Short Report: Transition to ICD-10 and the Prevalence of Autism in a Cohort of Healthcare Systems

Abstract

Currently, the prevalence of autism spectrum disorder (henceforth 'autism') is 1 in 36, an increasing trend from previous estimates. In 2015, the U.S. adopted a new version (ICD-10) of the World Health Organization coding system, a standard for classifying medical conditions. Our goal was to examine how the transition to this new coding system impacted autism diagnoses in ten healthcare systems.

We obtained information from electronic medical records and insurance claims data from July 2014 through December 2016 for each healthcare system. We used member enrollment data for 30 consecutive months to observe changes 15 months before and after adoption of the new coding system.

Overall, the rates of autism per 1,000 enrolled members was increasing for 0–5-year-old before transition to ICD-10 and did not substantively change after the new coding was in place. There was variation observed in autism diagnoses before and after transition to ICD-10 for other age groups.

The change to the new coding system did not meaningfully affect autism rates at the participating healthcare systems. The increase observed among 0–5-year-olds is likely indicative of an ongoing trend related to increases in screening for autism rather than a shift associated with the new coding.

Introduction

The current prevalence estimates of autism spectrum disorder (henceforth ‘autism’) among 8-year-old children is 1 in 36, an increase from 1 in 44 in 2018 (Maenner, Warren, Williams, et al., 2023; Maenner, Shaw, Bakian, et al., 2021). The autism diagnosis rate among children has been significantly increasing given comparable 2016 estimates of 1 in 54 children aged 8 years (Maenner, Shaw, Baio, et al., 2020), engendering growing concerns about possible causes. Some scholars have suggested that these increases are likely not due to increased screening alone but may represent true increases in prevalence (Rice, Rosanoff, Dawson, et al., 2012). Autism is characterized as a complex neurodevelopmental condition that impacts behavior, communication, and social interaction. The case definition of autism includes International Classification of Diseases (ICD) coding, a World Health Organization standard used to classify medical conditions. The United States transitioned to a new coding system, ICD Tenth Revision (ICD-10), in October 2015 after 35 years of using the ICD Ninth Revision, Clinical Modification (ICD-9) to comply with coding standards needed for healthcare services reimbursement (Centers for Medicaid and Medicare Services, 2014).

There have been concerns that adoption of the ICD-10 poses significant challenges to measuring changes in rates of health conditions over time within a healthcare system or population of interest (Yoon and Chow, 2017) because at least a third of ICD-9 codes do not have a corresponding ICD-10 code (Boyd et al., 2013). Some studies have stated the documentation for causes of death, suicidality, and birth defects before and after ICD-10 transition all showed changes attributable to the coding changes rather than actual changes in these health outcomes (Stewart, Crawford, and Simon, 2017; Salemi et al., 2019 and Anderson, Minino, Hoyert, et al., 2001). To our knowledge, no study has assessed whether the transition to ICD-10 coding affected rates of autism diagnosis over time. Therefore, the goal of this analysis was to explore the impact of the ICD-10 transition on autism diagnosis rates in a cohort of ten healthcare systems in the Mental Health Research Network (MHRN).

Methods

Autism diagnoses were extracted from a federated virtual data warehouse containing electronic medical records and insurance claims data (Ross et al., 2014) from July 2014 through December 2016 at each of ten participating sites of the MHRN: HealthPartners (Minnesota), Harvard Pilgrim Health Care (Massachusetts), Henry Ford Health System (Michigan), Baylor, Scott & White Health (Texas) & six Kaiser Permanente regions – Colorado, Georgia, Hawaii, Southern California, Washington and Northwest (Oregon/Southwest Washington). Participating healthcare systems serve an annual population of over 8 million members and reflect the demographic diversity of their geographic areas (Mental Health Research Network, 2015). Institutional Review Boards at each health system approved waivers of consent for this research use of aggregated data. There was no community involvement for this reported work.

Prior to the transition, ICD-9 codes for autism were classified as Autistic disorder 299; autism spectrum disorder 299.0, 299.00, 299.01; childhood disintegrative disorder 299.1, 299.10, 299.11; other specified pervasive developmental disorder 299.8, 299.80, 299.81; and unspecified pervasive developmental disorder 299.9, 299.90 & 299.91, grouped into distinct diagnostic codes, based on symptoms presented and subclassified to active or residual states. The revised ICD-10 autism diagnostic alphanumeric codes are simplified to the following classifications: autistic disorder F84.0, Asperger’s syndrome F84.5, childhood disintegrative disorder F84.3, pervasive developmental disorder, specified F84.8 or unspecified F84.9, without subclassifications of active or residual states. Definitions of autism conditions for ICD-10 were developed using the ICD-9 definitions used in MHRN and mapping tools from the Centers for Medicare and Medicaid Services, available from https://github.com/MHResearchNetwork/Diagnosis-Codes.

