Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2024 Jan 1.
Published in final edited form as: J Clin Child Adolesc Psychol. 2022 Dec 22;52(1):108–118. doi: 10.1080/15374416.2022.2154217

The Importance of Parent-Teacher Informant Discrepancy in Characterizing Autistic Youth: A Replication Latent Profile Analysis

Erin Kang 1, Matthew D Lerner 2, Kenneth D Gadow 3
PMCID: PMC9921175  NIHMSID: NIHMS1857627  PMID: 36548459

Abstract

Research about autism spectrum disorder (ASD) supports variation in symptom presentations across settings, and there is a growing literature that explicates how this variability may improve characterization of the autism phenotype. Capitalizing on a well-established literature on informant discrepancy as an index of contextual variability, research suggests that differing parent and teacher perceptions may impact treatment or education-related outcomes. A prior investigation by Lerner et al. (2017) suggests that parent-teacher discrepancy in ASD symptom ratings define discrete and clinically meaningful subgroups. However, replication in a larger sample is important to support the validity and utility of the subgroups for use in research and practice. The present paper sought to (1) replicate the Lerner et al. (2017) latent profile analysis (LPA) in a larger sample of 514 clinic-referred autistic youth, and (2) determine if parent-teacher informant discrepancies relate to clinical and functional correlates. We hypothesized that parent-teacher discrepancies in ASD symptom severity ratings would validly characterize ASD subgroups and predict clinical and functional correlates. Results of the LPA supported a 4-profile solution made up of two parent-teacher agreement groups (high parent-teacher, 21.2%, and low parent-teacher, 34.2%) and two parent-teacher discrepancy groups (high parent-low teacher, 18.1%, and moderate parent-high teacher, 26.5%), replicating findings from Lerner et al. (2017). Latent profile membership differentially predicted IQ, age, and educational outcomes of participants. Unique, clinically-useful information about the taxonomy and impact of ASD is obtained by considering informant discrepancies in symptom severity ratings, which underscores the importance of considering contextual variability assessed through multiple informants.

Keywords: autism spectrum disorder, informant discrepancies, latent profile analysis

Autism spectrum disorder (ASD) is characterized by difficulties in social communication and the presence of repetitive and restricted behaviors and interests (American Psychiatric Association, 2013). Recent autism research suggests great heterogeneity in symptom presentations across settings (Stratis & Lecavalier, 2015), and there is a growing literature that explicates how this variability may improve characterization of the autism phenotype. Obtaining accurate assessment of symptoms in the core domains of ASD is particularly important in understanding individual profiles for clinical purposes, such as planning and monitoring intervention and services for autistic youth. For this reason, ratings are typically obtained from informants who know the child, such as the child’s parents and the child’s teacher (Pearson et al., 2012).

The use of reports taken from multiple informants is indeed an important aspect of evidence-based assessments of psychopathology in children (Mash & Hunsley, 2005). Inconsistencies and poor agreement often arise in the reports of different informants, called informant discrepancies (De Los Reyes & Kazdin, 2005). Historically, informant discrepancy has been treated measurement error rather than important information (De Los Reyes, 2011). However, the utility of considering and understanding the relationship between informants’ reports has emerged as a rich domain of investigation that provides useful information above and beyond that of a single individual’s report alone (e.g., De Los Reyes et al., 2015). For example, there is a well-established literature that suggests considering reports from multiple informants is necessary for proper assessment of contextual variation in expression of child psychopathology (Achenbach et al., 1987; De Los Reyes et al., 2015), as well as reflecting youth clients’ needs that may vary within and across contexts (De Los Reyes et al., 2022). Taken together with the contextual variability in symptom presentation and corresponding service needs, multi-informant assessment is crucial, as informants (e.g., parents, teachers) may have differences in their level of expertise for assessing behaviors and needs in different contexts (e.g., home, school; De Los Reyes et al., 2022). Therefore, collecting reports from multiple informants and interpreting the degree of convergence and divergence are crucial for assessment in research settings, but also in determining intervention targets and monitoring progress in clinical and school settings (De Los Reyes & Makol, 2021).

Capitalizing on the body of literature on informant discrepancy as an index of contextual variability (De Los Reyes et al., 2015), several studies have examined the issue of agreement between parent and teacher rating in ASD research, with mixed findings. For instance, one study suggested that parents and teachers tend to agree on adaptive functioning but not in regards to autism symptomatology (Szatmari et al., 1994), whereas others have found that parent and teacher agree on severity of ASD symptoms (e.g., Constantino et al., 2007; Kanne et al., 2009). In addition to ASD symptoms, parents and teachers may disagree on emotional and behavioral symptoms of autistic youth or those with intellectual disability (Stratis & Lecavalier, 2015).

Indeed, research suggests that differing parent and teacher perceptions may impact treatment or services outcomes in autistic children. Parents and teachers may agree in valuing and selecting intervention targets in areas such as responsibility and cooperation, but may also disagree in determining intervention targets due to contextual differences between home and school settings (Frey et al., 2014). This is especially relevant in the diagnostic and treatment planning process for autistic individuals. For example, even when one informant determines that a child would benefit from receiving a specific intervention or treatment, it may be difficult to obtain these services when parents and teachers disagree. Therefore, it is especially valuable to explore and characterize pattern of parent-teacher informant discrepancies in autistic populations to see if such discrepancies relate to diagnostic, clinical, and functional correlates and outcomes in autistic children.

One particular outcome of interest is the youth’s special education designation and type of specialized classroom placement. The Individuals with Disabilities Education Act (IDEA; 2004) includes provisions for education with peers without disabilities to the maximum extent appropriate (i.e., “the least restrictive environment”). There are varying levels of inclusive education options that range from general education/regular classroom with no additional or specialized assistance or inclusive classroom settings with the special education teacher in a consultative role, a general education/regular classroom with resource room for specialized instructions in areas of need, a self-contained special education classroom within general education school, to placement in a separate school exclusively for special education. The most restrictive option includes education through residential or hospital instructional program.

