SEX DIFFERENCES IN THE AGE OF CHILDHOOD AUTISM DIAGNOSIS AND THE IMPACT OF CO-OCCURRING CONDITIONS

Abstract

Sex differences in the age of autism diagnosis during childhood have been documented consistently but remain poorly understood. In this study, we used electronic health records data from a diverse, academic medical center to quantify differences in the age of autism diagnosis between boys and girls and identify associations between the age of diagnosis and co-occurring neurodevelopmental, psychiatric, and medical conditions. An established computable phenotype was used to identify all autism diagnoses within the Duke University Health System between 2014 and 2021. Co-occurring neurodevelopmental and psychiatric diagnoses as well as visits to specific medical and supportive services were identified in the two years prior to autism diagnosis. Cox proportional hazards models were fitted to quantify associations between diagnosis age and sex with and without controlling for the presence of each co-occurring diagnosis and visit type. Records from 1438 individuals (1142 boys, 296 girls) were included. Girls were more likely to be diagnosed either before age 3 (χ²=497.720, p<0.001) or after age 11 (χ²=4.014, p=0.047), whereas boys were more likely to be diagnosed between ages 3 and 11 (χ²=5.532, p=0.019). Visits for anxiety (χ²=4.200, p=0.040) and mood disorders (χ²=7.033, p=0.008) were more common in girls, and associated with later autism diagnosis (HR=0.615, p<0.001; and HR=0.493, p<0.001). Visits for otolaryngology were more common in boys, and associated with earlier autism diagnosis (HR= 1.691, p<0.001). After controlling for these conditions, associations between sex and diagnosis age were reduced and not statistically significant. These results show that the age of autism diagnosis differs in girls compared to boys, but these differences were neutralized when controlling for co-occurring neurodevelopmental and psychiatric conditions prior to autism diagnosis. Understanding sex differences and the possible mediating role of other diagnoses may suggest targets for intervention to promote earlier and more equitable diagnosis.

LAY SUMMARY

Sex-related differences in the age of autism diagnosis remain poorly understood. In this study, we used electronic health record data to quantify these differences in a large, diverse, academic medical center and determine whether they were associated with the presence of co-occurring neurodevelopmental, psychiatric, and medical conditions in the two years preceding autism diagnosis. Results show that girls are diagnosed later than boys, and that these differences are partly explained by sex differences in rates of co-occurring neurodevelopmental and psychiatric diagnoses.

1. Introduction

The Centers for Disease Control estimate that 1 in 36 children received an autism diagnosis in 2018, an over 3-fold increase over estimated rates in 2002.¹ Diagnosis is approximately 4 times as common in boys than in girls², and boys are more likely than girls to be diagnosed before age 5 even after controlling for levels of autistic features.³ Another study similarly found that girls are diagnosed at a later age on average, with a median age of identification of 6.1 years compared to 5.6 years⁴. Conflicting findings have been observed, with some studies suggesting that girls tend to be evaluated and diagnosed earlier than boys² or at approximately the same age (around 5 years).⁵

Although sex-related differences in autism assessment and the age of diagnosis are increasingly recognized, further study is needed to understand and quantify these differences. Some studies suggest that disparities in diagnosis rates and age at diagnosis stem from sex differences in the clinical manifestations of autism. For example, sex-associated disparities in core autism features have been observed^6,7, and a study by Lai et al.⁸ suggested that in girls, autistic features and the need for supportive services often arise at a later developmental stage (e.g., in adolescence). Additionally, girls may be more likely to use specific, adaptive coping strategies that disguise their autism-related behaviors.⁹ There is also evidence of bias in commonly used screening and assessment measures,¹⁰ which may be less sensitive to autistic features in girls compared to boys as the use of clinical tools designed to fit the male autism phenotype.¹¹ Finally, provider or parent biases may be an important contributor, and it is unclear to what degree the hypothesized “female protective effect” reflects true biological differences between sexes versus differences in the rates at which autistic features are recognized.¹² In girls, a greater number of behavioral or cognitive problems tend to be present before a diagnosis of autism is made, which may suggest that these problems can overshadow the autism diagnosis.¹³ Each of these mechanisms may contribute to unequal diagnosis rates and the age of diagnosis between boys and girls, and their relative importance is uncertain.

