Results

1
The Social Responsiveness Scale Concurrence with
Diagnosis, Correlation with the ADI-R, and
Inter-Rater Agreement Sullivan, A. Risi, S.
Burke, J., Qiu, S. Lord, C. University of
Michigan Autism and Communication Disorders Center
Methods
Results
Abstract
Demographics for Groups 1 and 2

• INTRODUCTION
• Detection of Autism Spectrum Disorders (ASDs),
  specifically in those exhibiting milder autistic
  impairments, is a priority for both researchers
  and clinicians. A dimensional measure of autistic
  features appropriate for population studies is a
  valuable addition to the field.
• The SRS is a reliable, quickly-administered
  instrument capable of discriminating ASDs from
  general or clinic populations. It has been
  indicated for use both as a screener (general
  population use lower raw score cutpoints of 65
  and 70 for females and males, respectively
  clinical population use higher raw score
  cutpoint of 85 for both genders) and as an aid in
  clinical diagnosis. Approximately 15-20 minutes
  are required for a parent or teacher to complete
  this measure. The questionnaires use of a
  Likert-scale format, as opposed to binary queries
  regarding symptom presence or absence, allows for
  the measurement of severity across its five
  subdomains. Previous research has shown that all
  65 items on the SRS map onto a single factor.
• Validation of the SRS has been based in part on
  comparisons with the ADI-R, a diagnostic
  parent-report interview. High correlations with
  ADI-R Social Domain scores, strong inter-rater
  agreement, and concordance with clinical
  diagnosis have been reported in support of the
  SRSs usefulness as a tool for screening and
  research.

• Exploratory factor analyses with promax rotation
  were completed with M statistical software on
  the 65 items of the SRS. Models with up to 5
  components were tested, mirroring both previous
  factor analysis work and the SRSs five provided
  subdomains. A higher-functioning subset of Group
  1 (VIQs70 n177) was analyzed to replicate
  datasets used in published SRS factor analyses.
  Review of these analyses suggested a 2- or
  3-factor best-fit model. Confirmatory analysis
  reflected a satisfactory fit with a 3-factor
  model (Goodness of fit index CFI0.9, Root Mean
  Square Error of Approximation0.08 Table 7),
  although the meaning of each factor was not
  obvious. The first three factors accounted for
  28.56 of the variance.
• Additional analyses using all of Group 1 also
  indicated either a 2- or 3-factor best-fit
  model, with nearly identical item mapping onto
  the factors as was found in the
  higher-functioning group. Confirmatory analyses
  using Group 1, however, did not demonstrate an
  acceptable fit with either a 2- or 3-factor model
  (2-factor CFI0.8, RMSEA 0.09 3factor
  CFI0.8, RMSEA0.09).
• Moreover, confirmatory analyses demonstrated an
  unsatisfactory fit with a single factor model
  using both Group 1 (CFI0.6, RMSEA0.1) and the
  higher-functioning subset (CFI0.7, RMSEA0.1).

• SRS Total Raw Scores were significantly
  correlated (pScores. Correlations for Group 1 were lower than
  previously reported data (range 0.24-0.48 see
  Table 3 ADI-R Social Total in Fig. 1).
  Correlations completed for Group 2, a
  higher-functioning subset of participants, were
  closer to the original findings (range
  0.40-0.58 highest correlations in restricted and
  repetitive behaviors domain ADI-R Social Total
  in Fig. 2).

Table 1
Demographics for Group 3
Item Mapping of Three Factor Model

• Raw SRS scores from both parents and teachers
  were significantly correlated (r 0.42, pChi-square tests demonstrated low agreement when
  comparing diagnosis to inter-rater agreement on
  attainment of cutpoint criterion. Both the lower
  general (Table 4) and higher clinical
• (Table 5) SRS cutpoints were considered.

Table 2

• AIMS
• We sought to replicate the following aspects of
  the validation of the SRS (Constantino Gruber,
  2005 Constantino et al., 2004)
• Neither parent- or teacher-reported scores were
  correlated with participants full scale IQ
  scores (Constantino et al., 2003).
• High correlations (ranging from 0.52 - 0.79) were
  found between SRS Total Scores and ADI-R
  Algorithm Subdomain Totals (Constantino et al.,
  2003).
• Inter-rater agreement, both between two parents
  and between a parent and teacher, was also high
  (correlations ranging from 0.75-0.91)
  (Constantino et al., 2003).
• The measures concurrent validity with diagnosis
  was supported by high sensitivity (.77) and
  specificity (.75) associated with use of the
  lower, general population cutpoint. Use of the
  measures higher cutpoint, intended for clinical
  populations, lowered sensitivity but raised
  specificity to 90.
• A single, continuously-distributed factor was
  found underlying SRS items, which accounted for
  34.97 of the variance.

• Measures
• The Social Reciprocity Scale (SRS Constantino
  Gruber, 2005) was mailed to families completing
  diagnostic evaluations or participating in
  research. Questionnaires were typically
  completed by the mother however, teacher
  respondent data is also available from some
  participants.
• The Autism Diagnostic Interview-Revised (ADI-R
  Lord, Rutter, Le Couteur, 1994 Rutter, Le
  Couteur, Lord, 2003) was administered to the
  childs parents either in a clinical setting or
  in their homes. In the case of multiple ADI-R
  administrations for a single participant, data
  from the ADI-R completed nearest to the SRS
  evaluation date were used in analysis.
• The Autism Diagnostic Observation Scale (ADOS
  Lord et al., 2000) was also included as part of
  the standard diagnostic assessment. In the case
  of multiple ADOS administrations for a single
  participant, data from the ADOS completed nearest
  to the SRS evaluation date were used.