Autistic children are more likely to receive special education and be placed in a less inclusive classroom than children with other psychiatric diagnoses (e.g., learning disorders, language disorders, attention-deficit/hyperactivity disorder, anxiety disorders), and more severe ASD symptoms are a predictor of the likelihood of having a special education designation and less inclusive classroom placement (Spaulding et al., 2017). Moreover, as autistic children progress through grades in school, the proportion of common supportive services decreases when compared to total services, and individuals may be placed in progressively lower levels of inclusiveness (Spaulding et al., 2017). A more recent study showed that clinician and teacher evaluations of ASD severity are more closely associated with the frequency of school services, but only in children with greater ASD symptoms (Rosen et al., 2019). However, it is unclear how differing perceptions of parents and teachers contribute to decisions about classroom placement for autistic children.

In light of these findings, parsing out heterogeneity in the pattern of agreement among multi-informant ratings is warranted, instead of considering the autistic population as a homogeneous group. To this end, multiple studies have attempted to parse the heterogeneity among ASD using person-centered approaches, which are useful for this purpose (e.g., Azad et al., 2020; Dekker et al., 2021; Gotham et al., 2012; Kang, Gadow, et al., 2020; Kang, Lerner, et al., 2020; Landa et al., 2012; Lerner et al., 2017; Lord et al., 2012). While, in general, low agreement characterizes multiple informants’ reports on average, these person-centered approaches are especially meaningful when considering informant discrepancy, as variation in levels of discrepancies (some informant pairs reporting similar reports while others reporting very different reports) is needed to yield meaningful information about domains of interest (De Los Reyes et al., 2019).

In particular, a prior investigation in a sample of 283 children referred for an ASD evaluation revealed that the degree of discrepancy between parent and teacher regarding autism symptom ratings defines discrete and clinically meaningful subgroups of ASD (Lerner et al., 2017). Specifically, the findings suggest that children in distinct subgroups (e.g., moderate-to-large parent–teacher agreement vs. large parent-teacher discrepancy) differed in terms of medication status and receipt of school-based special education services, as well as meeting ADOS-2 diagnostic criteria for ASD. Specifically, youth with agreement profiles, in contrast to discrepancy profiles, were more likely to be receiving school-based special education supportive services, and to meet ADOS-2 diagnostic criteria for ASD. Youth with the Low Parent/Moderate Teacher Symptom Severity Profile were less likely to receive psychotropic medication than other profiles. Notably, these variations in service and diagnostic information could not be identified with single informant reports, but rather with symptom profiles characterized by the level of informant discrepancy. However, replication in an independent investigation with a larger sample is important in supporting the validity and utility of the subgroups for use in research and practice. Moreover, it is important to examine if resulting subgroups demonstrate clear differences in terms of other validators and outcomes, such as educational outcomes (e.g., special education) and psychotropic medication.

Current Study

We sought to characterize patterns of parent-teacher informant discrepancies among autistic children and adolescents. The primary goal of this study was to use a bottom-up approach to explore empirical ASD profiles, based on information from parents and teachers, by using a latent profile analysis (LPA) to uncover subgroups within the sample. The secondary goal of this study was to explore between-group differences in demographic variables and education-related outcomes to examine how the respective grouping based on informant discrepancy reveals meaningful differences.

It was hypothesized that LPA of parent and teacher ratings of ASD symptoms would reveal four groups, consistent with findings from Lerner et al (2017): two informant agreement groups with high or low ratings from both informants, and two informant discrepancy groups with high parent–low teacher ratings and low parent–high teacher ratings. The current study explored how the identified profiles may differ on IQ and age. In addition, it was hypothesized that special education designation and type of classroom, as well as use of psychotropic medication would differ based on pattern of agreement and disagreement profiles between parent’s and teacher’s perception of severity of child’s ASD symptoms. Consistent with the findings from Lerner et al. (2017), we hypothesized that the agreement profiles would be more likely to receive special education and to be on medication.

Methods

Participants

Case records for consecutive referrals to a university hospital developmental disabilities clinic located on Long Island, New York, were screened for children who were between 6 and 18 years old and assessed with prerequisite questionnaires completed by parent and/or teacher at the time of their initial diagnostic evaluation. The sample (N = 609) was composed of youth who were evaluated in a developmental disabilities specialty clinic and diagnosed as having an ASD according to DSM-IV diagnostic criteria, and had no known overlap with the original sample of Lerner et al. (2017). DSM-IV-based ASD diagnoses were confirmed by an expert diagnostician and based on five sources of information: (1) comprehensive developmental history, (2) clinician interview with child and caregiver(s), (3) direct observations of the child, (4) review of validated ASD rating scales including the Child and Adolescent Symptom Inventory-4R (Gadow, Schwartz, et al., 2008), and (5) in a subset of children (n = 347), the Autism Diagnostic Observation Schedule (ADOS; Lord et al., 2008) administered by a certified examiner. The ADOS testing was generally limited to youth who did not have a prior well-documented diagnosis of ASD (e.g., prior clinician or school evaluations), and who received all of the aforementioned assessments but not the ADOS.

Owing to missing data for parent- and teacher-rated symptoms on Child and Adolescent Symptom Inventory-4R (see Measures), the number of cases available for data analyses was 514. Participants (Mage = 9.31, SDage = 2.83) in the sample were largely male (83.2%) and Caucasian (90.4%). Participants’ IQ ranged from 19 and 140 (MIQ = 86.92, SDIQ = 24.53), with 98 children with IQ below 70.

This study was approved by a university Institutional Review Board, and appropriate measures were taken to protect youth and caregiver confidentiality.

