Search

The early detection of the acute deterioration of escalating illness severity is crucial for effective patient management and can significantly impact patient outcomes. Ambient sensing technology, such as computer vision, may provide real-time information that could impact early recognition and response. This study aimed to develop a computer vision model to quantify the number and type (clinician vs. visitor) of people in an intensive care unit (ICU) room, study the trajectory of their movement, and preliminarily explore its relationship with delirium as a marker of illness severity. To quantify the number of people present, we implemented a counting-by-detection supervised strategy using images from ICU rooms. This was accomplished through developing three methods: single-frame, multi-frame, and tracking-to-count. We then explored how the type of person and distribution in the room corresponded to the presence of delirium. Our designed pipeline was tested with a different set of detection models. We report model performance statistics and preliminary insights into the relationship between the number and type of persons in the ICU room and delirium. We evaluated our method and compared it with other approaches, including density estimation, counting by detection, regression methods, and their adaptability to ICU environments.

Abstract Sepsis is a leading cause of mortality and early identification improves survival. With increasing digitalization of health care data automated sepsis prediction models hold promise to aid in prompt recognition. Most previous studies have focused on the intensive care unit (ICU) setting. Yet only a small proportion of sepsis develops in the ICU and there is an apparent clinical benefit to identify patients earlier in the disease trajectory. In this cohort of 82,852 hospital admissions and 8038 sepsis episodes classified according to the Sepsis-3 criteria, we demonstrate that a machine learned score can predict sepsis onset within 48 h using sparse routine electronic health record data outside the ICU. Our score was based on a causal probabilistic network model—SepsisFinder—which has similarities with clinical reasoning. A prediction was generated hourly on all admissions, providing a new variable was registered. Compared to the National Early Warning Score (NEWS2), which is an established method to identify sepsis, the SepsisFinder triggered earlier and had a higher area under receiver operating characteristic curve (AUROC) (0.950 vs. 0.872), as well as area under precision-recall curve (APR) (0.189 vs. 0.149). A machine learning comparator based on a gradient-boosting decision tree model had similar AUROC (0.949) and higher APR (0.239) than SepsisFinder but triggered later than both NEWS2 and SepsisFinder. The precision of SepsisFinder increased if screening was restricted to the earlier admission period and in episodes with bloodstream infection. Furthermore, the SepsisFinder signaled median 5.5 h prior to antibiotic administration. Identifying a high-risk population with this method could be used to tailor clinical interventions and improve patient care.

Computer vision (CV), a type of artificial intelligence (AI) that uses digital videos or a sequence of images to recognize content, has been used extensively across industries in recent years. However, in the healthcare industry, its applications are limited by factors like privacy, safety, and ethical concerns. Despite this, CV has the potential to improve patient monitoring, and system efficiencies, while reducing workload. In contrast to previous reviews, we focus on the end-user applications of CV. First, we briefly review and categorize CV applications in other industries (job enhancement, surveillance and monitoring, automation, and augmented reality). We then review the developments of CV in the hospital setting, outpatient, and community settings. The recent advances in monitoring delirium, pain and sedation, patient deterioration, mechanical ventilation, mobility, patient safety, surgical applications, quantification of workload in the hospital, and monitoring for patient events outside the hospital are highlighted. To identify opportunities for future applications, we also completed journey mapping at different system levels. Lastly, we discuss the privacy, safety, and ethical considerations associated with CV and outline processes in algorithm development and testing that limit CV expansion in healthcare. This comprehensive review highlights CV applications and ideas for its expanded use in healthcare.

There is a lack of diagnostic performance measures associated with pulmonary embolism (PE). We aimed to explore the concept of the time to diagnostic certainty, which we defined as the time interval that elapses between first presentation of a patient to a confirmed PE diagnosis with computed tomography pulmonary angiogram (CT PA). This approach could be used to highlight variability in health system diagnostic performance, and to select patient outliers for structured chart review in order to identify underlying contributors to diagnostic error or delay. We performed a retrospective observational study at academic medical centers and associated community-based hospitals in one health system, examining randomly selected adult patients admitted to study sites with a diagnosis of acute saddle PE. One hundred patients were randomly selected from 340 patients discharged with saddle PE. Twenty-four patients were excluded. Among the 76 included patients, time to diagnostic certainty ranged from 1.5 to 310 hours. We found that 73/76 patients were considered to have PE present on admission (CT PA ≤ 48 hours). The proportion of patients with PE present on admission with time to diagnostic certainty of > 6 hours was 26% (19/73). The median (IQR) time to treatment (thrombolytics/anticoagulants) was 3.5 (2.5-5.1) hours among the 73 patients. The proportion of patients with PE present on admission with treatment delays of > 6 hours was 16% (12/73). Three patients acquired PE during hospitalization (CT PA > 48 hours). In this study, we developed and successfully tested the concept of time to diagnostic certainty for saddle PE.

