Search AMIA.org

This study aims to evaluate the impact of using a large language model (LLM) for generating draft responses to patient messages in the electronic health record (EHR) system on clinicians and support staff workload and efficiency.We partnered with Epic Systems to implement OpenAI's ChatGPT 4.0 for responding to patient messages. A pilot study was conducted from August 2023 to July 2024 across 13 ambulatory specialties involving 323 participants, including clinicians [...]

Fairness concerns stemming from known and unknown biases in healthcare practices have raised questions about the trustworthiness of Artificial Intelligence (AI)-driven Clinical Decision Support Systems (CDSS). Studies have shown unforeseen performance disparities in subpopulations when applied to clinical settings different from training. Existing unfairness mitigation strategies often struggle with scalability and accessibility, while their pursuit of group-level prediction performance parity does not effectively translate into fairness at the point of [...]

The CONCERN Early Warning System (CONCERN EWS) is an artificial intelligence-based clinical decision support system (AI-CDSS) for the prediction of clinical deterioration, leveraging signals from nursing documentation patterns. While a recent multisite randomized controlled trial (RCT) demonstrated its effectiveness in reducing inpatient mortality and length of stay, evaluating implementation outcomes is essential to ensure equitable results across patient populations.This study aims to (1) assess whether clinicians' usage of the CONCERN [...]

Retinopathy of prematurity (ROP) is the leading cause of preventable childhood blindness. Guidelines recommend screening for infants with gestational age at birth <31 weeks or birth weight ≤1,500 g. However, ensuring timely screening during readmissions after birth is challenging.To analyze the performance of an interruptive alert at a large academic pediatric hospital for identifying premature infants needing ROP screening upon hospital readmission and to describe how data informed the transition to a non-interruptive dashboard.The alert appeared for patients 1 to 365 days of age hospitalized in acute care or pediatric intensive care and instructed providers to order an ophthalmology consult from within the alert and to call ophthalmology for at-risk patients. For quality improvement, the clinical decision support (CDS) advisory group evaluated the effectiveness and efficiency of the alert. We extracted alert metrics from the hospital's enterprise data warehouse, including the user response and feedback, patient characteristics (age, birth gestational age, and birth weight), and any ophthalmology consultations. We analyzed the percentage of encounters seen by ophthalmology using a statistical process control chart during alert implementation and 6 months before and after.The alert appeared 3,309 times during 2,194 patient encounters usually. Users chose "Accept and place order" for 43% (943/2,194) of encounters, but only 11% (102/943) had an ophthalmology consult; 34% (53/155) of ophthalmology consultations occurred in encounters with a final response other than "Accept and place order." The intervention was redesigned using a non-interruptive surveillance dashboard with greater specificity, and the alert was de-implemented.Analysis of a failed interruptive alert for identifying patients at risk for ROP led to a transition to targeted surveillance using a dashboard. This case emphasizes the importance of aligning the CDS modality to the clinical workflow, information availability, and user decision-making needs and should be supported by governance.

Despite their potential, Clinical Decision Support (CDS) systems often lack alignment with clinicians' needs and are underutilized in practice. Pilot implementations can help to improve the fit between systems and local needs by engaging users in real-world testing and refinement. Although pilot implementations of CDS have been reported, limited evidence has explored the factors contributing to pilot success.This study aimed to explore the opportunities and challenges associated with the pilot [...]

Nursing documentation burden is a growing point of concern in the United States health care system. Documentation in the electronic health record (EHR) is a contributor to perceptions of burden. Efficiency tools like flowsheet macros are one development intended to ease the burden of documentation.This study aimed to evaluate whether flowsheet macros, a documentation efficiency tool in the EHR that consolidates documentation into a single click, reduces the time spent [...]

Systemic lupus erythematosus (SLE) is a chronic autoimmune disease that disproportionately affects women and racial/ethnic minority groups. Predicting disease flares is essential for improving patient outcomes, yet few studies integrate both clinical and social determinants of health (SDoH). We therefore developed FLAME (FLAre Machine learning prediction of SLE), a machine learning pipeline that uses electronic health records (EHRs) and contextual-level SDoH to predict 3-month flare risk, emphasizing explainability and fairness.

Recent advancements of Artificial Intelligence (AI) are rapidly transforming clinical research. While this technology offers exciting opportunities, it amplifies existing concerns regarding the need for transparent methodology that fosters patient engagement, and introduces new challenges. PCORI's Improving Methods portfolio has invested in methodological research to enhance rigor and transparency via patient-centered approaches in AI.