Measures

Cross-informant ratings of ASD symptoms.

Informants rated youths’ symptoms with either the parent or teacher version the Child and Adolescent Symptom Inventory-4R (CASI-4R; Gadow & Sprafkin, 2005). The CASI-4R covers a range of disorders, including the ASD subscale. For each item, parents or teachers rated which frequency ‘best describes this youth’s overall behavior.’ Individual items bear one-to-one correspondence with DSM-IV-TR symptoms and are rated from 0 (never) to 3 (very often). Items were summed to generate a symptom severity score. The last item in each symptom subscale addresses impairment by asking the informant “How often do the behaviors in [this category] interfere with youth’s ability to do schoolwork or get along with others”. This closely approximates the DSM-5 criterion for impairment, which in the case of ASD is phrased as follows: “There is clear evidence that the symptoms interfere with, or reduce the quality of, social, academic, or occupational functioning” (American Psychiatric Association, 2013).

Numerous studies indicate CASI-4R subscales demonstrate satisfactory psychometric properties. Specifically, individual symptom dimensions evidence satisfactory internal consistency (Cronbach’s alpha), test-retest reliability, and convergent and divergent validity with respective measures from a range of relevant assessment instruments and diagnostic procedures in community-based normative, clinic-referred non-autistic and autistic samples (see Gadow, 2022). Moreover, in a factor analytic study, the internal validity of the ASD symptom subscales were supported among autistic youth (Lecavalier et al., 2009).

Parent Questionnaire.

The Parent Questionnaire (Gadow, DeVincent, et al., 2008) was used to obtain information about child and family characteristics and developmental history from youth’s primary caregiver. Medical and treatment characteristics included whether the child was ever or is currently on medication. Education-related variables included special education designation (i.e., whether or not a student has been deemed eligible for, and is receiving, an Individualized Education Plan to address needs that interfere with access to the curriculum, binary: yes/no) and level of inclusiveness of classroom placement in which a child is assigned to spend the day: (a) inclusive classroom settings (i.e., spending less than 20% of time outside the regular classroom), (b) general education/regular classroom with resource room for specialized instructions in areas of need (i.e., 21%–60% of time spent outside the regular classroom), (c) a self-contained special education classroom (more than 60% of the school day in a separate class), (d) special education school, and (e) other.

Procedure

Prior to their initial diagnostic evaluation, parents of potential patients completed an intake assessment battery that included the CASI-4R, Parent Questionnaire, and permission for release of school reports. Parents delivered a similar packet of materials to the school with instructions that requested teachers to complete the CASI-4R, and the school to provide copies psycho-educational evaluations, special education evaluation records, and IQ testing results. Schools mailed their information directly to the clinic. Intake diagnostic evaluations included interviews with the children and their caregivers; informal observation of parent-child interaction; and review of the packet materials.

Data Analytic Plan

Latent profile analysis (Bartholomew, 1987) was conducted using Mplus, Version 7 (Muthén & Muthén, 2012) to explore heterogeneity within autistic children and identify profiles of autistic children with similar pattern of multi-informant ratings. Latent profile analysis is a person-centered and model-based statistical method (Nylund et al., 2007). The objective is to categorize the people into profiles using the observed continuous variables and identify items that best distinguish between profiles. A few advantages of this approach include relatively few assumptions in this technique, such as modest correlations among indicators. Moreover, a full information maximum likelihood estimation under the assumption that the data are missing at random is a widely accepted way of handling missing data (Muthén & Shedden, 1999). Ninety-six subjects were removed from data analysis due to missing data on latent profile indicators, which were parent ratings (1: social, 2: language, 3: repetitive and restrictive behaviors) and teacher ratings (4: social, 5: language, 6: repetitive and restrictive behaviors) on the CASI-4R. The mixture missing command was used in all analyses to account for missing data.

There are multiple statistical indicators of model fit. In order to determine the number of profiles that were best represented by the data, following criteria were considered in the current analyses: (1) the Bayesian information criterion (BIC; Schwarz, 1978) and the sample-size-adjusted BIC (SSABIC; Sclove, 1987), such that smaller values indicate a better fit model, as BIC and adjusted BIC are comparatively better indicators of the number of profiles than AIC (Nylund et al., 2007); (2) significance of the Vuong-Lo-Mendell-Rubin (VLMR) Likelihood Ratio Test, which assesses the fit between nested models that differ by one profile and provides a p value that indicates which model fits better, such that significant p value indicates whether the k – 1 profile model is rejected in favor of the k profile model; (3) entropy was also used as an index of confidence that adequate separation between profiles exist, such that values closer to 1 indicating better classification, in conjunction with other model fit indices; and (4) substantive interpretation of the results. In the present study, models specifying two, three, four, and five profiles were compared to determine best-fitting model for the data.

After specifying the profiles, a one-way Multivariate Analysis of Variance (MANOVA) in SPSS (SPSS, 2013) was used to determine if the groups derived from the latent profile analysis demonstrated meaningful differences. For these analyses, participants were assigned to their most probable profile on the basis of the LPA outcome as described above. MANOVA was conducted on two continuous dependent variables: IQ and age, with the latent profile derived by the LPA as the independent variable. MANOVA rather than two separate ANOVAs were used to account for the correlation between IQ and age and to reduce the possibility of Type I error. Assumptions of samples sizes in relation to group missing data, and homogeneity of variance-covariance matrices were tested. Presence of outliers, homogeneity of regression, multicollinearity and singularity were also checked. If significant results were obtained from MANOVA, a Roy-Bargmann stepdown analysis was performed on the prioritized Dependent Variables (DVs) to investigate the impact of each main effect on the individual DVs. The pattern of significant differences between pairs of means was examined by post-hoc comparisons by least significant difference (LSD) method.

