Introduction

Autism Spectrum Disorder (ASD) is a heterogeneous condition where core features are expressed differently within a person across the lifespan and between individuals with the same diagnosis. Though studies of core symptom trajectories are few, most autistic individuals show characteristics of ASD across the lifespan (Giserman-Kiss & Carter, 2019; Guthrie, Swineford, Nottke, & Wetherby, 2013; Seltzer, Shattuck, Abbeduto, & Greenberg, 2004; Wiggins et al., 2012). In contrast, symptom presentations may change between childhood and adolescence (Baghdadli et al., 2018; Georgiades, Pickles, & Lord, 2021; King & Bearman, 2009). Methodologies vary, including assessment tools, sample sizes, cognitive abilities, and the amount of time elapsed between diagnosis and follow-up (Lord, Bishop, & Anderson, 2015; Simonoff et al., 2019; Woodman, Smith, Greenberg, & Mailick, 2014). Here, we examine the phenotypic trajectories of a longitudinal cohort between the ages of 2 and 25 (Lord, McCauley, Pepa, Huerta, & Pickles, 2020), stratified by best estimate diagnoses.

Bal, Kim, Fok, and Lord (2019) explored change in parent-reported social-communication from ages 2 to 19 in the same sample described in this paper. Participants showed decreases in social-communication difficulties on the Autism Diagnostic Interview-Revised (ADI-R), with language development and maturation contributing to trajectories of skills. Modest decreases of ADI-R social-communication impairments between childhood and adulthood have similarly been reported in other longitudinal samples (Gillespie-Lynch et al., 2012; McGovern & Sigman, 2005; Shattuck et al., 2007; Woodman et al., 2014). Relatedly, in a longitudinal sample spanning ages 5–15, low language and high autism symptoms were risk factors for low growth trajectories of socialization and communication on the Vineland (Baghdadli et al., 2012).

In a sample overlapping with this study, using direct measures of social-communication and RRBs in the Autism Diagnostic Observation Schedule (ADOS), symptom change between 2 and 15 years of age occurred in about 7% of participants who were assigned to an improving class and 9% in a worsening class (Gotham et al., 2012). Using a less cognitively impaired cohort, Georgiades et al. (2021) identified a larger subset of participants (27% of sample) with ADOS trajectories that continuously improved between initial preschool diagnosis and age 10. For the remainder of the sample, a “turning point” emerged when improvements plateaued around 6 years old.

Several studies have gone beyond these findings to examine individuals who change diagnostic status from clearly defined autism in childhood to an adult presentation of no symptoms of ASD (Billstedt, Gillberg, & Gillberg, 2005; Fein et al., 2013; Zachor & Ben-Itzchak, 2020). In a previous report of the present sample, 9% of study participants no longer met diagnostic criteria for ASD (not including history) by age 19 (Anderson, Liang, & Lord, 2014). Other longitudinal follow-up studies have reported similar proportions of previously diagnosed individuals, including those with lower cognitive abilities, who no longer show autistic core features in adulthood (Baghdadli et al., 2018; Billstedt et al., 2005; Howlin, Goode, Hutton, & Rutter, 2004; Mawhood, Howlin, & Rutter, 2000). The alternative, symptom worsening such that individuals gain a diagnosis later in development, is also possible (Ozonoff et al., 2018). Often, studies of first diagnoses in adulthood assume that an earlier childhood diagnosis was missed, but, often standardized assessments of ASD were never received.

Fein et al. (2013) retrospectively characterized a group of 34 individuals with prior early childhood community diagnoses of ASD, who no longer met diagnostic criteria. Participants had more early, intensive applied behavioral analysis relative to people who retained a diagnosis of ASD (Orinstein et al., 2014). We did not find this relationship between outcomes at 19 and intensive treatment services among our longitudinal, but quite different sample (Anderson et al., 2014). Rather, the only treatment variable associated with very positive outcomes was minimal but regular amounts of treatment of any type in the year after diagnosis at 2 (Anderson et al., 2014).

