Increasing the Power and Generalizability of Late-Talker Prediction: An Integrative Data Analysis Approach

Vanderbilt University, Nashville TN

Investigators

Abstract

PROJECT SUMMARY The inability to meet developmental language milestones, known as late-talking or late-language emergence, is associated with developmental delays and deficits in a variety of academic, social, emotional, and other developmental skills. Current research in the field has identified some predictors of late-talking and subsequent language skills, but findings have generally lacked consistency, and issues with lack of statistical power, inconsistent measure use, and more have significantly limited the generalizability and reliability of these findings. The current project aims to address these issues through three steps. First, we aim to increase the availability of data on late-talkers that are available to researchers in the area. This will be done by curating and ingesting extant datasets on late-talkers into a learning and development domain-specific data repository known as LDBase. Making these data available to the public will increase the reach of each individual dataset, connect researchers with differing backgrounds and expertise to the datasets, and lead to more efficient use of the collected data. Second, we plan to integrate the contributing datasets using integrative data analysis (IDA), integrating the datasets by applying IDA to create a new consistently scaled language measure across the datasets. This serves to provide a dataset that addresses the issues of statistical power and inconsistent measure use. This also will provide the opportunity for new questions to be answered via the ability to analyze cross-study differences and the increased variance that exists in the bioecological factors assessed. Third, we will highlight the potential of the newly integrated dataset by analyzing the dataset to explore current gaps in knowledge related to the inconsistency of findings surrounding the bioecological factors that influence language development and predict late-talking. Specifically, we will link the original data with consistently reported and standardly measured GIS data describing the participants’ environments, and subsequently analyze how these measures and the bioecological factors measured in the contributing datasets predict the newly integrated language measure. This provides a more robust analysis of the prediction of the measure, given that the measure is now consistently scaled and invariant across studies, ensuring that the same consistent construct is being assessed and predicted across the various projects. We have leveraged our existing relationships with researchers in the area to obtain commitments from 5 sites to share their data from 11 projects, representing around 1,300 total individuals. We likewise will leverage our experience conducting similar studies and working with collaborating sites to quickly and efficiently carry out this project in the given timeframe, a timeline we believe would not be feasible for researchers without this direct experience. Finally, we will use this process to further disseminate open-science materials and best-practice documents for data sharing that we have developed via other NIH-funded initiatives.