In recent years, effective monitoring of categorical and count data has increasingly attracted attention of researchers in the area of statistical process control. However, most of the existing research model categorical and count data streams as independent and identically distributed data or serially correlated discrete time stochastic processes. Very limited research has been conducted for monitoring continuous-time stochastic processes (CTSPs). This paper develops a novel statistical monitoring method for CTSPs with a focus on queueing processes. The proposed method is based on detecting a change in the intensity function of such processes, using an approximate likelihood ratio test. The approximation method is both computationally easy for real-time implementation and well-suited for the introduction of penalization methods. Simulation results based on Markovian and non-Markovian queues show that the proposed methods effectively detect temporal changes in the queueing process. A case study focusing on monitoring the bed assignment process of patients visiting an emergency department demonstrates the efficacy of the proposed methods in a healthcare system. Note to practitioners The methods studied in this paper can be used by operations managers in service enterprises, such as healthcare and transportation industries, for monitoring the timeliness of service provided to customers. The proposed method requires arrival and departure timestamps of customers from a queueing system when the system is considered ideal. This data is then used to define a metric for evaluating the queueing system's performance in a real-time manner. The proposed method does not require the arrival rate of the customers to be time-homogeneous. The experimental results show that the method is agnostic to classical Markov process assumptions required in traditional performance modeling methods for queueing systems. The paper provides the details of applying the proposed method to a timeliness-of-care monitoring problem in the emergency department of a large academic medical center. This method is expected to be broadly applicable to other service systems as well.

OBJECTIVES:Electronic health records (EHR)-based triggers (eTriggers) have been used to study diagnostic errors in the emergency department (ED), often with suboptimal performance. Our objective was to investigate incremental value of multi-factor machine learning (ML) approaches to improve eTrigger performance. METHODS:Patients presenting to an academic ED were categorized into trigger-positive and trigger-negative using standard trigger (T) definitions: (T1) ED return visits resulting in admission within 10 days; (T2) care escalation from the inpatient unit to the ICU within 24 hours; and (T3) deaths within 24 hours of admission. We trained and evaluated 6 supervised ML models. RESULTS:A total of 124,053 consecutive encounters (5791 T-positive and 118,262 T-negative) were included. Among the T-positive, 4159 (72%) were associated with T1, 1415 (24%) with T2, and 217 (4%) with T3. The T-based positive predictive values (PPV) were 5.2% for T1, 8.2% for T2, and 6.5% for T3. ML models trained and evaluated on balanced training dataset and imbalanced test set had low classification performances (accuracy: 0.72-0.95; PPV: 0.00-0.16; F1-score: 0.00-0.23). Higher performances were observed in balanced test sets (accuracy: 0.80-0.97; PPV: 0.82-1.00; F1-score: 0.79-0.97). Comparing models trained on clinically annotated data with models trained on T-based labels identified other important factors. CONCLUSIONS:Utilizing machine learning to refine e-triggers slightly improves the identification of diagnostic errors, as evidenced by an increase in PPV values. We identified new potential factors contributing to ED diagnostic errors. These findings open new avenues to construct or modify more accurate e-triggers for diagnostic error identification.