Given that there are few longitudinal studies examining diagnostic stability into adulthood, this study seeks to create trajectories of the core features of ASD between 2 and 25 years of age among individuals with different cognitive levels. We also set out to establish the diagnostic stability of individuals who received repeated, in-person, and diagnostic evaluations between early childhood and young adulthood. We expected that a majority of participants diagnosed with ASD in childhood would continue to meet current diagnostic criteria in adulthood (Anderson et al., 2014).

Methods Participants

Participants were part of the Early Diagnosis Study (EDX; Lord et al., 2020), a longitudinal cohort initially recruited in 1990 as consecutive community-referrals for a diagnostic autism evaluation (Figure S1 for CONSORT). Initial aims of the study were to determine if diagnoses of autism or developmental delay made under age 3 were stable across time. Participants were drawn from 3 locales in the United States. In early childhood (Mage = 2.5 years, SD = 0.43, Range = 1.25–3.33), 124 participants enrolled in North Carolina (n = 74) and Chicago (n = 50). An additional 31 participants (Mage = 9.14 years, SD = 2.49, Range = 7.75–15.33) with prior early diagnostic evaluations (<3 years old) were recruited in Michigan and entered the study at age 9 (see Pickles, McCauley, Pepa, Huerta, & Lord, 2020). They were then followed concurrently with the same frequency as the rest of the sample into adulthood. All 155 individuals included in this paper (80.65% male) were seen at entry and participated in at least one assessment battery in young adulthood (Mage = 23.46 years; SD = 3.40; Range = 17.58–30.08).

Since baseline, attrition occurred due to unreachable status and refusals (Table S1). Between study entry (n = 253, inclusive of Michigan recruits) and adulthood (n = 155), attrition constituted 38.74% of the sample, and was not significantly associated with gender, diagnosis, IQ at baseline, or rural/urban status (25% rural). It was significantly higher for Black families (79% White, 19% Black, 2% Other) and those with lower levels of education (54% of caregivers had college degrees). Most participants (78.71%) had a diagnosis of ASD at some point during the study; 21.29% had a history of developmental delay but no ASD.

Over half (55.48%) of the sample was less cognitively able as measured at age 19 (i.e., Verbal Intelligence Quotient [VIQ] < 70). As adults, less cognitively able participants (n = 86) had significantly lower VIQ scores (MVIQ = 24.58, SD = 16.51), Non-Verbal Intelligence Quotient [NVIQ] scores (MNVIQ = 31.45, SD = 21.77), and adaptive skills (MVABC = 40.36, SD = 18.30) than more cognitively abled participants (n = 69; MVIQ = 102.56, SD = 16.97; MNVIQ = 101.73, SD = 17.74; MVABC = 80.19, SD = 18.30), with a bimodal distribution. Mean adult ADOS CSS scores met autism spectrum clinical thresholds (CSS = 4) for less cognitively able (M = 5.57; SD = 2.14, Range = 1–10) and more cognitively able participants (M = 4.45, SD = 2.46, Range = 1–10).

Procedure

Face-to-face diagnostic assessments were conducted between 1990 and 2018 with timepoints corresponding to mean participant ages (2, 3, 5, 9, 19, 25). Assessments (Table S1) were administered by clinicians who were aware that many, but not all, participants had an autism diagnosis. Clinicians were research-reliable and administered diagnostic batteries without knowledge of previous test results, histories, or diagnoses, making the assessment blinded. Research was IRB-approved and consent/assent forms were obtained for participants.

Measures Diagnostic assessments

The ADOS-2 (Lord et al., 2012), is a clinician-administered measure which distinguishes behaviors of autism from those in the typical population. Calibrated severity scores from ADOS modules, determined by age and language level are comparable across modules (CSS; Gotham, Pickles, & Lord, 2009), including the Adapted-ADOS administered to minimally verbal adults (Bal et al., 2020), which uses calibrated scores from Modules 1 and 2. The ADI-R (Lord, Rutter, le Couteur, & Free Hospital, 1994) is a standardized parent-report interview of social behaviors, communication, and repetitive interests. Current ADI-R social-communication scores were generated from the nonverbal algorithm (14-items administered at all ages between 2 and 19; see Bal et al., 2019). Parent-reported treatment hours prior to age 3 were aggregated using procedures from Anderson et al., 2014.

