Identifying Electronic Phenotypes associated with Patient Health Outcomes of Interhospital Transfer Patients

点击翻译按钮获取中文摘要

Project Summary/Abstract Approximately 1.6 million patients undergo interhospital transfer (IHT) each year in the United States. Of those IHTs, approximately 550,000 are conducted by air (i.e., helicopters or jets), or roughly one every 60 seconds, at an estimated annual cost of $6 billion. Contrary to common belief, patients who undergo IHT experience worse outcomes, which include double the length of stay, twice the cost, and higher mortality than non- transferred patients. While worse outcomes are not necessarily due to higher severity of illness, factors contributing to unfavorable outcomes have yet to be identified. For about 30% of the patients experiencing an immediately life-threatening condition such as trauma or heart attack, immediate IHT by air is beneficial; but for the other 70% of patients not experiencing a life-threatening condition, the benefit is less clear. The current decision-making process regarding who should be transferred and how (air vs. ground) is rarely guided by evidence or guidelines. Furthermore, patients and families rarely have input regarding how a patient is transferred. Empirical evidence is needed to identify meaningful indicators for IHT and to guide the decision regarding mode of transfer. Previous research efforts relied almost exclusively on data from the post- transport phase of illness, a major limitation when investigating what leads up to a patient needing transfer to another hospital. To investigate the factors that lead to IHT, we developed a data repository for patients who are transported from one hospital to another. This repository includes the electronic medical record of helicopter and jet transfers, as well as the sending and receiving hospital EMR data. The purpose of this study is to model complex patient-centered data that may predict those patients that will benefit from IHT by identifying pre-transport electronic phenotypes, and to determine who will benefit from air versus ground transfer. To fully leverage the data repository, this study will employ a data science approach that leverages statistical learning techniques to achieve the following aims: 1) Identify and rank specific combinations of comorbidities, active medical problems, and physiologic instability indicators according to frequency and impact on post-IHT mortality, and 2) Identify electronic phenotypes of IHT patients that benefit from air transfer according to health outcomes (hospital discharge status and functional status). After conducting exploratory data analysis and reducing variables with repeated measures, we will employ Association Rules to identify significant combinations of covariates to include in final model development (Aim 1). Then in Aim 2 we will use Random Forest to identify the most impactful variables from all of the available variables that will be analyzed via Classification and Regression Tree to identify distinct subgroups of patients that benefit from IHT by air transport. The individual characteristics identified from this data driven approach will provide the evidence needed to support future applications to develop clinical decision support for use by clinicians, patients, and families when making transfer decisions.