Herasevich, Vitaly

Objectives: Expanded coverage for telehealth during the COVID-19 pandemic allowed providers to bill for telemedicine services that were previously not reimbursable, including telemedicine critical care (TCC) services for critically ill patients. We aimed to characterize TCC billing practices among Medicare beneficiaries before, during, and after the COVID-19 pandemic. Design, Setting, and Patients: This was a serial cross-sectional study of adult Medicare Fee-For-Service beneficiaries with at least one bill for critical care at acute care hospitals from January 2018 to September 2024. TCC billing was identified using provider billing codes; multivariate regression models were used to determine characteristics associated with receipt of TCC. Key outcomes were patient-, provider-, and hospital-level characteristics associated with TCC billing. Interventions: None. Measurements and Main Results: Billing for TCC increased from 0.002% of critical care bills pre-pandemic to 0.01% of critical care bills during and after the pandemic. Patients billed for TCC were disproportionately likely to have COVID-19 but were otherwise relatively similar to critically ill patients not billed for TCC. Internal medicine/critical care providers accounted for the highest proportion of pandemic TCC bills (46.0%). TCC billing occurred more often at minor teaching hospitals (adjusted odds ratio [aOR], 1.21; 95% CI, 1.03–1.43) and at safety-net hospitals (aOR, 1.33; 95% CI, 1.04–1.70). TCC billing was less likely at small-sized hospitals (aOR, 0.39; 95% CI 0.26–0.58) and medium-sized hospitals (aOR, 0.67; 95% CI, 0.47–0.95), government-owned hospitals (aOR, 0.70; 95% CI, 0.57–0.86), for-profit hospitals (aOR, 0.58; 95% CI, 0.48–0.71), rural hospitals (aOR, 0.70; 95% CI, 0.55–0.89), and critical access hospitals (aOR, 0.59; 95% CI, 0.47–0.73). Conclusions: Billing for TCC among hospitalized critically ill Medicare beneficiaries increased during the pandemic but remained low as a proportion of all critical care bills. There was variability in utilization across subspecialties and lesser utilization at rural and critical access hospitals. Further studies are needed to characterize the clinical and economic consequences of this shift.

BACKGROUND. Survey research in critical care has dramatically increased over the past 25 years, and use of surveys within Delphi methods, which provide consensus definitions for heterogeneous disease states, has similarly expanded significantly. It is crucial for survey researchers to understand the nuances of survey development and expectations of information provided by surveys, to use them correctly. Few simple sources exist to guide development considerations to generate high-quality survey research. No studies to our knowledge exist that draw parallels between the principles of survey research as a guide within the Delphi process, as one potential conduit between a steering committee and expert panel. METHODS. We present a narrative review of and framework for rigorous design of survey research, highlighting the applicability to the design of surveys administered within Delphi methods. RESULTS AND CONCLUSIONS. The methodologic considerations presented for survey research and Delphi methods can inform high-quality critical care research.

Background/Objectives: Novel vision-language models (VLMs) can integrate patient textual data with image data to support medical diagnosis. Recent studies reported conflicting results regarding the performance of multimodal VLMs compared to other models and physician performance. This systematic review aims to assess the diagnostic performance of multimodal VLMs integrating both patient textual and image data across diverse real-world hospital settings. Methods: We performed comprehensive searches of eight resources, including Embase, MEDLINE, and SCOPUS, on 17 December 2025. Eligible studies reporting diagnostic performance of VLMs integrating both image and patient history textual data from real-world adult patients compared to that of other models and physicians were included. The review adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. The Prediction model study Risk Of Bias Assessment Tool + AI (PROBAST + AI) was used to assess the quality and risk of bias. The study protocol was registered in the PROSPERO database (CRD420251244054). This review received no external funding. Results: We screened 11,026 records, of which 18 studies met the inclusion criteria. Six studies comparing multimodal and unimodal models demonstrated the consistent superiority of the multimodal models. Four studies evaluating VLM accuracy as standalone agents compared with physician performance reported conflicting evidence. One study assessing VLMs as a clinical copilot demonstrated higher accuracy from the group of physicians using VLM assistance. A meta-analysis could not be performed due to the heterogeneity across study populations and outcomes. The majority of the studies were assessed as having a high risk of bias due to dataset quality. Primary limitations identified across studies include small sample size, a lack of external validation, and the need for prospective clinical deployment studies. No study provided documented considerations regarding model safety or data security. Conclusions: This systematic review suggests that multimodal VLMs consistently outperform unimodal models with access to only image or text. While model performance as standalone agents compared to humans remains inconclusive, a copilot model has demonstrated high diagnostic accuracy. Given substantial methodological concerns across studies, cautious interpretation is required, No firm clinical recommendation can be made regarding the use of standalone VLMs. Further research employing high-quality datasets is needed to ensure the reliability and clinical applicability of future VLMs.

