Autism Screening Tool STAT Shows High Sensitivity but Modest Specificity in Vietnamese Toddlers

A new cross-sectional study from Vietnam reports the first formal validation of the Screening Tool for Autism in Toddlers and Young Children (STAT) in the country, finding that the tool correctly identified most toddlers later diagnosed with autism but flagged a substantial number of children who did not receive the diagnosis. Among 86 children aged 24 to 36 months with language delay, the STAT achieved a sensitivity of 89.1 percent at its recommended cut-off score of 2 or higher, meaning it detected roughly nine of every ten children whose subsequent clinical evaluation confirmed autism spectrum disorder. Its specificity, however, was only 50.0 percent, so half of the children without a diagnosis were incorrectly classified as being at risk. The study, published in the Egyptian Pediatric Association Gazette, used DSM-5-based clinical diagnosis as the reference standard and was conducted at a tertiary pediatric referral hospital in Ho Chi Minh City between December 2022 and August 2023.

The STAT is an interactive, play-based, Level 2 screening instrument designed specifically for children aged 24 to 36 months. Level 2 tools are intended for targeted screening among children who have already been identified as having developmental concerns or an elevated likelihood of autism, rather than for universal screening of the general pediatric population. The instrument comprises 12 items grouped into four domains—play, requesting, directing attention, and imitation—each contributing a maximum of one point, with failed items contributing partial credit of 0.5 points in the Play and Requesting domains and 0.25 points in the Directing Attention and Imitation domains. The four domain scores are summed to produce a total score ranging from 0 to 4, with a score of 2 or higher indicating increased risk for autism. The assessment takes approximately 20 minutes and can be administered by trained professionals such as early intervention specialists and speech-language therapists, which makes it attractive for settings where comprehensive diagnostic resources are scarce.

The tool has already been translated and incorporated into Vietnam’s Ministry of Health 2021 national guidelines for early autism detection, yet its diagnostic validity had never been formally evaluated among Vietnamese children. This gap mattered because the performance of developmental screening and diagnostic instruments can vary across cultural and geographical contexts, where local perceptions of disability and developmental norms influence caregiver reporting and clinical interpretation. The STAT has been adapted and used in several countries, including the United States, Canada, Japan, and Taiwan, and previous research has shown moderate agreement between STAT classification and expert clinical diagnosis, as well as moderate concordance with ADOS-2 outcomes, with the strongest performance observed in younger age groups. The Vietnamese study set out to provide the empirical evidence needed to judge whether the tool’s original performance estimates hold in a lower-middle-income clinical setting.

The research team enrolled children attending the Department of Physiotherapy and Rehabilitation at a tertiary pediatric referral hospital in southern Vietnam. Inclusion criteria required children aged 24 to 36 months presenting with language delay plus at least one additional feature: absent or markedly limited eye contact, stereotyped behaviors, sensory abnormalities, or a sibling previously diagnosed with autism. Children were excluded if they had a prior autism diagnosis, confirmed genetic or neurological disorders such as cerebral palsy or Down syndrome, motor impairments that could interfere with the assessment, or incomplete data. Of 110 families approached, nine declined participation and 101 consented; four children did not complete the STAT assessment and 11 did not complete the diagnostic evaluation, leaving 86 children in the final analysis. The median age was 30 months, 65.1 percent of participants were male, and the prevalence of autism in the sample was high at 74.4 percent, reflecting the clinically enriched, referral-based nature of the cohort.

A key methodological strength of the study was the blinding of the diagnostic process. Certified speech therapists trained in the STAT administered the screening in a standardized, child-friendly setting, with each assessment lasting about 20 minutes. All children were then referred for DSM-5-based diagnostic evaluation regardless of their STAT result. A child psychologist with formal training conducted the diagnostic assessment, which included a parent or caregiver interview guided by DSM-5 criteria, direct clinical observation, and review of available developmental history and records. The psychologist was not informed of either the STAT total score or the individual item results, ensuring that screening outcomes could not bias the reference diagnosis. The psychologist did know that each child had been referred because of developmental concerns and had access to caregiver-reported history, since these are necessary components of a clinical assessment.

The diagnostic results painted a nuanced picture. Using the recommended cut-off of 2 or higher, 57 of 64 children diagnosed with autism were classified as at risk, yielding the sensitivity of 89.1 percent (95 percent confidence interval: 79.1 to 94.6 percent). Among the 22 children without autism, 11 were correctly classified as not at risk, giving the specificity of 50.0 percent (95 percent confidence interval: 30.7 to 69.3 percent). The positive predictive value was 83.8 percent and the negative predictive value was 61.1 percent. Overall classification accuracy was 79.1 percent. Receiver operating characteristic analysis produced an area under the curve of 0.7656 (95 percent confidence interval: 0.639 to 0.883), which the authors interpreted as fair overall discriminative ability. At this threshold, 18 children—20.9 percent of the cohort—were misclassified, including 11 false positives and 7 false negatives.

