# Early screening of childhood ASD at primary care hospitals in western China: a multi-center study in Chengdu, Sichuan Province

## TL;DR

This study evaluated four screening tools for autism in young children at primary care hospitals in Chengdu, finding that the ABC score and age were the strongest predictors of autism.

## Contribution

The study provides real-world evidence on the effectiveness of different ASD screening tools in primary care settings in western China.

## Key findings

- The ABC screening tool had the highest positive predictive value (94.57%) among the four tools tested.
- Child age was a strong independent predictor of an ASD diagnosis.
- The AWS and CHAT-23 tools also showed significant associations with ASD diagnosis in the adjusted model.

## Abstract

Autism spectrum disorder (ASD) prevalence is increasing rapidly, the screening efficacy of ASD at primary care hospital directly affects diagnostic efficacy. Currently, early ASD screening capacity in western China are limited. In this study we assessed the efficacies of four different ASD screening methods at primary care hospital in Chengdu.

We recruited children 18–48 months through 20 primary care hospitals in Chengdu. Warning sign checklist [including autism warning sign (AWS) and social behavior observation (SB)], the Chinese-validated version of the Checklist for Autism in Toddlers (CHAT23), and autism behavior checklist (ABC) were utilized to identify potential ASD cases. Confirmation diagnosis was conducted at tertiary hospitals. Multivariate linear regression were conducted to compare efficacies of screening methods and to identify factors for positive ASD diagnosis.

In this real-world study, a total of 14,298 children attended the early ASD screening from 20 primary care hospitals in Chengdu. 13,458 (94.1%) children provided sufficient information and 3,502 (26.0%) children were concurrently screened with four ASD screening tools. The screen-positive rates varied across tools, in descending order, was AWS (3.1%), SB (2.7%), CHAT-23 (2.18%), and the ABC (1%). The overall referral rate reached 66.88%. Among the screened population who completed diagnostic follow-up, the detected prevalence of ASD was 0.79%. Regarding screening accuracy, the positive predictive values (PPV) were ABC (94.57%), SB (91.95%), and AWS (87.06%), CHAT-23 (84.72%) from highest to lowest. The multivariable logistic regression analysis showed factors associated with a final ASD diagnosis, adjusted for age, gender, and primary hospital. The ABC demonstrated the strongest independent association with ASD diagnosis (β = 8.40, SE = 1.92, p < 0.01). Age was also a strong significant predictor (β = 6.12, SE = 2.00, p < 0.01). Both the AWS (β = 3.43, SE = 1.85, p < 0.05) and CHAT-23 screening result (β = 2.63, SE = 1.58, p < 0.05) were independently and positively associated with the diagnosis. SB (β = 2.39, SE = 1.79) and gender (β = 1.35, SE = 1.64) were not statistically significant in the adjusted model.

In summary, the ABC score and child’s age emerged as the most robust indicators of ASD in primary care hospitals. The effects of AWS and CHAT-23 support their utility as components in a multi-tool screening cascade. These findings support the practical value of integrated, tiered screening strategies in real-world public health settings.