Abstract Rationale Per the 2021 Sepsis Campaign, it is recommended that hospitals use a performance improvement program for sepsis screening, utilizing machine learning to improve specificity. Around 20% of sepsis at the Mayo Clinic, Rochester site is hospital acquired. As inpatient sepsis is usually detected by extreme physiologic derangement or clinical concern, we seek to leverage risk factors beyond traditional SIRS and sepsis criteria to create a machine and deep learning based risk stratification model to improve the precision of sepsis monitoring. Methods We designed a retrospective observational case control study, including patients diagnosed with sepsis as an inpatient at the Mayo Clinic, Rochester hospital and compared them to patients without a sepsis diagnosis over the last 7 years. Sepsis diagnosis was labeled by a combination of sepsis ICD codes and initial lactate elevation. We excluded transfers from outside hospitals, direct admissions for sepsis and patients diagnosed with sepsis in the emergency department. Data was extracted using Mayo Data Explorer, and data preparation and validation were performed using JMP Statistical Software and Python libraries. Feature selection was critical as we examined all data from previous contacts with healthcare, environmental factors, and dynamic trends of inpatient objective data; amongst 5,000 options of lab tests and vital measurements, we selected near 100 to reduce model complexity. We used XG Boost to train our model with a baseline cohort of 100 sepsis patients and 1000 patients without sepsis. Results We created a model that categorically classified patients into three sepsis risk level profiles (low-risk, medium-risk, high-risk) with performance metrics of a specificity of 97%, sensitivity of 65% and area under the receiver operating characteristic curve of 95%. Amongst laboratory data, lactate and creatinine had the highest feature importance (figure below). Our model also identified variables with the least importance to further refine our architecture. Conclusions Our model is one of the first sepsis risk stratification models whose algorithm is optimized for non-ICU inpatient wards, while the majority of similar studies were conducted on ICU and emergency department patients due to limited data availability. The model’s high specificity will allow for earlier sepsis recognition. By including our full cohort of 7,500 patients with sepsis and 35,000 patients without sepsis, we hope to improve class imbalance and increase sensitivity. By integrating transformer architecture, we seek to eliminate variables of low significance to create a model with more practical use, while maintaining high specificity. This abstract is funded by: None

