Exploratory and Confirmatory Factor Analyses of the Arabic Version of the Childhood Autism Rating Scale (CARS-2)

Overview of the CARS-2 Factor Analysis Study

• This research, entitled "Exploratory and confirmatory factor analyses of the Arabic version of the Childhood Autism Rating Scale," was conducted by Bander Alotaibi (University of Hail, Saudi Arabia) and Abdulhadi Alotaibi (Umm Al-Qura University, Saudi Arabia).
• The study evaluates the factor structure, reliability, and validity of the Arabic version of the Childhood Autism Rating Scale, Second Edition ($CARS-2$).
• The sample consisted of $301$ children diagnosed with Autism Spectrum Disorder ($ASD$) according to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition ($DSM-5$) criteria.
• Key Findings:     - Internal consistency reliability was found to be $0.79$.     - Inter-rater reliability was measured at $0.65$.     - Intraclass correlation coefficient ($ICC$) was $0.76$.     - Exploratory Factor Analysis ($EFA$) suggested a three-factor solution: Communications, Emotions, and Senses and Physical.     - Confirmatory Factor Analysis ($CFA$) confirmed a $14$-item, three-factor model that adequately fits the data ($RMSEA = 0.08$).     - The findings support the continued use and relevance of the Arabic version of $CARS-2$ in identifying and assessing $ASD$.

Context and Evolution of ASD Diagnosis

• Prevalence Estimates: According to the Autism and Developmental Disabilities Monitoring Network ($2018$), $ASD$ prevalence is estimated at one in $59$ children.
• Clinical Characteristics: $ASD$s are neurodevelopmental disorders defined by skill deficits, cognitive/motor delays, and sensory sensitivities.
• Diagnostic Transitions:     - $DSM-IV$ Criteria: Included Autistic Disorder ($AD$), Asperger Syndrome ($AS$), and Childhood Disintegrative Disorder ($CDD$) under the Pervasive Developmental Disorder category. Pervasive Developmental Disorder Not Otherwise Specified ($PDD-NOS$) was previously included.     - $DSM-5$ Changes: Removed impaired language as a primary marker; replaced it with deficits in social communication and restrictive/repetitive interests and behaviors. Social Communication Disorder ($SCD$) was added.     - Clinical Challenges: Evaluations are difficult due to continuous modifications in diagnostic criteria, specifically the move from $DSM-IV$ to $DSM-5$.

Assessment Tools: The CARS and CARS-2

• The original Childhood Autism Rating Scale ($CARS$) was published in $1980$ as a clinical observation tool utilizing a four-point scale.
• The $CARS-2$ (published in $2010$) includes two versions:     - Standard Form ($CARS2-ST$): Maintains the original scale format.     - High-Functioning ($CARS2-HF$): Designed for individuals aged six years and older with an estimated $IQ$ of $80+$ who communicate fluently.
• Methodology of Assessment: Ratings are based on direct observations by clinicians combined with collateral information from parents or teachers.
• Sensitivity Data:     - $CARS2-ST$ sensitivity based on $DSM-5$ is $0.84$, compared to $0.81$ for $DSM-IV-TR$.     - Diagnostic agreement between $CARS$ and $DSM-5$ is approximately $84\%$.
• Known Limitations: The $CARS$ has a historical propensity to classify young children with intellectual disabilities ($ID$) as having autism (Lord, $1995$).

Study Methodology and Participants

• Sample Location: The Unit of Evaluation and Diagnosis at Hail Charitable Association for Children with Disability in Hail, Saudi Arabia.
• Participant Demographics ($N = 301$):     - Gender: $77.4\%$ Male ($n = 233$), $22.6\%$ Female ($n = 68$).     - Age: $47.8\%$ were $2-5$ years old; $52.2\%$ were $6-12$ years old.     - Diagnoses: $56.8\%$ Autistic Disorder ($AD$); $43.2\%$ Social Communication Disorder ($SCD$).
• Exclusion Criteria: Participants with Attention Deficit Hyperactivity Disorder ($ADHD$) associated with $ID$ and stereotyped movements were excluded due to uncertain nosological status.
• Evaluation Team: At minimum, a licensed clinical psychologist, child psychiatrist, education specialist, speech pathologist, and occupational therapist (expertise ranging from $5$ to $15+$ years).
• Procedure:     - Psychiatrists provided $DSM-5$ clinical diagnoses independently.     - Clinical psychologists and speech pathologists rated the $CARS-2$ independently from the psychiatrist's diagnosis to reduce rater bias.     - Data derived from $CARS2-ST$ (for children aged $6+$) and $CARS2-HF$ (for children under $6$).

Statistical Analysis Framework

• Reliability Metrics: Cronbach’s $\alpha$ (internal consistency), Intraclass Correlation Coefficient ($ICC$), and Cohen’s $\kappa$ (agreement between $CARS-2$ and $DSM-5$).
• Diagnostic Accuracy: Receiver operating characteristic ($ROC$) analyses, sensitivity, specificity, and Area Under the Curve ($AUC$).
• Construct Validity:     - Data was randomly split into Subsample 1 ($n = 151$) and Subsample 2 ($n = 150$).     - Exploratory Factor Analysis ($EFA$): Used Principal Axis Factoring ($PAF$) with oblique Promax rotation. Kaiser–Meyer–Olkin ($KMO$) measure was $0.875$, exceeding the $0.6$ threshold. Bartlett’s test of sphericity was statistically significant.     - Confirmatory Factor Analysis ($CFA$): Used Analysis of Moment Structures ($Amos$) $25.0$ to evaluate model fit using indices like $\chi^2$, $CFI$, $TLI$, $RMSEA$, $AIC$, and $BIC$.
• Retention Criteria: Eigenvalues > 1.00; factor pattern loadings > 0.40.

