Applied Clinical Informatics最新文献

{"title":"Special Issue on CDS Failures: Development and Evaluation of ORCA, a Resilient Solution for Order Set Access During EHR Downtimes.","authors":"Stephon N Proctor, Bimal Desai","doi":"10.1055/a-2620-6221","DOIUrl":"https://doi.org/10.1055/a-2620-6221","url":null,"abstract":"<p><strong>Background: </strong>Clinical decision support systems (CDSS) are central to modern healthcare, but their effectiveness is compromised during system downtimes, which affect 96% of healthcare organizations. During these failures, clinicians lose access to critical decision-making tools like order sets, increasing the risk of medical errors. Traditional downtime solutions, such as paper-based protocols, are often impractical and difficult to maintain.</p><p><strong>Objectives: </strong>This study introduces and evaluates ORCA (Offsite Repository for Clinical Assets), a resilient web-based solution designed to maintain access to EHR order sets during system failures. We assessed its usability and effectiveness as a downtime decision support tool across various clinical settings.</p><p><strong>Methods: </strong>ORCA was developed based on analysis of previous downtime incidents, replicating essential order set functionality while ensuring offsite accessibility. We conducted usability testing with 16 clinicians from diverse specialties, using structured tasks and think-aloud protocols. User feedback was collected through the Usability Metric for User Experience (UMUX) questionnaire and thematic analysis of interview transcripts.</p><p><strong>Results: </strong>ORCA demonstrated strong usability (mean UMUX score: 86.2). Thematic analysis revealed key implementation challenges: system limitations (24.56% of responses), workflow integration (23.39%), and interface navigation (22.22%). Users valued ORCA's familiar interface and offsite accessibility but identified critical gaps in dynamic decision support capabilities.</p><p><strong>Conclusions: </strong>ORCA represents a viable approach to maintaining basic clinical decision support during downtimes. However, significant challenges remain in replicating dynamic CDS features and ensuring effective integration with existing downtime procedures. These findings inform future development of resilient CDS systems and highlight the importance of planned fallback pathways in clinical systems.</p>","PeriodicalId":48956,"journal":{"name":"Applied Clinical Informatics","volume":" ","pages":""},"PeriodicalIF":2.1,"publicationDate":"2025-05-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144152613","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"The Effect of Computerized Alerts on Prescribing and Patient Outcomes: A Systematic Review.","authors":"Brian Bell, Adam Khimji, Basharat Hussain, Anthony Avery","doi":"10.1055/a-2620-3244","DOIUrl":"https://doi.org/10.1055/a-2620-3244","url":null,"abstract":"<p><p>Background In recent years, there has been an expansion in the literature on the effects of computerized alerts on prescribing and patient outcomes. The aim of our study was to examine the impact of these systems on clinician prescribing and patient outcomes. Methods We searched three databases (Medline, Embase and PsychINFO) for studies that had been conducted since 2009 and included studies that examined the effects of alerts at the point of prescribing. We extracted data from 69 studies. Results Most studies reported a beneficial effect on prescribing of computerized alerts (n = 58, 84.1%), including all studies (n=4) that used passive alerts. Seven of the 10 studies that reported on patient outcomes showed a beneficial effect. Both randomized controlled trials (RCTs) and non-RCTS showed beneficial effects on prescribing across a range of different types of alert. In 43 studies it was possible to ascertain the effects of different types of alert; the interventions that were most frequently associated with improvements in prescribing were drug-laboratory alerts (9/11; 81.8%); dose range checking (6/7; 85.7%); formulary alerts (8/9; 88.9%) and drug-allergy alerts (4/4; 100%). However, most of the studies did not satisfy the quality criteria. Conclusion Most of the studies found a beneficial effect of computerized alerts on prescribing. We have also shown that these benefits are apparent for a range of different types of alert. These findings support continued development, implementation and evaluation of computerized alerts for prescribing.</p>","PeriodicalId":48956,"journal":{"name":"Applied Clinical Informatics","volume":" ","pages":""},"PeriodicalIF":2.1,"publicationDate":"2025-05-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144152614","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Typing proficiency among physicians in internal medicine: a pilot study of speed and performance.","authors":"Francois Bastardot, Vanessa Kraege, Julien Castioni, Alain Petter, David W Bates, Antoine Garnier","doi":"10.1055/a-2620-3147","DOIUrl":"https://doi.org/10.1055/a-2620-3147","url":null,"abstract":"<p><strong>Background: </strong>Electronic health records (EHR) are widely implemented and consume nearly half of physicians' work time. Despite the importance of efficient data entry, physicians' typing skills - potential contributors to documentation burden - remain poorly studied.