In addition, a series of χ2 tests were conducted in order to compare individuals within each profile across categorical educational outcomes (i.e. receiving special education, type of class, current medication status, ever on medication). Assumptions of sample size (at least 80% of contingency cells have expected frequency ≥ 5) and independence assumptions were tested.

Statistical significance was defined as p < .05 for these analyses. In multiple comparisons, Bonferroni corrections were used to reduce the risk of Type I error.

Results

Identification and Description of the Latent Profiles

A LPA was conducted to determine the optimal number of profiles and the characteristics associated with each profile. Six indicator variables were included in the analysis: Parent ratings of 1) social, 2) language, and 3) repetitive/restrictive/stereotyped interest and behavior symptoms and teacher ratings of 4) social, 5) language, and 6) repetitive or restrictive interest and behaviors symptoms.

Models positing two to five groups were evaluated in relation to the various measures of fit and the fit indices are summarized in Table 1. The AIC, BIC, and the SSABIC continued to decrease across the models considered. The results of the VLMR tests were highly significant for the two-profile solution but not for the three-profile solution. However, the VLMR value for the four-profile was also highly significant, suggesting that four profiles fit better than a three-profile model. On the other hand, the five-profile model was not significantly better than the four-profile model according to VLMR. The entropy was good for two profiles but dropped for three or more profiles, with an increase for the four-profile model. In addition, based on substantive interpretation, the four-profile solution emerged as the optimal fit for the data. As seen in Table 2, mean posterior probabilities of a person’s likelihood of being assigned to the group were good. This model was selected for closer examination and was used for all following analyses.

Table 1.

Criteria for Assessing Model Fit for Different Number of Profiles

1
Profile		 2
Profiles		 3
Profiles		 4
Profiles		 5
Profiles
AIC 16094.94 15664.95 15512.43 15396.12 15342.45
BIC 16145.85 15745.55 15622.73 15536.11 15512.14
Sample-size Adjusted BIC 16107.76 15685.24 15540.20 15431.37 15385.17
Entropy N/a 0.804 .766 .77 .752
VLMR N/a 2 v 1
p = .0001		 3 v 2
p = .1699		 4 v 3
p = .0019 		5 v 4
p = .087
N for each profile P1=514 P1=309 P1=257 P1=139 P1=125
P2=205 P2=135 P2=93 P2=81
P3=122 P3=174 P3=80
P4=108 P4=125
P5=103
Open in a new tab

Note. AIC = Akaike Information Criterion; BIC = Bayesian Information Criterion; VLMR = Vuong-Lo-Mendell-Rubin likelihood ratio test.

Table 2.

Average Latent Profile Probabilities for Most Likely Latent Profile Membership (Row) by Latent Profile (Column)

1 2 3 4
1 .85 .03 .07 .05
2 .03 .82 .13 .02
3 .06 .06 .88 .00
4 .06 .03 .00 .92
Open in a new tab

Figure 1 summarizes the prevalence and clinical characteristics of four latent profiles in terms of expected means of the parent and teacher-ratings in three ASD symptom domains (indicator variables). As depicted in Figure 1, Profiles 1 (26.5%) and 2 (18.1%) are best characterized as informant disagreement groups (moderate parent/high teacher disagreement group and high parent/low-moderate teacher disagreement group, respectively) and Profiles 3 (34.2%) and 4 (21.2%) are best characterized as informant agreement groups (low-moderate parent/teacher agreement group and high parent/teacher agreement group, respectively).

Figure 1.

Figure 1.

Open in a new tab

Latent profile solution with four profiles. Estimated means of Child & Adolescent Symptom Inventory-4R ratings in three symptom domains of autism spectrum disorder given profile membership: parent ratings (1: social, 2: language, 3: repetitive and restrictive behaviors) and teacher ratings (4: social, 5: language, 6: repetitive and restrictive behaviors). Profiles 1 and 2 represent cross-informant discrepancy groups (moderate severity parent/high severity teacher disagreement group and high severity parent/low-moderate severity teacher disagreement group, respectively) and Profiles 3 and 4 represent cross-informant agreement groups (low-moderate severity parent/teacher agreement group and high severity parent/teacher agreement group, respectively).

Follow-up Analyses

Participants were assigned to their most probable profile on the basis of the LPA outcome and these profiles were used to predict DVs of IQ and age. For the one-way MANOVA, total N of 514 was reduced to 491 with removal of cases with missing values for the IQ variable. There were no univariate or multivariate within-cell outliers at p < .001. Results of evaluation of assumptions of normality, homogeneity of variance-covariance matrices, linearity, and multicollinearity were satisfactory. Using Wilks’ λ criterion, the combined DVs were significantly related to the profile membership, F(6,830) = 20.04, p <.001. The results reflected a modest association between profile membership and the combined DVs, η2 = .24.

Given the significance of the overall test, a Roy-Bargmann stepdown analysis was conducted. Both DVs were judged to be sufficiently reliable to warrant stepdown analysis. Homogeneity of variance-covariance of matrices was achieved for all components of the stepdown analysis. In stepdown analysis, each DV was analyzed, with higher priority DV treated as covariate and with the higher priority DV tested in a univariate ANOVA. Priority was assigned first to IQ then age, according to their theoretical and practical considerations. Therefore, IQ was entered alone in step #1, then age was entered in a second step, with IQ as a covariate. A unique contribution to predicting differences between profiles was made by IQ, stepdown F(3,416) = 38.18, p < .001. After the pattern of differences measured by IQ was entered, a difference was also found on age, stepdown F(3,415) = 3.84, p = .01. Table 3 summarizes the results of the adjusted descriptive statistics of DVs of MANOVA. Post-hoc LSD comparisons indicated that participants in Profiles 2 (high severity parent/low-moderate severity teacher disagreement group) and 3 (low-moderate severity parent/teacher agreement group) were older than those in Profiles 1 (moderate severity parent/high severity teacher disagreement group) and 4 (high severity parent/teacher agreement group; all p <.02). Profile 3 (low-moderate severity parent/teacher agreement group) evinced higher IQ than all three other profiles, Profile 4 (high severity parent/teacher agreement group) evinced lower IQ than all three other profiles (all p < .004).

