Project Details - Ongoing
Grant Number:R01 HS027793
- Funding Mechanism(s):
AHRQ Funded Amount:$1,963,619
- Principal Investigator(s):
- Project Dates:9/30/2020 to 7/31/2021
- Care Setting:
- Medical Condition:
- Type of Care:
- Health Care Theme:
Emergency departments are a primary source of healthcare for many individuals and used for a variety of conditions, spanning non-urgent to time-sensitive, life-threatening conditions and illnesses. Providers delivering care in the emergency department are faced with making high-stakes clinical decisions in a high-volume, time-sensitive, and hectic environment. Artificial intelligence (AI) holds promise for extracting and analyzing relevant electronic health record (EHR) data for use in emergency point-of-care clinical decision support (CDS). However, methods for incorporating AI-derived insights into clinical care delivery are underdeveloped.
Acute kidney injury (AKI) is prevalent, associated with adverse clinical outcomes, and one of the most costly conditions in U.S. healthcare. Studies suggest detection and assessment of the severity of AKI in the emergency department setting is limited due to diagnostic criteria and lack of syndrome recognition. However, it is well suited for automated detection by predictive modeling and risk stratification. Diagnosis criteria and severity staging can be applied using data from the EHR. Using AKI as a reference case, a Johns Hopkins University research team will develop an AI-driven algorithm to detect AKI. The algorithm will be developed into a CDS tool in the EHR and optimized for use in the emergency department setting.
The specific aims of this research are as follows:
- Develop an AI-driven algorithm for promotion of AKI-focused clinical decision making in the emergency department.
- Translate the AI algorithm to an AKI-CDS system to enable indepth study of clinician-AI interactions in the emergency department.
- Perform a multisite effectiveness implementation-evaluation of the AKI-CDS system in the emergency department.
Building on previously developed AKI surveillance and prediction tools, the research team will develop an EHR-based algorithm to estimate AKI risk and flag patients for AKI CDS in the EHR. The algorithm will be validated using two groups of discharged patients. The AI algorithm will be translated into an AKI-CDS system and pilot tested among emergency department clinicians to examine AI trustworthiness and explainability. Findings will inform establishment of end-user requirements for the AKI-CDS system.
Lastly, the research team will evaluate the AKI-CDS system implemented at three study sites to understand the drivers of end-user trust and adoption of AI in the emergency department setting. In doing so, the team will develop methods that increase reliability of AI-driven predictions and promote trustworthiness, transparency, explainability, and usability of AI in the emergency department setting.