A Learning Health Sciences Approach to Understanding Clinical Documentation in Pediatric Rehabilitation Settings

Abstract

The work presented in this dissertation provides an analysis of clinical documentation that challenges the concepts and thinking surrounding missingness of data from clinical settings and the factors that influence why data are missing. It also foregrounds the critical role of clinical documentation as infrastructure for creating learning health systems (LHS) for pediatric rehabilitation settings. Although completeness of discrete data is limited, the results presented do not reflect the quality of care or the extent of unstructured data that providers document in other locations of the electronic health record (EHR) interface. While some may view imputation and natural language processing as means to address missingness of clinical data, these practices carry biases in their interpretations and issues of validity in results. The factors that influence missingness of discrete clinical data are rooted not just in technical structures, but larger professional, system level and unobservable phenomena that shape provider practices of clinical documentation. This work has implications for how we view clinical documentation as critical infrastructure for LHS, future studies of data quality and health outcomes research, and EHR design and implementation. The overall research questions for this dissertation are: 1) To what extent can data networks be leveraged to build classifiers of patient functional performance and physical disability? 2) How can discrete clinical data on gross motor function be used to draw conclusions about clinical documentation practices in the EHR for cerebral palsy? 3) Why does missingness of discrete data in the EHR occur? To address these questions, a three-pronged approach is used to examine data completeness and the factors that influence missingness of discrete clinical data in an exemplar pediatric data learning network will be used. As a use-case, evaluation of EHR data completeness of gross motor function related data, populated by providers from 2015-2019 for children with cerebral palsy (CP), will be completed. Mixed methods research strategies will be used to achieve the dissertation objectives, including developing an expert-informed and standards-based phenotype model of gross motor function data as a task-based mechanism, conducting quantitative descriptive analyses of completeness of discrete data in the EHR, and performing qualitative thematic analyses to elicit and interpret the latent concepts that contribute to missingness of discrete data in the EHR. The clinical data for this dissertation are sourced from the Shriners Hospitals for Children (SHC) Health Outcomes Network (SHOnet), while qualitative data were collected through interviews and field observations of clinical providers across three care sites in the SHC system.