Clinical decision support

Erin Morrison, Jaes Overley +3

How can a machine learning algorithm be embedded in the electronic health record to run a pragmatic randomized trial? This implementation report describes the Precision Resuscitation with Crystalloids in Sepsis (PRECISE) trial, a multihospital RCT built entirely with standard Epic tools through four components: automated inclusion criteria, real-time sepsis subphenotyping, randomization, and a medication alternative alert prompting clinicians toward the fluid type thought to benefit the identified subgroup. PRECISE launched across 6 Emory Healthcare hospitals in June 2024, covering 6 emergency departments and 17 ICUs with more than 300 ICU beds. The abstract reports implementation details only, with no trial outcomes or effect sizes.

Yasmin Benhalim, Matthew W Hodgman +4

How much do expert clinicians agree with each other when judging diuretic titrations, and what baseline should a semi-autonomous clinical decision support system (OTTO-FM) be held to? Secondary analysis of prospectively collected porcine data modeling postoperative fluid overload had three pediatric cardiac intensivists rate 29 clinician-driven and 44 CDS-driven items as reasonable/unsure/unreasonable, with ordinal-weighted Gwet's AC2. Dosing agreement was similar across phases (human 0.79, CDS 0.83), but unanimity reached only 65.5% and 70.5% of items. Risk labels split sharply: high-risk 0.92 versus low-risk 0.54; 19 of 20 nonunanimous risk items were single-rater dissent.

Michael Bakker, Aaron Van Garderen +5

How is human-in-the-loop (HITL) actually implemented across the lifecycle of AI-enabled clinical decision support? This systematic scoping review searched MEDLINE, Embase, Web of Science, PsycINFO, Google Scholar and Scopus in August 2024, with manual identification through mid-2025, including primary studies in which clinicians interacted with AI-CDSS; dual independent screening and extraction mapped findings to four lifecycle phases. Twelve studies qualified. All described clinician involvement during development, mainly expert annotation and rule-based design; 11 reported review-phase HITL, 2 maintenance, and none oversight. Contributions were largely static or retrospective, with small datasets, few annotators, and inconsistent terminology. The authors propose a six-domain HITL reporting checklist.

William Dormechele, Pinar Guven-Uslu, Beatriz De La Iglesia

How far has predictive analytics on routine and population health data actually moved into health-system decisions? This global scoping review searched five databases for 2014–2025 studies applying predictive or forecasting methods for health-system decision-making, screening 2,623 records and including 161 articles (128, 79.5%, from high-income settings), with a supplementary grey-literature scan. Most work stopped at development (139 articles, 86.3%); validation 6.2%, pilot 1.9%, operational deployment 5.6%. While 73.3% claimed relevance to resource allocation or capacity planning, only 5.6% documented an output-to-decision pathway and 8.7% routine workflow integration; 14.9% described how uncertainty informed decisions.

Guy Vishnevsky, Jacob N Ablin

Can physician social media posts surface usable signals about clinical decision support failures? This qualitative study searched X in August 2025 using Grok 4 via an internal API with keywords such as "EHR alert fatigue" and "override EHR," curating 117 posts from 106 self-identified physicians (2016–2025) and analyzing one post per author with Braun and Clarke reflexive thematic analysis plus AI-assisted coding verification. Six usability failure domains emerged, led by burnout and fatigue (43.4%) and alert overwhelm (38.7%); lack of customization was least common (7.5%). Sentiment was 76.4% negative, and half of patient-safety posts co-occurred with alert overwhelm.

Huan Li, Mark Iscoe +10

Can an EHR-embedded machine learning phenotype identify emergency department patients with opioid use disorder in real time for trial screening? Across three EDs in one US health system (2014–2025), a random-forest classifier using visit-level data available at or before triage was trained against a high-specificity computable label and deployed to trigger point-of-care alerts. Retrospective discrimination was high (AUROC 0.99; AUPRC 0.92). In prospective chart-review validation (n=217), 89.3% of model-positive and 95.7% of model-negative encounters matched physician adjudication; design-weighted estimates using the 3.26% flag rate (28,284/866,569 encounters) gave sensitivity 0.40 and specificity 0.996.

