Childhood obesity is a major public health problem across the globe as well as in the US. In 2019, the prevalence of obesity was 18.5% affecting almost 13.7 million US children and adolescents aged 18 or less, with children of lower socioeconomic status and from rural areas affected in a disparate manner. Childhood obesity often continues into adulthood and is known to be a major risk factor for chronic diseases such as diabetes, cancer, and cardiovascular diseases. Decades of rigorous research have shown that prevention and management of obesity are not easy. This is partly due to our limited understating of obesity and the complex interactions among a myriad of various factors, including biological and environmental ones, that are known to contribute to obesity. In such a complex domain, advanced predictive models are effective in informing decision-makers and providers in designing and delivering more effective interventions. While various predictive models have been developed for childhood obesity, existing models have been created using small populations, are only good for a specific age prediction, and most importantly, do not use the temporal data patterns of body weight changes across time. To address this important gap in the field, a set of predictive models of childhood obesity using a longitudinal dataset of children derived from the electronic health records (EHR) of a large pediatric healthcare system (Nemours) are being developed by the investigators of this project. Building on our progress in developing preliminary models, the current project pursues two main aims to expand these models. The first is improving the performance of these models by including additional variables related to family SES (socioeconomic status). While the current models include several SES factors such as the type of insurance, additional SES factors, including education and income levels and rurality, will be collected to be incorporated in the model and to inform applicability of the model to the underserved populations targeted by the IDeA States Pediatric Clinical Trials Network.
The second aim i s to extend the prediction duration of the models from current short-term (one to three years) windows of prediction to long-term (up to 20 years) prediction windows. Including additional SES factors is expected to improve the performance of the models and lengthening the prediction windows will make them more practical and useful in real settings, including identification of children who would most benefit from future clinical trials. Successful expansion of these predictive models will offer powerful tools that can inform both prevention and treatment interventions. Specifically, by identifying children at a higher risk of developing obesity, these predictions can facilitate engaging in appropriate interventions at earlier ages. Additionally, by offering an estimate of the disease severity for a current patient and relevant risk factors, the models can help providers and practitioners in making more informed decisions about which treatment strategy to choose.
Childhood obesity is a major public health problem affecting the US population especially those with lower socioeconomic status. Engaging in timely and early interventions to prevent and control this disease is a key factor in increasing the success rate of obesity interventions. In this project, we aim to create a set of predictive models using large-scale electronic health records that can assist in the early identification of children with higher risks of obesity.
Showing the most recent 10 out of 333 publications