BackgroundEmergency departments (EDs) are high-pressure environments where clinicians diagnose patients under significant constraints, including limited medical histories, severe time pressures, and frequent interruptions. Current ED care practices often inadequately support meaningful patient participation. Most interventions prioritize clinical workflow and health care provider communication, inadvertently overlooking patients’ needs. Additionally, patient-facing technologies in EDs are typically developed without meaningful patient input, leading to solutions that may not effectively address patients’ specific challenges. To enhance both patient-centered care practices and the diagnosis process in EDs, patient involvement in technology design is essential to ensure their needs during emergency care are understood and addressed. ObjectiveThis study aimed to invite ED patients to participatory design sessions, identify their needs during ED visits, and present potential design guidelines for technological interventions to address these needs. MethodsWe conducted 8 design sessions with 36 ED patients and caregivers to validate their needs and identify considerations for designing patient-centered interventions to improve diagnostic safety. We used 10 technological intervention ideas as probes for a needs evaluation of the study participants. Participants discussed the use cases of each intervention idea to assess their needs during the ED care process. We facilitated co-design activities with the participants to improve the technological intervention designs. We audio- and video-recorded the design sessions. We then analyzed session transcripts, field notes, and design sketches. ResultsOn the basis of ED patients’ feedback and evaluation of our intervention designs, we found the 3 most preferred intervention ideas that addressed the common challenges ED patients experience. We also identified 4 themes of ED patients’ needs: a feeling of inclusion in the ED care process, access to sources of medical information to enhance patient comprehension, addressing patient anxiety related to information overload and privacy concerns, and ensuring continuity in care and information. We interpreted these as insights for designing technological interventions for ED patients. Therefore, on the basis of the findings, we present five considerations for designing better patient-centered interventions in the ED care process: technology-based interventions should (1) address patients’ dynamic needs to promote continuity in care; (2) consider the amount and timing of information that patients receive; (3) empower patients to be more active for better patient safety and care quality; (4) optimize human resources, depending on patients’ needs; and (5) be designed with the consideration of patients’ perspectives on implementation. ConclusionsThis study provides unique insights for designing technological interventions to support ED diagnostic processes. By inviting ED patients into the design process, we present unique insights into the diagnostic process and design considerations for designing novel technological interventions to enhance patient safety. International Registered Report Identifier (IRRID)RR2-10.2196/55357

Objective:To report the first steps of a project to automate and optimize scheduling of multidisciplinary consultations for patients with longstanding dizziness utilizing artificial intelligence. Study Design:Retrospective case review. Setting:Quaternary referral center. Methods:A previsit self-report questionnaire was developed to query patients about their complaints of longstanding dizziness. We convened an expert panel of clinicians to review diagnostic outcomes for 98 patients and used a consensus approach to retrospectively determine what would have been the ideal appointments based on the patient's final diagnoses. These results were then compared retrospectively to the actual patient schedules. From these data, a machine learning algorithm was trained and validated to automate the triage process. Results:Compared with the ideal itineraries determined retrospectively with our expert panel, visits scheduled by the triage clinicians showed a mean concordance of 70%, and our machine learning algorithm triage showed a mean concordance of 79%. Conclusion:Manual triage by clinicians for dizzy patients is a time-consuming and costly process. The formulated first-generation automated triage algorithm achieved similar results to clinicians when triaging dizzy patients using data obtained directly from an online previsit questionnaire.

BackgroundVancomycin is a renally eliminated, nephrotoxic, glycopeptide antibiotic with a narrow therapeutic window, widely used in intensive care units (ICU). We aimed to predict the risk of inappropriate vancomycin trough levels and appropriate dosing for each ICU patient.MethodsObserved vancomycin trough levels were categorized into sub-therapeutic, therapeutic, and supra-therapeutic levels to train and compare different classification models. We included adult ICU patients (≥ 18 years) with at least one vancomycin concentration measurement during hospitalization at Mayo Clinic, Rochester, MN, from January 2007 to December 2017.ResultThe final cohort consisted of 5337 vancomycin courses. The XGBoost models outperformed other machine learning models with the AUC-ROC of 0.85 and 0.83, specificity of 53% and 47%, and sensitivity of 94% and 94% for sub- and supra-therapeutic categories, respectively. Kinetic estimated glomerular filtration rate and other creatinine-based measurements, vancomycin regimen (dose and interval), comorbidities, body mass index, age, sex, and blood pressure were among the most important variables in the models.ConclusionWe developed models to assess the risk of sub- and supra-therapeutic vancomycin trough levels to improve the accuracy of drug dosing in critically ill patients.

