Development and Validation of a Machine Learning Algorithm Using Clinical Pages to Predict Imminent Clinical Deterioration.

BACKGROUND: Early detection of clinical deterioration among hospitalized patients is a clinical priority for patient safety and quality of care. Current automated approaches for identifying these patients perform poorly at identifying imminent events. OBJECTIVE: Develop a machine learning algorithm using pager messages sent between clinical team members to predict imminent clinical deterioration. DESIGN: We conducted a large observational study using long short-term memory machine learning models on the content and frequency of clinical pages. PARTICIPANTS: We included all hospitalizations between January 1, 2018 and December 31, 2020 at Vanderbilt University Medical Center that included at least one page message to physicians. Exclusion criteria included patients receiving palliative care, hospitalizations with a planned intensive care stay, and hospitalizations in the top 2% longest length of stay. MAIN MEASURES: Model classification performance to identify in-hospital cardiac arrest, transfer to intensive care, or Rapid Response activation in the next 3-, 6-, and 12-hours. We compared model performance against three common early warning scores: Modified Early Warning Score, National Early Warning Score, and the Epic Deterioration Index. KEY RESULTS: There were 87,783 patients (mean [SD] age 54.0 [18.8] years; 45,835 [52.2%] women) who experienced 136,778 hospitalizations. 6214 hospitalized patients experienced a deterioration event. The machine learning model accurately identified 62% of deterioration events within 3-hours prior to the event and 47% of events within 12-hours. Across each time horizon, the model surpassed performance of the best early warning score including area under the receiver operating characteristic curve at 6-hours (0.856 vs. 0.781), sensitivity at 6-hours (0.590 vs. 0.505), specificity at 6-hours (0.900 vs. 0.878), and F-score at 6-hours (0.291 vs. 0.220). CONCLUSIONS: Machine learning applied to the content and frequency of clinical pages improves prediction of imminent deterioration. Using clinical pages to monitor patient acuity supports improved detection of imminent deterioration without requiring changes to clinical workflow or nursing documentation.

Balanced Crystalloids versus Saline in Critically Ill Adults.

BACKGROUND: Both balanced crystalloids and saline are used for intravenous fluid administration in critically ill adults, but it is not known which results in better clinical outcomes. METHODS: In a pragmatic, cluster-randomized, multiple-crossover trial conducted in five intensive care units at an academic center, we assigned 15,802 adults to receive saline (0.9% sodium chloride) or balanced crystalloids (lactated Ringer's solution or Plasma-Lyte A) according to the randomization of the unit to which they were admitted. The primary outcome was a major adverse kidney event within 30 days - a composite of death from any cause, new renal-replacement therapy, or persistent renal dysfunction (defined as an elevation of the creatinine level to ≥200% of baseline) - all censored at hospital discharge or 30 days, whichever occurred first. RESULTS: Among the 7942 patients in the balanced-crystalloids group, 1139 (14.3%) had a major adverse kidney event, as compared with 1211 of 7860 patients (15.4%) in the saline group (marginal odds ratio, 0.91; 95% confidence interval [CI], 0.84 to 0.99; conditional odds ratio, 0.90; 95% CI, 0.82 to 0.99; P=0.04). In-hospital mortality at 30 days was 10.3% in the balanced-crystalloids group and 11.1% in the saline group (P=0.06). The incidence of new renal-replacement therapy was 2.5% and 2.9%, respectively (P=0.08), and the incidence of persistent renal dysfunction was 6.4% and 6.6%, respectively (P=0.60). CONCLUSIONS: Among critically ill adults, the use of balanced crystalloids for intravenous fluid administration resulted in a lower rate of the composite outcome of death from any cause, new renal-replacement therapy, or persistent renal dysfunction than the use of saline. (Funded by the Vanderbilt Institute for Clinical and Translational Research and others; SMART-MED and SMART-SURG ClinicalTrials.gov numbers, NCT02444988 and NCT02547779 .).

Etiology and Timing of Postoperative Rapid Response Team Activations.

In this retrospective cohort study we sought to evaluate the association between the etiology and timing of rapid response team (RRT) activations in postoperative patients at a tertiary care hospital in the southeastern United States. From 2010 to 2016, there were 2,390 adult surgical inpatients with RRT activations within seven days of surgery. Using multivariable linear regression, we modeled the correlation between etiology of RRT and timing of the RRT call, as measured from the conclusion of the surgical procedure. We found that respiratory triggers were associated with an increase in time after surgical procedure to RRT of 10.6 h compared to activations due to general concern (95% CI 3.9 - 17.3) (p = 0.002). These findings may have an impact on monitoring of postoperative patients, as well as focusing interventions to better respond to clinically deteriorating patients.

Intensive Care Unit Enhanced Recovery Pathway for Patients Undergoing Orthotopic Liver Transplants Recipients: A Prospective, Observational Study.

BACKGROUND: Liver transplant recipients continue to have high perioperative resource utilization and prolonged length of stay despite improvements in perioperative care. Enhanced recovery pathways have been shown in other surgical populations to produce reductions in hospital resource utilization. METHODS: A prospective, observational study was performed to examine the effect of an enhanced recovery pathway for postoperative care after liver transplantation. Outcomes from patients undergoing liver transplantation from November 1, 2013, to October 31, 2014, managed by the pathway were compared to transplant recipients from the year before pathway implementation. Multivariable regression analysis was used to assess the association of the clinical pathway on clinical outcomes. RESULTS: The intervention and control groups included 141 and 106 patients, respectively. There were no demographic differences between the control and intervention group including no differences between the length of surgery and cold ischemic time. Median intensive care unit length of stay was reduced from 4.4 to 2.6 days (P < .001). The intervention group had a higher likelihood of earlier discharge (hazard ratio [95% CI], 2.01 [1.55-2.62]; P < .001), and a 69% and 65% lower odds of receiving a plasma (P < .001) or packed red blood cell (P < .001) transfusion. There was no significant effect on hospital mortality (P = .40), intensive care unit readmission rates (P = .75), or postoperative infections (urinary traction infections: P = .09; pneumonia: P = .27). CONCLUSIONS: An enhanced recovery pathway focused on milestone-based elements of intensive care unit management and predetermined management triggers including hemodynamic goals, fluid therapy, perioperative antibiotics, glycemic control, and standardized transfusion triggers led to reductions in intensive care unit length of stay without an increase in perioperative complications.

