD subgroups based on COCs have relatively unique risk and preventative factors identified by the use of adjusted logistic regression on prenatal medical claims. Future research should investigate these differentiating factors to discern the possible biological pathways associated with ASD subgroups. By doing so, it can be envisioned that targeted treatment and preventative strategies can be determined for families with an increased risk of an ASD diagnosis, particularly if these are associated with certain COCs.