Table 7

• DISCUSSION
• These data demonstrate the impact of
  heterogeneity within a dataset. Many of the
  discrepancies found between the current and
  published findings may be due to differences in
  VIQ distribution. While previous samples
  included only higher-functioning participants,
  nearly 30 of our larger sample had a VIQ
• Inter-rater correlations, while smaller than
  previous studies, demonstrated moderate agreement
  between parents and teachers. Parent and teacher
  scores may differ because of discrepant reference
  points used to rate an individual childs
  behavior. In this sample, findings support the
  use of the SRS as a qualitative, rather than a
  categorical, measure of autistic features.
• Researchers looking to recruit homogeneous
  samples will be interested in the high
  specificity (.87) and positive predictive value
  (89.5) seen in the group for which independent
  raters (in this case, a parent and a teacher)
  agreed on SRS higher cutpoint attainment (raw
  score of 85). They should be aware, however, of
  the low sensitivity (.25) and consequent
  potential for exclusion of appropriate cases
  resulting from use of this same criterion.
• Findings supporting a multiple- as opposed to a
  single-factor model may be due in part to
  differences in dataset homogeneity. Whereas our
  higher-functioning sample has similar IQs to
  previously published datasets (Constantino et
  al., 2004), the discrepant findings may be linked
  to the less-distinct nature of our nonASD sample,
  which most likely demonstrated more severe
  communication and social difficulties than seen
  in earlier samples.

• Parent and teacher cutpoint agreement, using
  general and clinical criteria (labeled SRS PT
  Table 6), was analyzed using clinical diagnosis
  as the gold standard. Use of inter-rater
  agreement criterion resulted in relatively low
  sensitivity (.55) and specificity (.67) for the
  lower cutpoint. Similar to previous analyses, we
  found low sensitivity (.25) and high specificity
  (.87) when using agreement on the higher SRS
  cutpoint as the criterion.

• RESULTS
• As suggested for research analyses (Constantino
  Gruber, 2005), only raw scores from the SRS were
  used. Missing SRS item data were prorated
  according to manual guidelines protocols were
  excluded if more than 15 items were missing.
• Similar to previous findings, analyses using
  Group 1 demonstrated no strong correlations
  between SRS total scores and verbal or nonverbal
  IQ scores (VIQ and NVIQ). While the correlation
  between NVIQ scores and SRS total scores was
  statistically significant, the r-value was very
  low (Table 3).

• METHODS
• Participants
• Data were collected from 313 participants between
  the ages of 4-18 years as part of a standard
  diagnostic evaluation. Children were referrals
  for autism clinics or recruited through research
  projects to be included in comparison groups
  (Group 1, Table 1). These assessments included
  administrations of the ADI-R, ADOS, psychometric
  tests, and various questionnaires.
• Of these 313 participants, parents of a subset of
  154 individuals completed the SRS before the
  ADI-R was administered (Group 2, Table 1).
• Data were also obtained from both the parents and
  teachers of another subset of 84 children
  participating in a longitudinal study (Group 3,
  Table 2). Individuals in this group received
  multiple SRS questionnaires protocols were
  chosen for analysis when teacher response data
  was available from the same time period.

Correlations between IQ, SRS Total Score, and
ADI-R Subdomain Scores for Group 1

• Results of the SRS (single rater, Table 6), using
  clinical diagnosis as the gold standard, were
  compared to the sensitivity and specificity of
  the ADI-R and ADOS. Current findings
  demonstrated similar SRS sensitivity (.79), but
  lower specificity (.50), than previous research.
  Analyses were also completed for combinations of
  the SRS with the ADOS and ADI-R. Combination with
  the ADOS maintained moderate sensitivity (.73)
  and increased specificity (.87). Combination
  with the ADI-R resulted in moderate sensitivity
  (.71) and specificity (.73).

References
Constantino, J.N., Davis, S.A., Todd, R.D.,
Schindler, M.K., Gross, M.M., Brophy, S.L., et
al. (2003). Validation of a brief quantitative
measure of autistic traits Comparison of the
Social Responsiveness Scale with the Autism
Diagnostic Interview Revised. Journal of Autism
and Developmental Disorders, 33(4), 427-433.
Constantino, J.N., Gruber, C.P., Davis, S.,
Hayes, S., Passanante, N., Przybeck T. (2004).
The factor structure of autistic traits. Journal
of Child Psychology and Psychiatry, 45(4),
719-726. Constantino, J., Gruber, C. (2005).
Manual Social Responsiveness Scale. Los
Angeles Western Psychological Services. Lord,
C., Rutter, M., Le Couteur, A. (1994). Autism
Diagnostic Interview-Revised A revised version
of a diagnostic interview for caregivers of
individuals with possible pervasive
developmental disorders. Journal of Autism and
Developmental Disorders, 24(5 ), 659-85. Lord,
C., Risi, S., Lambrecht, L., Cook, E. H.,
Leventhal, B. L., DiLavore, P. C., Pickles, A.,
Rutter, M. (2000). The Autism Diagnostic
Observation Schedule-Generic A standard measure
of social and communication deficits associated
with the spectrum of autism. Journal of Autism
and Developmental Disorders, 30(3), 205-223.
Rutter, M., Le Couteur, A., Lord, C. (2003).
Autism Diagnostic Interview-Revised. Los
Angeles Western Psychological Services.
- Correlation is significant to the 0.01 level
(2-tailed). Table 3
This work was supported by grants NIMH R01
MH066496 and R01 MH46865