OBJECTIVES:. To investigate whether a novel acute care multipatient viewer (AMP), created with an understanding of clinician information and process requirements, could reduce time to clinical decision-making among clinicians caring for populations of acutely ill patients compared with a widely used commercial electronic medical record (EMR). DESIGN:. Single center randomized crossover study. SETTING:. Quaternary care academic hospital. SUBJECTS:. Attending and in-training critical care physicians, and advanced practice providers. INTERVENTIONS:. AMP. MEASUREMENTS AND MAIN RESULTS:. We compared ICU clinician performance in structured clinical task completion using two electronic environments—the standard commercial EMR (Epic) versus the novel AMP in addition to Epic. Twenty subjects (10 pairs of clinicians) participated in the study. During the study session, each participant completed the tasks on two ICUs (7–10 beds each) and eight individual patients. The adjusted time for assessment of the entire ICU and the adjusted total time to task completion were significantly lower using AMP versus standard commercial EMR (–6.11; 95% CI, –7.91 to –4.30 min and –5.38; 95% CI, –7.56 to –3.20 min, respectively; p < 0.001). The adjusted time for assessment of individual patients was similar using both the EMR and AMP (0.73; 95% CI, –0.09 to 1.54 min; p = 0.078). AMP was associated with a significantly lower adjusted task load (National Aeronautics and Space Administration-Task Load Index) among clinicians performing the task versus the standard EMR (22.6; 95% CI, –32.7 to –12.4 points; p < 0.001). There was no statistically significant difference in adjusted total errors when comparing the two environments (0.68; 95% CI, 0.36–1.30; p = 0.078). CONCLUSIONS:. When compared with the standard EMR, AMP significantly reduced time to assessment of an entire ICU, total time to clinical task completion, and clinician task load. Additional research is needed to assess the clinicians’ performance while using AMP in the live ICU setting.

Pulmonary auscultation is essential for detecting abnormal lung sounds during physical assessments, but its reliability depends on the operator. Machine learning (ML) models offer an alternative by automatically classifying lung sounds. ML models require substantial data, and public databases aim to address this limitation. This systematic review compares characteristics, diagnostic accuracy, concerns, and data sources of existing models in the literature. Papers published from five major databases between 1990 and 2022 were assessed. Quality assessment was accomplished with a modified QUADAS-2 tool. The review encompassed 62 studies utilizing ML models and public-access databases for lung sound classification. Artificial neural networks (ANN) and support vector machines (SVM) were frequently employed in the ML classifiers. The accuracy ranged from 49.43% to 100% for discriminating abnormal sound types and 69.40% to 99.62% for disease class classification. Seventeen public databases were identified, with the ICBHI 2017 database being the most used (66%). The majority of studies exhibited a high risk of bias and concerns related to patient selection and reference standards. Summarizing, ML models can effectively classify abnormal lung sounds using publicly available data sources. Nevertheless, inconsistent reporting and methodologies pose limitations to advancing the field, and therefore, public databases should adhere to standardized recording and labeling procedures.

Abstract Background ICU operational conditions may contribute to cognitive overload and negatively impact on clinical decision making. We aimed to develop a quantitative model to investigate the association between the operational conditions and the quantity of medication orders as a measurable indicator of the multidisciplinary care team’s cognitive capacity. Methods The temporal data of patients at one medical ICU (MICU) of Mayo Clinic in Rochester, MN between February 2016 to March 2018 was used. This dataset includes a total of 4822 unique patients admitted to the MICU and a total of 6240 MICU admissions. Guided by the Systems Engineering Initiative for Patient Safety model, quantifiable measures attainable from electronic medical records were identified and a conceptual framework of distributed cognition in ICU was developed. Univariate piecewise Poisson regression models were built to investigate the relationship between system-level workload indicators, including patient census and patient characteristics (severity of illness, new admission, and mortality risk) and the quantity of medication orders, as the output of the care team’s decision making. Results Comparing the coefficients of different line segments obtained from the regression models using a generalized F-test, we identified that, when the ICU was more than 50% occupied (patient census > 18), the number of medication orders per patient per hour was significantly reduced (average = 0.74; standard deviation (SD) = 0.56 vs. average = 0.65; SD = 0.48; p