Research Results: Diagnostic Accuracy and Reliability

• Participant Scores: Mean $CARS-2$ score was $36.50$ ($SD = 5.27$), ranging from $15$ to $49$.
• Ideal Cutoff Point: A cutoff score of $\ge 26$ was determined for children younger than $13$.
• Diagnostic Efficacy of Cutoff $26$:     - Sensitivity: $0.96$.     - Specificity: $0.70$.     - Correctly classified $250$ of $301$ children.     - Positive Predictive Value ($PPV$): $76.2\%$.     - Negative Predictive Value ($NPV$): $97.2\%$.     - $AUC$: $0.65$ ($95\%\,CI = 0.53-0.78$, $P = 0.03$).
• Reliability Statistics:     - Internal consistency (total scores): $\alpha = 0.79$.     - Inter-rater agreement (Cohen’s $\kappa$): $0.65$ (moderate agreement).     - Intraclass Correlation Coefficient ($ICC$): $0.76$.

Research Results: Validation and Factor Structure

• The EFA produced a three-factor solution accounting for $55.83\%$ of the common variance.
• Factor 1: "Communications"     - Variance: $40.40\%$.     - Items: Imitation ($0.758$), Verbal communication ($0.743$), Non-verbal communication ($0.570$), Level and consistency of intellectual response ($0.484$), Visual Response ($0.424$), Relation to people ($0.429$).     - Internal consistency: $\alpha = 0.70$.
• Factor 2: "Emotions"     - Variance: $8.29\%$.     - Items: Emotional response ($0.584$), Fear and nervousness ($0.532$), General impressions ($0.449$), Adaptation to change ($0.400$).     - Internal consistency: $\alpha = 0.67$.
• Factor 3: "Senses and Physical"     - Variance: $7.13\%$.     - Items: Body use ($0.463$), Object use ($0.889$), Taste, smell, touch response and use ($0.773$), Activity level ($0.592$).     - Internal consistency: $\alpha = 0.65$.
• Non-loading Item: Listening Response did not exceed the $0.40$ loading threshold.
• Factor Correlations: Medium to large correlations between factors.     - Communications/Emotions: $0.643$.     - Communications/Senses Physical: $0.613$.     - Emotions/Senses Physical: $0.569$.

Confirmatory Factor Analysis (CFA) Fit Indices

• The CFA tested several models on Subsample 2 ($n = 150$).
• Initial Model (15 items): $\chi^2(df=87) = 195.097$, $CFI = 0.86$, $TLI = 0.84$, $RMSEA = 0.09$.
• 14-Item Model (excluding Listening Response): $\chi^2(df=74) = 158.159$, $CFI = 0.89$, $TLI = 0.86$, $RMSEA = 0.08$.
• Final Model (with 14 items and two error covariance paths):     - $\chi^2(df=72) = 141.270$     - $CFI = 0.91$ (adequate fit)     - $TLI = 0.88$ (adequate fit)     - $RMSEA = 0.08$ (good fit)     - $GFI = 0.89$     - $NFI = 0.83$     - $AIC = 207.270$     - $CAIC = 339.621$
• Paths were added between error terms for items $1$ and $2$, and items $3$ and $4$ to improve the model.

Discussion and Significance

• Relevance to Cultures: Results suggest the $CARS-2$ factor structure is appropriate for the Arabic context, similar to Western and other non-Western cultures like Swedish, Japanese, and Indian populations.
• Symptom Domains: The three-factor solution aligns more closely with $DSM-5$ symptom domains rather than $DSM-IV$.
• Item Ambiguity:     - Listening Response: Wording such as "use his senses" is vague; clinicians may interpret it as a social/communicative tool or a sensory pattern. Its loading was split between Communications ($0.21$) and Senses and Physical ($0.37$).     - Visual Response: Loaded on Senses and Physical, differing from previous studies (e.g., Moulton et al., $2016$) where it loaded on Communications.     - Level and Consistency of Intellectual Response: Loading on Communications suggests clinicians view intellectual variability as a core feature of the disorder.
• Cutoff Comparison: The ideal Arabic cutoff of $26$ is consistent with Lebanese studies (Akoury-Dirani et al., $2013$) but lower than suggested cutoffs in Japan ($30$) and India ($33$).

Limitations and Future Directions

• Sample Bias: The sample was drawn from a specific clinical population and did not compare results against other measures like the Autism Diagnostic Observation Schedule ($ADOS$) or the Autism Behavior Checklist ($ABC$).
• Test-Retest: Test-retest reliability was not determined in this study.
• Age Variability: While the wide age range ($2-12$) is a strength, it may obscure findings specific to narrow age brackets.
• Future Recommendations:     - Distinguish more clearly between sensory and social concepts in item terminology.     - Use concurrent administration of tests specifically for study purposes.     - Directly compare factors identified across different discrete age groups.