Cognitive and adaptive assessments

NVIQ and VIQ were derived depending on the age and skills of the participant. Measures included the Wechsler Adult Intelligence Scale-IV (Wechsler, 2008), Wechsler Abbreviated Scale of Intelligence (Wechsler, 1999), Differential Ability Scales (Elliott, 2007), or the Mullen Scale of Early Learning (Mullen, 1995). Adaptive skills were assessed via the Vineland Adaptive Behavior Composite (VABC; Sparrow, Cicchetti, & Balla, 2005).

Social and behavioral assessments

The Aberrant Behavior Checklist-Community (ABC; Aman, Singh, Stewart & Field, 1985) measures behavior problems in individuals with developmental delay. The Social and Emotional Functioning Interview (Rutter et al., 1988) was clinician-administered independently to parents and more cognitively able participants. Information on aspects of adult functioning, including education, living status, driving status, and finances were gathered. The Well-Being Questionnaire (Ryff, 1989) was a self-report measure of personal growth, purpose in life, and self-acceptance. Adult outcomes were characterized by Pickles et al. (2020), who generated four empirically derived latent classes from a range of adult functioning measures for this sample (1- “Best Outcome Class”; 2-“High-IQ ASD”; 3-“Low-IQ ASD without Behavioral Problems”; 4-“Low-IQ ASD with Behavioral Problems”) and by McCauley and Sigman (2020) who conceptualized outcomes a priori as a count variable encompassing autonomy, social relationships, and purpose (range = 0–3).

Diagnosis

After each in-person assessment, a best estimate autism diagnosis was assigned by research-reliable clinicians who conducted in-person testing with multiple informants and reviewed all available testing information (including cognitive functioning, ADOS-2, ADI-R, adaptive functioning, and mood/behavior screens). If the final diagnosis conflicted with the ADOS-2 and ADI-R (current), senior researchers reviewed all records, watched the ADOS-2 recording and discussed the case with the examiners to reach a consensus diagnosis. Importantly, these were not DSM-5 diagnoses, where historical information can be included; diagnoses were solely based on current functioning using DSM-IV criteria at early ages and DSM-5 from age 19 on.

Data analysis

Participants were divided into four groups based on diagnostic impressions in adulthood. Of those who received an ASD diagnosis (autism or Pervasive Developmental Disorder-Not Otherwise Specified [PDD-NOS]) in early childhood, two groups were identified: Retained Diagnosis and Lost Diagnosis. Of those who had never received an ASD diagnosis in early childhood, an additional two groups emerged: Never Had Diagnosis and Gained Diagnosis. Descriptive statistics characterized individuals in each diagnostic category, separating those with VIQs above 70 or VIQs below 70. The two cognitive groups have a bimodal distribution and differing ability to self-report. Concerns most relevant to families and outcomes among those with average IQ and those with moderate–severe intellectual disability vary greatly. Group level differences were assessed via one-way ANOVAs or Chi-Square Tests.

Mixed models (MIXED in STATA 16) indicated the rate of change over time in ADOS CSS (total, social-communication, and RRB), ADI-R Social-Communication current, and VABC scores among individuals in the four diagnostic groupings. The marginal mean estimates of diagnosis group were compared at each timepoint. The linear mixed model flexibly handled participant missing data using maximum likelihood estimation. Diagnostic groupings were compared at initial starting point (intercept) and rate of change (linear slopes). Time was centered at zero (i.e., the start of the study when children were approximately 2-years old). All models included a subject level random intercept to account for the nonindependence of the data caused by repeated measures of the same individuals over time. Race and maternal education were included as covariates.