Objective: To develop machine learning tools for automated hypertrophic cardiomyopathy (HCM) case recognition from echocardiographic metrics, aiming to identify HCM from standard echocardiographic data with high performance. Patients and Methods: Four different random forest machine learning models were developed using a case-control cohort composed of 5548 patients with HCM and 16,973 controls without HCM, from January 1, 2004, to March 15, 2019. Each patient with HCM was matched to 3 controls by sex, age, and year of echocardiography. Ten-fold crossvalidation was used to train the models to identify HCM. Variables included in the models were demographic characteristics (age, sex, and body surface area) and 16 standard echocardiographic metrics. Results: The models were differentiated by global, average, individual, or no strain measurements. Area under the receiver operating characteristic curves (area under the curve) ranged from 0.92 to 0.98 for the 4 separate models. Area under the curves of model 2 (using left ventricular global longitudinal strain; 0.97; 95% CI, 0.95-0.98), 3 (using averaged strain; 0.96; 95% CI, 0.94-0.97), and 4 (using 17 individual strains per patient; 0.98; 95% CI, 0.97-0.99) had comparable performance. By comparison, model 1 (no strain data; 0.92; 95% CI, 0.90-0.94) had an inferior area under the curve. Conclusion: Machine learning tools that analyze echocardiographic metrics identified HCM cases with high performance. Detection of HCM cases improved when strain data was combined with standard echocardiographic metrics.

Background Emergency departments (EDs) are complex and fast-paced clinical settings where a diagnosis is made in a time-, information-, and resource-constrained context. Thus, it is predisposed to suboptimal diagnostic outcomes, leading to errors and subsequent patient harm. Arriving at a timely and accurate diagnosis is an activity that occurs after an effective collaboration between the patient or caregiver and the clinical team within the ED. Interventions such as novel sociotechnical solutions are needed to mitigate errors and risks. Objective This study aims to identify challenges that frontline ED health care providers and patients face in the ED diagnostic process and involve them in co-designing technological interventions to enhance diagnostic excellence. Methods We will conduct separate sessions with ED health care providers and patients, respectively, to assess various design ideas and use a participatory design (PD) approach for technological interventions to improve ED diagnostic safety. In the sessions, various intervention ideas will be presented to participants through storyboards. Based on a preliminary interview study with ED patients and health care providers, we created intervention storyboards that illustrate different care contexts in which ED health care providers or patients experience challenges and show how each intervention would address the specific challenge. By facilitating participant group discussion, we will reveal the overlap between the needs of the design research team observed during fieldwork and the needs perceived by target users (ie, participants) in their own experience to gain their perspectives and assessment on each idea. After the group discussions, participants will rank the ideas and co-design to improve our interventions. Data sources will include audio and video recordings, design sketches, and ratings of intervention design ideas from PD sessions. The University of Michigan Institutional Review Board approved this study. This foundational work will help identify the needs and challenges of key stakeholders in the ED diagnostic process and develop initial design ideas, specifically focusing on sociotechnological ideas for patient-, health care provider–, and system-level interventions for improving patient safety in EDs. Results The recruitment of participants for ED health care providers and patients is complete. We are currently preparing for PD sessions. The first results from design sessions with health care providers will be reported in fall 2024. Conclusions The study findings will provide unique insights for designing sociotechnological interventions to support ED diagnostic processes. By inviting frontline health care providers and patients into the design process, we anticipate obtaining unique insights into the ED diagnostic process and designing novel sociotechnical interventions to enhance patient safety. Based on this study’s collected data and intervention ideas, we will develop prototypes of multilevel interventions that can be tested and subsequently implemented for patients, health care providers, or hospitals as a system. International Registered Report Identifier (IRRID) DERR1-10.2196/55357

In the Emergency Department (ED), healthcare providers face extraordinary pressures in delivering accurate diagnoses and care, often working with fragmented or inaccessible patient histories while managing severe time constraints and constant interruptions. These challenges and pressures may lead to potential errors in the ED diagnostic process and risks to patient safety. With the advances in technology, technological interventions have been developed to support ED providers in such pressured settings. However, these interventions may not align with the current practices of ED providers. To better design provider-centered interventions, identifying the needs of ED providers in the diagnostic process is critical. This study aims to identify providers’ needs in the ED diagnostic process by inviting them to participatory design sessions and to present potential design guidelines for provider-centered technological interventions that support decision-making and reduce errors. We conducted a participatory design study with ED providers to validate their needs and identify considerations for designing provider-centered interventions to improve diagnostic safety. We used nine technological intervention ideas as storyboards to evaluate the study participants' needs. We had participants discuss the use cases of each intervention idea to assess their needs during the ED care process and facilitated co-design activities with the participants to improve the technological intervention designs. We audio- and video-recorded the design sessions. We then analyzed session transcripts, field notes, and design sketches. In total, we conducted six design sessions with 17 ED frontline providers. Through design sessions with ED providers, we identified four key needs of providers in the diagnostic process: information integration, patient prioritization, provider-patient communication, and care coordination. We interpreted them as insights for designing technological interventions for ED patients. Hence, we discuss the design implications for technological interventions in four key areas: 1) Enhancing provider-provider communication, 2) Enhancing provider-patient communication, 3) Optimizing the integration of advanced technology, and 4) Unleashing the potential of AI tools in the ED to improve diagnosis. This work offers evidence-based technology design suggestions for improving diagnostic processes. This study provides unique insights for designing technological interventions to support ED diagnostic processes. By inviting ED providers into the design process, we present unique insights into the diagnostic process and design considerations for designing novel technological interventions to meet ED providers’ needs in the diagnostic process. RR2-10.2196/55357