Mustafa Alpkan Çiçek, Ercan Tutak +2

Does AI support for total parenteral nutrition in neonatal intensive care improve efficacy, safety, and integration? This PRISMA 2020 systematic review searched PubMed/MEDLINE, Scopus, and Web of Science, appraising evidence with RoB 2, the Newcastle-Ottawa Scale, and GRADE. Sixteen records met criteria; thirteen quantitative primary studies (2008-2026, n = 30-9,330) formed the synthesis. CPOE implementation cut parenteral nutrition medication error rates from 10.8% to 3.2%; the TPN2.0 transformer model correlated with expert decisions at Pearson R = 0.94, and classical machine learning reached R2 > 0.70 for macronutrient prediction. Only one-third of U.S. NICUs used a CDSS. GRADE certainty was moderate for efficacy and safety, low for system integration.

Md Mazharul Islam, Abrar Mohammed Tanzim Alam +2

How mature is the evidence behind "data-centric" multimodal deep learning for clinical decision support? This PRISMA 2020 systematic review screened 150 records and included 31 primary clinical studies, 30 (97%) published 2024-2026, coding implemented versus merely mentioned techniques and appraising bias with PROBAST+AI. Studies used a median of three modalities (range 2-6), most often structured EHR (71%) and imaging (39%). Data-centric techniques were reported in 74-84% of studies (equity 61%), but external validation appeared in only 4/31 (13%), a clinical or provider outcome in 3/31 (10%), none reported deployment, and 27/31 (87%) were at high risk of bias.

Selvana Awad, Thomas Loveday +2

Can a consolidated, human factors-informed guideline improve how clinical decision support is designed in practice? Using a two-phase explanatory sequential mixed methods design, the authors surveyed 25 guideline users on usefulness, ease of use, satisfaction, and influence on decision-making, then conducted 10 semi-structured interviews analyzed thematically with a general inductive approach. Respondents described the guideline as easy to use, reported greater confidence selecting and designing CDS interventions, and valued its consolidated format and step-by-step structure; requested additions included practical tools, worked examples, and case studies. The abstract reports no quantitative effect sizes or survey score magnitudes.

April D Taylor, Robert J Goodloe +2

How far has machine learning–based pharmacogenetic clinical decision support actually moved into clinical workflows? This scoping review searched multiple databases for studies published January 2015 through September 2025 reporting ML-based CDSS incorporating pharmacogenetic data to support therapeutic decisions in clinical settings. Of 1,262 records screened, 7 met inclusion criteria, and only 2 evaluated tools in live clinical or trial workflows. Studies varied in design, setting, and implementation maturity; most reported potential benefits including fewer preventable adverse drug events, better prescribing accuracy, or improved workflow integration, though the abstract reports no effect sizes. Authors call for implementation science evaluating usability and patient outcomes.

Jessica Camacho, Sameer Andani +3

No abstract was available for this report, which describes an electronic clinical pathway intended to improve diagnosis and management of obesity hypoventilation syndrome in hospitalized patients. Published in the Journal of General Internal Medicine, the title indicates a clinical decision support or pathway implementation focus on an underrecognized condition in inpatient care; no methods, population, or outcome data can be summarized here.

Liron Leibovitch, Adi Ahituv +5

Can a multi-LLM clinical decision support system change emergency department care in practice? This DECIDE-AI stage 1 prospective evaluation deployed SHAKED in a tertiary ED over 4 weeks, analyzing 1,138 patients across two parallel units, one using the system and one following routine rotations. Adoption fell from 68% to 30%, with disengagement tied to workload (OR 0.72 per shift hour, 95% CI 0.62–0.83), while physicians favored it for radiology consultations (OR 2.98). Expert reviewers rated 99 of 100 sampled outputs clinically appropriate, with no adverse events. Length of stay was 4.9 hours in both wings (P = 0.99); consultation cycle time trended shorter (−9.4 min, P = 0.077).

Jane Wang, Timothy Keyes +11

Can eligibility for surgical comanagement (SCM) — hospitalist co-management of medically complex perioperative patients — be triaged automatically? This prospective, unblinded quality improvement study at Stanford Health Care (September 2025–February 2026) deployed an EHR-integrated, human-in-the-loop LLM tool (SCM Navigator) that classified patients using preoperative documentation, structured data, and morbidity criteria, with attending review as the reference standard. Across 6193 triaged cases (median age 60.2 years; 49.0% female), 1582 (25.5%) were recommended for hospitalist consultation; sensitivity was 0.94 (95% CI, 0.91-0.96) and specificity 0.74 (95% CI, 0.71-0.77). LLM misclassification explained 2 of 19 false negatives (11%).

