Diagnostic suspicion bias and machine learning: Breaking the awareness deadlock for sepsis detection.

Many early warning algorithms are downstream of clinical evaluation and diagnostic testing, which means that they may not be useful when clinicians fail to suspect illness and fail to order appropriate tests. Depending on how such algorithms handle missing data, they could even indicate "low risk" simply because the testing data were never ordered. We considered predictive methodologies to identify sepsis at triage, before diagnostic tests are ordered, in a busy Emergency Department (ED). One algorithm used "bland clinical data" (data available at triage for nearly every patient). The second algorithm added three yes/no questions to be answered after the triage interview. Retrospectively, we studied adult patients from a single ED between 2014-16, separated into training (70%) and testing (30%) cohorts, and a final validation cohort of patients from four EDs between 2016-2018. Sepsis was defined per the Rhee criteria. Investigational predictors were demographics and triage vital signs (downloaded from the hospital EMR); past medical history; and the auxiliary queries (answered by chart reviewers who were blinded to all data except the triage note and initial HPI). We developed L2-regularized logistic regression models using a greedy forward feature selection. There were 1164, 499, and 784 patients in the training, testing, and validation cohorts, respectively. The bland clinical data model yielded ROC AUC's 0.78 (0.76-0.81) and 0.77 (0.73-0.81), for training and testing, respectively, and ranged from 0.74-0.79 in four hospital validation. The second model which included auxiliary queries yielded 0.84 (0.82-0.87) and 0.83 (0.79-0.86), and ranged from 0.78-0.83 in four hospital validation. The first algorithm did not require clinician input but yielded middling performance. The second showed a trend towards superior performance, though required additional user effort. These methods are alternatives to predictive algorithms downstream of clinical evaluation and diagnostic testing. For hospital early warning algorithms, consideration should be given to bias and usability of various methods.

Association between negative-pressure room utilization and hospital-acquired Aspergillus rates in patients with coronavirus disease 2019 (COVID-19) in two academic hospitals.

Hospital-acquired Aspergillus rates among coronavirus disease 2019 (COVID-19) patients were initially higher at a hospital with high negative-pressure room utilization compared to a similar hospital with low utilization but with otherwise identical infection control policies. After the index hospital decreased negative-pressure utilization, hospital-acquired Aspergillus case rates at the 2 hospitals converged.

Validation of a 'Usual Care' Model for Vasopressor Initiation in a Cohort of Emergency Department Patients with Sepsis.

Usual care regarding vasopressor (VP) initiation is ill-defined. We aimed to further validate a quantitative model for usual care in the Emergency Department (ED) regarding the timing of VP initiation in sepsis. We retrospectively studied a cohort of adult critically-ill ED patients who also received antibiotics in the ED. We applied a multivariable model previously developed from another patient cohort which distinguishes between time points at which patients were or were not subsequently started on a continuous VP infusion. The model has six independently significant predictors (respiratory rate, Glasgow Coma Scale score, systolic blood pressure, SpO2, administered intravenous fluids, and elapsed time). The outcome was initiation of VP infusion, either within the ED or within 6 hours after leaving the ED. We applied the model to all time points, beginning when all model input parameters were first available for a given patient, and ending when either VP were first started, or the patient left the ED. Out of 55,963 adult ED patients during the two-year study interval, we identified 1,629 who met our inclusion criteria. The area under the receiver operating characteristic curve was 0.81 for all patients, and 0.72 for the subset with at least one hypotensive blood pressure measurement. At a model threshold with sensitivity and specificity 0.74 and 0.74, respectively, the median advance detection time was 170.5 minutes (IQR 53 - 363).