Table 3.

Means and Standard Deviations for Dependent Variable by Profile

Profile 1 95% CI Profile 2 95% CI Profile 3 95% CI Profile 4 95% CI
Mean (SE) Mean (SE) Mean (SE) Mean (SE)
IQ 86.60 (2.02) 82.63–90.58 88.31 (2.48) 83.43–93.19 97.15 (1.75) 93.71–100.59 63.93 (2.57) 58.88–68.98
Age 8.97 (0.26) 8.44–9.48 9.99 (0.32) 9.35–10.62 9.81 (0.23) 9.35–10.25 8.86 (0.33) 8.20–9.52
Open in a new tab

A series of χ2 tests of independence was performed to examine the relation between group membership and educational outcomes (i.e., receiving special education, type of class) and medication status (i.e., currently on medication, ever on medication). All χ2 tests of independence and sample size assumptions were met for categorical variables considered. The groups demonstrated significant differences in whether or not receiving special education (χ2(3) = 31.31, p < .001; Table 4) and in type of class (χ2(15) = 59.56, p < .001; Table 5). Specifically, Profile 4 (high severity parent/teacher agreement group) evinced higher proportion than the other profiles of receiving special education, and of being placed in special school, resource room, or inclusion classrooms. The groups did not differ in medication status currently or ever (both χ2(3) < 4.91, p > .179; Table 6).

Table 4.

Results of χ2 Test and Frequency for Receiving Special Education by Latent Profile Membership

Profile
Special Education 1 2 3 4
Yes 119 (86.2%) 78 (83.9%) 139 (79.9%) 106 (99.1%)
No 19 (13.8%) 15 (16.1%) 35 (20.1%) 1 (0.2%)
Open in a new tab

Note. χ2 = 21.31, df = 3. Numbers in parentheses indicate column percentages.

Table 5.

Results of χ2 Test and Frequency for Type of Class by Latent Profile Membership

Class
Type of Class 1 2 3 4
Inclusion class 19 (16.0%) 13 (16.0%) 25 (18.0%) 4 (3.8%)
Resource room 24 (20.2%) 14 (17.9%) 38 (27.3%) 6 (5.7%)
Self-contained class 43 (36.1%) 30 (38.5%) 41 (29.5%) 42 (39.6%)
Special School 24 (20.2%) 14 (17.9%) 18 (12.9%) 45 (42.5%)
Other 9 (7.6%) 6 (7.2%) 17 (41.5%) 9 (8.5%)
Open in a new tab

Note. χ2 = 59.56, df = 15. Numbers in parentheses indicate column percentages.

Table 6.

Results of χ2 Tests and Frequencies of Currently or Ever on Medications by Latent Profile Membership

Profile
1 2 3 4
Currently on Medications Yes 54 (39.7%) 39 (42.4%) 78 (45.3%) 33 (32.0%)
No 82 (60.3%) 53 (57.6%) 94 (54.7%) 70 (68.0%)
Ever on Medications Yes 71 (51.8%) 54 (58.7%) 99 (57.2%) 62 (60.2%)
No 66 (48.2%) 38 (41.3%) 74 (42.8%) 41 (39.8%)
Open in a new tab

Note. χ2 = 4.91 for currently on medication and 2.01 for ever on medication, df = 3. Numbers in parentheses indicate column percentages.

Discussion

The current study was carried out in order to refine the classification of autistic children based on empirically-derived clinical profiles of ASD symptoms provided by parents and teachers in a large sample. Results from LPA comparing two- to five-profile models generally supported a four-profile model solution, consisting of two cross-informant disagreement groups (moderate severity parent/high severity teacher disagreement group and high severity parent/low-moderate severity teacher disagreement group) and two cross-informant agreement groups (low-moderate severity parent/teacher agreement group and high severity parent/teacher agreement group), as we hypothesized. Latent profile membership differentially predicted IQ and age of participants, and was associated with educational outcomes (type of class and special education), but not with medication status.

The findings from this study largely replicate in a larger sample the four-profile solution found in the Lerner et al. (2017) LPA, in which two profiles represented cross-informant agreement (moderate and high severity ratings, respectively), and two profiles represented cross-informant discrepancy (moderate parent/low teacher and moderate teacher/low parent severity ratings, respectively). Results of the two studies differed slightly in that two discrepant profiles in Lerner et al. (2017) were characterized by discrepancy reflecting low vs. moderately-elevated ratings for each ASD symptoms, whereas in the current study, the two discrepant profiles reflect low-moderate vs. high ratings similar to the level reported in the high-severity agreement profile (Profile 4). While the two samples did not differ markedly in terms of IQ or gender composition, the current sample was younger on average than the Lerner et al (2017) sample. Given that across both studies, profiles with younger participants on average reflected moderate to high symptoms endorsed by parents and/or teacher, it is possible that having more younger participants in the current study sample may relate to elevated scores by one informant in the discrepant profiles. Nevertheless, both studies are in line with prior work examining reporting patterns in psychosocial domains that reflect (1) informant pairs’ convergent report of high levels of symptoms, (2) informant pairs’ convergent report of low levels of symptoms, and (3) informant pairs’ divergent reports wherein one informant reporting higher symptoms than the other informant (see De Los Reyes et al., 2015, 2019 for reviews).