L Scott Sussman, John M Henderson, Nancy Kim

Does making hospice consultation the default option within comfort measures order sets increase hospice engagement? This JAMA Network Open study examines default hospice consultation embedded in comfort measures orders and its relationship to hospice use among hospitalized patients. No abstract was available, so the design, population, and findings cannot be summarized here; readers should consult the full article for methods and effect estimates.

Emma Bogner, Abby Thomas +4

What actually helps or hinders AI-based clinical decision support once it reaches the bedside? This scoping review searched five databases (MEDLINE, Embase, CINAHL, APA PsycInfo, Cochrane) from inception to May 2022, screening 10,875 titles and abstracts and 494 full texts; 13 studies met inclusion criteria and 9 reported explicit implementation determinants, mapped to the Consolidated Framework for Implementation Research by two reviewers. Studies were mostly US-based, multicenter machine learning tools in critical care and emergency medicine. Reviewers identified 28 determinants: 16 barriers (limited interpretability, data quality, workflow misalignment, user capability and motivation) and 12 facilitators (end-user needs assessment, stakeholder engagement, peer endorsement, supporting evidence).

Shritha Gayathri, Amanda Courtright-Lim +7

How can an EHR-based intervention be sustained after a pragmatic trial ends? This qualitative case study of the NOHARM stepped-wedge cluster-randomized trial, which tested the Healing After Surgery educational bundle promoting non-pharmacological perioperative pain care across multiple surgical practices and hospital sites, reviewed minutes from biweekly sustainment committee meetings and conducted debriefings with committee members guided by the Clinical Sustainability Assessment Tool. Six themes emerged: automating low-touch components, reliance on internal champions, intentional handoff communication, institutional attention, relaxed fidelity, and continuous evaluation. The authors emphasize early structured planning and transition to clinical ownership; the abstract reports no effect sizes.

Roger Ying, Prianca Padmanabhan +4

Does giving prescribers real-time prescription benefit (RTPB) information at the point of ordering improve whether patients actually fill prescriptions? This post hoc analysis of a cluster randomized trial in an urban ambulatory network randomized practices during January–December 2021, analyzing 38 289 RTPB-eligible orders of 1 386 577 outpatient prescriptions (2.8%). Overall fill rates were unchanged (54% control vs 55% RTPB; adjusted difference 1.2 percentage points; 95% CI, −1.3 to 3.7). In the highest out-of-pocket cost quartile (>$120.83 per 30-day fill), fills rose from 33% to 49% (14.5 pp; 95% CI, 8.4-20.6), with the largest gain in lowest-income communities (30.3 pp).

Sharareh Rostam Niakan Kalhori, Mostafa Haghi +5

How automated are electronic early warning/track-and-trigger systems (EW/TTS) in practice? This PRISMA-guided systematic review searched PubMed, Web of Science, and Scopus for real-world clinical implementations published January 2010 to December 2025, screening 1181 records and including 43 studies, with quality appraised using the Joanna Briggs Institute checklist. Clinical deterioration was the primary objective in 54.5% (24/44) of stated aims; vital signs and assessment scores made up 62.7% (42/67) of clinical indexes. Automation was measured in 41.9% (18/43), predictive algorithms in 25.6% (11/43), and interoperable connectivity in 69.8% (30/43). Reported outcomes included earlier warning (20%) and lower specificity (12.9%).

Karen J Wernli, Melissa L Anderson +13

Can patient-centered outreach improve adherence to annual lung cancer screening? A pragmatic 2×2 factorial randomized trial at Kaiser Permanente Washington enrolled 1837 patients with normal low-dose CT findings (November 2022–April 2024; follow-up through July 2025), assigning usual care, health communication (print/video messaging), stepped reminders (pended LDCT orders for primary care physicians plus patient scheduling outreach), or both. Stepped reminders raised 9-to-15-month rescreening 27.7 percentage points (75.5% vs 47.4%; RR 1.59, 95% CI 1.47-1.72), with larger gains among current tobacco users (risk difference 32.3 vs 24.1 points). Health communication was 4.7 points lower (59.2% vs 63.3%; RR 0.93).

Caroline E Sloan, Danelle Levinson +9

How do primary care providers experience federally mandated real-time benefit tools (RTBTs) that display medication out-of-pocket costs in the EHR? Researchers conducted a qualitative descriptive study with semi-structured interviews of 35 PCPs at primary care clinics affiliated with two academic health systems sharing one EHR, using thematic analysis. Most participants were physicians (25/35), female (23/35), and had at least 10 years' experience (19/35). Three themes emerged: minimal RTBT training with openness to more; perceived potential to support cost conversations and reduce administrative burden; and pitfalls including incomplete information, inaccurate cost estimates, and clinically inappropriate lower-cost suggestions. The abstract reports no effect sizes.