Abstract Introduction Large Language Models (LLMs) have proven effective at analyzing unstructured and structured data from electronic health records (EHR). Their reliance on static EHR data means they miss the dynamic, high-resolution physiological signals that are the first indicators of patient decline, creating a fundamental data lag as clinical documentation of an event may occur hours after its onset. This represents a critical gap in the ability to develop truly effective triage tools. Furthermore, while LLMs can synthesize multi-source data for acuity assessment, they exhibit known limitations, including hallucinations that constrain autonomous use in safety-critical settings. In contrast, a Vision-Language Model (VLM) with strong image-grounded clinical reasoning can interpret vital sign trend images, surpassing the limitations of current EHR-based models. This approach provides immediate triage evaluations that lead to earlier detection of patient deterioration. Methods This mixed-methods study deployed a VLM (Gemini 2.5) integrated within the eICU dashboard. We used near-real-time two-hour trends in heart rate, blood pressure, respiratory rate, and oxygen saturation from consecutive patient cases during alerts in hospital floor units monitored by eICU that admit acute medical and surgical patients. To increase the diversity of the sample, we sought routine cases from patients without alerts and emergent cases from patients admitted to the medical, surgical, trauma, and mixed ICUs. A structured context prompt was used to generate a standardized summary including clinical interpretation, data quality and artifact assessment, triage recommendation, and the model’s chain-of-thought reasoning. Results We included patients from Mayo Clinic Rochester, MN, using an 800-patient cohort for prompt development and a 100-patient cohort for validation against clinically adjudicated results. In the validation cohort the VLM achieved 80% accuracy (80/100). An analysis of the 20 incorrect predictions revealed a clear pattern of conservative over-triage: 80% (16/20) of errors were clinically acceptable adjacent-class discrepancies (routine↔urgent or urgent↔emergent). Only 4/20 discrepancies represented a substantive misclassification. Most errors therefore reflected conservative prioritization rather than incorrect assessment. The model demonstrated a perfect safety profile, underestimating patient acuity in 0% of cases, and thus avoiding the most clinically dangerous error. These results confirm the feasibility of VLM-based approach and provide a strong foundation for the proposed work. Conclusion This VLM performs cross-modal reasoning by fusing deep visual perception with broad clinical context. The result is a fully contextualized and explainable triage prioritization, which provides direct decision support and integrates into the clinical workflow, allowing for faster and more confident acute care assessments. This abstract is funded by: none

Abstract Objective This study aimed to evaluate how differences in case complexity affect information-processing burden, as measured by eye-tracking, as well as the efficiency of electronic health record (EHR) use among healthcare providers in the intensive care unit. Materials and Methods This cross-sectional study recruited providers from 4 U.S. medical centers that use 2 prominent EHR systems (Epic and Oracle). After reporting demographic information, participants reviewed 2 complex cases and 2 standard cases in their institution’s EHR system and then responded to 5 questions about each case, yielding a performance score. Information-processing burden was assessed by measuring the number of eye fixations via eye-tracking software. The efficiency of EHR use was assessed by measuring the task completion time, number of mouse clicks per minute, number of EHR screens viewed, and performance score. Results Eighty-one providers were included for analysis. Providers exhibited significantly more eye fixations (P < .001) and longer task completion times (P < .001) for complex cases than for standard cases. There were also significantly fewer mouse clicks per minute during complex cases (P < .001). Reviewing a complex case first led to significantly more eye fixations (P = .015) and longer task completion times (P < .01) than when a standard case was presented first. Case complexity did not significantly affect performance scores or the number of EHR screens viewed. Discussion Higher case complexity was shown to be associated with greater information-processing burden and less efficient EHR use. These findings have implications for enhancing the efficiency of EHR use, thereby leading to improved clinical decision-making and patient safety. Furthermore, reviewing complex cases first led to a greater information-processing burden, suggesting that providers could benefit from “warming up” with standard cases before reviewing complex cases.

