Clinician Perception of a Machine Learning–Based Early Warning System Designed to Predict Severe Sepsis and Septic Shock* Objective: To assess clinician perceptions of a machine learning–based early warning system to predict severe sepsis and septic shock (Early Warning System 2.0). Design: Prospective observational study. Setting: Tertiary teaching hospital in Philadelphia, PA. Patients: Non-ICU admissions November–December 2016. Interventions: During a 6-week study period conducted 5 months after Early Warning System 2.0 alert implementation, nurses and providers were surveyed twice about their perceptions of the alert’s helpfulness and impact on care, first within 6 hours of the alert, and again 48 hours after the alert. Measurements and Main Results: For the 362 alerts triggered, 180 nurses (50% response rate) and 107 providers (30% response rate) completed the first survey. Of these, 43 nurses (24% response rate) and 44 providers (41% response rate) completed the second survey. Few (24% nurses, 13% providers) identified new clinical findings after responding to the alert. Perceptions of the presence of sepsis at the time of alert were discrepant between nurses (13%) and providers (40%). The majority of clinicians reported no change in perception of the patient’s risk for sepsis (55% nurses, 62% providers). A third of nurses (30%) but few providers (9%) reported the alert changed management. Almost half of nurses (42%) but less than a fifth of providers (16%) found the alert helpful at 6 hours. Conclusions: In general, clinical perceptions of Early Warning System 2.0 were poor. Nurses and providers differed in their perceptions of sepsis and alert benefits. These findings highlight the challenges of achieving acceptance of predictive and machine learning–based sepsis alerts.

A Machine Learning Algorithm to Predict Severe Sepsis and Septic Shock: Development, Implementation, and Impact on Clinical Practice* Objectives: Develop and implement a machine learning algorithm to predict severe sepsis and septic shock and evaluate the impact on clinical practice and patient outcomes. Design: Retrospective cohort for algorithm derivation and validation, pre-post impact evaluation. Setting: Tertiary teaching hospital system in Philadelphia, PA. Patients: All non-ICU admissions; algorithm derivation July 2011 to June 2014 (n = 162,212); algorithm validation October to December 2015 (n = 10,448); silent versus alert comparison January 2016 to February 2017 (silent n = 22,280; alert n = 32,184). Interventions: A random-forest classifier, derived and validated using electronic health record data, was deployed both silently and later with an alert to notify clinical teams of sepsis prediction. Measurement and Main Result: Patients identified for training the algorithm were required to have International Classification of Diseases, 9th Edition codes for severe sepsis or septic shock and a positive blood culture during their hospital encounter with either a lactate greater than 2.2 mmol/L or a systolic blood pressure less than 90 mm Hg. The algorithm demonstrated a sensitivity of 26% and specificity of 98%, with a positive predictive value of 29% and positive likelihood ratio of 13. The alert resulted in a small statistically significant increase in lactate testing and IV fluid administration. There was no significant difference in mortality, discharge disposition, or transfer to ICU, although there was a reduction in time-to-ICU transfer. Conclusions: Our machine learning algorithm can predict, with low sensitivity but high specificity, the impending occurrence of severe sepsis and septic shock. Algorithm-generated predictive alerts modestly impacted clinical measures. Next steps include describing clinical perception of this tool and optimizing algorithm design and delivery.

National Trends in Timing of Death Among Patients With Septic Shock, 1994–2014 Objectives: To assess trends in timing of mortality among patients with septic shock. Design: Retrospective cohort study. Setting: Agency for Healthcare Research and Quality’s Healthcare Cost and Utilization Project’s National Inpatient Sample, 1994–2014. Patients: Hospitalized adults (≥ 18 yr) with International Classification of Diseases, 9th Edition, Clinical Modification codes consistent with septic shock; secondary analysis: adults with International Classification of Diseases, 9th Edition, Clinical Modification codes consistent with acute respiratory failure receiving invasive mechanical ventilation and patients with both septic shock and acute respiratory failure receiving invasive mechanical ventilation. Interventions: None. Measurements and Main Results: From 1994 to 2014, 48-hour mortality rates decreased among patients with septic shock (21.2% to 10.8%) and septic shock with acute respiratory failure receiving invasive mechanical ventilation (19.1% to 13.4%) but increased among patients with acute respiratory failure receiving invasive mechanical ventilation (7.9% to 9.8%; p value for all trends, < 0.001). Three-to-14-day mortality decreased among patients with septic shock (22.1% to 15.5%), septic shock with acute respiratory failure receiving invasive mechanical ventilation (28.7% to 22.4%) and acute respiratory failure receiving invasive mechanical ventilation (16.8% to 15.0%; p value for all trends, < 0.001). Mortality after 14 days decreased among all groups (septic shock: 12.6% to 6.7%; septic shock with acute respiratory failure receiving invasive mechanical ventilation: 20.3% to 11.3%; and acute respiratory failure receiving invasive mechanical ventilation: 12.7% to 5.8%; p value for all trends, < 0.001). Cox proportional hazard ratio for declining risk in mortality per year (adjusted for patient and hospital characteristics) was 0.96 (95% CI, 0.96–0.96) for septic shock, 0.97 (0.97–0.97) for acute respiratory failure receiving invasive mechanical ventilation and septic shock, and 0.99 (0.99–0.99) for acute respiratory failure receiving invasive mechanical ventilation. Conclusions: Septic shock 48-hour, 3–14-day and greater than 14-day mortality declined markedly over two decades; in contrast, patients with acute respiratory failure only experienced marked decreases in greater than 14-day in-hospital mortality rates.

