Utility of the 3Di short version in the identification and diagnosis of autism in children at the Kenyan coast

Abstract

Introduction: The precise epidemiological burden of autism is unknown because of the limited capacity to identify and diagnose the disorder in resource- constrained settings, related in part to a lack of appropriate standardised assessment tools and health care experts. We assessed the reliability, validity, and diagnostic accuracy of the Developmental Diagnostic Dimensional Interview (3Di) in a rural setting on the Kenyan coast. Methods: Using a large community survey of neurodevelopmental disorders (NDDs), we administered the 3Di to 2,110 children aged between 6 years and 9 years who screened positive or negative for any NDD and selected 242 who had specific symptoms suggestive of autism based on parental report and the screening tools for review by a child and adolescent psychiatrist. On the basis of recorded video, a multi-disciplinary team applied the Autism Diagnostic Observation Schedule to establish an autism diagnosis. Internal consistency was used to examine the reliability of the Swahili version of the 3Di, tetrachoric correlations to determine criterion validity, structural equation modelling to evaluate factorial structure and receiver operating characteristic analysis to calculate diagnostic accuracy against Diagnostic Statistical Manual of Mental Disorders (DSM) diagnosis. Results: The reliability coefficients for 3Di were excellent for the entire scale {McDonald’s omega (w) = 0.83 [95% confidence interval (CI) 0.79–0.91]}. A higher-order three-factor DSM-IV-TR model showed an adequate fit with the model, improving greatly after retaining high-loading items and correlated items. A higher-order two-factor DSM-5 model also showed an adequate fit. There were weak to satisfactory criterion validity scores [tetrachoric rho = 0.38 (p = 0.049) and 0.59 (p = 0.014)] and good diagnostic accuracy metrics [area under the curve = 0.75 (95% CI: 0.54–0.96) and 0.61 (95% CI: 0.49–0.73] for 3Di against the DSM criteria. The 3Di had a moderate sensitivity [66.7% (95% CI: 0.22–0.96)] and a good specificity [82.5% (95% CI: 0.74–0.89)], when compared with the DSM-5. However, we observed poor sensitivity [38.9% (95% CI: 0.17–0.64)] and good specificity [83.5% (95% CI: 0.74–0.91)] against DSM-IV-TR. Conclusion: The Swahili version of the 3Di provides information on autism traits, which may be helpful for descriptive research of endophenotypes, for instance. However, for accuracy in newly diagnosed autism, it should be complemented by other tools, e.g., observational clinical judgment using the DSM criteria or assessments such as the Autism Diagnostic Observation Schedule. The construct validity of the Swahili 3Di for some domains, e.g., communication, should be explored in future studies.