Applied Clinical Informatics最新文献

Electronic health records (EHRs) are intended to improve clinical practice, but excessive alerts can be counterproductive, leading to workarounds. The Mortality Surprise Question (MSQ), a tool for identifying patients who might benefit from early end-of-life discussions, was integrated into the Emergency Department (ED) EHR admission process.This study investigated how the staged implementation of a clinical decision support tool at the point of admission order entry affected ED clinician admission order practices.This retrospective cohort study examined ED admission orders from 2023 across three EDs. Clinicians used either the Quicklist or Disposition tab in the Epic EHR for admissions. The MSQ was introduced in two phases, first to the Quicklist on May 31, 2023, and then to the Disposition tab on September 11, 2023. Admissions from both tabs were analyzed pre- and post-MSQ implementation. Statistical analysis included chi-square testing to compare the admission source in the EHR after each phase of implementation of the MSQ to examine changes in the clinicians' admission workflow, with further categorization based on clinician EHR experience.Overall, 53,897 patients were admitted from the ED, with 29,542 (55%) admissions via the Quicklist and 24,355 (45%) via the Disposition tab. A statistically significant difference was found in Quicklist admission proportions before and after MSQ implementation in both workflows. As compared with clinicians with less than 2 years of experience with the EHR, clinicians with 2 to 4 years of EHR use were less likely to use the Quicklist after MSQ implementation, whereas those with over 4 years of use were more likely to use it.The MSQ disrupted established workflows, prompting clinicians to initially adopt more effortful alternatives to avoid the new cognitive task. Embedding the MSQ into these alternatives reduced resistance, highlighting that removing optionality promotes adoption. Accounting for clinician habits and potential workarounds can enhance the integration and efficiency of new quality improvement measures.

Many academic medical centers (AMCs) rely on systems like the Vizient Quality and Accountability Scorecard to track quality metrics such as the observed-to-expected (O/E) mortality ratio. The O/E mortality ratio calculation relies on clinical documentation. Missed documentation of diagnoses and risk factors for mortality leads to an underestimated expected mortality, which negatively affects the O/E metric.We aimed to reduce our O/E mortality ratio from a median of 1.08 (± 0.10) to a median well below 0.90 within 12 months by improving the accuracy of clinical documentation.We used a continuous quality improvement process that began with creating a rule-based tool within a standardized documentation template. The tool was designed to pull pertinent discrete electronic health record data into clinician documentation. The tool only pulled in data that were present on admission, and it especially prioritized inclusion of frequently missed risk factors according to prior coding query data. We then formed a multidisciplinary mortality review committee where providers reviewed mortality cases, made suggestions for documentation clarification, and found potential diagnoses and risk factors that the patient had which were missing from the documentation. We then leveraged the committee's expertise and feedback to improve the rule-based clinical tool.Over the 21-month period following implementation, the median O/E mortality ratio decreased by 30%, from 1.08 (± 0.10) to 0.72 (± 0.13) and consistently remained below the prior levels. Importantly, the intervention also led to a reduction in the total number of coding queries sent to clinicians, indicating a lower administrative burden for clinicians and coders.Our interventions showed a clear improvement in the O/E mortality ratio at our AMC and in the expected mortality percentage compared with other similar institutions without significantly increasing burden on clinicians or coding specialists.