Sex-related differences in the age at which autism is diagnosed may also stem from broader differences in health-related trajectories between boys and girls. Strong associations between the existence of a co-occurring condition and age at first diagnosis of autism have been reported.¹⁴ Autism and co-occurring neurodevelopmental and psychiatric conditions are often diagnosed closely after each other, suggesting that diagnosis of one condition may lead to recognition of others.¹⁴ Autism is associated with higher rates of a wide range of psychiatric and neurodevelopmental conditions including mood disorder, anxiety, and attention-deficit hyperactivity disorder (ADHD); and 20% of autistic children have two or more co-occurring psychiatric conditions.¹⁵ ADHD is the most common of these conditions in children with autism; followed by specific phobia (SPH), obsessive compulsive disorder (OCD), and oppositional defiant disorder (ODD).¹⁶

Importantly, population-level analysis has shown that co-occurring psychiatric conditions are more common in girls,¹² suggesting that understanding the impact of co-occurring diagnoses on the diagnostic process may be key to understanding and mitigating sex-related differences. Whereas boys with autism are at greater risk for externalizing disorders such as ADHD and oppositional defiant disorder, girls are at especially high risk for internalizing psychopathology.¹⁷ In adolescence specifically, the prevalence of anxiety disorder is lower in boys than in girls,¹⁸ and girls with autism appear to be more vulnerable to internalizing problems due to a combination of genetic, hormonal, and psychosocial factors.¹⁹ In contrast, externalizing disorders as well as behavioral characteristics such as stereotypic and disruptive behavior are more prominent among boys, which may increase the likelihood of clinical assessment.²⁰

While less explored compared to the impact of co-occurring psychiatric conditions, sex differences in rates of medical conditions and referral to specific medical services may further contribute to differences in the age of autism diagnosis. Autistic children often receive physical, occupational, and speech therapy, and they also visit specific clinical or medical services for possible interventions at higher rates,^21,22 all of which may help signal the need for autism evaluation.

Understanding the relationships between the age of autism diagnosis and the presence of co-occurring conditions is challenging partly due to the difficulty of obtaining reliable information about the timing of both the diagnosis autism and of co-occurring psychiatric diagnoses. In their analysis of autism diagnosis age based on the Autism and Developmental Disabilities Monitoring Network, for example, Shattuck, Paul T et al.³ pointed out that the age of autism identification cannot be directly confirmed because information about participants’ diagnostic history may be incomplete. Similarly, information about psychiatric comorbidities is based on parent-report and may therefore be incomplete or biased.²² Across a range of clinical applications, the use of electronic health record (EHR) data has led to reduction in medical errors and made improvements in patient-level measures that document diagnoses and services.²³ Although EHRs are not free from bias, in many patients they provide a more complete picture of children’s health trajectories, which in turn can facilitate further investigation on differences between boys and girls in the autism diagnostic process.

This study aimed to examine sex-associated differences in the age of autism diagnosis by analyzing large-scale EHR data from a large, diverse academic medical center. We also sought to understand how differences in the age of diagnosis relate to the presence or absence of specific co-occurring psychiatric conditions prior to autism diagnosis, which might in turn mediate differences in the age of diagnosis between boys and girls. Finally, we also explored associations between diagnosis age and other co-occurring medical conditions and the utilization of common autism-related services. By understanding how these events contribute to sex differences in the age of first autism diagnosis, we hope to provide targets for intervention to promote earlier and more equitable diagnosis.

2. Methods

2.1. Data extraction and Autism cohort identification

Study procedures were approved by the Duke Health Institutional Review Board. Participant consent was waived due to the minimal risk posed by the study procedures and infeasibility of obtaining consent in a large retrospective cohort. EHR data were extracted from Duke’s Clinical Research Datamart (CRDM), which provides access to curated and characterized Duke patient data within the Duke University Health System (DUHS), a large academic medical center based in Durham, North Carolina. DUHS provides care to a substantial majority (approximately 85%) of children in the City of Durham and surrounding Durham County, which has diverse racial, ethnic, and socioeconomic composition.²⁴

The CRDM includes data for all Duke patients beginning in 2014 organized into 22 source tables that support retrospective data analysis, and contains diagnostic data, patients’ detailed demographic and Duke specific clinic information. All data were analyzed within the Duke Protected Analytics Computing Environment (PACE) which is a highly protected virtual network space for analyzing identifiable EHR data.

