(MEnD-AKI) Multicenter Implementation of an Electronic Decision Support System for Drug-associated AKI

点击翻译按钮获取中文摘要

Project Summary Acute Kidney Injury (AKI) in the US has increased by 38% over the last eight years, with drugs as a major contributor to AKI in hospitalized patients. Drug-associated AKI (D-AKI) results in severe consequences with approximately 40% of patients experiencing in-hospital death or dialysis dependence. We have determined that many patients often continue to receive nephrotoxic drugs until AKI becomes severe. The goal of the parent project is to assess the effectiveness of a clinical decision support system (CDSS) augmented with real-time predictive analytics to support a pharmacist-led intervention to reduce the progression and complications of D-AKI. Specifically, we aim to 1) optimize the clinical performance of risk-alerts generated by a CDSS; 2) test whether an advanced CDSS coupled with a pharmacist-led intervention improves outcomes for patients with D-AKI; and 3) determine physician acceptance and cost-effectiveness of our intervention. This requires harmonization and cross validation of electronic AKI phenotypes and interpretable deep learning AKI transition model using data from two different EMR platforms, EPIC used by the University of Florida Health (UFH) and Cerner used by the University of Pittsburgh Medical Center (UPMC). While data integration, harmonization, and standardization processes are being developed for demographics and medical history, medications, laboratory results, and vital signs, it is lacking AI/ML ready datasets with social determinants of health (SDOH) exposome data and clinical notes that may carry important information about patient heath status and access to health care that may improve performance of the models. The proposed supplement project will develop integration, standardization, and processing tools and pipeline to create multimodal AI/ML ready datasets with SDOH and text data with aims: Aim 1: Preparation of AI/ML ready SDOH data. We will develop and assess tools for a) extracting, cleaning, imputing, preprocessing and representing data for various exposures contributing to a person’s SDOH exposome, b) integration of SDOH data to databases of University of Florida (UF) and University of Pittsburgh (UPitt) for them to be used in D-AKI risk model development and validation. Aim 2: Preparation of multimodal AI/ML ready data that includes unstructured text data. We will develop and assess tools for a) extracting, cleaning, preprocessing and representing unstructured text data b) integration of clinical data and unstructured text data to prepare multimodal AI/ML ready datasets at UF. The proposed supplement project will extend the aims of the parent project by including development of (1) a tool for integration of SDOH (2) a tool for extraction and integration of unstructured text data (3) tools for cleaning, imputing, preprocessing data and generalizable methods for improving the AI/ML-readiness of datasets through consensus-driven AI datasheets. The completion of these aims will provide multimodal AI/ML ready data with additional data elements, that require effort and budget not covered by parent project, to improve efficiency and performance of the D-AKI risk prediction models and other AI applications.