Optimizing Electronic Alerts for Acute Kidney Injury

点击翻译按钮获取中文摘要

Project Summary The overarching goal of this proposal is to determine how alerts for acute kidney injury (AKI) can be engineered to provide benefit to patients and whether that benefit can be enhanced with intelligent targeting. AKI is a common complication in hospitalized patients, and carries with it a substantially increased risk of morbidity and mortality. International guidelines suggest prompt diagnostic evaluation and avoidance of nephrotoxins once AKI develops, but multiple studies and our preliminary data have documented that the diagnosis of AKI is often delayed or missed altogether. Even when recognized, diagnostic actions (such as urinalysis) and therapeutic actions (such as discontinuing potentially nephrotoxic agents) are infrequently taken. Failure to engage in best practices is associated with increased morbidity and mortality among AKI patients. The lack of recognition of AKI, and its direct clinical consequences, has prompted several health systems in the US and the entire National Health Service of Great Britain to institute automated, electronic alerts for AKI. However, our pilot randomized trial of a basic alert system did not demonstrate a significant benefit of alerting for AKI on clinical outcomes such as progression of AKI, dialysis, or death. This proposal describes studies that build on our prior experience to determine how AKI alerts coupled with clinical best practice actions, and intelligent targeting of those alerts, can improve care and patient outcomes across multiple centers. We have designed a series of large, multi-center randomized trials of AKI alerts to identify modes of delivery and targeting that will best improve the care and clinical outcomes of patients. Across 6 hospitals, with a total of more than 10,000 AKI cases per year, we will evaluate 1) the clinical efficacy of an AKI alert tied to a clinical best practices order set, 2) whether targeting alerts to patients who have received specific drug-classes can modify prescription behavior and improve patient outcome and 3) whether progressive targeting of alerts through the use of uplift modeling can increase effectiveness while decreasing alert fatigue. In support of this application, we have collected preliminary data regarding the rates and outcomes of AKI at the study hospitals. We have also documented surprisingly low rates of potentially nephrotoxic-drug discontinuation among those with AKI. In addition, we have reanalyzed data from our original pilot alert trial to demonstrate that uplift modeling can be used to identify individuals most likely to benefit from alerts. This predictive modeling approach is an example of personalized medicine, where large amounts of patient data are used to target alerts to a narrow population, avoiding unnecessary alerting and thus reducing alert fatigue. Beyond informing the use of AKI alerts in the future, these studies will provide an infrastructure and a statistical foundation for the rigorous assessment and targeting of best-practice alerts across multiple fields of medicine.