Development and validation of a postnatal risk score that identifies children with prenatal alcohol exposure

Background This study aimed to develop an efficient and easily calculable risk score that can be used to identify an individual's risk of having been exposed to alcohol prenatally. Methods Data for this study were collected as part of the Collaborative Initiative on Fetal Alcohol Spectrum Disorders, Phases 2 and 3. Two cohorts (ages 5 to 17 years) completed a comprehensive neurobehavioral battery and a standard dysmorphology exam: a development cohort (DC; n = 325) and a comparative cohort (CC; n = 523). Both cohorts included two groups: those with histories of heavy prenatal alcohol exposure (AE-DC, n = 121; AE-CC, n = 177) and a control group that included subjects with minimal or no prenatal alcohol exposure (CON-DC, n = 204; CON-CC, n = 346). Behavioral assessments and physical exam data were combined using regression techniques to derive a risk score indicating the likelihood of prenatal alcohol exposure. Subjects were then divided into two subgroups: (1) low risk and (2) high risk. Chi-square (chi(2)) determined classification accuracy and ROC curves were produced to assess the predictive accuracy. Correlations between risk scores and intelligence quotient and executive function scores were calculated. Results Subjects were accurately classified in the DC (chi(2) = 78.61, p < 0.001) and CC (chi(2) = 86.63, p < 0.001). The classification model also performed well in the DC (ROC = 0.835 [SE = 0.024, p < 0.001]) and CC (ROC = 0.786 [SE = 0.021, p < 0.001]). In the AE-CC and CON-CC, there were modest but significant associations between the risk score and executive function (AE-CC: r = -0.20, p = 0.034; CON-CC: r = -0.28, p < 0.001) and intelligence quotient (AE-CC: r = -0.20, p = 0.034; CON-CC: r = -0.28, p < 0.001). Conclusion(s) The risk score significantly distinguished alcohol-exposed from control subjects and correlated with important cognitive outcomes. It has significant clinical potential and could be easily deployed in clinical settings.