Delay from Screening to Diagnosis in Autism Spectrum Disorder: Results from a Large National Health Research Network

Abstract

To examine delay from developmental screening to autism diagnosis, we used real-world healthcare data from a national research network to estimate the time between these events. We found an average delay of over 2 years from first screening to diagnosis, with no significant differences observed by sex, race, or ethnicity.

Autism spectrum disorder (ASD) is a developmental condition characterized by social communication challenges and restricted, repetitive behaviors and interests. Autism-specific services have been shown to improve significantly developmental outcomes for autistic children, particularly when implemented at early ages (1). Universal screening for autism and developmental delays aims to improve early detection and facilitate access to early intervention. The American Academy of Pediatrics (AAP) recommends universal autism-specific screenings at 18- and 24-month well-child visits, as well as developmental screening at 9-, 18-, and 30-month visits (2). Despite increased uptake of standardized screening tools (3), the median age at first ASD diagnosis has remained stubbornly stable between 4 and 5 years old (4). This discrepancy demonstrates the considerable time delay that many autistic children face from screening to diagnosis. Moreover, these delays disproportionately affect non-White, non-male autistic children, evidenced by longer, more onerous wait times from initial concern to diagnosis for Black and Latinx children compared with their White peers (5, 6) and for females compared with males (7).

While the practices of developmental screening and first ASD diagnosis have been examined separately (4, 8), few studies have reported specifically the length of time between screening and first diagnosis. A prolonged waiting period to ASD diagnosis is acknowledged as a barrier to timely access to services (9), yet it is still not well documented in the literature. Quantifying these delays is challenging, as such requires detailed longitudinal health records of autistic children over long observation periods, which may not be practically feasible among small samples or from only a few healthcare networks. In the current study, we leveraged a large health record dataset obtained from multiple US healthcare organizations (HCOs) to examine the delay in time from screening to ASD diagnosis and determine possible disparity patterns.

METHODS

Data Source

We obtained de-identified, patient-level health service encounter data from the TriNetX Research Network, which contains electronic health record (EHR) data for over 105 million patients from 67 HCOs. These HCOs are primarily large academic medical centers in the US (10). TriNetX data have been widely used in health service research (11–13), including for the pediatric population (14, 15). Details about the TriNetX platform and its data are described elsewhere (16). The dataset contained patients’ demographics (eg, sex, birth year, race, and ethnicity) and encounter-level information, including diagnoses (International Classification of Diseases [ICD]-9/10 diagnosis codes), and procedures (Current Procedural Terminology [CPT] and Healthcare Common Procedure Coding System [HCPCS] codes) through February 2022. Because this study used only de-identified patient records and did not involve the collection, use, or transmittal of individually identifiable data, this study was exempted from Institutional Review Board approval.

Study Sample

We constructed the study cohort based on the following inclusion criteria: (1) born between 2010–2014, (2) at least two separate encounters with an ASD diagnosis code under 8 years of age, and (3) at least one developmental screening or ASD-specific screening between the ages of 10 and 38 months, prior to receiving the initial ASD diagnosis (Table 1; online). The first inclusion criterion ensured that all included children were at least 8 years of age at the data cutoff time, allowing a sufficiently long follow-up period to capture an existing ASD diagnosis. The second criterion was consistent with the established approach for reliably identifying ASD cases from medical claims or EHR data that has been commonly employed in autism service research (17, 18).

The third criterion was designed to include conservatively all potential developmental screenings and ASD-specific screenings at 18-, 24-, and 30-month well-child visits following the AAP guidelines (2). Given that TriNetX data include only birth year for deidentification purposes, it was deemed not possible to determine the exact age in months at each encounter. Instead, we calculated a proxy age by considering July 1 of the child’s birth year as the birth date, following the same approach of the half-cycle correction from the decision-analytic modeling literature (19–21). Assuming dates of birth were uniformly distributed throughout a year, the proxy age equally underestimates and overestimates the exact age by up to 6 months for all individuals. Thus, the estimated mean proxy age is expected to approximate closely the mean of the exact age. Of note, using proxy age to approximate the mean exact age does not affect the calculation of the time interval between any two specific events, such as the period between screening and diagnosis (see Developmental Screenings and ASD Diagnosis Section). Given this, we limited the inclusion of screening encounters to those occurring between the cutoff proxy age of 10 months to 38 months. These conservative cutoff ages were determined by the earliest and the latest age for recommended developmental and ASD-specific screening (at 18-month and 30-month well-child visits, respectively), up to 2-month variability in scheduling the well-child visit, and the up to the 6-month margin of error for the proxy ages.

