Author ORCID Identifier

Date of Award

Summer 8-9-2022

Degree Type


Degree Name

Master of Public Health (MPH)


Public Health

First Advisor

Brian Barger

Second Advisor

Terri Pigott


INTRODUCTION: Early identification and early childhood intervention are known to improve cognitive, social, and adaptive functioning and improve the quality of life of children diagnosed with autism spectrum disorders. Since early intervention and treatment success depends on early identification, early childhood screening with high-quality screening tools is critical. Recovering missing data and adjusting the data for epidemiologically reported prevalence can provide more accurate estimates of missed cases when studies report results without follow-up data to verify missed cases.

AIM: The purpose of this study was to conduct an umbrella review and meta-analysis of diagnostic accuracy (MADA) of autism spectrum disorder (ASD) screening tools to evaluate the effect of three missing data and prevalence adjustment methods on pooled diagnostic accuracy metrics.

METHODS: This study selected a final sample size of 28 previously reviewed population-level studies for inclusion that focused on children ages 6 to 72 months (total screened = 205,934). The three data adjustments included: recovering missing or unreported data using standard diagnostic accuracy formulas; epidemiologically adjusting the recovered data using reported ASD prevalence by year and nation of origin; epidemiologically adjusting the recovered data using CDC yearly reported ASD prevalence. Using a bivariate random effects Reitsma model, the study calculated pooled sensitivity, specificity, and a pooled diagnostic odds ratio for each data adjustment approach. The bivariate analysis also applied metaregression models to simultaneously compare the pooled sensitivity and false positive rate for each data adjustment.

RESULTS: Sensitivity systematically decreased for each data source in both the univariate and bivariate analysis. Recovering missing data increased sensitivity in the univariate and bivariate analysis (Se = 0.732 and Se = 0.759, respectively). Univariate sensitivity decreased after the national (Se = 0.628) and CDC prevalence (.422) adjustments. Sensitivity in the bivariate analysis decreased after the national (Se = .652) and CDC prevalence (Se = .384) adjustments. The current results suggest that recovering missing data and adjusting for national prevalence may be an appropriate data adjustment method to obtain unbiased assessments of diagnostic accuracy and help detect screening tools weaker than those promoted from cross-sectional studies.


File Upload Confirmation