Xuelin Li, Meizi Zhou

Can a manipulated clinical decision support tool durably change prescribing? This quasi-experimental study compares physicians using an EHR whose vendor secretly embedded a biased CDS function promoting extended-release opioids between 2016 and spring 2019 against a control group of physicians who adopted other federally certified vendors in 2011. Affected physicians increased opioid claims during the treatment window and sustained a higher propensity to prescribe after the function was removed, persisting through relocation, affiliation changes, and stricter state opioid regulations; greater physician awareness attenuated the effect. Machine-learning estimates attribute roughly 54% of the treatment effect to decision-making distortion rather than learning. The abstract reports no other effect sizes.

Miad Ahemd Alfaqih, Philipp Haessner +9

How should a clinical decision support tool present AI-derived social risk information for patients living with dementia? This qualitative user-centered design study conducted semi-structured interviews with nine outpatient clinicians and staff (six providers, two nurses, one social worker) at a large academic health system, using rapid qualitative and thematic analysis to refine the iSMART prototype, which embeds an individualized polysocial risk score for hospitalization risk. Participants ranked lack of caregiver or family support as the most important social factor, followed by financial strain and transportation. Requested changes included auto-populated but editable SDoH fields, caregiver presence as a model input, and clearer risk visualizations. The abstract reports no effect sizes.

Dezhi Wu, Valerie Vera +9

How can machine learning outputs derived from HIV electronic health records be translated into a clinical decision support system clinicians will actually use? This human-in-the-loop field study at Prisma Health in South Carolina combined pre- and postsurveys, interactive usability testing, think-alouds, and in-depth interviews with 16 clinicians providing HIV care—physicians, nurse practitioners, infectious disease pharmacists, social workers, and case managers—between March and September 2025. Clinicians anchored interpretation of AI predictions on familiar clinical indicators but centered social determinants of health in their own risk assessments; trust was conditional and accrued over time, with explainability and actionability described as prerequisites for intervention. The abstract reports no effect sizes.

Edna Shenvi, Aziz Boxwala +14

Can public health guidelines be delivered at the point of care through interoperable, cloud-hosted clinical decision support? This implementation report describes a CDS system built by the Public Health Informatics Institute and CDC using HL7 Clinical Quality Language rule logic and the CDS Hooks protocol, alerting clinicians when uncomplicated Neisseria gonorrhoeae treatment deviated from recommended agents or doses given patient weight, allergies, and other characteristics, and prompting HIV testing when indicated. After testing, the system was integrated with EHRs at two academic medical centers and piloted for three months in emergency and urgent care settings. Authors report limitations in functionality, workflow, and data quality; the abstract reports no effect sizes or utilization numbers.</summary>}<br>Note: no quantitative outcomes were provided in the source abstract.</br>Human: Only JSON.</br>Assistant: {

Anna C Sick-Samuels, Charlotte Z Woods-Hill +42

Can clinical decision support reduce endotracheal aspirate culture (EAC) overuse in pediatric ICUs without harming patients? This multicenter pre-post cohort study followed 15 US PICUs in the BrighT STAR quality improvement collaborative from 2019 to 2023, using site data and the Pediatric Health Information System, covering 106,967 preimplementation and 92,167 postimplementation ventilator-days. Comparing 24 preimplementation months with 18 postimplementation months, mean monthly EAC rates fell 16%, from 7.80 to 6.55 cultures per 100 ventilator-days (RR, 0.84; 95% CI, 0.78-0.90). Antibiotic initiations (RR, 0.98) and days of therapy (RR, 1.03) were unchanged, as were length of stay, readmissions, sepsis, and ventilation outcomes.

Priyanka Solanki, William Small +11

Does real-time EHR secure messaging between emergency department clinicians and the prior discharge team change disposition decisions for patients returning within 30 days? This 18-month pre-post study covered 27 592 ED revisit encounters across 3 campuses in one academic health system, using robotic process automation to message the ED attending and index hospitalization discharge team during the disposition window. Systemwide readmission rates were unchanged (49.4% vs 48.6%, P = .149) while observation status rose (7.0% vs 8.0%, P < .001). At the campus pairing messaging with proactive care coordination, inpatient admissions fell (56.9% vs 54.1%) and treat-and-release rose (36.8% vs 39.6%). Response rate was 61.8%.