Impact of Critical Illness on Resource Utilization: A Comparison of Use in the Year Before and After ICU Admission* Objectives: Increasingly, patients admitted to an ICU survive to hospital discharge; many with ongoing medical needs. The full impact of an ICU admission on an individual’s resource utilization and survivorship trajectory in the United States is not clear. We sought to compare healthcare utilization among ICU survivors in each year surrounding an ICU admission. Design: Retrospective cohort of patients admitted to an ICU during one calendar year (2012) in a multipayer healthcare system. We assessed mortality, hospital readmissions (categorized by ambulatory care sensitive conditions and emergency department), and outpatient visits. We compared the proportion of patients with visits during the pre-ICU year versus the post-ICU year. Patients: People admitted to an Intermountain healthcare ICU for greater than 48 hours in the year 2012 Interventions: None. Measurements and Main Results: Among 4,074 ICU survivors, 45% had increased resource utilization. Readmission rates at 30-day, 90-day, and 1-year were 15%, 26%, and 43%. The proportion of patients with a hospital admission increased significantly in the post-ICU period (43% vs 29%; p < 0.001). Of patients with a readmission in the post-ICU period, 24% were ambulatory care sensitive condition. Patients with increased utilization differed by socioeconomic status, insurance type, and severity of illness. Sixteen percent of patients had either an emergency department or inpatient admission, but no outpatient visits during the post-ICU period. Conclusions: An ICU admission is associated with increased resource utilization including hospital readmissions, with many due to an ambulatory care sensitive condition. Lower socioeconomic status and higher severity of illness are associated with increased resource utilization. After an ICU visit patients seem to use hospital resources over outpatient resources. Interventions to improve and coordinate care after ICU discharge are needed.

Ultra-Protective Ventilation Reduces Biotrauma in Patients on Venovenous Extracorporeal Membrane Oxygenation for Severe Acute Respiratory Distress Syndrome* Introduction: Ventilator settings for patients with severe acute respiratory distress syndrome supported by venovenous extracorporeal membrane oxygenation are currently set arbitrarily. The impact on serum and pulmonary biotrauma markers of the transition to ultra-protective ventilation settings following extracorporeal membrane oxygenation implantation, and different mechanical ventilation strategies while on extracorporeal membrane oxygenation were investigated. Design: Randomized clinical trial. Settings: Nine-month monocentric study. Patients: Severe acute respiratory distress syndrome patients on venovenous extracorporeal membrane oxygenation. Interventions: After starting extracorporeal membrane oxygenation, patients were switched to the bi-level positive airway pressure mode with 1 second of 24 cm H2O high pressure and 2 seconds of 12 cm H2O low pressure for 24 hours. A computer-generated allocation sequence randomized patients to receive each of the following three experimental steps: 1) high pressure 24 cm H2O and low pressure 20 cm H2O (very high positive end-expiratory pressure–very low driving pressure); 2) high pressure 24 cm H2O and low pressure 5 cm H2O (low positive end-expiratory pressure–high driving pressure); and 3) high pressure 17 cm H2O and low pressure 5 cm H2O (low positive end-expiratory pressure–low driving pressure). Plasma and bronchoalveolar lavage soluble receptor for advanced glycation end-products, plasma interleukin-6, and monocyte chemotactic protein-1 were sampled preextracorporeal membrane oxygenation and after 12 hours at each step. Measurements and Main Results: Sixteen patients on ECMO after 7 days (1–11 d) of mechanical ventilation were included. “Ultra-protective” mechanical ventilation settings following ECMO initiation were associated with significantly lower plasma sRAGE, interleukin-6, and monocyte chemotactic protein-1 concentrations. Plasma sRAGE and cytokines were comparable within each on-ECMO experimental step, but the lowest bronchoalveolar lavage sRAGE levels were obtained at minimal driving pressure. Conclusions: ECMO allows ultra- protective ventilation, which combines significantly lower plateau pressure, tidalvolume, and driving pressure. This ventilation strategy significantly limited pulmonary biotrauma, which couldtherefore decrease ventilator-induced lung injury. However, the optimal ultra-protective ventilation strategy once ECMO is initiated remains undetermined and warrants further investigations. (Crit Care Med 2019; 47:1505–1512)