The initial cohort included all patients with autism-related ICD-10-CM diagnosis (F84.0, F84.5, F84.8 or F84.9) associated with encounters before 1st June 2021. Following a computable phenotype previously investigated at DUHS,²⁵ our inclusion criteria were (a) diagnosis codes related to autism associated with ⩾2 encounters; (b) ⩾1 documented encounter of any type at least two years prior to the first autism code; and (c) an observed autism diagnosis code before age 18. Criterion (a) was designed to ensure patients received routine care within DUHS for a substantial period of time prior to autism diagnosis. This was important to increase the likelihood that psychiatric, neurodevelopmental, and medical conditions prior to diagnosis were identified. For children diagnosed prior to age 3 years, the criterion implies that the child was seen within DUHS within their first year of life. Criterion (b) was designed to ensure that the first occurrence of an autism diagnosis code reflects the age of autism diagnosis rather than the age at which the child first received autism-related services at DUHS. Criterion (c) was designed to limited the age interval from childhood through adolescence. Sensitivity analyses were conducted to ensure results were robust to small changes in these criteria, including relaxing requirement (b) to require ⩾1 documented encounter at least one year (rather than two years) prior to the first autism code.

For each individual meeting inclusion criteria, we then identified co-occurring psychiatric conditions, autism-related services, and medical conditions taking place within a 2-year window prior to first autism diagnosis (i.e. first occurrence of an autism-related ICD code). All conditions were identified using computable phenotypes based on billing codes from the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) and Tenth Revision, Clinical Modification (ICD-10-CM) (see sTable 1 for details). Anxiety disorder, mood disorder, deficit hyperactivity disorder (ADHD), conduct disorder, intellectual disability, genetic neurodevelopmental conditions, and other neurodevelopmental conditions were selected as common psychiatric comorbidities related to autism diagnosis. Speech therapy, audiology and physical/occupational therapy were determined to be the most relevant clinical services. Outpatient visits to neurology, ophthalmology, otolaryngology, gastroenterology, cardiology and endocrinology were identified as related medical services, as they have previously been shown to be associated with later autism diagnosis.²⁶ Participants’ use of these services was documented when it occurred within DUHS, which includes a large system of outpatient facilities in and around Durham County, including ⩾30 primary care locations.

2.2. Statistical Analyses

Sex differences in demographic variables (Race, Ethnicity) were assessed via Chi-square test. The age of autism diagnosis was defined as the age (in years) when the autism-related diagnosis code was observed. Sex differences in age at autism diagnosis were first assessed by Mann-Whitney U test. Cox proportional hazards models were fitted to assess the effect of sex on diagnosis age while controlling for race (Black or African American, Asian, White, Other race), and ethnicity (Hispanic or Latino, not Hispanic or Latino, refuse to answer). Cox proportional hazards and models and Kaplan-Meier curves were estimated using the “coxph” and “survfit” functions, respectively, from the R package “survival”.²⁷

Analyses then focused on the occurrence rates of each co-occurring psychiatric condition, neurodevelopmental condition, and clinical and medical service. Differences between boys and girls in the rates of these conditions and use of these services were examined by Chi-square test. The effect of each of these factors of interest on the age of autism diagnosis was assessed by fitting additional Cox proportional hazards models that included that factor, as well a term to quantify its potential interaction with sex, while controlling for the same demographic factors. Based on the number of individuals within each sub-group of sex, race and ethnicity, the most common sub-groups were selected as the reference groups. Log-likelihood ratio tests were used to compare nested models (e.g., with and without interaction terms), and differences between boys and girls for each event was assessed by including the interaction between them.

Summary tables and forest plots were generated to display the hazard ratios, 95% confidence intervals, and relevant p-values. The p-value for each model parameter was obtained by testing the null hypothesis that this hazard ratio is 1 with a two-tailed z-test. Raw p-values have been reported throughout, but statistical significance was determined by applying Bonferroni correction with an initial threshold of α = 0.05.