Developmental Screenings and ASD Diagnosis

We identified the following key events from the longitudinal health records of each child: (1) first developmental screening, (2) latest developmental screening before the first encounter with ASD diagnosis, and (3) first encounter with ASD diagnosis. If only one developmental screening was completed before the ASD diagnosis, the first screening encounter was also considered the latest screening encounter. If both ASD diagnosis and screening codes were recorded on the same day, we considered them a single screening encounter by omitting the ASD diagnosis code and assuming the subsequent encounter with an ASD diagnosis code was the initial ASD diagnosis encounter. We made this assumption primarily because, in practice, it is uncommon to diagnose a child definitively with ASD during a screening visit; rather, it is more likely that an ASD diagnosis code was recorded to note ASD-related concerns or a positive ASD screening result. To assess the potential impact of this assumption on our results, we performed a sensitivity analysis by excluding individuals with ASD screening and diagnosis codes at the same visit (6.9% of the total cohort). We further stratified the analyses for children with single and multiple screenings separately to explore the differences in ages at screenings and diagnosis by the number of screening encounters.

Delay in ASD Diagnosis

To examine the time lag in the ASD diagnosis process, we defined the primary outcome of our analysis, diagnostic delay, as the months (1) between the first developmental screening and the initial ASD diagnosis, and (2) between the latest developmental screening and the initial ASD diagnosis. Of note, although proxy ages were calculated for individual screening and diagnosis encounters due to the lack of the exact values of birth month and day in the data, calculations of the delay from screening to diagnosis were exact values, as they were not affected by the assumption of the birth date. For all outcome measures, summary statistics and 95% confidence intervals (CIs) were reported. Stratified estimates by patient characteristics were calculated to examine potential disparities in diagnostic delay. Statistical differences between subgroups were assessed using two-sample t-tests.

RESULTS

We identified 1,915 children from the TriNetX Research Network who met the inclusion criteria (Table 2; Figure 1, online). The resulting cohort exhibited a male-to-female ratio of 4:1, consistent with previous reports of diagnostic sex differences. Most children had the race (79.3%) data available, among which 56.4% were White, 37.1% Black, and 5.7% Asian. Among those (89.1%) with ethnicity data available, 25.5% were Hispanic. A growing number of children met the inclusion criteria with each additional birth year, with the most recent birth cohort comprising 41.0% of the total sample.

The proxy ages at the first and latest developmental screening encounters were estimated at 21.0 months (95% CI, 20.6–21.3) and 26.9 months (95% CI, 26.6–27.2), respectively (Table 3; online). Further, the proxy age at first ASD diagnosis was estimated to be 47.9 months (95% CI, 47.1–48.7). Nearly half (48.4%) had only one of the five recommended screenings documented prior to their ASD diagnosis (Table 4; online). The single-screening group was screened at an average age of 25.2 months (95% CI, 24.7–25.6), whereas the multiple-screening group was screened at an average age of 17.1 months (95% CIs, 16.7–17.4) and 28.5 months (95% CIs, 28.1–28.8) for the first and latest screening, respectively. However, the age of the first ASD diagnosis did not significantly differ between these two groups (47.5 months [95% CI, 46.4–48.7] for the single-screening group versus 48.3 months [95% CI, 47.2–49.4] for the multiple-screening group).

The average diagnostic delay was 26.9 months (95% CI, 26.1–27.8) from the first developmental screening and 21.0 months (95% CI, 20.2–21.9) months from the latest screening (Table 5). In the stratified analyses, delays to diagnosis from first and latest screening varied between 25.2 to 27.8 months since the first screening and 17.4 to 22.1 months since the latest screening across sex, race, and ethnicity subgroups, but none of the differences were statistically significant. Diagnostic delays did not significantly differ across most birth year cohorts except for the birth cohort of the year 2013. There were no clear patterns of increasing or decreasing delays over time (Figure 2; online). Sensitivity analyses showed similar estimates when excluding children with developmental screening and ASD diagnosis at the same encounter (Table 6; online).