Physician Judgment and Circulating Biomarkers Predict 28-Day Mortality in Emergency Department Patients* Objectives: To determine whether biomarkers of endothelial activation and inflammation provide added value for prediction of in-hospital mortality within 28 days when combined with physician judgment in critically ill emergency department patients. Design: Prospective, observational study. Setting: Two urban, academic emergency departments, with ≈80,000 combined annual visits, between June 2016 and December 2017. Patients: Admitted patients, greater than 17 years old, with two systemic inflammatory response syndrome criteria and organ dysfunction, systolic blood pressure less than 90 mm Hg, or lactate greater than 4.0 mmol/L. Patients with trauma, intracranial hemorrhage known prior to arrival, or without available blood samples were excluded. Interventions: Emergency department physicians reported likelihood of in-hospital mortality (0–100%) by survey at hospital admission. Remnant EDTA blood samples, drawn during the emergency department stay, were used to measure angiopoietin-1, angiopoietin-2, tumor necrosis factor receptor-1, interleukin-6, and interleukin-8. Measurements and Main Results: We screened 421 patients and enrolled 314. The primary outcome of in-hospital mortality within 28 days occurred in 31 (9.9%). When predicting the primary outcome, the best biomarker model included angiopoietin-2 and interleukin-6 and performed moderately well (area under the curve, 0.72; 95% CI, 0.69–0.75), as did physician judgment (area under the curve, 0.78; 95% CI, 0.74–0.82). Combining physician judgment and biomarker models improved performance (area under the curve, 0.85; 95% CI, 0.82–0.87), with area under the curve change of 0.06 (95% CI, 0.04–0.09; p < 0.01) compared with physician judgment alone. Conclusions: Predicting in-hospital mortality within 28 days among critically ill emergency department patients may be improved by including biomarkers of endothelial activation and inflammation in combination with emergency department physician judgment.

U.K. Intensivists’ Preferences for Patient Admission to ICU: Evidence From a Choice Experiment Objectives: Deciding whether to admit a patient to the ICU requires considering several clinical and nonclinical factors. Studies have investigated factors associated with the decision but have not explored the relative importance of different factors, nor the interaction between factors on decision-making. We examined how ICU consultants prioritize specific factors when deciding whether to admit a patient to ICU. Design: Informed by a literature review and data from observation and interviews with ICU clinicians, we designed a choice experiment. Senior intensive care doctors (consultants) were presented with pairs of patient profiles and asked to prioritize one of the patients in each task for admission to ICU. A multinomial logit and a latent class logit model was used for the data analyses. Setting: Online survey across U.K. intensive care. Subjects: Intensive care consultants working in NHS hospitals. Measurements and Main Results: Of the factors investigated, patient’s age had the largest impact at admission followed by the views of their family, and severity of their main comorbidity. Physiologic measures indicating severity of illness had less impact than the gestalt assessment by the ICU registrar. We identified four distinct decision-making patterns, defined by the relative importance given to different factors. Conclusions: ICU consultants vary in the importance they give to different factors in deciding who to prioritize for ICU admission. Transparency regarding which factors have been considered in the decision-making process could reduce variability and potential inequity for patients.

Association Between Sepsis and Microvascular Brain Injury* Objectives: Many survivors of sepsis suffer long-term cognitive impairment, but the mechanisms of this association remain unknown. The objective of this study was to determine whether sepsis is associated with cerebral microinfarcts on brain autopsy. Design: Retrospective cohort study. Setting and Subjects: Five-hundred twenty-nine participants of the Adult Changes in Thought, a population-based prospective cohort study of older adults carried out in Kaiser Permanente Washington greater than or equal to 65 years old without dementia at study entry and who underwent brain autopsy. Measurements and Main Results: Late-life sepsis hospitalization was identified using administrative data. We identified 89 individuals with greater than or equal to 1 sepsis hospitalization during study participation, 80 of whom survived hospitalization and died a median of 169 days after discharge. Thirty percent of participants with one or more sepsis hospitalization had greater than two microinfarcts, compared with 19% participants without (χ2p = 0.02); 20% of those with sepsis hospitalization had greater than two microinfarcts in the cerebral cortex, compared with 10% of those without (χ2p = 0.01). The adjusted relative risk of greater than two microinfarcts was 1.61 (95% CI, 1.01–2.57; p = 0.04); the relative risk for having greater than two microinfarcts in the cerebral cortex was 2.12 (95% CI, 1.12–4.02; p = 0.02). There was no difference in Braak stage for neurofibrillary tangles or consortium to establish a registry for Alzheimer’s disease score for neuritic plaques between, but Lewy bodies were less significantly common in those with sepsis. Conclusions: Sepsis was specifically associated with moderate to severe vascular brain injury as assessed by microvascular infarcts. This association was stronger for microinfarcts within the cerebral cortex, with those who experienced severe sepsis hospitalization being more than twice as likely to have evidence of moderate to severe cerebral cortical injury in adjusted analyses. Further study to identify mechanisms for the association of sepsis and microinfarcts is needed.