Abstract Background Palliative care is a medical specialty centered on improving the quality of life (QOL) of patients with complex or life-threatening illnesses. The need for palliative care is increasing and with that the rigorous testing of triage tools that can be used quickly and reliably to identify patients that may benefit from palliative care. Methods To that aim, we will conduct a two-armed stepped-wedge cluster randomized trial rolled out to two inpatient hospitals to evaluate whether a machine learning algorithm accurately identifies patients who may benefit from a comprehensive review by a palliative care specialist and decreases time to receiving a palliative care consult in hospital. This is a single-center study which will be conducted from August 2019 to November 2020 at Saint Mary’s Hospital & Methodist Hospital both within Mayo Clinic Rochester in Minnesota. Clusters will be nursing units which will be chosen to be a mix of complex patients from Cardiology, Critical Care, and Oncology and had previously established relationships with palliative medicine. The stepped wedge design will have 12 units allocated to a design matrix of 5 treatment wedges. Each wedge will last 75 days resulting in a study period of 12 months of recruitment unless otherwise specified. Data will be analyzed with Bayesian hierarchical models with credible intervals denoting statistical significance. Discussion This intervention offers a pragmatic approach to delivering specialty palliative care to hospital patients in need using machine learning, thereby leading to high value care and improved outcomes. It is not enough for AI to be utilized by simply publishing research showing predictive performance; clinical trials demonstrating better outcomes are critically needed. Furthermore, the deployment of an AI algorithm is a complex process that requires multiple teams with varying skill sets. To evaluate a deployed AI, a pragmatic clinical trial can accommodate the difficulties of clinical practice while retaining scientific rigor. Trial registration ClinicalTrials.gov NCT03976297 . Registered on 6 June 2019, prior to trial start.

OBJECTIVES:. Administrative and clinical efforts to improve hospital mortality and intensive care utilization commonly focus on patient rescue, where deteriorating patients are systematically identified and intervened upon. Patient rescue is known to depend on hospital context inclusive of technologic environment, structural features, and hospital organizational behavioral features. With widespread adoption of electronic medical records, early warning score (EWS) systems, which assign points to clinical data elements, are increasingly promoted as a tool for timely patient rescue by referencing their prediction of patient deterioration. We describe the extent to which EWS intervention studies describe the hospital environment of the intervention—details that would be critical for hospital leaders attempting to determine the real-world utility of EWSs in their own hospitals. DATA SOURCES:. We searched CINAHL, PubMed, and Scopus databases for English language EWS implementation research published between 2009 and 2021 in adult medical-surgical inpatients. STUDY SELECTION:. Studies including pediatric, obstetric, psychiatric, prehospital, outpatient, step-down, or ICU patients were excluded. DATA EXTRACTION:. Two investigators independently reviewed titles/abstracts for eligibility based on prespecified exclusion criteria. DATA SYNTHESIS:. We identified 1,434 studies for title/abstract screening. In all, 352 studies underwent full-text review and 21 studies were summarized. The 21 studies (18 before-and-after, three randomized trials) detailed 1,107,883 patients across 54 hospitals. Twelve reported the staff composition of an EWS response team. Ten reported the proportion of surgical patients. One reported nursing ratios; none reported intensive care staffing with in-house critical-care physicians. None measured changes in bed utilization or availability. While 16 qualitatively described resources for education/technologic implementation, none estimated costs. None described workforce composition such as team stability or culture of safety in the hospitals. CONCLUSIONS:. Despite hundreds of EWS-related publications, most do not report details of hospital context that would inform decisions about real-world EWS adoption. To make informed decisions about whether EWS implementation improves hospital quality, decision-makers may require alternatives such as peer networks and implementation pilots nested within local health systems.

