Performance of LASSO logistic regression and random forest models in prediction of autism spectrum disorder
| Settings and prediction model | AUROC (95% CI) | AUPRC (95% CI) | F1 (95% CI) | Sensitivity target, % (95% CI) | Specificity, % (95% CI) | PPV, % (95% CI) |
| At age of 24 months (base case) | ||||||
| LASSO logistic regression | 0.758 (0.755 to 0.762) | 0.101 (0.098 to 0.104) | 0.193 (0.188 to 0.197) | 40* | 90.0 (89.7 to 90.4) | 8.7 (8.4 to 9.0) |
| 50 | 83.7 (83.2 to 84.2) | 6.7 (6.5 to 6.9) | ||||
| 70 | 66.1 (65.4 to 66.8) | 4.6 (4.5 to 47) | ||||
| Random forest | 0.775 (0.771 to 0.779) | 0.143 (0.138 to 0.148) | 0.246 (0.240 to 0.251) | 40 | 93.0 (92.6 to 93.5) | 12.1 (11.4 to 12.8) |
| 50 | 87.3 (86.7 to 87.9) | 8.4 (8.1 to 8.8) | ||||
| 70 | 69.6 (68.7 to 70.4) | 5.1 (4.9 to 5.2) | ||||
| At age of 18 months (younger) | ||||||
| LASSO logistic regression | 0.720 (0.716 to 0.723) | 0.066 (0.064 to 0.068) | 0.128 (0.124 to 0.132) | 50 | 78.4 (77.9 to 78.9) | 5.1 (5.0 to 5.2) |
| Random forest | 0.717 (0.714 to 0.721) | 0.067 (0.065 to 0.069) | 0.130 (0.125 to 0.134) | 50 | 78.8 (78.3 to 79.4) | 5.1 (5.0 to 5.2) |
| At age of 30 months (older) | ||||||
| LASSO logistic regression | 0.800 (0.797 to 0.803) | 0.148 (0.143 to 0.153) | 0.255 (0.249 to 0.261) | 50 | 90.4 (90.0 to 90.8) | 11 (10.6 to 11.5) |
| Random forest | 0.832 (0.828 to 0.835) | 0.234 (0.227 to 0.240) | 0.326 (0.322 to 0.331) | 50 | 95.6 (95.3 to 95.8) | 21.0 (20.2 to 21.8) |
*The sensitivity threshold of 40% was selected to be comparable with the estimated sensitivity of 33%–39% for the existing autism-specific screening tools from real-world clinical settings.7 ,8
AUPRC, area under precision-recall curve; AUROC, area under receiver operator characteristic curve; LASSO, least absolute shrinkage and selection operator; PPV, positive predictive value.