BACKGROUND AND OBJECTIVES:Understanding factors affecting the timing of critical clinical events in ALS progression. METHODS:We captured ALS progression based on the timing of critical events (tollgates), by augmenting 6366 patients' data from the PRO-ACT database with tollgate-passed information using classification. Time trajectories of passing ALS tollgates after the first visit were derived using Kaplan-Meier analyses. The significant prognostic factors were found using log-rank tests. Decision-tree-based classifications identified significant ALS phenotypes characterized by the list of body segments involved at the first visit. RESULTS:Standard (e.g., gender and onset type) and tollgate-related (phenotype and initial tollgate level) prognostic factors affect the timing of ALS tollgates. For instance, by the third year after the first visit, 80-100% of bulbar-onset patients vs. 43-48% of limb-onset patients, and 65-73% of females vs. 42-49% of males lost the ability to talk and started using a feeding tube. Compared to the standard factors, tollgate-related factors had a stronger effect on ALS progression. The initial impairment level significantly impacted subsequent ALS progression in a segment while affected segment combinations further characterized progression speed. For instance, patients with normal speech (Tollgate Level 0) at the first visit had less than a 10% likelihood of losing speech within a year, while for patients with Tollgate Level 1 (affected speech), this likelihood varied between 23 and 53% based on additional segment (leg) involvement. CONCLUSIONS:Tollgate- and phenotype-related factors have a strong effect on the timing of ALS tollgates. All factors should be jointly considered to better characterize patient groups with different progression aggressiveness.

BACKGROUND:Few studies have described the insights of frontline health care providers and patients on how the diagnostic process can be improved in the emergency department (ED), a setting at high risk for diagnostic errors. The authors aimed to identify the perspectives of providers and patients on the diagnostic process and identify potential interventions to improve diagnostic safety. METHODS:Semistructured interviews were conducted with 10 ED physicians, 15 ED nurses, and 9 patients/caregivers at two separate health systems. Interview questions were guided by the ED-Adapted National Academies of Sciences, Engineering, and Medicine Diagnostic Process Framework and explored participant perspectives on the ED diagnostic process, identified vulnerabilities, and solicited interventions to improve diagnostic safety. The authors performed qualitative thematic analysis on transcribed interviews. RESULTS:The research team categorized vulnerabilities in the diagnostic process and intervention opportunities based on the ED-Adapted Framework into five domains: (1) team dynamics and communication (for example, suboptimal communication between referring physicians and the ED team); (2) information gathering related to patient presentation (for example, obtaining the history from the patients or their caregivers; (3) ED organization, system, and processes (for example, staff schedules and handoffs); (4) patient education and self-management (for example, patient education at discharge from the ED); and (5) electronic health record and patient portal use (for example, automatic release of test results into the patient portal). The authors identified 33 potential interventions, of which 17 were provider focused and 16 were patient focused. CONCLUSION:Frontline providers and patients identified several vulnerabilities and potential interventions to improve ED diagnostic safety. Refining, implementing, and evaluating the efficacy of these interventions are required.

