Machine learning validation of medication regimen complexity for critical care pharmacist resource prediction

点击翻译按钮获取中文摘要

Intensive care unit (ICU) patients are at heightened risk of adverse drug events (ADEs) and poor outcomes. Critical care pharmacists (CCPs) prevent ADEs, improve patient outcomes, and reduce healthcare costs through performing medication interventions. However, CCPs are an underused healthcare resource due to lack of health information technology (IT)-based predictive tools to allocate the care they provide to ICU patients. Currently, there are no validated health IT tools for CCPs available to optimize patient-centered care. The central hypothesis of this R21 Health Information Technology to Improve Health Care Quality and Outcomes Award, based on preliminary data, is that data-driven methods applied to the MRC-ICU Scoring Tool will out-perform predictions of a rules-based model in predicting CCP interventions that can improve patient outcomes and may serve as the foundation for development of novel health IT tools that optimize the patient-centered care provided by CCPs. The MRC-ICU Scoring Tool is the first tool designed to measure medication regimen complexity in ICU patients. To be scaled-up, this tool requires thorough validation and IT based automation. The objective of this work is to apply machine learning (ML) methodology to multi-center data to create prediction tools for integration into visualization dashboards that answer vital questions including (1) what are the predicted number of CCP interventions per patient; (2) what is the risk of real-time modifiable outcomes (e.g., fluid overload); (3) what are the predicted outcomes (e.g., mortality, length of stay). The long- term goal of the proposed work is to establish validated prediction models that can be embedded into dashboards in the electronic health record (EHR) to help guide CCP resource deployment. The rationale for this work is that it will establish the MRC-ICU Scoring Tool as a means of synthesizing patient data for integration across EHR systems. The central hypothesis will be tested using large, multi-center data of ICU patients via these specific aims: (1) Apply ML-based prediction methods to develop a new model of medication regimen complexity as a metric for predicting CCP interventions and patient outcomes; (2) Compare the performance of different models to predict CCP interventions and patient outcomes; (3) Design a web-based dashboard (ICView) to visualize medication regimen complexity-based predictions. The health IT product will result in a Web-based dashboard (ICView) that houses a real-time, automated MRC-ICU Scoring Tool in addition to prediction models for CCP interventions that can improve patient outcomes. This innovative approach applies state-of-the-art ML methodology to the novel MRC-ICU Scoring Tool. The proposed work is significant because any advances in the understanding of how CCPs improve outcomes would have a profound public health impact due to their established role on the interprofessional healthcare team. The health IT products provide the necessary foundation for a future R18 application for a multi-center, prospective trial to evaluate MRC-ICU based CCP resource allocation strategies.