Objective: To reliably improve diagnostic fidelity and identify delays using a standardized approach applied to the electronic medical records of patients with emerging critical illness. Patients and Methods: This retrospective observational study at Mayo Clinic, Rochester, Minnesota, conducted June 1, 2016, to June 30, 2017, used a standard operating procedure applied to electronic medical records to identify variations in diagnostic fidelity and/or delay in adult patients with a rapid response team evaluation, at risk for critical illness. Multivariate logistic regression analysis identified predictors and compared outcomes for those with and without varying diagnostic fidelity and/or delay. Results: The sample included 130 patients. Median age was 65 years (interquartile range, 56-76 years), and 47.0% (52 of 130) were women. Clinically significant diagnostic error or delay was agreed in 23 (17.7%) patients (κ=0.57; 95% CI, 0.40-0.74). Median age was 65.4 years (interquartile range, 60.3-74.8) and 9 of the 23 (30.1%) were female. Of those with diagnostic error or delay, 60.9% (14 of 23) died in the hospital compared with 19.6% (21 of 107) without; P

Understanding temporal dynamics of COVID-19 symptoms could provide fine-grained resolution to guide clinical decision-making. Here, we use deep neural networks over an institution-wide platform for the augmented curation of clinical notes from 77,167 patients subjected to COVID-19 PCR testing. By contrasting Electronic Health Record (EHR)-derived symptoms of COVID-19-positive (COVIDpos; n = 2,317) versus COVID-19-negative (COVIDneg; n = 74,850) patients for the week preceding the PCR testing date, we identify anosmia/dysgeusia (27.1-fold), fever/chills (2.6-fold), respiratory difficulty (2.2-fold), cough (2.2-fold), myalgia/arthralgia (2-fold), and diarrhea (1.4-fold) as significantly amplified in COVIDpos over COVIDneg patients. The combination of cough and fever/chills has 4.2-fold amplification in COVIDpos patients during the week prior to PCR testing, in addition to anosmia/dysgeusia, constitutes the earliest EHR-derived signature of COVID-19. This study introduces an Augmented Intelligence platform for the real-time synthesis of institutional biomedical knowledge. The platform holds tremendous potential for scaling up curation throughput, thus enabling EHR-powered early disease diagnosis.

Abstract Background Quantitative studies have demonstrated several factors predictive of readmissions to intensive care. Clinical decision tools, derived from these factors have failed to reduce readmission rates. The purpose of this study was to qualitatively explore the experiences and perceptions of physicians and nurses to gain more insight into intensive care readmissions. Methods Semi-structured interviews of intensive care unit (ICU) and general medicine care providers explored work routines, understanding and perceptions of the discharge process, and readmissions to intensive care. Participants included ten providers from the ICU setting, including nurses (n = 5), consultant intensivists (n = 2), critical care fellows (n = 3) and 9 providers from the general medical setting, nurses (n = 4), consulting physicians (n = 2) and senior resident physicians (n = 3). Principles of grounded theory were used to analyze the interview transcripts. Results Nine factors within four broad themes were identified: (1) patient factors – severity-of-illness and undefined goals of care; (2) process factors – communication, transitions of care; (3) provider factors – discharge decision-making, provider experience and comfort level; (4) organizational factors – resource constraints, institutional policies. Conclusions Severe illness predisposes ICU patients to readmission, especially when goals of care were not adequately addressed. Communication, premature discharge, and other factors, mostly unrelated to the patient were also perceived by physicians and nurses to be associated with readmissions to intensive care. Quality improvement efforts that focus on modifying or improving aspects of non-patient factors may improve outcomes for patients at risk of ICU readmission.

Background Electronic Health Record (EHR)-based sepsis alert systems have failed to demonstrate improvements in clinically meaningful endpoints. However, the effect of implementation barriers on the success of new sepsis alert systems is rarely explored. Objective To test the hypothesis time to severe sepsis alert acknowledgement by critical care clinicians in the ICU setting would be reduced using an EHR-based alert acknowledgement system compared to a text paging-based system. Study Design In one arm of this simulation study, real alerts for patients in the medical ICU were delivered to critical care clinicians through the EHR. In the other arm, simulated alerts were delivered through text paging. The primary outcome was time to alert acknowledgement. The secondary outcomes were a structured, mixed quantitative/qualitative survey and informal group interview. Results The alert acknowledgement rate from the severe sepsis alert system was 3% (N = 148) and 51% (N = 156) from simulated severe sepsis alerts through traditional text paging. Time to alert acknowledgement from the severe sepsis alert system was median 274 min (N = 5) and median 2 min (N = 80) from text paging. The response rate from the EHR-based alert system was insufficient to compare primary measures. However, secondary measures revealed important barriers. Conclusion Alert fatigue, interruption, human error, and information overload are barriers to alert and simulation studies in the ICU setting.