The ED-SED Study: A Multicenter, Prospective Cohort Study of Practice Patterns and Clinical Outcomes Associated With Emergency Department SEDation for Mechanically Ventilated Patients Objectives: To characterize emergency department sedation practices in mechanically ventilated patients, and test the hypothesis that deep sedation in the emergency department is associated with worse outcomes. Design: Multicenter, prospective cohort study. Setting: The emergency department and ICUs of 15 medical centers. Patients: Mechanically ventilated adult emergency department patients. Interventions: None. Measurements and Main Results: All data involving sedation (medications, monitoring) were recorded. Deep sedation was defined as Richmond Agitation-Sedation Scale of –3 to –5 or Sedation-Agitation Scale of 2 or 1. A total of 324 patients were studied. Emergency department deep sedation was observed in 171 patients (52.8%), and was associated with a higher frequency of deep sedation in the ICU on day 1 (53.8% vs 20.3%; p < 0.001) and day 2 (33.3% vs 16.9%; p = 0.001), when compared to light sedation. Mean (sd) ventilator-free days were 18.1 (10.8) in the emergency department deep sedation group compared to 20.0 (9.8) in the light sedation group (mean difference, 1.9; 95% CI, –0.40 to 4.13). Similar results according to emergency department sedation depth existed for ICU-free days (mean difference, 1.6; 95% CI, –0.54 to 3.83) and hospital-free days (mean difference, 2.3; 95% CI, 0.26–4.32). Mortality was 21.1% in the deep sedation group and 17.0% in the light sedation group (between-group difference, 4.1%; odds ratio, 1.30; 0.74–2.28). The occurrence rate of acute brain dysfunction (delirium and coma) was 68.4% in the deep sedation group and 55.6% in the light sedation group (between-group difference, 12.8%; odds ratio, 1.73; 1.10–2.73). Conclusions: Early deep sedation in the emergency department is common, carries over into the ICU, and may be associated with worse outcomes. Sedation practice in the emergency department and its association with clinical outcomes is in need of further investigation.

Optimal Arterial Blood Oxygen Tension in the Early Postresuscitation Phase of Extracorporeal Cardiopulmonary Resuscitation: A 15-Year Retrospective Observational Study* Objectives: Hyperoxia could lead to a worse outcome after cardiac arrest. Few studies have investigated the impact of oxygenation status on patient outcomes following extracorporeal cardiopulmonary resuscitation. We sought to delineate the association between oxygenation status and neurologic outcomes in patients receiving extracorporeal cardiopulmonary resuscitation. Design: Retrospective analysis of a prospective extracorporeal cardiopulmonary resuscitation registry database. Setting: An academic tertiary care hospital. Patients: Patients receiving extracorporeal cardiopulmonary resuscitation between 2000 and 2014. Interventions: None. Measurements and Main Results: A total of 291 patients were included, and 80.1% were male. Their mean age was 56.0 years. The arterial blood gas data employed in the primary analysis were recorded from the first sample over the first 24 hours in the ICUs after return of spontaneous circulation. The mean Pao2 after initiation of venoarterial extracorporeal membrane oxygenation was 178.0 mm Hg, and the mean Pao2/Fio2 ratio was 322.0. Only 88 patients (30.2%) demonstrated favorable neurologic status at hospital discharge. Multivariate logistic regression analysis indicated that Pao2 between 77 and 220 mm Hg (odds ratio, 2.29; 95% CI, 1.01–5.22; p = 0.05) and Pao2/Fio2 ratio between 314 and 788 (odds ratio, 5.09; 95% CI, 2.13–12.14; p < 0.001) were both positively associated with favorable neurologic outcomes. Conclusions: Oxygenation status during extracorporeal membrane oxygenation affects neurologic outcomes in patients receiving extracorporeal cardiopulmonary resuscitation. The Pao2 range of 77 to 220 mm Hg, which is slightly narrower than previously defined, seems optimal. The Pao2/Fio2 ratio was also associated with outcomes in our analysis, indicating that both Pao2 and the Pao2/Fio2 ratio should be closely monitored during the early postcardiac arrest phase for postextracorporeal cardiopulmonary resuscitation patients.