A study conducted by Tein and MacKinnon explored the possibility of conducting mediation analysis with survival data under Cox proportional hazards models.²⁸ In our case, other diagnoses and medical services may act as mediators between sex and diagnosis age. While mediation analysis with time-to-event outcomes is still underdeveloped,²⁹ our analyses report the difference in coefficients associated with sex for models that include versus exclude possible mediating variables of interest, and as such can be interpreted as a form of mediation analysis.

3. Results

3.1. Description of cohort

A total of 7916 unique patients had at least one encounter associated with an autism spectrum disorder diagnosis code before 1^st June 2021, and 5118 had at least two such encounters, thus satisfying our computable phenotype. Of these, 1819 had at least one encounter of any type two years prior to first autism diagnosis, and 381 of these patients were excluded because they were >18 when first diagnosed with autism, leaving 1438 patients diagnosed with autism who met criteria. Demographics by group are presented in Table 1.

Table 1:

Participant demographics and age at autism diagnosis

Variable	Value	Male	Female	p-value
Total	N	1142	296
Race	White	601 (52.6%)	155 (52.4%)	0.00508
	Asian	25 (2.2%)	18 (6.1%)
	Black or African American	337 (29.5%)	83 (28.0%)
	Other	179 (15.7%)	40 (10.8%)
Ethnicity	Hispanic	118 (10.3%)	32 (14.8%)	0.875
	Not Hispanic	962 (84.2%)	246 (83.1%)
	Refuse to answer	62 (5.4%)	18 (6.1%)
AGE	Mean (SD)	8.28 (4.44)	8.70 (4.71)	0.251
	Median [Min, Max]	7.46 [2.00, 18.0]	7.82 [2.00, 17.9]

Among patients with autism, 1142 (79.4%) were boys and 296 (20.6%) were girls. Co-occurring anxiety disorder appeared in 209 (14.5%) of the patients prior to first autism diagnosis, whereas 41 (2.9%) had mood disorder, 183 (12.7%) had ADHD, 87 (6.1%) had conduct disorder, 39 (2.7%) had intellectual disability, 401 (27.9%) had other neurodevelopmental conditions, and 8 (0.6%) had genetic neurodevelopmental conditions, respectively, co-occurring with autism.

There were 12 patients with no outpatient encounters for autism-related services. Among the remaining 1426 (1438-12=1426), the ratio between boys and girls was 3.85. There were 217 (15.2%) patients with documented encounters with speech therapy, 352 (24.7%) with audiology, and 442 (31.0%) with physical or occupational therapy (PT/OT). Among the medical services, 237 (16.6%) patients had encounters with neurology compared to 310 (21.7%) with ophthalmology, 191 (13.4%) with otolaryngology, 143 (10.0%) with gastroenterology, 117 (8.2%) with cardiology, and 88 (6.2%) with endocrinology. These results were robust when conducting sensitivity analyses to explore the impact of our inclusion criteria.

3.2. Sex differences in age at autism diagnosis

Differences between boys and girls in the age of autism diagnosis were statistically significant (U = 1024, p< 0.001), with boys being diagnosed earlier, but this association was reduced to a trend (see Table 2) when controlling for race and ethnicity (HR=0.897 (0.788, 1.020); p=0.10). A more detailed analysis of the age of diagnosis by year (see Figure 1) revealed that girls were more likely to be diagnosed either before age 3 (χ²=497.720, p<0.001) or after age 11 (χ²=4.014, p=0.047), whereas boys were more likely to be diagnosed between ages 3 and 11 (χ²=5.532, p=0.019). Kaplan-Meier analysis (see Figure 2) illustrated that the Kaplan-Meier curves cross between ages 3.5 and 4, confirming that around this age that there is a transition from autistic girls being more likely to be diagnosed to autistic boys being more likely to be diagnosed.

Table 2:

Results of the multivariate cox proportional hazards model using variables Sex, Race, and Hispanic Ethnicity.

Characteristic	HR	95% CI	p-value
Race
White	-	-	-
Asian	0.930	(0.679, 1.274)	0.7
Black or African American	1.295	(1.149, 1.460)	<0.001
Other	1.351	(1.123, 1.626)	0.001
Ethnicity			0.875
Not Hispanic	-	-	-
Hispanic	1.211	(0.89, 1.489)	0.071
Refuse to answer	0.950	(0.746, 1.209)	0.7
SEX
Male	-	-	-
Female	0.897	(0.788, 1.020)	0.10

Figure 1:

Frequency of age of autism diagnosis by sex.