Resulting subgroups evidenced clear differences in terms of IQ and age, as well as educational outcomes. Specifically, autistic children with high parent/low-moderate teacher ratings (Profile 2) and low-moderate ratings from both informants (Profile 3) were older than children who received moderate-parent/high-teacher ratings (Profile 1) and those with high ratings from both parents and teacher (Profile 4). Those in the low-moderate parent/teacher agreement group (Profile 3) evinced higher IQ than all three other profiles, whereas those with high parent-teacher agreement group (Profile 4) evinced lower IQ than all three other profiles.

It is notable that while all but one (99%) of youth with Profile 4 (high parent-teacher agreement group) membership received special education designation, much smaller percentages of youth in Profiles 1 and 2 (informant discrepancy groups) received such designation even though parents or teachers in these groups reported high severity of symptoms at comparable levels to ratings in Profile 4 (particularly for social and repetitive and restricted behaviors domains). It is plausible that autistic youth who evince higher levels of agreement between parents and teachers may be considered as particularly in need of special education services, while autistic children with contextual variability in symptom presentation may have some of their special education needs overlooked due to disagreement between perspectives of parents and teachers, or as only one source of information is considered in school-based special education services. These findings appear to add support to Diverging Operations in the Operations Triad Model (De Los Reyes et al., 2013), where multiple informants’ reports yield discrepant findings and the discrepancies reflect meaningful variation in the behaviors being assessed, namely variations that relate to school-based services and delivery.

Moreover, children in different profiles were likely to be placed in different classroom types. Prior work on classroom placements of autistic youth suggests that greater ASD severity relates to less inclusive classroom placement of youth (Rosen et al., 2019; Spaulding et al., 2017). Consistent with these findings, most of the youth for whom parents and teachers both reported high symptoms were placed in special schools, resource rooms, and inclusion classroom, compared to other three profiles where youth were distributed across varying levels of inclusion. Importantly, these youth also had lower IQ than those in other profile groups. Thus, it may be important to consider the specific and distinct roles IQ and ASD severity may play in classroom placements, such as additional impairment associated with intellectual functioning that may relate to more intense and specialized services for autistic youth.

The relationship between parent and teacher ratings suggests that core ASD symptoms can be perceived similarly or differently by parents and teachers, and that different social contexts can also elicit different behaviors. It is also notable that despite the between-informant variability by profiles, within-informant ratings were consistent across the three domains of ASD symptoms in all profiles. This finding is important because it provides a possible explanation for the inconsistent findings regarding agreement between parent and teacher ratings of core ASD symptoms (Kanne et al., 2009; Szatmari et al., 1994), and underscores the importance of obtaining multiple informants’ ratings in autistic children when characterizing profiles of psychopathology and treatment response, rather than relying on informant-specific variance and ignoring one informant (e.g., parent’s report) based on the other informant (e.g., teachers or school psychologists).

While the degree of agreement and discrepancy in parent-teacher ratings meaningfully related to variations in correlates and outcomes relevant to school-based services, it did not relate to differences in medication status, whether currently or ever by history. This is in contrast to the findings from Lerner et al. (2017) that latent profiles were marginally associated with psychotropic medication status, where youth in subgroups with parent/teacher agreement on ASD symptoms were more likely to receive psychotropic medication than those in discrepancy subgroups. It is possible that psychotropic medication is relatively more influenced by parental factors due to the role parents play in seeking and filling prescriptions, whereas special education placement and level of inclusion are relatively more influenced by school factors. These results highlight the significance of informant discrepancy in “matching” variations in behavior associated with the phenomena relevant to the constructs they are reporting in specific contexts (De Los Reyes et al., 2019), which then has implications for services relevant to those contexts (e.g., educational outcomes).

The present study has a number of strengths, including a large, well-characterized, intellectually-diverse sample, and well-validated measures of ASD symptom severity completed by two different informants. Moreover, the use of the model-based clustering procedure of LPA adds to the growing body of literature on identifying subtypes of ASD to parse the heterogeneity in presentation in this population, as well as using these approaches to reveal the significance of informant discrepancy in providing meaningful information about youths’ functioning. Nevertheless, our results are subject to several limitations. The study sample was comprised of youth referred for outpatient evaluation and therefore may not be representative of community-based samples. In addition, the sample was predominantly white, with relatively high socioeconomic status. Future studies should aim to replicate these results in more diverse community or epidemiological samples. Moreover, the number of outcome variables available for analysis was limited in this sample. For instance, specific data for ADOS scores were not available for this sample, as the examiner simply recorded whether ADOS criteria were met or not. Incorporating a wider variety of demographic, clinical, and functional correlates and outcomes, such as behavioral indices, symptom severity, and impairment in functioning associated with other comorbid psychiatric disorders, will help to validate and better characterize groups and also provide greater understanding of ASD psychopathology and lead to refinement of treatment.

Conclusions

Overall, these results suggest that coherent patterns of converging and diverging parent and teacher ratings of autism symptom severity can yield distinct subgroups that show meaningful variations in terms of key characteristics such as IQ and age, as well as variations in outcomes relevant to school-based educational outcomes. These findings are especially significant, as they largely replicate findings from Lerner et al. (2017). Taken together, these results support the hypothesis that subtyping of autistic youth based on informant discrepancies may be empirically, clinically, and theoretically fruitful, including across diverse ages and levels of intellectual functioning when assessing severity of symptoms. Moreover, these investigations add to the rich body of literature conceptualizing informant discrepancies as reflecting information about contexts and factors that relate to differing needs of youth, and, in turn, can inform educational services and outcomes in contexts in which they demonstrate these needs.

Acknowledgements:

The authors wish to thank Dr. John Pomeroy, M.D., for directing the ASD diagnoses and Carla DeVincent, Ph.D., for coordinating data collection.

Financial and Material Support:

This study was supported, in part, by the Matt and Debra Cody Center for Autism and Developmental Disorders. The funder had no role in study design; the data collection, analysis, and interpretation; manuscript writing; and the decision to submit the article for publication.