Queuing networks (QNs) are widely used stochastic models for service systems include healthcare systems, transportation systems, and computer networks. While existing literature has extensively focused on modeling and optimizing resource allocation in QNs, very little research has been done on developing systematic statistical monitoring methods for QNs. This paper proposes cumulative sum (CUSUM) control charts that monitor the queuing information collected in real-time from the QN. We compare the proposed methods with existing statistical monitoring methods to demonstrate their ability to quickly detect a change in the service rate of one or more queues at the nodes in the QN. Simulation results show that the proposed CUSUM charts are more effective than existing statistical monitoring methods. The motivation for this research comes from the need to monitor the performance of a hospital emergency department (ED) with the goal of monitoring delays experienced by patients visiting the ED. A case study using the data from the ED of a large academic medical center shows that proposed methods are a promising tool for monitoring the timeliness of care provided to patients visiting the ED.

PurposePatient length of stay (LOS) is an important indicator of emergency department (ED) performance. Investigating factors that influence LOS could thus improve healthcare delivery and patient safety. Previous studies have focused on patient-level factors to explain LOS variation, with little research into service-related factors. This study examined the association between LOS and multi-level factors including patient-, service- and organization-level factors.Design/methodology/approachThis study uses a retrospective observational design to identify a cohort of patients from arrival to discharge from ED. A year-long data regarding patients flow trhoguh ED were analyzed using analytics techniques and multi-regression models. The response variable was patient LOS, and the independent variables were patient characteristics, service-related factors and organizational variables.FindingsThe findings of this study showed that older patients, middle triage and hospitalization were all associated with longer LOS. Service-related factors such as complexity of care provided, initial ward designation and ward transfer had a significant impact as well. Finally, prolonged LOS was associated with a higher ratio of patients per medical doctor and per nurse. In contrast, a higher number of residents in the ED were associated with longer patient LOS.Originality/valuePrevious studies on patient LOS have focused on patient-level factors, with little research on service-related factors. This study has addressed that gap by examining the association between LOS and multi-level factors including patient-, service- and organization-level factors. Patient-level factors included demographics, acuity, arrival shift, arrival mode and discharge type. Service-level factors consisted of first ward, ward transfer and complexity of care provided. Organizational factors consisted of three ratios: patients per MD, patients per nurse and patients per resident. The results add to the current understanding of factors that increase patient LOS in EDs and contribute to the body of knowledge on ED performance, operation management and quality of care. The study also provides practical and managerial insights that could be used to improve patient flow in EDs and reduce LOS.

Diagnostic decision-making in the emergency department (ED) is a complex cognitive process involving high uncertainty, making it susceptible to diagnostic errors. The use of data-centric approaches can aid with the identification of factors contributing to diagnostic errors. One of these approaches is Classification and Regression Tree (CART). Our objective in this project is to apply previously validated diagnostic error triggers to patient ED encounters and use machine learning to compare trigger-positive and trigger-negative cases and help identify factors that influence diagnostic safety. We evaluated a cohort of ED visits (all ages, 2017-2019) to identify factors contributing to ED diagnostic error in two phases. In phase one we used the SAS HPSPLIT procedure (CART) to identify important features related to a visit being trigger-positive or trigger-negative. The model was initially based on all data to identify important features using a 10-fold cross-validation method. In the second phase we built a model using the features from phase one and applied the trained model to the manually reviewed cases. We predicted being trigger-positive or trigger-negative and compared them against cases with confirmed errors (Error Yes/No) to compute diagnostic test accuracy and positive and negative predictive values (PPV, NPV). There were 125,342 unique ED encounters including 119,456 trigger-negative and 5,886 trigger-positive. A total of 720 events were manually reviewed (291 trigger-negative, 429 trigger-positive), and 32 of these cases had an identified diagnostic error. The overall rate of diagnostic error for these reviewed cases was 4.4% (95% CI 3.2 to 6.2%). After performing the first phase, an accuracy of 0.953, an F1 score of 0.036, and an AUC of 0.677 was achieved. The second phase resulted in an accuracy of 0.672, F1 score of 0.737, and AUC of 0.713. After the second phase, our PPV was 5.1% (CI 3.2 to 7.7%), NPV was 96.6% (CI 93.8 to 98.3%), sensitivity was 68.8%, and specificity was 40.8%. Top predictive factors are presented in Figure. Our proposed classifier based on phase two had an accuracy of 42% in separating the error-positive and error-negative cases. It was successful in highlighting predictive performance of multiple factors including ICD codes, chief complaints, age, and number of labs and lab panels completed in the ED. Diagnostic errors were uncommon, and application of predicted features did not improve the PPV. Transparency in the reporting of the methods is key for future implementation of science and dissemination of findings. Future work would be to evaluate the performance of this model on a larger set of clinically annotated data.