The effective operation of magnetic resonance imaging (MRI) systems relies on physical interactions with complex imaging environments, equipment, and user interfaces (UIs). However, there is limited empirical data evaluating how physical interactions with MRI equipment and accessories, workspace configuration, MRI UI design, and technical proficiency influence clinical workflow.In this study, a cross-sectional survey was conducted among MRI end-users, across public and private health facilities (n = 13), using a structured questionnaire to assess demographics, patient positioning and equipment handling, MRI workspace adequacy, interface usability (guided by Nielsen's heuristics), and self-reported MRI skill proficiency.The predominant field strength of scanners in current use was 1.5T. General Electric was the most frequently used MRI scanner brand. Most respondents received their MRI training from nonvendor sources-such as academic institutions or peer-based instruction-rather than directly from equipment manufacturers. High ease-of-use ratings were reported for patient positioning and equipment handling tasks. Workspace adequacy was mostly rated as very adequate to highly adequate. Computed Tomography-experienced users showed moderate-to-high proficiency in MRI pulse sequencing and image optimization. However, lower proficiency was noted in quality assurance and physiologic monitoring. Help documentation within the MRI interface received the lowest usability scores. No significant differences in usability or proficiency were found between those trained by vendors versus nonvendors (U = 8.5-15.0; p = 0.376-0.921).Opportunities exist to enhance clinical workflow and patient throughput by refining error-handling features, improving support documentation, reinforcing ongoing professional development, and re-evaluating training delivery by incorporating iterative, multimedia-based learning modules and regular postinstallation refresher sessions. End-user input in UI design and user feedback analysis should be prioritized to improve system usability and clinical efficiency.

The use of real-time clinical decision support (CDS), such as Our Practice Advisory (OPAs), augments clinical decisions while helping to reduce errors and ensuring compliance with organizational best practices.1 In complex large health systems, processes for standardization and adherence to emergency department (ED)-based suicide screening practices are challenging and may benefit from the use of CDS-based tools adhering to the five rights of CDS.2 To improve suicide screening compliance for the ED to 95% by implementing a contextually appropriate CDS-based tool within the electronic health record (EHR).A multidisciplinary group of quality and ED nursing leadership aimed to develop a chief complaint-driven OPA that improved adherence to and completion of suicide screening in the ED. Using an iterative design process over 3 months, a series of two distinct suicide screening OPAs were developed with varying levels of interruption, but both relied on rule-based logic to identify if an ED patient met one of the 57 predefined "Reasons for Visit" or chief complaints requiring suicide screening. Use of chief complaint-driving CDS removed the need for manually remembering complex criteria while contributing to meeting regulatory and organizational standards.The ED suicide screening compliance improved from 64.96 to 77.66% with the initial implementation of the noninterruptive OPA. Subsequently, an interruptive OPA (pop-up window based on a defined trigger that stops the clinician and requires a response), was introduced which further increased screenings being completed to 91.69%. The use of CDS interruptive OPAs significantly improved compliance with suicide screening by including the Columbia Suicide Severity Rating Scale tool directly in the OPA.Use of contextually relevant information, such as reason for visit or chief complaint, and interruptive CDS tools embedded into EHR workflows may improve ED-based suicide screening.

Despite its morbidity, mortality, and financial burden, in-hospital malnutrition remains underdiagnosed and undertreated. Artificial intelligence (AI) offers a promising clinical informatics solution for identifying malnutrition risk and one that can be coupled with clinician-delivered patient care.The objectives of the study were to evaluate an AI-based hospital malnutrition screening model in a large and diverse inpatient population and to compare it to the currently used clinician-delivered malnutrition screening tool.We studied the performance of a gradient-boosted decision tree model incorporating a large language model (LLM) for feature extraction using the electronic medical record data of 106,449 patients over 3.75 years.The model's area under the receiver operating curve was 0.92 (95% confidence interval [CI]: 0.91-0.92) on the first day of hospitalization and rose to 0.95 (95% CI: 0.95-0.96) using the maximum risk predicted for each patient throughout hospitalization, indexed against discharge-coded malnutrition. Similar results were observed when indexed against dietitian-recorded malnutrition. The model outperformed the nurse-administered, modified version of the Malnutrition Screening Tool (MST) that was used in practice. Patients identified by the model had higher likelihoods of readmission and death compared with patients identified by the nurse-administered screener.Our study findings provide validation for a novel model's use in the prediction of in-hospital malnutrition.

