Patients and healthcare providers rely on devices that use artificial intelligence or machine learning in diagnostics, treatment, and monitoring. This study utilizes adverse event reports submitted to the FDA's Manufacturer and Use Facility Device Experience (MAUDE) database for machine learning-enabled devices. Mammography was implicated in 69% of reports, and the majority were near-miss events.

Health information technology (HIT), such as electronic health records (EHRs) or computerized provider order entry (CPOE) systems, are important approaches to improving safety. This scoping review of 45 articles found that machine learning and statistical modeling are the most commonly used automated, HIT-based methods for early detection of safety threats. Machine learning was often used to detect errors occurring in laboratory test results, prescriptions, and patient records. Statistical modeling was used to detect issues with clinical decision support systems.

This article discusses the influence of artificial intelligence tools and cyber-social systems on human decisions in healthcare. Opportunities to use cyber-social systems in public and population health (e.g., disease tracking) and primary care (e.g., patient-facing technologies, such as symptom checkers) as well as approaches to exploit cyber-social systems within a learning health system are discussed.

This simulation study compared medical students' performance of electronic prescribing with and without clinical decision support. Students were less likely to access outside references to verify that medications were safe when decision support was in place, even when the decision support was incorrect. The authors conclude that electronic prescribing should be redesigned to facilitate external verification of medication safety.

Health information technology (IT) implementation is a complex endeavor that requires a sociotechnical orientation to succeed. This article outlines nine key challenges to safety that must be addressed across the three stages of health IT: design and development; implementation and use; and monitoring, evaluation, and optimization.

This analysis of a previous simulation study of electronic prescribing examined the effect of cognitive load, or demand on working memory, on errors. This study found that physician participants who reported a lower cognitive load were more likely to make errors of omission, suggesting that they were not paying sufficient attention to the task. The authors conclude that errors may arise from a mismatch in allocating cognitive resources (how much attention is paid to a task) and the cognitive requirement needed to safely accomplish that task.

Artificial intelligence can improve practice by making synthesized data available in real time to inform frontline decision-making. This commentary describes factors clinicians should consider as artificial intelligence becomes more prevalent in health care and discusses how this technology can enable clinicians to focus on helping patients navigate complex care choices.

Electronic documentation burden can contribute to physician burnout. Speech recognition software has been touted as a potential strategy to address this concern. However, this study found that speech recognition was more error prone and time consuming than using a keyboard and mouse for various emergency physicians' electronic tasks. This study finding adds to existing concerns that speech recognition may lead to clinically significant errors.

This systematic review of studies describing safety problems associated with health information technology identified mostly qualitative data and case reports. The authors suggest that their framework, an information value chain which extends from the interaction with technology to the patient outcome, could be used to enhance the literature on health information technology and safety.

… recent book . … Magrabi F, Liaw ST, Arachi D, Runciman W, Coiera E, Kidd MR. Identifying patient safety problems …

Providers and policymakers have raised concerns about risks associated with health information technology (IT). This commentary spotlights the importance of considering human factors and cognition when designing health IT systems to understand how human–computer interaction can contribute to error.

Health information technology can both improve patient safety and introduce risks. This analysis examined all safety events associated with the United Kingdom's national program for health information technology. The researchers found that while most events were technical failures, incidents involving human errors had a higher chance of causing harm to patients. Technical failures affecting 10 or more patients accounted for nearly 25% of events and were more likely to impact care delivery. These results underscore the concerns in prior reports about the unintended consequences of implementing health information technology on patient safety. The findings also lend weight to the Institute of Medicine recommendations that errors related to health information technology be reported and investigated in the United States. A past AHRQ WebM&M perspective explored the promised benefits of health information technology alongside the challenges of implementation and idiosyncrasies of available systems.

This commentary combines two models of systems safety concepts—Reason's Swiss Cheese model and the Iceberg theory—and applies variations of the hybrid concepts to explain hazards in care delivery. The authors suggest that these models demonstrate how oversimplifying complex health systems may inhibit understanding about patient safety.

Many studies have shown that patients admitted to the hospital on the weekend experience more preventable complications and are at increased risk for mortality. The mechanism for this finding is unknown and could be due to health care system factors (i.e., lower weekend staffing and availability of clinical services) or patient factors (i.e., those admitted on weekends could be more complex and at higher risk of death than weekday admissions). This population-based analysis from New South Wales, Australia sought to determine the contribution of health care system and patient factors to elevated weekend mortality by analyzing daily mortality rates for the 7-day period following weekend admission. For certain diagnoses, such as pulmonary embolism, the risk of death was elevated during the first 48 hours after weekend admission but declined thereafter, implying that health care system factors are the primary driver of the weekend effect. However, patients with cancer-related diagnoses continued to have elevated mortality risk for the full 7 days after weekend admission, implying that patient factors such as illness severity are the major contributor to excess mortality. Other diagnoses, such as stroke, showed a mixed pattern of system and patient factors. By providing a nuanced view of the types of diagnoses and factors associated with the weekend effect, this study demonstrates the need for tailored solutions for this well-documented problem.

Syndromic surveillance—an approach used to identify infectious disease epidemics—can be applied to monitor health information technologies for early signs of unreliability or failure.

A 2011 Institute of Medicine report found that existing health information technology (IT) systems have several problems that seriously compromise their ability to improve the safety and quality of care, and the report recommended standardizing measures of adverse events associated with health IT. This study discusses a novel method of identifying health IT–related adverse consequences within an existing database. Using the Food and Drug Administration's Manufacturer and User Facility Device Experience (MAUDE) database, which includes voluntarily reported safety incidents relating to medical devices, the authors developed and iteratively tested queries for identifying health IT–related adverse events. These queries can be used for earlier detection of patient care problems associated with health IT, in a manner analogous to post-approval drug safety surveillance. This study also demonstrates how "big data" can be analyzed to improve patient care (the dataset included more than 44 billion individual data elements generated from more than 500,000 separate incidents). A wrong-patient error resulting from poor health IT interoperability is discussed in an AHRQ WebM&M commentary.

Implementation of health information technology (IT) has presented several unanticipated patient safety issues, particularly related to computerized provider order entry. Health IT vendors generally have a "hold harmless" clause that removes their liability for patient harm due to health IT failures. This review explores patient safety initiatives for health IT systems across six countries, including the United States. Significant gaps in health IT safety initiatives were identified, including inadequate regulatory oversight in the US for these medical devices. The authors conclude that greater standardization and oversight is required to ensure safety for all types of health IT systems. An AHRQ WebM&M interview with Dr. David Bates discusses the benefits and potential risks of health IT.

According to data presented in this research letter, tests ordered on the day of discharge represent a disproportionate number of tests that are not followed-up.

This commentary discusses interruption research in health care, challenges to understanding its impact, and approaches to reducing interruptions in care delivery.

This study reviewed nearly 900,000 reports from the FDA Manufacturer and User Facility Device Experience database (MAUDE) and identified 678 reports describing health information technology issues. Investigators uncovered problems with software functionality, system configuration, interface with devices, and network configuration as new categories to the existing classification system.