The beginning and endpoints of the trajectories over time were set by our best estimate diagnoses. Group differences on measures from the start and end of our study are to be expected because they contributed to the diagnoses that formed the categories. Therefore, of greatest empirical interest are the trajectories of autism symptoms between early childhood and adulthood.

Results More cognitively able group composition

Most participants (71.01%) with VIQ > 70 had stable diagnostic impressions from early childhood into young adulthood (Retained Diagnosis, n = 31 and Never Had Diagnosis, n = 18). For a minority of the sample (28.99%), diagnostic perceptions differed from what was assigned earlier in childhood (Lost Diagnosis, n = 13 and Gained Diagnosis, n = 7). Early in the study, PDD-NOS diagnoses for those with VIQ>70 was common (72%) and associated with the four diagnostic groups. It should be noted, however, that for the purposes of this paper, PDD-NOS and autism classifications were collapsed into a greater ASD diagnosis per DSM-5.

Thirteen individuals with VIQ > 70 comprised the Lost Diagnosis group, which was exclusively male and mostly White (84.62%). All 13 participants were diagnosed prior to age 3 (Mage = 2.51years; SD = 0.39). We can estimate when autistic symptom severity dropped below the clinical threshold by the first face-to-face visit when the diagnosis was not autism, recognizing that we did not do face-to-face assessments between 9 and 19 and that there were missing data in some cases. Given these limitations, our estimates are that two participants moved from autism to nonspectrum at the 5-year-old assessment, one participant was identified as nonspectrum at the 9-year-old assessment, four participants moved out of ASD between 9 and 19, and six participants were identified as nonspectrum between the 19 and 25-year-old assessments. Thus, nearly half (46.15%) of the individuals who moved out of ASD did so after age 18. We recognize that between the ages of 9 and 18, classification switched from DSM-IV-TR to DSM-5. Even though this period of time represented a change in DSM diagnostic criteria, the four individuals who were identified as nonspectrum (current) between 9 and 19 did not meet DSM-IV criteria for autism or PDD-NOS or DSM-5 criteria for ASD.

The Gained Diagnosis group (n = 7) was entirely male with the majority (85.71%) White. All received ASD diagnoses in adulthood, with five assigned this categorization for the first time at the 19-year-old assessment and the remaining two given the diagnosis at age 21. For all individuals in both the Lost and Gained diagnosis groups, once a changed classification was made, diagnoses were sustained in subsequent assessments.

See Table 1 for adult more cognitively able demographic characteristics. Residence in adulthood differed by diagnosis group, χ2 (1, 54) = 19.109, p < .001 such that those who retained their diagnosis lived with their parents significantly more (ps < .001) than those who never had a diagnosis or acquired a diagnosis in adulthood. 32.14% of the sample was still enrolled in college during their most recent adult assessment. When diagnostic groups were compared to Pickles et al. (2020) outcome groups, significance emerged (p = .007). However, individuals belonging to Class 1 (i.e., least ASD symptomology, high IQs, and good functional/behavioral outcomes) were found in each diagnostic group (Retained Diagnosis = 37.5%, Never Had Diagnosis = 54.54%, Lost Diagnosis = 100%, Gained Diagnosis = 50%). Using McCauley et al.s' (2020) definition, outcomes were also related to diagnostic group (p = .003). All of the Lost Diagnosis group achieved at least 2 of the three outcomes (i.e., independent living, employment, and friendships), whereas 36% of the Retained, 54.54% of the Never Had, and 50% of the Gained Diagnosis groups had. Thus, while good outcomes were more frequent in those who no longer had autism symptoms, some participants who retained or even gained autism diagnoses also were doing well.