Background: Paroxysmal atrial fibrillation (AF) is a common arrhythmogenic complication in the intensive care unit (ICU) associated with significant costs and morbidity. AF detection models in the literature have utilized entropic analyses of telemetry to diagnose AF, but they have not been utilized to forecast its onset. Hypothesis: Deep learning techniques applied to high-frequency telemetry data can be utilized to forecast AF with clinically significant lead time. Methods: Telemetry data were collected from 72 ICU monitored beds in a tertiary care center from 4/2018-12/2023. Coefficient of sample entropy (COSEn), a metric of heart rate variability complexity, was calculated on successive 1-minute windows and utilized to identify areas of AF initiation as well as control windows without AF. ECG lead II (256Hz) and beat-to-beat (B2B) measurements were analyzed by the MOMENT time-series foundational model to generate representation embeddings. A binary classification multilayer perceptron (MLP) was trained on these embeddings to forecast AF onset 5m in advance. All results were reported on separate test sets. Group-based trajectory modelling (GBTM) was performed on risk score trajectories from the 20m before AF onset or absence. Results: Over 230,000 hours of B2B data from 2263 patients was analyzed using COSEn to identify 287 transitions to AF from 136 patients. The same number of AF-absent windows were randomly selected. COSEn alone as a forecaster produced an area under the receiver operating characteristic curve (AUC) of 0.65 (95% CI: 0.61-0.69). The MOMENT + MLP model (MMM) achieved an AUC of 0.81 (0.78-0.83) (Fig. 1). COSEn and MMM were subsequently applied to the two hours preceding AF onset or absence. For COSEn, the mean risk trajectory dipped in the baseline period whereas the MMM risk trajectory sustained a high level through the baseline, highlighting its capacity to forecast AF (Fig. 2). For AF patients, GBMT yielded two types of trajectories: one with elevated risk throughout, and another with a rising trend 5 minutes prior to AF onset (Fig. 3). Conclusions: Deep learning techniques can forecast the onset of AF with up to 5 minutes of lead-time and GBTM allowed for the identification of different pre-AF risk trajectory phenotypes.The ability to forecast AF onset has potential to clinically prevent poor outcomes in the ICU, though further refinements to augment the amount of lead time are warranted.

Abstract Rationale: Ongoing healthcare staffing shortages have prompted new care paradigms that incorporate remote care delivery to support timely detection of patient physiological deterioration through mechanisms other than clinician bedside evaluations. In partnership with institutional Information Technology, our team has developed a unique clinical informatics environment, known as the Advanced Medical Practice (AMP) platform, that incorporates data from multiple sources and platforms and supports remote observation of patients from the electronic intensive care unit(eICU). AMP takes advantage of technologies outside of vendor-supported EHRs, and its advanced user-interface design reduces information overload while incorporating real-time data from novel devices and sensors. We partnered with the hospital practice to co-develop a score within the AMP platform(Clinical Events Detection and Response[CEDAR]) that incorporates heart rate, blood pressure, respiratory rate, and SpO2 to stratify patient risk and generate deterioration alerts for the electronic ICU(eICU) team. This formative evaluation employed an implementation science approach to refine CEDAR and ensure it is acceptable to users and can be feasibly implemented in practice. Methods: The study team conducted this pilot work in the eICU at Mayo Clinic (Rochester, MN). We observed the eICU workflow and documented feedback from eICU staff in fieldnotes. eICU staff were asked to log issues in a tracking file during shifts. Issues were reviewed daily and discussed at multidisciplinary team meetings weekly. Observation and tracking data were coded to constructs in the Consolidated Framework for Implementation Research(CFIR 2.0) framework and mapped to Expert Recommendations for Implementation Change(ERIC) strategies that targeted key determinants. Results: Between 9/23/24 and 10/25/24 we completed 25 days of observations. Most identified issues related to technical/design issues like unclear vital trigger alerts (CFIR: Innovation trialability and design); staff knowledge of CEDAR functionality and attitudes about functionality, complexity, or relative advantage compared to current systems (CFIR: Inner setting access to knowledge and information and relative advantage); and difficulties incorporating CEDAR into workflow(CFIR: Inner setting/compatibility). Implementation strategies mapped to CFIR determinants included promoting adaptability of customized interface features; staging implementation scale-up with iterative improvements; providing local technical assistance; assessing for readiness and identifying facilitators and barriers; and conducting educational outreach visits. Conclusions: Formative evaluation during pilot studies are a critical step in the process of intervention development and testing. User engagement and frameworks from implementation science can identify issues of feasibility and inform strategies to ensure acceptability prior to intervention scale-up in clinical practice and readiness for conducting larger implementation studies.