For every 30 minutes a provider spends seeing a patient, they spend 36 minutes charting in the electronic health record (EHR). Clinical documentation burden in U.S. health care is driven by increasing administrative tasks associated with EHRs, regulatory demands, and workflow inefficiencies. This burden contributes to increased cognitive load, fragmented care, and staff burnout. No comprehensive conceptual framework guides researchers addressing these challenges.This study aimed to develop a conceptual framework clarifying the interplay between psychological factors, technology, and documentation attributes-usability, effort, and perceived burden-among health care providers.Data were collected from a cross-sectional survey using a convenience sample of hospital- and ambulatory-based physicians, advanced practice registered nurses, and physician assistants. A newly constructed questionnaire was used, incorporating elements from well-established instruments. Descriptive and exploratory factor analysis was performed to identify significant findings and develop the preliminary Clinical Documentation Burden Framework.The analysis revealed three main factors underpinning clinical documentation burden: Poor usability, perceived task value, and excessive mental exertion. These factors were significantly correlated with professional dissonance (PD) and burnout, underscoring the complex interplay between time requirements, design challenges, task engagement, and cognitive load. The resulting conceptual framework highlights the importance of aligning documentation tasks with provider values to mitigate burden.The study offers new insights into the complex phenomenon of documentation burden affecting health care providers by incorporating key psychological factors. This conceptual framework provides a preliminary foundation for understanding this multifaceted problem. Like prior burnout research, conceptual clarity is key to creating shared definitions and a dedicated measurement instrument to support effective interventions. Given that the sample was predominantly advanced practice providers with underpowered subgroup comparisons, the framework should be interpreted as preliminary. This new appreciation of the dimensionality of documentation burden expands the potential levers available to alleviate operational strain and reduce PD and burnout.

Rare B-rapidly accelerated fibrosarcoma gene (BRAF) mutations in advanced melanoma, and other malignancies, represent a significant clinical challenge due to sparse evidence on the efficiency of targeted therapy. Conventional genomic databases do not integrate detailed outcome data on treatments for patients with these mutations, requiring innovative informatics approaches.For the use case of patients with rare BRAF-mutated melanoma, we developed a "Treatment Outcome Tool" as a web-based database on rare cancers that aggregates anonymized, expert-validated clinical data. Unstructured interviews with dermato-oncologic experts guided the design, ensuring that the system allows users to query specific or combined rare BRAF mutations and retrieve key outcome measures, such as progression-free survival, overall response rate, and disease control rate with BRAF and/or mitogen-activated proteinkinase kinase (MEK) inhibition. Data are collected via a structured input form. After rigorous review and quality assurance by dedicated experts, data are then transferred to an externally accessible R/Shiny platform, where they can be assessed. The usability of the developed database was then evaluated by the System Usability Scale (SUS) of contributing dermato-oncologic experts.The first productive database version was implemented in October 2024. As of May 2025, the database contained data from 130 patients with 23 BRAF mutations. Evaluation of the "Treatment Outcome Tool" by 14 international dermato-oncologic experts yielded a median SUS score of 92.5, confirming excellent usability.Our database fills a critical gap in personalized oncology therapy by directly correlating rare BRAF mutation profiles with treatment outcomes. Our tool had usability and was found to be of high clinical value. The generic informatics framework chosen by us has the potential to be expanded to other rare tumors, ultimately enhancing evidence-based clinical practice and fostering international collaboration in cancer research.

This study aimed to evaluate and compare the diagnostic responses generated by two artificial intelligence (AI) models developed 54 years apart, and encourage physicians to explore the use of large language models (LLMs) like GPT-4o in clinical practice.A clinical case of metabolic acidosis was presented to GPT-4o, and the model's diagnostic reasoning, data interpretation, and management recommendations were recorded. These outputs were then compared with the responses from Schwartz's 1970 AI model built with a decision-tree algorithm using Conversational Algebraic Language (CAL). Both models were given the same patient data to ensure a fair comparison.GPT-4o generated an advanced analysis of the patient's acid-base disturbance, correctly identifying likely causes and suggesting relevant diagnostic tests and treatments. It provided a detailed, narrative explanation of the metabolic acidosis. The 1970 CAL model, while correctly recognizing the metabolic acidosis and flagging implausible inputs, was constrained by its rule-based design. CAL offered only basic stepwise guidance and required sequential prompts for each data point, reflecting a limited capacity to handle complex or unanticipated information. GPT-4o, by contrast, integrated the data more holistically, although it occasionally ventured beyond the provided information.This comparison illustrates substantial advances in AI capabilities over five decades. GPT-4o's performance demonstrates the transformative potential of modern LLMs in clinical decision-making, showcasing abilities to synthesize complex data and assist diagnosis without specialized training, yet necessitating further validation, rigorous clinical trials, and adaptation to clinical contexts. Although innovative for its era and offering certain advantages over GPT-4o, the rule-based CAL system had technical limitations. Rather than viewing one as simply "better," this study provides perspective on how far AI in medicine has progressed while acknowledging that current AI tools remain supplements to-not replacements for-physician judgment.