Dr. Matthew Lerner received support from National Institute of Mental Health (grant R01MH110585; principal investigator, Dr. Lerner), the Health Resources and Services Administration (grant T73MC42026; principal investigator, Michelle Ballan, Ph.D.) during the course of preparing this manuscript.

Footnotes

Author Conflicts of Interest/Disclosures: Kenneth D. Gadow is shareholder in Checkmate Plus, publisher of the Child and Adolescent Symptom Inventory. Erin Kang and Matthew D. Lerner report no potential conflicts of interest.

References

  1. Achenbach TM, McConaughy SH, & Howell CT (1987). Child/adolescent behavioral and emotional problems: Implications of cross-informant correlations for situational specificity. Psychological Bulletin, 101(2), 213–232. 10.1037/0033-2909.101.2.213 [DOI] [PubMed] [Google Scholar]
  2. American Psychiatric Association. (2013). Diagnostic and statistical manual of mental disorders (5th ed.). Washington DC: Author. [Google Scholar]
  3. Azad GF, Dillon E, Feuerstein J, Kalb L, Neely J, & Landa R (2020). Quality of life in school-aged youth referred to an autism specialty clinic: A latent profile analysis. Journal of Autism and Developmental Disorders, 50(4), 1269–1280. 10.1007/s10803-019-04353-x [DOI] [PubMed] [Google Scholar]
  4. Bartholomew DJ (1987). Latent variable models and factors analysis. Oxford University Press, Inc. [Google Scholar]
  5. Constantino JN, Lavesser PD, Zhang Y, Abbacchi AM, Gray T, & Todd RD (2007). Rapid quantitative assessment of autistic social impairment by classroom teachers. Journal of the American Academy of Child and Adolescent Psychiatry, 46(12), 1668–1676. 10.1097/chi.0b013e318157cb23 [DOI] [PubMed] [Google Scholar]
  6. Dekker V, Nauta MH, Timmerman ME, Mulder EJ, Hoekstra PJ, & de Bildt A (2021). Application of latent class analysis to identify subgroups of children with autism spectrum disorders who benefit from social skills training. Journal of Autism and Developmental Disorders, 51(6), 2004–2018. 10.1007/s10803-020-04678-y [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. De Los Reyes A (2011). Introduction to the special section: More than measurement error: Discovering meaning behind informant discrepancies in clinical assessments of children and adolescents. Journal of Clinical Child & Adolescent Psychology, 40(1), 1–9. 10.1080/15374416.2011.533405 [DOI] [PubMed] [Google Scholar]
  8. De Los Reyes A, Augenstein TM, Wang M, Thomas SA, Drabick DAG, Burgers DE, & Rabinowitz J (2015). The validity of the multi-informant approach to assessing child and adolescent mental health. Psychological Bulletin, 141(4), 858–900. 10.1037/a0038498 [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. De Los Reyes A, Cook CR, Gresham FM, Makol BA, & Wang M (2019). Informant discrepancies in assessments of psychosocial functioning in school-based services and research: Review and directions for future research. Journal of School Psychology, 74(June), 74–89. 10.1016/j.jsp.2019.05.005 [DOI] [PubMed] [Google Scholar]
  10. De Los Reyes A, & Makol BA (2021). Interpreting convergences and divergences in multi-informant, multimethod assessment. In The Oxford Handbook of Personality and Psychopathology Assessment, Second Edition (Issue January). 10.1093/oxfordhb/9780190092689.013.33 [DOI] [Google Scholar]
  11. De Los Reyes A, Talbott E, Power TJ, Michel JJ, Cook CR, Racz SJ, & Fitzpatrick O (2022). The Needs-to-Goals Gap: How informant discrepancies in youth mental health assessments impact service delivery. Clinical Psychology Review, 92(December 2021), 102114. 10.1016/j.cpr.2021.102114 [DOI] [PubMed] [Google Scholar]
  12. De Los Reyes A, Thomas S. a, Goodman KL, & Kundey S. M. a. (2013). Principles underlying the use of multiple informants’ reports. Annual Review of Clinical Psychology, 9(October), 123–149. 10.1146/annurev-clinpsy-050212-185617 [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. De Los Reyes A, & Kazdin AE (2005). Informant discrepancies in the assessment of childhood psychopathology: A critical review, theoretical framework, and recommendations for further study. Psychological Bulletin, 131(4), 483–509. 10.1037/0033-2909.131.4.483 [DOI] [PubMed] [Google Scholar]
  14. Frey JR, Elliott SN, & Kaiser AP (2014). Social Skills intervention planning for preschoolers: Using the SSiS-Rating Scales to identify target behaviors valued by parents and teachers. Assessment for Effective Intervention, 39(3), 182–192. 10.1177/1534508413488415 [DOI] [Google Scholar]
  15. Gadow KD (2022). Checkmate Plus Symptom Inventories: An annotated research bibliography. In 2022. Checkmate Plus. (available from www.checkmateplus.com) [Google Scholar]
  16. Gadow KD, DeVincent C, & Schneider J (2008). Predictors of psychiatric symptoms in children with an autism spectrum disorder. Journal of Autism and Developmental Disorders, 38(9), 1710–1720. 10.1007/s10803-008-0556-8 [DOI] [PubMed] [Google Scholar]
  17. Gadow KD, Schwartz J, DeVincent C, Strong G, & Cuva S (2008). Clinical utility of autism spectrum disorder scoring algorithms for the Child Symptom Inventory-4. Journal of Autism and Developmental Disorders, 38(3), 419–427. 10.1007/s10803-007-0408-y [DOI] [PubMed] [Google Scholar]
