Applying Machine Learning in the Prediction and Identification of Children Affected by Prenatal Alcohol Exposure

点击翻译按钮获取中文摘要

Project summary Fetal alcohol spectrum disorders (FASD), which are caused by prenatal alcohol exposure, occur in up to 5% of the population in the United States, and are associated with lifelong disability. There are multiple difficulties in obtaining an accurate diagnosis of FASD, including subtlety of physical features and heterogeneity in presentation. Consequently, FASD is grossly under-recognized, and the majority of affected children never receive a diagnosis. If FASD could be diagnosed earlier and with more reliability, many years of beneficial intervention would not be lost. The objective of this research is to apply machine learning to high-dimensional data in well-characterized data sets to predict or characterize children with FASD. The central hypothesis of this research is that the application of machine learning will accurately predict and recognize FASD compared with expert clinical diagnosis. To test this hypothesis, machine learning will be employed to: 1) characterize FASD based on the presence of non-cardinal malformations, 2) establish multivariate predictors of FASD in preschool aged children, and 3) identify diagnosis specific neurodevelopmental markers that distinguish alcohol related neurodevelopmental deficits from neurodevelopmental deficits without prenatal exposure. Two secondary data sources will be used in this proposal; a prospective study of 400 pregnant women and their offspring in Ukraine (half of whom consumed high amounts of alcohol) with full clinical evaluations for FASD, and a cross- sectional study of over 2,900 first grade children in four regions of the U.S., all with clinical FASD evaluations. Upon successful completion of the proposed research, the expected contribution is for more accurate prediction and recognition of children with FASD. The proposed research is innovative, as it represents a departure from current practice by incorporating machine learning techniques into predictive models of FASD. As a perinatal epidemiologist, I have a strong foundation in analytic techniques, and the advanced training in machine learning will further enhance these skills. Additionally, the disease-focused training in dysmorphology and neurodevelopment will provide a strong foundation to make significant contributions to the field of FASD research. Finally, training and mentoring in grant writing and the responsible conduct of research will provide a strong foundation to transition to an independent researcher. This proposed research builds on previous NIAAA funded research by my interdisciplinary mentoring team, who are all strongly supportive of this research and training plan. This seminal application of machine learning to FASD research will demonstrate its capacity to predict and identify affected children, ultimately leading to earlier intervention of children prenatally exposed to alcohol.