Figure 2: Age of autism diagnosis by sex.

Kaplan-Meier survival curves of time to autism diagnosis among patients meeting inclusion criteria. A 95% confidence interval (estimated from the log hazard) is presented in the shaded area.

3.3. Sex differences in rates and age of co-occurring psychiatric and neurodevelopmental conditions

Rates of co-occurring psychiatric and neurodevelopmental conditions preceding autism diagnosis differed between girls and boys (Figure 3). More autistic girls were diagnosed with anxiety disorder, mood disorder, ADHD, intellectual disability, and genetic neurodevelopmental conditions. Among these conditions, a trend was observed for anxiety disorder (χ²=4.200, p=0.040) and a statistically significant difference was observed for mood disorder (χ²=7.033, p=0.008).

Figure 3: Rates of psychiatric and neurodevelopmental conditions by sex.

The bar graphs compare rates of psychiatric ((Anxiety Disorder, Mood Disorder, Conduct Disorder)) and neurodevelopmental ((ADHD, Intellectual Disability, Genetic Neurodevelopmental Condition, Other Neurodevelopmental Condition)) conditions documented prior to autism diagnosis between boys and girls.

Among all multivariate Cox PH models, likelihood ratio tests indicated statistically significant interactions with sex were observed only for co-occurring genetic neurodevelopmental conditions (p=0.008), which accelerated diagnosis in girls only (HR=12.229, p<0.05). Therefore, this interaction term was not included in the final Cox PH models for anxiety disorder, mood disorder, conduct disorder, ADHD, intellectual disability, and other neurodevelopmental conditions. Forest plots (Figure 4) provided a visual representation of the hazard ratios, their corresponding confidence intervals, p-values for each psychiatric and neurodevelopmental condition, race, sex, and interactions with sex within Cox proportional hazards models predicting the age of autism diagnosis. Forest plots indicated that the diagnosis of anxiety disorder (HR=0.615, p<0.001), mood disorder (HR=0.493, p<0.001), and ADHD (HR=0.608, p<0.001) delayed the age of autism diagnosis, on the contrary intellectual disability (HR=1.883, p<0.001), and other neurodevelopmental conditions (HR=2.317, p<0.001) accelerated the diagnostic process while controlling for sex, race, and Hispanic ethnicity. After controlling for co-occurring conditions, associations with sex were no longer statistically significant.

Figure 4: Forest plot.

Forest plot showing hazard ratios, confidence intervals, and P-values associated with the occurrence of each psychiatric (Anxiety Disorder, Mood Disorder, Conduct Disorder) and neurodevelopmental (ADHD, Intellectual Disability, Genetic Neurodevelopmental Condition, Other Neurodevelopmental Condition) condition, race, sex, and interactions with sex in Cox proportional hazards models predicting the age of autism diagnosis.

3.4. Sex differences in rates and age of autism-related clinical and medical services

Statistically significant differences in rates of girls versus boys with documented encounters to ophthalmology (χ²=11.736, p<0.001) and endocrinology (χ²=5.168, p=0.023), but not other medical or autism-related services included in our analysis (see Figures 5–6).

Figure 5: Rates of outpatient visits to autism related clinical services.

The bar graphs compare rates of referral to each clinical service (Audiology, PT/OP, Speech Therapy) between boys and girls diagnosed with autism.

Figure 6: Rates of outpatient visits to autism related medical services.

The bar graphs compare rates of referral to each medical service (Cardiology, Endocrinology, Gastroenterology, Neurology, Ophthalmology, Otolaryngology) between boys and girls diagnosed with autism.