For a large portion of mental health patients, the Emergency Department is the first point of contact when in crisis and in need of urgent acute care. Unfortunately, those who have already received an admission disposition may wait hours or days after before being placed in a psychiatric inpatient (IP) care unit. Known as ED boarding, one primary contributor is the inability to locate an available IP bed for transferred patients. In this study, we develop a discrete event simulation modeling patients arriving at numerous EDs throughout the region, as they're either internally placed in a psychiatric IP care unit or externally transferred to an IP unit located outside the ED they originally arrived. This simulation is then used to investigate the effect of three proposed interventions to the IP bed placement process on key performance indicators like patient treatment delay, a metric incorporating both the patient's ED boarding period and time to travel to their eventual IP destination.

Screening for diagnostic errors in the emergency department (ED) includes the use of electronic health record (EHR) triggers to identify patients with certain patterns of care, such as escalation of care or return ED visits. Once errors are identified, data analytics and machine learning techniques can be applied to evaluate factors associated with trigger positive and negative cases and evaluate the accuracy with confirmed diagnostic errors. Association rule mining (ARM) is a data mining technique that uses prior knowledge of frequent item set and aims to extract frequent item set, meaningful correlation, or causal structure within data. Our objectives were to extract rules that govern the relationship between the patient and systems factors and risk of being trigger-positive. A total of 119,456 trigger-negative (T-neg) cases were matched against 6,127 trigger-positive (Tpos) cases, resulting in a total of 12,254 observations. A small subset of this data was reviewed for existence of diagnostic errors (Yes: 33, No: 682). Patient demographic information (eg age, gender) and ED encounter parameters (eg time, Chief complaints, labs, diagnosis codes) were extracted from the EHR for each ED encounter. All data was passed through a series of preprocessing steps, including discretization of continuous variables, converting string values into categorical, and summarizing categorical data into similar groups, and finally replacing all null values with a new "unknown" categorical value. After preprocessing, we performed two experiments. In the first experiment we used the matched dataset to train an Apriori algorithm (arules package in R) to identify the governing rules that separate T-pos from T-neg labels. The trained rules (model) were then tested on the reviewed dataset to investigate the performance of predictions against the annotated error labels. In the second experiment, we ran the Apriori algorithm on the reviewed cases and set the target prediction to be the "Error" column. We then explored how well the top extracted features were in predicting the right error label on the same dataset. In the first experiment, the extracted rules were able to predict the T-neg and T-pos labels with a high predictive performance (PPV) (PPV: 0.82, Acc: 0.45). The same model performed poorly with respect to the Error label (PPV: 0.05, Acc: 0.72). The model in the second experiment was much more accurate (Acc: 1, PPV: 1) using all extracted rules suggesting overfitting. We then enforced the number of rules to only the top 50 rules, to investigate the amount of information carried by these selected rules, where there was a significant drop in the performance (Acc: 0.91, PPV: 0.32). We investigated the use of ARM techniques on a trigger-labeled dataset and its potential application in extraction of a small subset of governing rules that can be used in identification of diagnostic errors in the ED. The fact that our model (extracted rules) had high PPV of 0.82 in separating T-pos from Tneg but had a very low PPV of 0.05 with respect to Error indicates potential challenges in the definition of Trigger labels. The hope was to increase the prediction performance, so more reliable sets of rules can be identified. Our experiments also highlight the effect of having a very limited number of confirmed Error-positive cases in limiting our learning process and propose further in-depth analyses.

