Development and validation of a risk score model for predicting autism based on pre- and perinatal factors.

Background: The use of pre- and perinatal risk factors as predictive factors may lower the age limit for reliable autism prediction. The objective of this study was to develop a clinical model based on these risk factors to predict autism.

Methods: A stepwise logistic regression analysis was conducted to explore the relationships between 28 candidate risk factors and autism risk among 615 Han Chinese children with autism and 615 unrelated typically developing children. The significant factors were subsequently used to create a clinical risk score model. A chi-square automatic interaction detector (CHAID) decision tree was used to validate the selected predictors included in the model. The predictive performance of the model was evaluated by an independent cohort.

Results: Five factors (pregnancy influenza-like illness, pregnancy stressors, maternal allergic/autoimmune disease, cesarean section, and hypoxia) were found to be significantly associated with autism risk. A receiver operating characteristic (ROC) curve indicated that the risk score model had good discrimination ability for autism, with an area under the curve (AUC) of 0.711 (95% CI=0.679-0.744); in the external validation cohort, the model showed slightly worse but overall similar predictive performance. Further subgroup analysis indicated that a higher risk score was associated with more behavioral problems. The risk score also exhibited robustness in a subgroup analysis of patients with mild autism.

Conclusion: This risk score model could lower the age limit for autism prediction with good discrimination performance, and it has unique advantages in clinical application.