</p><p><strong>Objective: </strong>To evaluate the typing skills of physicians and their associations with demographic characteristics and professional roles.</p><p><strong>Methods: </strong>This cross-sectional pilot study included a convenience sample of physicians (residents, chief residents, and attending physicians) from the internal medicine division of an academic hospital. Participants completed a one-minute typing test under supervised conditions. The primary outcome was raw typing speed, measured in words per minute (WPM). Secondary outcome was a performance score calculated by subtracting 50 points for each error from the total number of characters typed per minute.</p><p><strong>Results: </strong>Participation rate was 100% (82/82 physicians). Mean age 33.7 ± 7.3 years; 7.2 ± 7.1 years since graduation; 45.1% female. Mean typing speed was 53.4 WPM (range: 31-91 WPM), with 57.3% (47/82) of participants exceeding 50 WPM, a threshold commonly considered as professional. Bivariable analysis showed significant negative association with age (Spearman's ρ = -0.281, p = 0.011), which was not sustained in the multivariable analysis. No significant association was observed with sex, country of diploma, or role. Upon multivariable analysis, performance score showed significant negative association with age (β = -17.724, p = 0.009) but positive association with years since graduation (β = 16.850, p = 0.021), suggesting a generation- and experience-related interaction.</p><p><strong>Conclusions: </strong>Nearly half of physicians exhibited professional-level typing skills, yet overall performance varied widely and was influenced by both generational factors and clinical experience. Given that documentation burden affects clinicians across all skill levels, both individual and systemic strategies-such as improved EHR design and alternative input methods-should be explored.</p>","PeriodicalId":48956,"journal":{"name":"Applied Clinical Informatics","volume":" ","pages":""},"PeriodicalIF":2.1,"publicationDate":"2025-05-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144152528","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Special Topic Burnout: The Digital Workload Divide: Investigating Gender Differences in EHR Messaging Among Primary Care Clinicians.","authors":"Julianne Scholes, Lauren Schiff, Alicia Ann Jacobs, Michelle Cangiano, Marie Sandoval","doi":"10.1055/a-2618-4580","DOIUrl":"https://doi.org/10.1055/a-2618-4580","url":null,"abstract":"<p><strong>Background: </strong>Electronic Health Record (EHR) patient portal messaging has become as essential tool for patient-clinician communication by improving accessibility to primary care. While messaging is beneficial for patients, it can increase clinicians' workloads. Female clinicians receive a greater number of EHR messaging, resulting in an increased workload.</p><p><strong>Objectives: </strong>This evaluation explores the factors in clinician gender disparity in EHR messaging burden.</p><p><strong>Methods: </strong>The first phase of the evaluation included a retrospective analysis of the messages to 267 primary care clinicians in the University of Vermont Health Network (UVMHN). The second phase analyzed patient demographics and panel complexity. Statistical analysis was performed on the types of patient-initiated messages to primary care clinicians and subsequently on all messages across the UVMHN.</p><p><strong>Results: </strong>Female clinicians received significantly more patient-initiated medical advice requests than their male counterparts (68.28 vs. 49.22 messages/month, p=0.005) and spent more time managing messages (1.85 vs. 1.35 min/day, p=0.006). Despite this increased workload, response times remained similar between genders. Female clinicians have a higher proportion of female patients, and analysis of all messages sent across the organization demonstrated that female patients send more messages than male patients. (59 vs. 52 messages/female vs. male, p=0.001). Panels size and complexity were similar for both male and female providers.</p><p><strong>Conclusions: </strong>These findings highlight an unequal messaging burden for female clinicians, largely due to patient demographics. Patient panel complexity and clinician full-time equivalent (FTE) were similar between genders. Disparities in message volumes appear to be driven primarily by patient communication behavior differences between genders rather than differences in workload allocation. These findings likely contribute to increased burnout risk among female clinicians. Addressing this imbalance through workflow optimization and AI-driven message triage systems may help to mitigate the burden on female clinicians and promote greater equity in primary care.