Computable social factor phenotyping using EHR and HIE data

点击翻译按钮获取中文摘要

Most health systems attempt to measure patients' social risk factors, but such data collection is typically fraught with operational and conceptual difficulties. Multi-domain screening questionnaires face reliability, validity, and workflow challenges. Area-level data are not valid proxies for individual characteristics. Diagnosis codes are underutilized. The day-to-day use of natural language processing (NLP) to extract social factors from text is beyond the capacity of most organizations. Thus, health care organizations need more implementable and valid approaches to measuring social factors. With implementable and valid approaches, health systems will more effectively address the negative cost, quality and health outcomes associated with patients' social risk factors. The objective of this proposal is to assess the validity of patient-level computable social factor phenotypes for use in predicting patients' risk of increased healthcare costs and utilization. Computable phenotypes are com- posites of characteristics defined through single data elements or a collection of data elements, observations or events. Because these phenotypes derive from existing healthcare operations and electronic data systems, they are well-positioned for widespread implementation. Our central hypothesis is that phenotypes computed from existing structured demographic, clinical, and business operations data will support equally or more valid infer- ences about patient social risks than other measurement approaches. Building upon strong preliminary data and direction from experts in the field, we will determine the validity and usefulness of six novel social factor pheno- types computed from already collected information within EHRs and health information exchanges (HIE) through the following aims: Aim 1, Assess the concurrent validity of patient-level computable social factor phenotypes, compares the concurrent validity of computed phenotypes, multi-domain questionnaires, and NLP against gold standard measures of social factors in two health systems. Aim 2, Assess the predictive validity of patient-level computable social factor phenotypes, will assess the validity of computable phenotypes, multi-domain question- naires, NLP, and combined approaches in predicting costs and utilization. Aim 3, Assess the reliability (bias) of patient-level computable social factor phenotypes across patient gender, race, ethnicity, and age, assesses the reproducibility of measurement approaches across underserved populations. We will employ a multi-method research approach to identify and mitigate potential bias. This project will lead to more valid and implementable approaches to patient social factor measurement. The proposed research is significant because it directly ad- dresses the challenges organizations face in addressing patients' social risks and will provide key inputs to support organizations efforts at achieving a learning health system. This proposal is innovative by advancing the psychometrics of social factors and identifying novel usages of EHR and HIE data. By working with multiple and diverse populations, we address the priority populations of socioeconomically disadvantaged, racial minority populations, and the elderly.