  18. Gadow KD, & Sprafkin J (2005). Child and adolescent symptom inventory-4R. In Stony Brook, NY: Checkmate Plus. [Google Scholar]
  19. Gotham K, Pickles A, & Lord C (2012). Trajectories of autism severity in children using standardized ADOS scores. Pediatrics, 130, 1278. 10.1542/peds.2011-3668 [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Kang E, Gadow KD, & Lerner MD (2020). Atypical communication characteristics, differential diagnosis, and the autism spectrum disorder phenotype in youth. Journal of Clinical Child & Adolescent Psychology, 49(2), 251–263. 10.1080/15374416.2018.1539912 [DOI] [PubMed] [Google Scholar]
  21. Kang E, Lerner MD, & Gadow KD (2020). Atypical communication characteristics among clinic-referred youth with and without autism spectrum disorder: Stability and associations with clinical correlates. Development and Psychopathology, 32(4), 1240–1253. 10.1017/S095457942000070X [DOI] [PubMed] [Google Scholar]
  22. Kanne SM, Abbacchi AM, & Constantino JN (2009). Multi-informant ratings of psychiatric symptom severity in children with autism spectrum disorders: The importance of environmental context. Journal of Autism and Developmental Disorders, 39(6), 856–864. 10.1007/s10803-009-0694-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Landa RJ, Gross AL, Stuart EA, & Bauman M (2012). Latent class analysis of early developmental trajectory in baby siblings of children with autism. Journal of Child Psychology and Psychiatry and Allied Disciplines, 53(9), 986–996. 10.1111/j.1469-7610.2012.02558.x [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Lecavalier L, Gadow KD, DeVincent CJ, & Edwards MC (2009). Validation of DSM-IV model of psychiatric syndromes in children with autism spectrum disorders. Journal of Autism and Developmental Disorders, 39(2), 278–289. 10.1007/s10803-008-0622-2 [DOI] [PubMed] [Google Scholar]
  25. Lerner MD, De Los Reyes A, Drabick DAG, Gerber AH, & Gadow KD (2017). Informant discrepancy defines discrete, clinically useful autism spectrum disorder subgroups. Journal of Child Psychology and Psychiatry, 58(7), 829–839. 10.1111/jcpp.12730 [DOI] [PubMed] [Google Scholar]
  26. Lord C, Luyster R, Guthrie W, & Pickles A (2012). Patterns of developmental trajectories in toddlers with autism spectrum disorder. Journal of Consulting and Clinical Psychology, 80(3), 477–489. 10.1037/a0027214 [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Lord C, Rutter M, Dilavore PC, & Risi S (2008). ADOS: Autism diagnostic observation schedule. Hogrefe. [DOI] [PubMed] [Google Scholar]
  28. Mash EJ, & Hunsley J (2005). Evidence-based assessment of child and adolescent disorders: Issues and challenges. Journal of Clinical Child and Adolescent Psychology, 34(3), 362–379. 10.1207/s15374424jccp3403_1 [DOI] [PubMed] [Google Scholar]
  29. Muthén BO, & Muthén LK (2012). Software Mplus Version 7. [Google Scholar]
  30. Muthén B,O, & Shedden K (1999). Finite mixture modeling with mixture outcomes using the EM algorithm. Biometrics, 55(2), 463–469. 10.1111/J.0006-341X.1999.00463.X [DOI] [PubMed] [Google Scholar]
  31. Nylund KL, Asparouhov T, & Muthén BO (2007). Deciding on the number of classes in latent class analysis and growth mixture modeling: a Monte Carlo simulation study. Structural Equation Modeling, 14(4), 535–569. 10.1080/10705510701575396 [DOI] [Google Scholar]
  32. Pearson DA, Aman MG, Arnold LE, Lane DM, Loveland KA, Santos CW, Casat CD, Mansour R, Jerger SW, Ezzell S, Factor P, Vanwoerden S, Ye E, Narain P, & Cleveland LA (2012). High concordance of parent and teacher attention-deficit/hyperactivity disorder ratings in medicated and unmedicated children with autism spectrum disorders. Https://Home.Liebertpub.Com/Cap, 22(4), 284–291. 10.1089/CAP.2011.0067 [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Rosen TE, Spaulding CJ, Gates JA, & Lerner MD (2019). Autism severity, co-occurring psychopathology, and intellectual functioning predict supportive school services for youth with autism spectrum disorder. Autism, 23(7), 1805–1816. 10.1177/1362361318809690 [DOI] [PubMed] [Google Scholar]
  34. Schwarz G (1978). Estimating the dimension of a model. The Annals of Statistics, 6(2), 461–464. [Google Scholar]
  35. Sclove SL (1987). Application of model-selection criteria to some problems in multivariate analysis. Psychometrika, 52(3), 333–343. [Google Scholar]
  36. Spaulding CJ, Lerner MD, & Gadow KD (2017). Trajectories and correlates of special education supports for youth with autism spectrum disorder and psychiatric comparisons. Autism, 21(4), 423–435. 10.1177/1362361316645428 [DOI] [PubMed] [Google Scholar]
  37. SPSS, I. B. M. (2013). IBM SPSS statistics 22. Algorithms. Chicago: IBM SPSS Inc. [Google Scholar]
  38. Stratis EA, & Lecavalier L (2015). Informant agreement for youth with autism spectrum disorder or intellectual disability: A meta-analysis. Journal of Autism and Developmental Disorders, 45(4), 1026–1041. 10.1007/s10803-014-2258-8 [DOI] [PubMed] [Google Scholar]
  39. Szatmari P, Archer L, Fisman S, & Streiner DL (1994). Parent and teacher agreement in the assessment of pervasive developmental disorders. Journal of Autism and Developmental Disorders, 24(6), 703–717. 10.1007/BF02172281 [DOI] [PubMed] [Google Scholar]

RESOURCES