Cox PH models assessing the relationship between encounters with autism-related clinical and medical services revealed statistically significant interactions only between sex and cardiology (p=0.005), therefore interaction terms with sex were omitted in all other models. Forest plots (Figure 7 and 8) provided a visual representation of the hazard ratios, their corresponding confidence intervals, and p-values associated with each type of clinical and medical visit and its interaction with sex in Cox proportional hazards models predicting the age of autism diagnosis. Encounters with all clinical services including speech therapy (HR=1.945, p<0.001), audiology (HR=2.842, p<0.001) and physical and occupational therapy (HR=1.351, p<0.001) did accelerate the diagnosis of autism significantly when controlling for sex, race, and Hispanic ethnicity. Similar results were observed for otolaryngology (HR=1.691, p<0.001), whereas associations with ophthalmology and gastroenterology were not statistically significant, and both HRs were close to 1. By contrast, encounters with neurology (HR=0.800, p=0.002), cardiology (HR=0.816, p=0.067) and endocrinology (HR=0.660, p<0.001) were associated with delays in diagnosis. Particularly among patients referred to cardiology, girls were diagnosed earlier than boys. In summary, visits to autism-related clinical services (speech therapy, audiology, and physical and occupational therapy) were associated with earlier diagnosis, whereas results for medical services were mixed: otolaryngology visits were associated with earlier diagnosis, while visits to neurology, cardiology, and endocrinology were associated with delays. Importantly, however, differences in diagnosis age between boys and girls remained after taking these outpatient visits into account, with girls diagnosed later than boys.

Figure 7: Forest plot.

Forest plot showing hazard ratios, confidence intervals, and P-values associated with outpatient visits to related clinical services (Audiology, PT/OP, Speech Therapy) in Cox proportional hazards models predicting the age of autism diagnosis.

Figure 8: Forest plot.

Forest plot showing hazard ratios, confidence intervals, and P-values associated with the occurrence of each outpatient medical visits (Cardiology, Endocrinology, Gastroenterology, Neurology, Ophthalmology, Otolaryngology), race, and sex in Cox proportional hazards models predicting the age of autism diagnosis.

4. Discussion

This study investigated differences in the age at which autism is diagnosed between boys and girls and associations between this age and the presence of co-occurring psychiatric conditions, medical conditions, and the use of autism-related services. Our results confirm that in a large, diverse academic medical center, girls are diagnosed later than boys. However, this relationship disappears when controlling for co-occurring psychiatric conditions and related health services. These services, in turn, are strongly associated with sex, which may suggest that differential rates of co-occurring diagnoses and clinical services may contribute to observed sex differences in the age of autism diagnosis. Our analyses were based on large-scale EHR data extracted over an extended collection window, which provided us with increased statistical power to identify sex-related differences in the diagnostic process in a large, clinically and demographically diverse population.

Differences between boys and girls in the age of autism diagnosis were observed not only in adolescence, but also in childhood. In our cohort, boys were more often diagnosed between ages 3 and 11, but surprisingly, more girls were diagnosed before age 3. These results are consistent with the findings of Shattuck et al, which found a median diagnosis age 6.1 years in girls compared to 5.6 years in boys;³ however, our findings further suggest that this trend may vary by age, with girls being more likely to be diagnosed either very early or very late. Our results also suggest that the median age of diagnosis is later than age 6 in both sexes: 7.82 years for girls and 7.46 years for boys. This difference in the median age of diagnosis compared to the findings of Shattuck et al. may be partly attributable to advantages of our EHR-based design, which provides longer follow-up and allows us to capture diagnoses occurring at any age.

Motivated by findings that co-occurring psychiatric and neurodevelopmental conditions are associated with delays in autism diagnosis, a major emphasis of this work has been to determine whether sex disparities in age at diagnosis persisted in subgroups of autistic children previously diagnosed with anxiety, mood disorder, ADHD, conduct disorder, intellectual disability, genetic neurodevelopmental conditions, and other neurodevelopmental conditions. Our results confirm earlier results showing that these conditions are associated with differences in age at diagnosis; and further show that within these subgroups, associations with sex are neutralized, with hazard ratios near 1. This finding suggests that effects of sex may be mediated by diagnosis of other conditions, and it is consistent with hypothesis that features of autism are masked by co-occurring psychiatric conditions differently in girls versus boys.¹² In contrast, conduct disorder³⁰ and neurodevelopmental conditions more often seen in boys³¹ tend to accelerate rather than slow the diagnostic process. Overall, co-occurring conditions in boys appear to contribute to more prompt diagnosis, whereas those more common in girls contribute to delays. Moreover, the absence of statistically significant interaction terms suggest that effects of these conditions do not differ between sexes when they are present; rather, it is differences in rates of these conditions that are responsible for the overall effect.