The comparatively low specificity stands out against earlier validation work. In the original validation study, the STAT showed sensitivity of 92 percent and specificity of 85 percent at the same cut-off in a clinic-referred sample, and a Taiwanese adaptation reported sensitivity of 97 percent and specificity of 93 percent when differentiating autism from developmental delay. The authors attribute the discrepancy largely to case mix: their cohort consisted exclusively of children with language delay and additional developmental concerns, a group in which symptom overlap between autism and other neurodevelopmental conditions is common and differentiation is inherently more difficult. Non-autistic children in this setting may display developmental and social-communication difficulties that resemble autism features, driving false positives. Differences in diagnostic framework may also contribute, since earlier studies used relatively small clinic-based samples assessed under DSM-IV criteria, whereas this study applied DSM-5.

In an exploratory analysis, the researchers used the Youden Index to identify the cut-off with the best combined sensitivity and specificity in their sample. A data-derived threshold of 2.5 performed best, with sensitivity of 73.4 percent and specificity of 68.2 percent and a Youden Index of 0.416. As the cut-off increased, sensitivity decreased while specificity increased across the range of thresholds examined. The authors emphasize that this threshold was identified and evaluated in the same relatively small cohort, so it should be regarded as hypothesis-generating and requires independent validation before any change to the recommended cut-off is considered. They note that the 2.5 threshold aligns with findings from Taiwan, where a threshold of 2.5 was identified as optimal for toddlers aged 18 to 24 months, though that study included younger participants and a comprehensive multidisciplinary diagnostic process. Other work has reported an even higher optimal cut-off of 2.75 for children under 24 months, suggesting that optimal thresholds may vary across developmental stages.

The authors caution that the predictive values observed should not be extrapolated to community screening. Positive and negative predictive values depend on prevalence, and the 74.4 percent autism prevalence in this clinically enriched cohort is far higher than what would be seen in the general toddler population. In a high-prevalence setting, the positive predictive value of 83.8 percent indicates a clinically meaningful referral yield, but the negative predictive value of 61.1 percent means a negative STAT result cannot be used to rule out autism—seven children with confirmed diagnoses had negative screening results. The study also has other limitations: recruitment from a tertiary referral hospital produced a pre-selected high-risk sample, the final sample fell short of the prespecified target of 95 participants, only 22 children were classified as non-autistic, producing wide confidence intervals around specificity, and inter-rater reliability of STAT administration was not formally assessed.

Despite these caveats, the findings offer practical support for current screening practice in Vietnam and similar settings. The national guidelines recommend a STAT cut-off of 2 or higher to prioritize sensitivity and facilitate early referral, and the authors conclude that their results support retaining this threshold for primary screening within high-risk clinical populations. In resource-limited settings where gold-standard diagnostic tools such as the ADOS-2 are constrained by cost, time, and specialized training requirements, the STAT’s brief administration time and interactive, observational format may reduce reliance on parent-report bias, which can affect parent-report instruments like the M-CHAT-R/F. The authors stress that the STAT is a screening instrument intended to support early risk identification and referral, not to establish or exclude a diagnosis, and that children with persistent developmental concerns should continue to receive developmental surveillance and, when clinically indicated, comprehensive diagnostic evaluation regardless of the screening result. Larger multicenter and community-based studies, they conclude, are needed to confirm these findings and assess how well the tool’s performance generalizes beyond tertiary referral clinics.

Subject of Research: Validation of the STAT autism screening tool in Vietnamese toddlers with language delay

Article Title: Validating the Screening Tool for Autism in Toddlers and Young Children (STAT) in children aged 24–36 months in a low- to middle-income setting

Article References: Nguyen, T. B. N., Ngo, H. K. T., Dinh, G., & Nguyen, A. N. (2026). Validating the Screening Tool for Autism in Toddlers and Young Children (STAT) in children aged 24–36 months in a low- to middle-income setting. Egyptian Pediatric Association Gazette, 74(1), Article 116. https://doi.org/10.1186/s43054-026-00615-3

Image Credits: AI Generated

DOI: 10.1186/s43054-026-00615-3

Keywords: autism spectrum disorder, STAT, screening tool, validation study, Vietnam, toddlers, language delay, DSM-5, diagnostic accuracy, sensitivity, specificity, low- and middle-income countries