Statistical Analysis

For each site, monthly autism diagnosis rates, not differentiated into specific types, were calculated as the total number of enrolled members receiving at least one diagnosis of autism during that calendar month divided by the total number of members enrolled during that calendar month. This rate was then standardized to number of diagnoses per 1,000 enrolled members. Monthly time series rates were constructed by specific age ranges and by site. A total of 30 observations were available for analysis (15 months before and after the transition). Two sites (Sites 8 & 9) were missing enrollment data for Months 25 to 30 and Months 28 to 30, respectively in the post-ICD-10 period. To analyze the impact of the transition, we utilized a quasi-experimental approach, interrupted time series analysis (Penfold and Zhang, 2013). A segmented regression model was fit to each time series resulting in 4 model parameters of interest: (1) average value of the time series at the start of the observation period, i.e., the intercept; (2) monthly change in the autism diagnosis rate pre-ICD-10 conversion, i.e., baseline slope ; (3) shift in the average value of the time series immediately following the conversion to ICD-10, i.e., level change; and (4) change in slope following ICD-10 conversion. Given the correlated nature of these monthly rates, estimates were obtained using autoregressive error models. Model diagnostics were examined and included partial autocorrelation plots, white noise probability plots, and tests for autocorrelation. Analyses were performed with SAS Enterprise Guide software, Version 8.2 (SAS Institute Inc., Cary, NC), and statistical significance was assessed at the 0.05 level.

Results

Figure 1 and Table 1 show overall trends in autism diagnosis rates among the 4 age groups assessed. For 0-5-year-olds aggregated across all sites, the average autism diagnosis rates were 4.302 and 6.082 per 1,000 enrolled members, pre- and post-ICD-10, respectively. However, as shown in Figure 1a and Table 1; we observed an increasing trend in the monthly rate of diagnosis pre-ICD-10 for 0-5-year-olds (B = 0.109 (SE: 0.003); p < 0.0001). Post-ICD-10 transition, the observed rate of diagnosis did not significantly change and continued the same trajectory (ΔΒ = 0.005 (SE 0.004); p = 0.251). When examining trends within sites for the 0-5 age group (Supplemental Table 1), pre-ICD-10, all but one site (Site 5) showed similar, significantly increasing rates of autism diagnosis monthly. Following ICD-10 transition, rates of diagnosis varied across the 10 sites. Four sites (Sites 2, 3, 9 & 10) found no significant rate change suggesting that the transition to ICD-10 did not impact autism diagnosis rates. In contrast, 4 sites (Sites 4, 5, 6 & 8) showed an increase in the monthly diagnosis rate, and 2 sites (Sites 1 & 7) found a decrease in the monthly rate of autism diagnosis (although the overall trend was still positive).

Figure 1.

Overall Trends from July 2014 – Dec 2016 in Autism Diagnosis (Dx) Rates per 1000 Members in 10 Health Systems by Age Group

Table 1.

Summary of Parameter Estimates (per 1000 Enrolled Members) from Interrupted Time Series Models in 10 Health Systems

ERA	Parameter	Age 0 – 5 Years		Age 6 – 11 Years		Age 12 – 17 Years		Age 18+ Years
		Est. (SE)	P- value	Est. (SE)	P-value	Est. (SE)	P-value	Est. (SE)	P-value
Pre ICD-10 07/2014 – 9/2015	Intercept	3.430 (0.030)	<.0001	6.125 (0.064)	<.0001	4.046 (0.054)	<.0001	0.351 (0.004)	<.0001
	Trend (B)	0.109 (0.003)	<.0001	0.009 (0.007)	0.2245	−0.007 (0.006)	0.2324	−0.003 (0)	<.0001
Post ICD-10 Transition 10/2015 - 11/2016	Level change	0.030 (0.044)	0.5075	0.042 (0.091)	0.6521	0.216 (0.076)	0.0098	0.010 (0.006)	0.1198
	Trend change (ΔB)	0.005 (0.004)	0.2510	−0.012 (0.009)	0.2126	−0.031 (0.007)	0.0003	0.002 (0.001)	0.0024

Among children aged 6–11 years (Figure 1b), we observed no significant change in the overall autism diagnosis rate. Aggregated across all sites, the average autism diagnosis rates were 6.189 and 6.075 per 1,000 enrolled members, pre- and post-ICD-10 transition, respectively. For participating sites (Supplemental Table 1), we observed varying trends both pre- and post-ICD-10 transition. Only four sites, pre-ICD-10 (Sites 1, 2, 4 & 5) and post-ICD-10 (Sites 4, 5, 8 & 9), respectively, showed an increasing monthly autism diagnosis rate while other participating sites saw decreasing or no change in monthly autism diagnosis rate.