</p>","PeriodicalId":48956,"journal":{"name":"Applied Clinical Informatics","volume":" ","pages":""},"PeriodicalIF":2.1,"publicationDate":"2025-05-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144128079","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Summarize-then-Prompt: A Novel Prompt Engineering Strategy for Generating High-Quality Discharge Summaries.","authors":"Eyal Klang, Jaskirat Gill, Aniket Sharma, Evan Leibner, Moein Sabounchi, Robert Freeman, Roopa Kohli-Seth, Patricia Kovatch, Alexander Charney, Lisa Stump, David Reich, Girish Nadkarni, Ankit Sakhuja","doi":"10.1055/a-2617-6572","DOIUrl":"https://doi.org/10.1055/a-2617-6572","url":null,"abstract":"<p><strong>Background: </strong>Accurate discharge summaries are essential for effective communication between hospital and outpatient providers but generating them is labor-intensive. Large language models (LLMs), such as GPT-4, have shown promise in automating this process, potentially reducing clinician workload and improving documentation quality. A recent study using GPT-4 to generate discharge summaries via concatenated clinical notes found that while the summaries were concise and coherent, they often lacked comprehensiveness and contained errors. To address this, we evaluated a structured prompting strategy, summarize-then-prompt, which first generates concise summaries of individual clinical notes before combining them to create a more focused input for the LLM.</p><p><strong>Objectives: </strong>The objective of this study was to assess the effectiveness of a novel prompting strategy, summarize-then-prompt, in generating discharge summaries that are more complete, accurate, and concise in comparison to the approach that simply concatenates clinical notes.</p><p><strong>Methods: </strong>We conducted a retrospective study comparing two prompting strategies: direct concatenation (M1) and summarize-then-prompt (M2). A random sample of 50 hospital stays was selected from a large hospital system. Three attending physicians independently evaluated the generated hospital course summaries for completeness, correctness, and conciseness using a 5-point Likert scale.</p><p><strong>Results: </strong>The summarize-then-prompt strategy outperformed direct concatenation strategy in both completeness (4.28 ± 0.63 vs. 4.01 ± 0.69, p < 0.001) and correctness (4.37 ± 0.54 vs. 4.17 ± 0.57, p = 0.002) of the summarization of the hospital course. However, the two strategies showed no significant difference in conciseness (p = 0.308).</p><p><strong>Conclusion: </strong>Summarizing individual notes before concatenation improves LLM-generated discharge summaries, enhancing their completeness and accuracy without sacrificing conciseness. This approach may facilitate the integration of LLMs into clinical workflows, offering a promising strategy for automating discharge summary generation and could reduce clinician burden.</p>","PeriodicalId":48956,"journal":{"name":"Applied Clinical Informatics","volume":" ","pages":""},"PeriodicalIF":2.1,"publicationDate":"2025-05-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144121223","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Extracting social determinants of health from dental clinical notes.","authors":"Farhana Pethani, Alec Chapman, Mike Conway, Xiang Dai, Demiana Bishay, Victor Jun Xiang Choh, Alexander He, Su-Elle Lim, Huey Ying Ng, Tanya Mahony, Albert Yaacoub, Sarvnaz Karimi, Heiko Spallek, Adam G Dunn","doi":"10.1055/a-2616-9858","DOIUrl":"https://doi.org/10.1055/a-2616-9858","url":null,"abstract":"<p><p>Objective In dentistry, social determinants of health (SDoH) are potentially recorded in the clinical notes of Electronic Dental Records (EDRs). The objective of this study was to examine the availability of SDoH data in dental clinical notes and evaluate NLP methods to extract SDoH from dental clinical notes. Methods A set of 1,000 dental clinical notes was sampled from a dataset of 105,311 patient visits to a dental clinic and manually annotated for information pertaining to sugar, tobacco, alcohol, methamphetamine, housing, and employment. Annotations included temporality, dose, type, duration, and frequency where appropriate. Experiments were to compare extraction using fine-tuned pre-trained language models (PLMs) with a rule-based approach. Performance was measured by F1-score. Results For identifying SDoH, the best performing PLM method produced F1-scores of 0.75 (sugar), 0.69 (tobacco), 0.67 (alcohol), 0.42 (housing), and 0 (employment). The rule-based method produced F1-scores of 0.70 (sugar), 0.69 (tobacco), 0.53 (alcohol), 0.44 (housing), and 0 (employment). The overall difference between PLMs and rule-based methods was F1-score of 0.04 (95% confidence interval -0.01, 0.09). SDoH were relatively rare in dental clinical notes, from sugar (9.1%), tobacco (3.9%), alcohol (1.2%), housing (1.2%), employment (0.2%), and methamphetamine use (0%). Conclusions The main challenge of extracting SDoH information from dental clinical notes was the frequency with which they are recorded, and the brevity and inconsistency where they are recorded. Improved surveillance likely needs new ways to standardise how SDoH are reported in dental clinical notes.