Our objective was to identify barriers to implementing a custom clinical decision support (CDS) alert to randomize individuals in a pragmatic study, specifically those with a positive antinuclear antibody (ANA) test.We integrated a validated logistic regression model into the electronic health record to predict the risk of developing autoimmune disease for individuals with a positive ANA (titer ≥ 1:80). A custom CDS alert was created to randomize eligible individuals into a pragmatic study evaluating whether the risk model reduces time to autoimmune disease diagnosis. The custom CDS alert runs silently in the background and is not visible to providers. Individuals were randomized to either an intervention or control arm. In the intervention arm, the study team reviewed risk model results, notified providers of high-risk scores, and offered expedited rheumatology referrals to high-risk individuals in addition to standard of care. The control arm received standard care only. The study team accessed a daily Epic report containing randomization assignments and model variables.Starting in June 2023, the risk model assessed 3,961 individuals and successfully randomized 2,105 individuals to date. Technical challenges that prevented the custom CDS alert from firing included an unanticipated change in the laboratory testing vendor and reporting due to a broken laboratory machine, followed by a change in the laboratory test name.This case report showcases the successful implementation of a laboratory-based custom CDS alert to randomize individuals for a pragmatic study. This approach enabled our study to be feasible across a large health care system. Key lessons learned included the importance of close collaboration with the laboratory team and thorough understanding of the laboratory testing, workflow, and reporting to ensure successful execution of the laboratory-based custom CDS alert.

Patient portals are increasingly used to support digital health engagement, but little is known about how caregivers used patient portals before, during, and after the coronavirus disease 2019 (COVID-19) pandemic.This study aimed to examine longitudinal changes in caregiver engagement with pediatric patient portals, focusing on logins, session duration, messaging behaviors, and provider response times across prepandemic, pandemic, and postpandemic periods.We conducted a retrospective cohort study using deidentified MyChart data from caregivers of children aged 0 through 11 who received care at four pediatric primary care clinics in the Southeastern United States between March 2018 and March 2023. Generalized linear models were used to compare portal engagement across prepandemic, pandemic, and postpandemic periods. Outcomes included login frequency, session duration, message volume, message types and recipients, and provider response times, all normalized per user per year.Among 478 caregivers, portal logins and session duration increased significantly during and postpandemic, with 16-fold increases postpandemic compared with prepandemic (p < 0.001). Message volume declined substantially during the pandemic (p < 0.001) but returned to baseline levels. Provider response times shortened during the pandemic and remained lower than prepandemic levels (p = 0.032). Messaging to primary care declined and did not recover fully, while specialty care messaging increased across all periods. Appointment and medical advice messages declined during the pandemic, with only the latter rebounding. Customer service inquiries rose significantly and remained elevated, and medication renewal messages increased markedly postpandemic.The COVID-19 pandemic initiated lasting changes in caregivers' engagement with pediatric patient portals, including deeper engagement, quicker provider responses, and shifts in messaging patterns. Findings can be used to guide and optimize caregiver-centered digital health strategies in pediatrics. Future work should explore potential provider burnout from increased portal workload, incorporate multicenter studies, and link portal use to clinical characteristics to better inform digital health interventions.