Autism Spectrum Disorder Screening Instruments for Very Young Children: A Systematic Review
Table 5
Research summaries for the Screening Test for Autism in Two-Year-Olds (STAT), Parent Observation of Early Milestones Scale (POEMS), and Autism Detection in Early Childhood (ADEC).
Small: 19 had ASD. All participants were of high risk because of diagnosed older sibling.
All participants were of high risk because of diagnosed older sibling.
All participants were of high risk based on being scheduled for an evaluation at a specialty clinic.
Did investigators use a sample matched for developmental level?
No.
No.
Yes.
Screening instrument
Was there anything about how the screener was administered that would be different from its intended use in a nonresearch, community setting?
No.
The POEMS was filled out by families every three months. Giving the POEMS many times could sensitize parents to ASD behaviors, especially since they would already be so because of their older child with ASD. However, this is consistent with its intended use within this study.
The team administering the screener received some training.
Were there any issues regarding the way it is scored in the study?
No.
No.
See evaluation domain below.
Reference standard
Did all children receive a BED from in-person evaluations? How extensive was the information available to the clinician making the Best Estimate Diagnosis?
Yes; the information available included ADOS and Mullen.
No direct examination for diagnostic status. They relied on parent report of community diagnosis. They were able to give ADI-R to 3 out of the 9 children with ASD to confirm.
BED with cognitive assessment, ADOS, and ADI-R if AD or PDD-NOS diagnosis considered. “77.5% had an independent confirmatory diagnosis from either two other independent professionals who were recognized by the state’s autism association or other medical professionals such as pediatricians and psychologists.”
Were the reference standard evaluators blind to the screener risk status of the children?
Not reported.
Most likely, considering they were clinicians in the community who were independent of the study.
Yes.
What diagnostic outcome categories were used to test prediction from screener to reference standard?
Autism, PDD-NOS, DD, LI (language impairment), BAP (Broader Autism Phenotype), and no diagnosis.
Only categories were ASD versus no-ASD. This is a departure from most studies, which also include other DDs. This appeared to be a function of the study methods, which involved using reports the parents obtained from the community.
For initial analyses, ASD (AD + PDD-NOS), other DDs, and TD. However, authors indicated that the ADEC is intended to detect Autistic Disorder, so PDD-NOS was left out for ROC analysis and this suggests that it will detect more severe children on the spectrum.
Timing and flow
Was there excessive attrition through any phase of screening and evaluation?
No.
No, sample size was adequate.
N/A.
Were there conditions besides attrition that filtered the negative and positive screens from the original screening to the reference standard diagnostic testing phase?
No.
No, sample size was adequate.
N/A.
Evaluation
How were performance/predictive values calculated?
The groups were combined as follows: autism and PDD-NOS were all ASD; the others were non-ASD. With this categorization, the most false positives were found for 12-13-month-olds, so Se and Sp were calculated both with and without them. They achieved acceptable Se and Sp levels by raising the cutoff score compared to that for the 24–36-month-olds.
Predictive validity was first explored by forming two groups: infant siblings who were confirmed to have ASD at age of 36 months () and those who were not (). They then compared how the POEMS score diverged over the different age levels.
Investigators left the PDD-NOS group out and compared AD to Other Developmental Disabilities (ODD) with and without the TD group. This can inflate performance compared to studies that include milder children.
Was performance/prediction for younger versus older children explored?
Yes—reported false positives for three different groups between 12 and 24 months. More false positives for the 12-13-month group than older children.
Yes—see below. The sensitivity got higher as age progressed over 3, 6, 9, 12, 18, and 24 months. Sensitivity reached the acceptable level at 18 months.
Yes, for 12–24 versus 24–36 with no differences found. Did not look at the youngest children (under 18 months).
Performance
What were the performance/predictive values?
Using a cutoff of 2.75: Se = .95, Sp = .73, PPV = .56, NPV = .97. Excluding the 12-13-month-olds: Se = .93, Sp = .83, PPV = .68, NPV = .97.
A cutoff score of 70 resulted in a mean sensitivity (across all age groups) of .74 and specificity of .87; PPV overall was .21. At 12 months, Se = .71 and Sp = .68. At 18 months, Se = .89 and Sp = .65.
Using a cutoff score of 11.
Unmatched AD versus ODD: Se = 1.0; Sp = .77. AD versus ODD + TD: Se = 1.0; Sp = .89.
Matched AD versus ODD: Se = 1.0; Sp = .74. AD versus ODD + TD: Se = 1.0; Sp = .90.
What was the developmental level of children detected?
The sample included higher functioning children; at mean of 24 months, MSEL Early Learning Composite: M = 93.5; SD = 23.3.
Not reported.
Not reported.
Of the false positives for ASD, what proportion had other developmental or learning disabilities?
50% of false positives had other DD diagnoses.
Not reported.
10/70.
STAT = Screening Test for Autism in Two-Year-Olds; POEMS = Parent Observation of Early Milestones Scale; ADEC = Autism Detection in Early Childhood; ASD = autism spectrum disorder; BED = Best Estimate Diagnosis; ADOS = Autism Diagnostic Observation Schedule; ADI-R = Autism Diagnostic Interview-Revised; AD = Autistic Disorder; PDD-NOS = Pervasive Developmental Disorder-Not Otherwise Specified; DD = developmental delay; LI = language impairment; BAP = Broader Autism Phenotype; TD = typically developing; ROC = receiver operator characteristics; Se = sensitivity; Sp = specificity; ODD = Other Developmental Disabilities; PPV = positive predictive value; NPV = negative predictive value; MSEL = Mullen scales of early learning; M = mean; SD = standard deviation.