</p>","PeriodicalId":48956,"journal":{"name":"Applied Clinical Informatics","volume":" ","pages":""},"PeriodicalIF":2.1,"publicationDate":"2025-05-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144121216","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Special Edition on CDS Failures: Challenges with Implementing Predictive Models for Inpatient Hypoglycemic Events in CDS.","authors":"Sarah Stern, Richa Bundy, Lauren Witek, Adam Moses, Christopher Kelly, Matthew Gorris, Cynthia Burns, Ajay Dharod","doi":"10.1055/a-2617-6522","DOIUrl":"https://doi.org/10.1055/a-2617-6522","url":null,"abstract":"<p><p>Background Inpatient hypoglycemia is associated with increased length of stay and mortality. There have been several models developed to predict a patient's risk of inpatient hypoglycemia. Objectives Describe the barriers to implementing a model that we developed to predict inpatient hypoglycemic events informing a clinical decision support tool. Methods A logistic regression model was trained on inpatient hospitalizations of diabetic patients receiving insulin at Atrium Health Wake Forest Baptist Medical Center, an academic medical center in the Southeastern United States, from January 2020 to December 2021. The model was developed to predict a hypoglycemic event (glucose < 70 mg/dL) within 24 hours of a patient's first borderline-low glucose measurement (70-90 mg/dL). Results The model area under the curve (AUC) was 0.69 on the validation dataset, however we chose not to implement the model in clinical practice. Conclusions We decided not to implement our predictive model into clinical decision support due to a variety of factors including limitations in the predictiveness of the model and several contextual factors. Through this work we learned that it is not always feasible to use predictive analytics in clinical decision support especially when attempting to predict low incidence events for which some important predictors are not documented in the EHR in a structured way.</p>","PeriodicalId":48956,"journal":{"name":"Applied Clinical Informatics","volume":" ","pages":""},"PeriodicalIF":2.1,"publicationDate":"2025-05-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144121220","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Special Issue on CDS Failures: Right Idea, Wrong time: Focusing on alert timing for effective decision support.","authors":"Averi Wilson, Andrew Patrick Bain, Janet Webb, Christoph Ulrich Lehmann, Brett Moran, Nainesh Shah, Ellen O'Connell","doi":"10.1055/a-2605-4510","DOIUrl":"https://doi.org/10.1055/a-2605-4510","url":null,"abstract":"<p><p>1.1.</p><p><strong>Background: </strong>Effective CDS interventions improve adherence to care guidelines, reduce prescribing errors, and, in some settings, decrease patient mortality. However, misalignment with the \"Five Rights\" framework, particularly regarding CDS timing in clinical workflows, can lead to implementation failures, alert fatigue, and physician burnout. 1.2.</p><p><strong>Objectives: </strong>This case series aimed to evaluate and redesign three interruptive CDS alerts at a large safety-net health system to better align with clinician workflows, reduce interruptions, and improve compliance with care guidelines. 1.3.</p><p><strong>Methods: </strong>We analyzed three interruptive alerts using data from Epic's SlicerDicer tool, focusing on alert frequency, contributors to alert triggering, and user responses before and after intervention. Alerts were modified to improve their timing and relevance within the workflow. 1.4.</p><p><strong>Results: </strong>Modifications included retiming an HIV screening alert to trigger during laboratory test orders, reducing alert firings by 87% while increasing monthly screening orders from 3,561 to 4,547 (p<0.001). An administrative alert's firing frequency decreased by 86% through the introduction of a four-hour lockout period, maintaining compliance rates. Finally, restricting a pediatric head circumference discrepancy alert to in-person visits only eliminated interruptions during telehealth encounters, addressing a major source of clinician frustration. 1.5.</p><p><strong>Conclusions: </strong>Aligning CDS tools with clinical workflows through adherence to the \"Five Rights\" framework reduces interruptions and improves outcomes. Iterative review, user feedback, and proactive redesign are essential to ensure CDS effectiveness, particularly as healthcare evolves to include novel care delivery models like telehealth.</p>","PeriodicalId":48956,"journal":{"name":"Applied Clinical Informatics","volume":" ","pages":""},"PeriodicalIF":2.1,"publicationDate":"2025-05-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144041901","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Parental Access to Adolescent Online Healthcare Portals: Benefits, Problems, and Barriers.","authors":"Bryan A Sisk, Alison Antes, Christine Bereitschaft, Fabienne Bourgeois, James DuBois","doi":"10.1055/a-2605-4893","DOIUrl":"https://doi.org/10.1055/a-2605-4893","url":null,"abstract":"<p><strong>Objective: </strong>Online healthcare portals provide access to electronic health information and support clinical communication. Almost no studies have examined perspectives on parental portal access. We aimed to characterize parental and adolescent perspectives on parental portal access.