Table 1. Adult demographic variables (VIQ > 70; n = 69) Sample characteristics Retained diagnosis (n = 31) Lost diagnosis (n = 13) Never had diagnosis (n = 18) Gained diagnosis (n = 7) Unadj. p-value Male (%) 93.55 100 61.11 100 .002 Site (%) North Carolina 48.39 46.15 38.89 14.29 <.001 Chicago 32.26 53.85 5.56 0 Michigan 19.35 0 55.56 85.71 Urban 74.19 80 66.67 100 .363 Rural 25.81 20 33.33 0 Race (%) White 74.19 92.31 88.89 85.71 .105 Black 25.80 7.69 5.56 0 Other 0 0 5.56 14.29 Cognitive ability (M, SD) VIQ 101.36 (17.37) 113.33 (12.44) 94.59 (16.00) 108.29 (15.61) .019 NVIQ 100.11 (15.97) 114.09 (11.80) 98.94 (20.50) 95.29 (19.50) .071 Autism features (M, SD) CSS Total 6.04 (2.01) 1.92 (0.76) 3.06 (1.95) 6.14 (0.90) <.001 CSS Soc. Affect 6.07 (2.09) 2.31 (1.32) 3.71 (2.34) 6.43 (1.40) <.001 CSS RRB 7.03 (1.64) 3.54 (2.50) 3.47 (2.45) 6.14 (2.54) <.001 ADI-R Sociala 17.63 (8.13) 11.17 (7.47) 5.50 (3.72) 10.14 (6.47) <.001 ADI-R Comm.a 14.48 (5.91) 8.92 (7.28) 5.64 (3.99) 9.57 (6.71) <.001 ADI-R RRBa 6.52 (2.82) 3.67 (1.50) 2.29 (3.34) 2.29 (1.80) <.001 Adaptive skills (M, SD) Vineland ABC 73.93 (14.56) 90.92 (10.56) 83.07 (17.36) 78.29 (20.20) .014 Independent living (%) Drives 59.26 84.62 71.43 71.43 .261 Lives on own 14.81 84.62 28.57 57.14 <.001 Paid Job 65.38 92.31 53.33 71.43 .08 ABC (M, SD) Irritability 6.41 (7.27) 0.14 (0.38) 5.53 (4.88) 1.00 (1.73) .067 Social withdrawal 10.36 (8.64) 0.71 (1.25) 3.60 (3.68) 1.33 (2.31) .001 Stereotypy 3.77 (4.42) 0.43 (1.13) 0.87 (1.41) 0.33 (0.58) .020 Hyperactivity/noncompliance 7.50 (7.37) 0.57 (1.13) 5.13 (4.21) 0.33 (0.58) .024 Inap. speech 3.05 (2.89) 0 0.67 (1.13) 0 .001 Well-being (M, SD) 187.73 (27.49) 202.33 (22.47) 154.50 (63.59) 144.67 (69.79) .095 a ADI-R scores were from the 19-year-old assessment only.

Although diagnosed early, participants did not receive high amounts of intervention prior to 3 years (Anderson et al., 2014). Individuals seen in early childhood in the Lost Diagnosis group received more cumulative hours of one-to-one structured teaching prior to 36 months than individuals in the Retained Diagnosis group, χ2 (1, 44) = 4.60, p = .032. Between ages 2 and 3, those in the Lost Diagnosis group received on average 45-min of structured teaching per week, whereas those in the Retained Diagnosis group received a weekly average less than 10-min. Nobody in the Gained Diagnosis group received one-to-one structured teaching before age 3.

Trajectories of autism features across time (VIQ > 70)

ASD features as measured by the ADOS were examined across time for members in each diagnostic group. Growth mixture modeling revealed a significant time by group interaction for individuals in the Lost and Gained diagnosis groups only (Table 2, Figure 1). Race and education were included in the model as predictors to account for attrition but were not significant. There were steady, incremental improvements in CSS scores over time for those in the Lost Diagnosis group and gradual worsening in the Gained Diagnosis group.