DISCUSSION

This study is among the first to leverage EHR data drawn from a large number of HCOs to examine and document the period of time between screening and diagnosis among children with ASD diagnoses. Despite substantial improvements in the adoption of early screening practices (3), these results suggest that, on average, autistic children still experience a delay of over 2 years to receiving a diagnosis after undergoing developmental screening.

Our estimated average ages at screening and diagnosis closely approximate autism screening recommendations (ie, 18 and 24 months of age) (2) and previous findings about the average age of first diagnosis (approximately 4 years old) (4). The age of the first diagnosis did not differ significantly for children screened once versus multiple times, suggesting that multiple screenings may not have directly translated to an earlier diagnosis in this cohort. This finding also supports the need for timely follow-up actions after positive screenings or parent reporting of concerns.

Distributions of sex, race, and ethnicity in our study cohort were largely consistent with recent national autism prevalence reports (4). Among our cohort, we found no significant differences in the average delay time by sex, race, or ethnicity. These findings appear misaligned with much of the literature on racial and ethnic disparities in autism identification that suggest children from non-White backgrounds are disproportionately identified later (8, 22, 23). The absence of disparities in diagnostic delays should be investigated further to determine whether national gaps in the age of diagnosis are indeed closing for all children, as has been suggested by recent studies (4, 24), or rather if disparities exist but were not detected with these data, methods, and chosen metric (e.g., average age) (25). For example, using cumulative incidence as the metric, Shaw et al (8) found disparities for Black, non-Hispanic, and Hispanic children without intellectual disability, but Wallis et al (24) found no disparities in the age of first diagnosis. It is also possible that our cohort was not sufficiently representative of the distributions of race and ethnicity in the exact communities from which these data were drawn, r that small sample sizes of certain racial subgroups may have limited our ability to detect differences. In addition, we did not have access to other factors which may also have influenced time to diagnosis, such as the types of insurance and healthcare access, socioeconomic status, or urban versus rural settings (26–28).

On the other hand, differences in the length of time from developmental screening to diagnosis did reach significance between some of the birth year cohorts; however, there were no clear trends of increasing or decreasing delays over time. A longer delay was observed for the birth cohort of 2013 only, which is isolated and could be due to natural variation over time, particularly given varying sample sizes. Follow-up studies are warranted to examine further whether the time from screening to diagnosis improves over time, particularly for Black and Hispanic children as well as under-resourced children, known to experience delayed diagnosis.

There are several data limitations in our analysis. Information about individual screening encounters was limited such that there were no data about the specific screening tool used (ie, autism-specific or general developmental screening), the result of screening, or whether clinical referrals were made. Follow-up studies with these data available are warranted to determine if the specific clinical details of autism screening encounters (eg, impetus for screening, whether autism-specific screening was performed, and outcome of screening) are associated with the length of diagnostic delay in real-world settings. Additionally, for age-related estimates, our analyses were limited by the absence of the exact birth dates in the data; however, given the relatively large sample size and the assumption of uniformly distributed errors for underestimating or overestimating the exact age, these approximated age estimates can be interpreted with confidence but warrant further validation in other independent data sources. Lastly, our data did not include information regarding patients’ insurance type, socioeconomic status, and urban/rural residence, which are all potentially important factors affecting the access to health services needed for completing ASD diagnosis and can be used to elucidate disparities in the diagnosis delay outcomes. Future research should leverage EHR or all-payer claims data that include more detailed patient information in order to better understand the existing patterns and disparities in ASD early detection using more comprehensive measures, such as cumulative incidences beyond the point estimate of median (25).

In conclusion, our study examined healthcare encounters for a large sample of autistic children in the US and provides robust estimates of the length of the critical period between screening and the first diagnosis of ASD. Documenting this length of time specifically is a critical step in reducing the lengthy delay experienced by many autistic children. There are likely many factors that contribute to this delay, including systemic barriers to accessing services, limited availability of diagnostic services, parental hesitancy, and co-occurring developmental or medical conditions receiving clinical priority over autism concerns. Examination of possible factors that delay referrals and completing diagnosis was outside the scope of this multi-site EHR-based study but nevertheless is an important next step for future work. Establishing an estimate for the length of delay autistic children face from first screening to diagnosis should inform future efforts to understand and reduce this delay, in turn promoting earlier diagnosis, earlier access to intervention services, and optimal developmental outcomes.