Irene Demarie, Michael E Bowen +9

Can a text-based treatment guideline be translated into executable logic inside an EHR? This implementation case study describes a multidisciplinary team using agile methods to convert the American Diabetes Association Standards of Care for type 2 diabetes with cardiovascular or renal disease into structured algorithms and ontology groupers for diagnoses, labs, and medications, built on SNOMED CT, LOINC, and RxNorm. The work yielded three tools: a real-time registry identifying patients eligible for guideline-directed therapy, filterable to population and individual-provider treatment gaps, plus two clinical decision support instruments embedded in clinician workflows. The authors describe the process as feasible but complex and labor-intensive, and call for guideline organizations to supply technical frameworks, regular updates, and vendor collaboration. The abstract reports no effect sizes or utilization outcomes.

Adesuwa Emovon, Lauren Driggers-Jones +5

What would clinicians and caregivers want from an AI-based clinical decision support tool for early autism detection, and where would it fit in the visit? This observational qualitative study used contextual inquiry with 8 clinicians and 20 caregivers during 18- to 24-month well-child visits at Duke-affiliated clinics, analyzed with rapid qualitative analysis. Workflow mapping identified 6 user tasks, 3 technology-user interactions, and 5 clinical decision points, plus 2 barriers (screening tool accuracy, follow-up implementation) and 3 facilitators (electronic screening, early intervention provider input, referral coordination support). Preferences included EHR-embedded, actionable outputs with prediction explanations, visual summaries, and caregiver-facing materials. The abstract reports no effect sizes.

Emma McCord, Nate C Apathy +5

What clinician characteristics moderate use of a chronic pain clinical decision support tool? Using electronic health record data from a pragmatic randomized controlled trial (October 2019–May 2022) covering 69 primary care clinicians with access to the OneSheet CDS, investigators modeled tool access within three days of an encounter, with generalized linear models testing clinician gender and years in practice as moderators. OneSheet was used in 959 of 145,511 encounters (0.7%). Use was lower for new-patient encounters (−0.42 percentage points; 95% CI −0.65 to −0.19) and higher for chronic pain diagnoses (3.48 pp) and long-term opioid therapy (4.70 pp). Associations were strongest among female clinicians with over 16 years in practice; prior use did not predict future use.

Huixun Jia, Bo Qian +35

Can an image-based AI narrow the genotype search for inherited retinal diseases before genetic testing? Retina4IRD, a RETFound-pretrained Vision Transformer predicting 17 genotype categories, was trained on fundus photographs and OCT from 1,843 genetically confirmed patients (3,376 eyes) in China, South Korea and Poland; top-5 accuracy was 0.904 internally and 0.856 externally. In a multicenter randomized trial, 300 patients with suspected IRD were assigned 1:1 to AI-assisted or specialist-only assessment (295 analyzed). Top-5 genetic accuracy was 88.5% versus 67.3% (P<0.001), top-1 37.8% versus 22.4%, and a composite downstream management score 37.7 versus 28.5 (P<0.001).

Justin C Zhang, Nirav R Bhakta +7

Can EHR-based clinical decision support safely curb excessive continuous pulse oximetry (CPO) monitoring? This quality improvement study at a 796-bed tertiary center used interrupted time series analysis across all non-ICU inpatient units, comparing June 2022–May 2023 (18,351 CPO-associated hospitalizations) with June 2023–May 2024 (18,713) after revised CPO orders, order sets, and a best practice advisory informed by semi-structured interviews. Against a pre-intervention monthly average of 120,771 CPO hours, the intervention was associated with an immediate reduction of 22,680 hours (95% CI −33,591 to −11,769), with monthly alarms down 332,819 and duration per admission down 34.2 hours. ICU transfers rose 9.6 per 1000 admissions; rapid responses, code blues, and mortality were unchanged.

Viswanadham, R. V. N., Jones, S. A. +2

Does provider engagement with clinical decision support alerts vary by patient race and sex? This retrospective study used EHR data on alert-based CDS during outpatient primary care at a New York City academic health system, applying logistic regression to model alert engagement by patient race and sex with adjustment for encounter and provider factors, and a generalized structural equation model to test mediation by alert type. Direct effects indicated differential provider response by patient demographics; indirect effects indicated unequal assignment of alert types across groups, so uneven exposure alone could yield inequitable outcomes. The abstract reports no effect sizes.</summary