Multi-hospital Electronic Decision Support for Drug-associated Acute Kidney Injury

Researchers will randomize 38 hospital service clusters to receive either: 1) a Cerner electronic medical record (EMR)-based AKI passive alert which is standard of care at UPMC: this alert provides decision support within the EMR for the diagnosis and basic staging of AKI but without specific recommendations for management (Usual Care Arm); or 2) protocolized stage-based intervention delivered to the physician by a pharmacist for consideration and approval. The intervention uses an automated alerting system to identify patients: 1) receiving a high-risk drug or drug combination associated with D-AKI and at low-risk for progression to either stage 2 AKI or stage 3 AKI per KDIGO criteria (Level A) and 2) patients without AKI or stage 1 AKI receiving a high-risk drug or drug combination associated with D-AKI and at high risk for progression to either stage 2 AKI or stage 3 AKI per KDIGO criteria, and patients with AKI stage 2 or stage 3 receiving a high-risk drug or drug combination associated with D-AKI or a medication that requires renal dose adjustment (Level B). This patient specific risk-profile will be coupled with recommendations for medication management and delivered to the physician by a pharmacist for consideration and approval. Additionally, the investigators will assess cost-effectiveness and physicians' perception of the pharmacist-led service.

The primary outcome is Major Adverse Kidney Events within 30 days of randomization (MAKE30), defined as defined as a composite of death, new kidney replacement therapy, or final serum creatinine ≥150% of reference at the earliest of hospital discharge or 30 days from study enrollment, whichever occurs first. Key secondary outcomes include: progression of AKI from time of Level B intervention (first alert generated) to hospital discharge, AKI intensity (duration of AKI by all stages, duration of AKI stage 2, and duration of AKI stage 3), and nephrotoxic burden.

The intervention uses an automated alerting system to identify patients of enrolled physicians: 1) receiving a high-risk drug or drug combination associated with AKI and at low-risk for progression to either stage 2 AKI or stage 3 AKI (Level A) and 2) patients without AKI or stage 1 AKI receiving a high-risk drug or drug combination associated with AKI and at high risk for progression to either stage 2 AKI or stage 3 AKI, and patients with AKI stage 2 or stage 3 receiving a high-risk drug or drug combination associated with AKI or a medication that requires renal dose adjustment (Level B). This patient specific risk-profile will be coupled with recommendations for medication management and delivered to the physician by a pharmacist for consideration and approval.

Other: Level A · Other: Level B

A Cerner EMR-based AKI passive alert which is standard of care at UPMC.

Other: Passive Alert

Pharmacy personnel will generate a general recommendation based on the AKI KDIGO management guidelines to the physician.

The pharmacist will make nephrotoxic/renally eliminated medication management recommendations to the attending physician (or designee). Recommendations may include stopping or changing a drug, changing dose or schedule, ordering laboratory tests, taking no action, or other. The pharmacist will record details of the interaction with the physician and whether recommendations were accepted.

Passive Cerner alert provides decision support within the EMR for the diagnosis and basic staging of AKI but without specific recommendations for management.

Time frame: up to 30 days

Composite of death, new kidney replacement therapy, or final serum creatinine greater than or equal to 150 percent of reference at the earliest of hospital discharge or 30 days from study enrollment, whichever occurs first. Patient encounters for the analyses reflects those meeting the criteria for the retrospective cohort. It was anticipated that patients would receive a considerable number of non-actionable alerts so \~ 3000 patients were included for the per protocol analysis to be impactful.

Time frame: up to 30 days

Percent of high-risk patients without AKI at time of first level B alert who subsequently progress to maximum AKI severity stages 1, 2, or 3 before hospital discharge or 30 days, whichever comes first.

Percent of high-risk patients with stage-1 AKI at the time of level B alert who subsequently progress to maximum severity stages 2 or 3 AKI before hospital discharge or 30 days, whichever comes first.

Percent of high-risk patients with stage-2 AKI at the time of level B alert who subsequently progress to maximum severity stage 3 AKI before hospital discharge or 30 days, whichever comes first.

The machine learning algorithm operated silently for patient encounters in the Usual Care arm to determine whether they would have had a Level A or Level B alert, but no alert was delivered to study pharmacists and no intervention was received by study physicians. Patient encounters in the Usual Care arm represent the subset of patient encounters that would have had a qualifying Level B alert.

Time frame: up to 30 days

AKI intensity defined as the estimated marginal mean number of inpatient days spent in AKI overall at any stage and by each stage of AKI among patients cared for by enrolled physicians.

Time frame: up to 30 days

Drug.days\* in both study arms for those drugs considered possible/probable, probable and definitely related to AKI in adult, non-ICU patients.

\*Drug.days calculation: each drug and each day of therapy increases the burden by 1 drug.day.

Multi-hospital Electronic Decision Support for Drug-associated Acute Kidney Injury (MEnD-AKI)