All Content

Many medications and medical devices can result in preventable harm in pediatric patients. This article describes one hospital’s efforts to implement explicit, structured processes and huddles to increase situational awareness regarding high-risk therapies among the care team and family members. After implementation, the percentage of electronic health record (EHR) alerts correctly describing high-risk therapies increased from 11% to 96%.

Based on the SaferDx and SPADE frameworks, researchers applied a symptom-disease pair-based electronic trigger (e-trigger) to identify patients hospitalized for stroke who had been previously discharged from the emergency department with a diagnosis of headache or dizziness in the preceding 30 days. Analyses show that the e-trigger identified missed diagnoses of stroke with a modest positive predictive value.

This single-site retrospective cohort found that vulnerable populations (defined by race/ethnicity, insurance status, and limited English proficiency) were generally not at increased risk of patient safety events. However, stratified analyses comparing events identified via automated versus voluntary incident reporting systems found voluntary systems may undercount events in some racial/ethnic populations.

This article describes the implementation of a new system for identifying, communicating, and resolving safety reports pertaining to COVID-19 in one academic tertiary care center through the use of electronic safety reporting systems and daily huddles.

This study assessed the performance of an automated emergency department (ED) trigger tool designed to identify a more efficient sample of adverse event cases for chart review. Beginning with a set of 97 candidate triggers, researchers identified those triggers associated with adverse events and arrived at a narrowed set of 30 triggers, eliminating almost half of the population of records eligible for manual review. This computerized query may eliminate the need for manual screening for triggers.  

This systematic review explored how artificial intelligence (AI) based on machine learning algorithms and natural language processing is used to address and report patient safety outcomes. The review suggests that AI-enabled decision support systems can improve error detection, patient stratification, and drug management, but that additional evidence is needed to understand how well AI can predict safety outcomes.  

Efforts to track hospital quality and safety result in data and incentive complexities that detract from effective leadership decision making to improve safety. This article examines the juxtaposition of three emerging technologies to capture safety metrics and the pressures they bring to bear on effective management of adverse events and patient compensation schemes. The author suggests roles for leadership and Medicare to drive improvements.   

Use of data can improve the response of clinicians to patient concerns and deterioration. This article discusses how data surveillance can provide insights at the point of care to inform action and improve safety.   

Building upon prior research developing trigger tools for identifying preventable errors in dentistry, this study reviewed 1,885 electronic health records (EHR) across four dental practices and found that 16% contained an adverse event. The most common events were pain (27.5%), hard tissue (14.8%) or soft tissue injuries (14.8%) and nerve injuries (13.3%). An EHR-based trigger tool can be an effective approach to identifying safety incidents and measuring the quality of care.

Underreporting of adverse drug reactions (ADRs) is an international patient safety problem. This systematic review of studies assessed how various strategies designed to improve ADR reporting impacted ADR rates. While all strategies increased ADR reporting, particularly those using electronic reporting tools, the quality of the studies was generally low. The authors expressed the need for higher quality studies to focus on how electronic methods might improve ADR reporting.

This study evaluated the efficacy of reviewing ED return visits that result in an ICU admission to determine if they were associated with deviations in care and to understand the common errors. They found that of patients who were return ED visits and admitted to the ICU, 44% (223 cases) returned for reasons associated with the index visit and, in those, 14% (31 cases) had a deviation in care at the index visit. Implementing a standard diagnostic process of care framework to those 31 cases with a deviation in care, 47.3% had a failure in the initial diagnostic pathway. The authors concluded reviewing 14 day returns with ICU admissions contribute to better understanding of diagnostic and systems errors.

An alternative to manual chart review, natural language processing (NLP) can efficiently analyze narrative text to identify adverse events. This systematic review identified 35 studies demonstrating that NLP can be used to classify narrative text according to incident type and harm severity and many NLP models can perform classification with similar outcomes to manual human classification.  

Deviations or variations in diagnostic fidelity, including diagnostic errors and delays, can lead to serious adverse events or death, yet measurement tools and reporting processes for ensuring diagnostic fidelity are underdeveloped. This single-site retrospective study found that these errors and delays can be reliably identified using EMR data, and that variations in diagnostic fidelity are linked to increased morbidity and mortality. 

Advances in technology enable real-time intraoperative data capture to prevent adverse events and improve patient safety and recovery. This commentary describes a surgical innovation that combined artificial intelligence, video technology, and clinical decision support and was designed to flag potential bleeding events in the surgical suite.

Cognitive task analysis is a human factors engineering method used to evaluate individuals' thinking to better understand safety. This study examined medication safety through the lens of cognitive task analysis and concluded that the method identifies actionable safety gaps and should be more widely used in health care.

Auditory warnings to flag patients at risk for sepsis can have unintended consequences, such as alert fatigue or distraction. Although heightened awareness of sepsis is crucial due to its potential for harm, the authors call for rigorous study and testing of these systems to reduce their negative effects. They highlight how recently published negative results illustrate the importance of designing sepsis alerting functions that are safe and effective. A WebM&M commentary discussed a case involving a misdiagnosis of sepsis.

Look-alike and sound-alike (LASA) drugs are a well-established source of medication errors that place patients at risk for adverse drug events. Prior research has shown that these medications can be automatically identified using diagnostic codes at the time of electronic prescribing. Using electronic health record data on medication orders and diagnostic claims data from a single academic medical center as well as data on medication indications, researchers developed an algorithm to identify LASA prescribing errors. Although the algorithm was able to identify LASA prescribing errors that may not have been found by other means, the positive predictive value was 12.1% and the false-positive rate was greater than 75%. The authors advocate for further research to improve specificity and sensitivity of this approach. A past WebM&M commentary discussed a case involving the mix-up of two medications with similar names.

Alerts designed to prevent inappropriate prescribing of medications are frequently overridden and contribute to alert fatigue. This study describes the use of machine learning to improve the clinical relevance of medication error alerts in the inpatient setting.