Evaluating developmental screening in clinical practice

OBJECTIVE To demonstrate a method of evaluating accuracy of developmental screening modeled on the evidence-based medical literature. METHOD A retrospective review was performed on 418 children screened with the Denver II by a trained technician. Two models for analyzing screening data were examined, using predictive values and likelihood ratios (LR+ and LR-). RESULTS The technician, working at 20% time, screened 44% of eligible children. There were 129/418 (31%) children with Suspect Denver II results, 115/418 who were referred, 81/115 (70%) who were evaluated by Early Intervention, and 64/81 (79%) who qualified for services. The uncorrected positive predictive value for the Denver II alone (44%) was insufficient to meet the preset standard of 60%, but the LR+ of 4.16 indicated a significant contribution of test information to improving predictive value. Combining test results with information from the parent-technician conference to achieve a referral decision resulted in an uncorrected predictive value of 56%, which rose with correction for children referred but not evaluated to 72% (LR+ 10.33). Negative predictive values and likelihood ratios of a negative test and a non-referral decision achieved recommended levels. Parents who expressed concern were significantly more likely to complete recommended evaluation than those who did not (82% vs 58%, p < .01). Results were in the same range as in published studies with other screening tests but showed three areas for improvement: screening more children, more carefully supervising some referral decisions, and getting more children to evaluation. CONCLUSION Levels of predictive accuracy above 60% can be obtained by combining different types of information about development to make decisions about referral for more complete evaluation. Systematic study of such combinations could lead to improved predictive accuracy of screening programs and support attempts to close the gap between referral and evaluation.