Table 2. Change in autistic symptoms from 2–25 (VIQ > 70; n = 69) Predictors Social affect CSS coefficient (SE) RRB CSS coefficient (SE) Total CSS coefficient (SE) ADI social-communication coefficient (SE)b Intercept 6.29 (0.34)** 6.25 (0.36)** 6.10 (0.35)** 14.10 (0.92)** Group Retained – – – Never had −3.49 (0.66)** −3.09 (0.69)** −3.62 (0.66)** −9.75 (1.79)** Lost −0.86 (0.57) −0.75 (0.59) −0.96 (0.57) −2.40 (1.43) Gained −2.95 (1.11)** −3.33 (1.19)** −3.46 (1.09)** −7.55 (4.76) Time −0.03 (0.02) 0.05 (0.02)* −0.01 (0.02) −0.53 (0.09)** Interaction Time*Retained – – – Time*Never Had 0.07 (0.04) −0.03 (0.04) 0.04 (0.04) 0.36 (0.16)* Time*Lost −0.11 (0.03)** −0.12 (0.04)** −0.14 (0.03)** −0.12 (0.15) Time*Gained 0.16 (0.06)* 0.14 (0.07)* 0.17 (0.06)** 0.53 (0.34) White – – – Minority 0.59 (0.51) −0.86 (0.52) 0.22 (0.52) −0.94 (1.24) Parent college educated – – – Parent not college educated 0.03 (0.51) 0.23 (0.52) 0.22 (0.52) 0.88 (1.26) Variance Random effects Intercept 1.18 (0.37) 1.04 (0.40) 1.35 (0.39) 3.79 (2.21) Residuala 3.78 (0.37) 4.72 (0.46) 3.45 (0.34) 30.49 (3.33) *p < .05, **p < .01. a LR test versus linear model suggested support for random effect. b ADI-R only administered until 19.

Autism symptom change (VIQ > 70; n = 69). ADI-R, Autism Diagnostic Interview-Revised; ADOS, Autism Diagnostic Observation Schedule; CSS, Calibrated Severity Score; NV, non-verbal; RRB, restricted and repetitive behavior; SA, social affect

Individuals who were assigned to the Retained Diagnosis and Lost Diagnosis group in adulthood could first be distinguished from one another both on social affect CSS scores (p = .02) and RRB CSS scores (p = .04) as early as age 5 (but not at age 2 or 3), with significant group differences continuing at each assessment thereafter (Figure 1). Thus, for those diagnosed with autism in early childhood, both social affect and RRB trajectories may begin diverging prior to school age. See Figure S2a,b for individual trajectories.

Social communication scores on the ADI-R replicated the divergence in autistic behaviors beginning at age 5 for those in the Retained and Lost diagnosis groups (p = .02). Social-communication improvements over time occurred across the sample (Table 2, Figure 1) with only one significant diagnosis by time interaction (Never Had Diagnosis group), whose social communication scores worsened relative to the Retained group. With respect to adaptive skills, a significant diagnosis by time trajectory emerged for the Lost Diagnosis group (p = .003; Table S2) with adaptive skill differences between the Lost and Retained Diagnosis groups emerging beginning at age 9 (p = .007).

Less cognitively able group composition

Fewer participants with lower cognitive abilities had changing diagnostic categorizations in adulthood, resulting in very small group sizes. Four participants with VIQ < 70 were characterized by “blinded” clinicians as not autistic in adulthood, even though they had repeated diagnoses of autism in childhood (MIQ = 11.75, SD = 9.57). All of them received non-ASD diagnoses of Profound Intellectual Disability and had lower mean VIQ and adaptive scores than the Retained Diagnosis group (Table 3). The majority (75%) of this group were female and Black. An additional four individuals were identified as gaining a diagnosis of ASD in adulthood (MIQ = 31.50, SD = 8.19), with 50% male and 100% White. Unlike the more cognitively able participants, individuals in the Lost Diagnosis group continued to have mean ADOS CSS scores above the clinical threshold (Table 3). There were no differences in treatment among groups.

Table 3. Adult demographic variables (IQ < 70; n = 86) Sample characteristics Retained diagnosis (n = 62) Lost diagnosis (n = 4) Never had diagnosis (n = 16) Gained diagnosis (n = 4) Unadj. p-value Male (%) 87.10 25 50 50 .001 Site (%) North Carolina 48.39 75.00 62.50 50 <.001 Chicago 50.00