Our analyses of autism related clinical and medical services provide further insight into the timing of diagnosis. Outpatient visits to all autism-related services (i.e., speech therapy, audiology, and physical and occupational therapy) as well as otolaryngology specialists were associated with accelerated autism diagnosis, suggesting a consistent finding that autistic children often have developmental delays at a very young age, manifested as delayed language or motor skills or hearing loss which lead to more prompt recognition of autism. However, in contrast to the psychiatric conditions, encounters with these types of clinical and medical services occur at similar rates in boys and girls. Thus, after taking these outpatient visits into account, differences between boys and girls remain, with girls diagnosed later than boys. In other words, even when girls are referred to autism associated services such as audiology, they are diagnosed later, which may suggest that sex differences or biases contributing to differences in age of diagnosis persist in these settings.

Interpreting these findings is challenging partly because we do not know how many diagnoses were missed, or whether the proportion of missed diagnoses differs in boys versus girls. Our EHR-based design is advantageous in this regard, as our extended observation window improves the likelihood that autism cases were recognized. However, the implications of these findings depend to some degree on the number of missed diagnoses, which in turn depends on the true difference in rates of autism between boys and girls. As an example, our results show that the proportion of known autism diagnoses identified before age 3 was higher in girls. However, this does not necessarily imply that the proportion of true autism diagnoses identified before age 3 is higher in girls, because the number of missed diagnoses may differ between the sexes.

Several of these findings are directly relevant to clinical decision-making regarding the need for autism evaluation. In particular, the association between delayed autism diagnosis and mood and anxiety diagnoses suggests that when making these diagnoses, it is important to consider the potential role of autistic features and possible need for autism evaluation. Further work is needed to understand to what degree this association indicates masking of autistic features by mood and anxiety disorders versus misdiagnosis of these conditions in autistic children.

5. Limitations

Our study has important limitations. Firstly, co-occurring psychiatric and neurodevelopmental diagnoses, medical diagnoses, and autism-related services were coded as present or absent in the two years preceding diagnosis, and the specific timing of these events, the number of codes and their pattern over time, and interactions between multiple co-occurring diagnoses were not considered due to limitations of the data source and sample size. As a result, further work is needed understand possible interactions between the age of autism diagnosis and the specific age at which co-occurring conditions are identified. This will become more feasible as additional EHR data are collected, which will allow analysis of longer-term trajectories of autistic children. Secondly, while our models incorporated demographic characteristics believed to be relevant, there may be confounding variables not observed or not included in our models that could have influenced our interpretation had they been included. Thirdly, all analyses are conducted using EHR data, which is subject to known biases in the data collection process that are difficult to quantify, including details of the reporting of race and ethnicity. In particular, our autism computable phenotype, while validated at DUHS, does not reflect gold standard autism diagnoses. Further, our criteria for identifying co-occurring psychiatric, neurodevelopmental, and medical conditions has not been validated, and in some cases may reflect provider concerns rather than diagnosis. Fourthly, the population of DUHS, while large and diverse, is not reflective of the US population as a whole, and some demographic groups are not adequately represented. Finally, details of encounters taking place outside of DUHS were not available. Our inclusion criteria were designed to limit our population to individuals receiving routine care within DUHS. This helped ensure we accurately captured age of diagnosis and co-occurring conditions, but it also limited our analyses to individuals with an established record of receiving care through DUHS, which is a subset of all patients receiving autism-related care within DUHS.

6. Conclusions

In a large, diverse, academic health system, the age of autism diagnosis differed between boys and girls. Whereas boys were more often diagnosed between ages 3 and 11, girls were more diagnosed before age 3 or after age 11. Co-occurring psychiatric and neurodevelopmental conditions as well as outpatient visits to specific medical services were also strongly associated with differences in the age of autism diagnosis. The inclusion of these factors individually in multivariable Cox PH models substantially diminished associations of diagnosis age with sex, suggesting that diagnoses and services received prior to autism diagnosis may partly mediate sex-associated differences in its timing. These results underscore the importance of understanding the reasons for sex differences and may suggest potential targets for intervention to promote earlier and more equitable diagnosis.

Data Availability Statement

This study is based on large-scale electronic health record data from the Duke University Health System, and we do not have permission or the necessary resources to make these data available to others.