</p><p><strong>Materials and methods: </strong>Semi-structured interviews with 51 dyads of parents and adolescents (102 total interviews). We stratified sampling for equal proportions of adolescents with and without chronic illnesses. We analyzed interview transcripts using thematic analysis.</p><p><strong>Results: </strong>Parents and adolescents identified several benefits of parental portal access: improving understanding and access to information; supporting parents in managing adolescent's health and logistics; supporting parents in teaching adolescents about their health. Parents and adolescents identified the following problems: threatening the adolescent's privacy; creating or exacerbating tension within the family; struggling to understand medical information; creating emotional distress for parents. Parents described the following barriers to portal use: difficulties with enrollment and maintaining access; interface challenges; lack of awareness; lack of interest. Some parents preferred to maintain access after their child was legally an adult. Although the portal has potential to support collaborative care management between parents and adolescents, few parents used this tool collaboratively with their adolescent.</p><p><strong>Discussion: </strong>Parents and adolescents identified multiple benefits, problems, and barriers to parents accessing the adolescent portal. Parents need sufficient access to health-related information in the portal to help them manage their adolescent's health and illness, especially for adolescents with chronic illness. Future efforts could better leverage the portal as a way of supporting collaboration in care between parents and adolescents.</p><p><strong>Conclusion: </strong>Portals offer several potential benefits to parents and adolescents. However, these benefits are impeded by technological limitations and lack of engagement of the adolescent.</p>","PeriodicalId":48956,"journal":{"name":"Applied Clinical Informatics","volume":" ","pages":""},"PeriodicalIF":2.1,"publicationDate":"2025-05-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144032795","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Special Issue on CDS Failures: Performance Degradation between Development and Deployment of a Predictive Model for Central-Line Associated Blood Stream Infections in Hospitalized Children.","authors":"Jonathan M Beus, Mark Mai, Nikolay P Braykov, Swaminathan Kandaswamy, Edwin Ray, D Brad Cundiff, Paulette Djachechi, Sarah A Thompson, Azade Tabaie, Ryan Birmingham, Rishi Kamaleswaran, Evan Orenstein","doi":"10.1055/a-2605-1847","DOIUrl":"https://doi.org/10.1055/a-2605-1847","url":null,"abstract":"<p><strong>Background: </strong>Central line-associated bloodstream infections (CLABSIs) are associated with substantial pediatric morbidity and mortality. The capacity to predict which children with central lines are at greatest risk of CLABSI could inform surveillance and prevention efforts. Our team previously published in silico predictive models for CLABSI.</p><p><strong>Objective: </strong>To prospectively implement a pediatric CLABSI predictive model and achieve adequate performance in offline validation for implementation in clinical practice.</p><p><strong>Methods: </strong>The most performant predictive models were deep learning models requiring substantial pre-processing of many features into 8-hour windows including the current day and up to 56 days prior for the current admission. To replicate this pre-processing, we created novel infrastructure to (1) organize current-day data for all the relevant features and (2) create a staged historical data store for those same features with application programming interfaces to connect the two. We compared predictive performance of these scores for CLABSI in the next 48 hours with two labels, one based on manual review of positive blood cultures in children with central lines and another based on positive blood culture and receipt of at least 4 days of new IV antibiotics.</p><p><strong>Results: </strong>The area under the receiver-operating characteristic (AUROC) fell from 0.97 from retrospective data to <0.60 despite multiple iterations of troubleshooting. Primary root causes included train/serve skew, feature leakage, and overfitting. Hypothesized secondary drivers were complex model specification, poor data governance and inadequate testing, challenging feature translation between real-time and historical data models, limited monitoring and logging infrastructure for troubleshooting, and suboptimal handoff between the model development and deployment teams.</p><p><strong>Conclusion: </strong>To bridge the gap from predictive model development to clinical deployment requires early and close coordination between data governance, data science, clinical informatics, and implementation engineers. Balancing predictive performance with implementation feasibility can accelerate the adoption of predictive clinical decision support systems.</p>","PeriodicalId":48956,"journal":{"name":"Applied Clinical Informatics","volume":" ","pages":""},"PeriodicalIF":2.1,"publicationDate":"2025-05-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144035943","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}