Purpose: We developed and validated two parsimonious algorithms to predict the time of diagnosis of any stage of acute kidney injury (any-AKI) or moderate-to-severe AKI in clinically actionable prediction windows.Materials and methods: In this retrospective single-center cohort of adult ICU admissions, we trained two gradient-boosting models: 1) any-AKI model, predicting the risk of any-AKI at least 6 h before diagnosis (50,342 ad-missions), and 2) moderate-to-severe AKI model, predicting the risk of moderate-to-severe AKI at least 12 h before diagnosis (39,087 admissions). Performance was assessed before disease diagnosis and validated prospectively.Results: The models achieved an area under the receiver operating characteristic curve (AUROC) of 0.756 at six hours (any-AKI) and 0.721 at 12 h (moderate-to-severe AKI) prior. Prospectively, both models had high positive predictive values (0.796 and 0.546 for any-AKI and moderate-to-severe AKI models, respectively) and triggered more in patients who developed AKI vs. those who did not (median of 1.82 [IQR 0-4.71] vs. 0 [IQR 0-0.73] and 2.35 [IQR 0.14-4.96] vs. 0 [IQR 0-0.8] triggers per 8 h for any-AKI and moderate-to-severe AKI models, respectively).Conclusions: The two AKI prediction models have good discriminative performance using common features, which can aid in accurately and informatively monitoring AKI risk in ICU patients.

BACKGROUNDWe aimed to determine if patient symptoms and computed tomography enterography (CTE) and magnetic resonance enterography (MRE) imaging findings can be used to predict near-term risk of surgery in patients with small bowel Crohn's disease (CD).METHODSCD patients with small bowel strictures undergoing serial CTE or MRE were retrospectively identified. Strictures were defined by luminal narrowing, bowel wall thickening, and unequivocal proximal small bowel dilation. Harvey-Bradshaw index (HBI) was recorded. Stricture observations and measurements were performed on baseline CTE or MRE and compared to with prior and subsequent scans. Patients were divided into those who underwent surgery within 2 years and those who did not. LASSO (least absolute shrinkage and selection operator) regression models were trained and validated using 5-fold cross-validation.RESULTSEighty-five patients (43.7 ± 15.3 years of age at baseline scan, majority male [57.6%]) had 137 small bowel strictures. Surgery was performed in 26 patients within 2 years from baseline CTE or MRE. In univariate analysis of patients with prior exams, development of stricture on the baseline exam was associated with near-term surgery (P = .006). A mathematical model using baseline features predicting surgery within 2 years included an HBI of 5 to 7 (odds ratio [OR], 1.7 × 105; P = .057), an HBI of 8 to 16 (OR, 3.1 × 105; P = .054), anastomotic stricture (OR, 0.002; P = .091), bowel wall thickness (OR, 4.7; P = .064), penetrating behavior (OR, 3.1 × 103; P = .096), and newly developed stricture (OR: 7.2 × 107; P = .062). This model demonstrated sensitivity of 67% and specificity of 73% (area under the curve, 0.62).CONCLUSIONSCTE or MRE imaging findings in combination with HBI can potentially predict which patients will require surgery within 2 years.

Recent use of noninvasive and continuous hemoglobin (SpHb) concentration monitor has emerged as an alternative to invasive laboratory-based hematological analysis. Unlike delayed laboratory based measures of hemoglobin (HgB), SpHb monitors can provide real-time information about the HgB levels. Real-time SpHb measurements will offer healthcare providers with warnings and early detections of abnormal health status, e.g., hemorrhagic shock, anemia, and thus support therapeutic decision-making, as well as help save lives. However, the finger-worn CO-Oximeter sensors used in SpHb monitors often get detached or have to be removed, which causes missing data in the continuous SpHb measurements. Missing data among SpHb measurements reduce the trust in the accuracy of the device, influence the effectiveness of hemorrhage interventions and future HgB predictions. A model with imputation and prediction method is investigated to deal with missing values and improve prediction accuracy. The Gaussian process and functional regression methods are proposed to impute missing SpHb data and make predictions on laboratory-based HgB measurements. Within the proposed method, multiple choices of sub-models are considered. The proposed method shows a significant improvement in accuracy based on a real-data study. Proposed method shows superior performance with the real data, within the proposed framework, different choices of sub-models are discussed and the usage recommendation is provided accordingly. The modeling framework can be extended to other application scenarios with missing values.