Among 12–17-year-olds (Figure 1c), the overall rate of autism diagnosis was 3.990 and 3.388 per 1,000 enrolled members, pre- and post-ICD-10 transition, respectively. Prior to ICD-10, the rate of autism diagnosis was stable with a relatively flat trend line. After transition to ICD-10, there was an immediate, but small increase in the diagnosis rate, followed by a significant decrease in the pre-ICD trend rate (ΔB = −0.031 (SE: 0.007); p = 0.0003). These results suggest the rate autism diagnosis rate started to decline after the transition to ICD-10. When examining trends in monthly autism diagnosis rate within sites (Supplemental Table 2), we observed variability in this age group too with 2 sites (9 & 10) and 3 sites (Sites 4, 5 & 10) showing increases in monthly autism diagnosis rate, pre-ICD-10 & post-ICD-10 respectively. Other participating sites found no change or had a decrease in the monthly autism diagnosis rate.

In the 18+ years group, we observed relatively low but similar rates of autism diagnoses during both eras. Rates of autism diagnosis per 1,000 enrolled members were 0.327 and 0.338, pre- and post-ICD-10 transition, respectively. Figure 1d shows a decreasing trend pre-ICD-10 (B = −0.003 (SE: 0); p = <0.0001) and an increase in the trend post-ICD-10 transition (ΔB = 0.002 (SE: 0.001); p = 0.0024) such that the overall change in the monthly diagnosis rate was essentially 0 after the transition. Similar to the other age groups, we observed site variation in rates of diagnosis (Supplemental Table 2). Monthly autism diagnosis rates increased for 2 sites (Sites 5 & 9) pre-ICD-10 and for 4 sites (Sites 1, 4, 7 & 8) post-ICD-10 with other sites showing no change or decreasing trend in the monthly autism diagnosis rate.

Discussion

Trends in the rates of autism diagnoses were not meaningfully different after the transition from ICD-9 to ICD-10 for any age group. Overall, autism rates for 0-5-year-olds across all sites showed an increasing trend prior to ICD-10 transition and no changes in trend were observed post-ICD-10. This suggests that the modest increases observed between the two eras are indicative of an ongoing trend rather than a shift associated with coding changes. That is, a gradual increase over the entire study period rather than an abrupt change at the time of the coding transition. The overall increasing rate in this age group is not surprising given the current emphasis on early screening to decrease the age at diagnosis to improve outcomes and quality of life for young children diagnosed with autism.

The current autism prevalence of 1 in 59 (17.0 per 1000) among children aged 4 years and 1 in 44 (23.0 per 1000) among children aged 8 years increased 9% and 24%, respectively, from previous estimates; this may be partly attributable to early intervention efforts (Shaw, Maenner, Bakian, et al., 2021). Also, States’ mandates of private health insurance coverage for some autism-related healthcare have made it possible for families to seek covered services (Chatterji, Decker and Markowitz, 2015; Choi, Knight, Stein and Coleman, 2020), and providers may be more likely to formally diagnose and recommend required services that families can afford through state mandated health insurance coverage. Furthermore, more research has been done to understand presentation and management of autism, while public awareness and advocacy efforts to reduce stigma have increased access to screening and services for kids and their families.

One site (Site 4) consistently showed accelerated rates of autism diagnosis across all age groups post-ICD-10, and that may be partly attributed to mandated state insurance for autism-related behavioral health services put into effect just prior to the ICD-10 transition. Other sites saw variation in autism diagnosis rates across all age groups. Variation observed in the older age groups pre- and post-ICD-10 across all sites could partly be due to small sample sizes. Alternately, it may be (1) young children who were able to access early intervention services no longer fit the diagnostic criteria due to improvements in their condition, (2) the previous diagnosis was a misdiagnosis during early screening, or (3) there are limited services available in the health care system for older children. Thus, if access to services serves as motivation for screening/diagnosis, receipt of an autism diagnosis would not provide any advantages. These additional explanations warrant future study.

One of the limitations of this assessment is that no stratified analysis of autism diagnosis code subclasses was completed to examine possible differences in subclass codes within age groups and/or across sites. Further, these analyses are based on data from integrated healthcare delivery systems and warranting caution when generalizing to other contexts. However, the study sample was large and geographically and racially/ethnically diverse.

Conclusion

We found no evidence of sudden changes in autism diagnosis before and after the transition to ICD-10 in all youth groups; the increases observed in the youngest age group are likely indicative of a more general trend toward increased early screening and intervention rather than a coding transition effect. Thus, our analysis demonstrates that comparisons of historical trends in autism can be made without confounding by the ICD-10 transition in these health systems.