Daily Lowest Hemoglobin and Risk of Organ Dysfunctions in Critically Ill Patients.

OBJECTIVES: To determine the association between hemoglobin levels and the daily risk of individual organ dysfunctions in critically ill patients. DESIGN: Post hoc analysis of prospectively collected data. SETTING: Vanderbilt University Medical Center and Saint Thomas Hospital Medical and Surgical ICUs. PATIENTS: Medical and surgical ICU patients admitted with respiratory failure or shock. INTERVENTIONS: Baseline demographic data, and detailed in-ICU and hospital data, including daily lowest hemoglobin, were collected up to hospital day 30. We assessed patients daily for brain dysfunction (delirium, using Confusion Assessment Method for ICU), for renal and respiratory dysfunction (using the ordinal renal and respiratory Sequential Organ Failure Assessment score), and for ICU mortality. Associations between the lowest hemoglobin on a given day and organ dysfunctions the following day were assessed using multivariable regressions, adjusting for age, Acute Physiology and Chronic Health Evaluation II score, Charlson comorbidity index, Framingham Stroke Risk Profile, ICU day, ICU type, sepsis, and current organ dysfunction status. A sensitivity analysis further adjusted for daily transfusions and fluid balance in a subset of our patients. MEASUREMENTS AND MAIN RESULTS: We enrolled 821 patients with a median (interquartile range) age of 61 (51-71) years, Acute Physiology and Chronic Health Evaluation II score of 25 (19-31), and hemoglobin level of 10.0 (9.0-11.1) g/dL. There was no evidence of an association between lowest daily hemoglobin and brain dysfunction (p = 0.69 for delirium), renal dysfunction (p = 0.30), or ICU mortality (p = 0.95). The lowest hemoglobin on a given day was significantly associated with the respiratory Sequential Organ Failure Assessment score the following day; for each increasing hemoglobin unit, the odds of worsened respiratory Sequential Organ Failure Assessment score the following day were decreased by 36% (OR, 0.64; 95% CI, 0.53-0.77; p < 0.001). The sensitivity analysis including daily transfusions and fluid balance (in a subset of 518 patients) did not qualitatively change any of these associations. CONCLUSIONS: In this study in ICU patients, lower hemoglobin was associated with a higher probability of worsening respiratory dysfunction scores the following day. There was no evidence of association between hemoglobin and brain or renal dysfunction, or ICU mortality. The possible differential effects of anemia on organ dysfunctions seen in this hypothesis-generating study will have to be studied in a larger prospective study before any alterations to present restrictive transfusion guidelines can be recommended.

Creation and Execution of a Novel Anesthesia Perioperative Care Service at a Veterans Affairs Hospital.

Physician-led perioperative surgical home models are developing as a method for improving the American health care system. These models are novel, team-based approaches that help to provide continuity of care throughout the perioperative period. Another avenue for improving care for surgical patients is the use of enhanced recovery after surgery pathways. These are well-described methods that have shown to improve perioperative outcomes. An established perioperative surgical home model can help implementation, efficiency, and adherence to enhanced recovery after surgery pathways. For these reasons, the Tennessee Valley Healthcare System, Nashville Veterans Affairs Medical Center created an Anesthesiology Perioperative Care Service that provides comprehensive care to surgical patients from their preoperative period through the continuum of their hospital course and postdischarge follow-up. In this brief report, we describe the development, implementation, and preliminary outcomes of the service.

An Electronic Tool for the Evaluation and Treatment of Sepsis in the ICU: A Randomized Controlled Trial.

OBJECTIVES: To determine whether addition of an electronic sepsis evaluation and management tool to electronic sepsis alerting improves compliance with treatment guidelines and clinical outcomes in septic ICU patients. DESIGN: A pragmatic randomized trial. SETTING: Medical and surgical ICUs of an academic, tertiary care medical center. PATIENTS: Four hundred and seven patients admitted during a 4-month period to the medical or surgical ICU with a diagnosis of sepsis established at the time of admission or in response to an electronic sepsis alert. INTERVENTIONS: Patients were randomized to usual care or the availability of an electronic tool capable of importing, synthesizing, and displaying sepsis-related data from the medical record, using logic rules to offer individualized evaluations of sepsis severity and response to therapy, informing users about evidence-based guidelines, and facilitating rapid order entry. MEASUREMENTS AND MAIN RESULTS: There was no difference between the electronic tool (218 patients) and usual care (189 patients) with regard to the primary outcome of time to completion of all indicated Surviving Sepsis Campaign 6-hour Sepsis Resuscitation Bundle elements (hazard ratio, 1.98; 95% CI, 0.75-5.20; p = 0.159) or time to completion of each element individually. ICU mortality, ICU-free days, and ventilator-free days did not differ between intervention and control. Providers used the tool to enter orders in only 28% of available cases. CONCLUSIONS: A comprehensive electronic sepsis evaluation and management tool is feasible and safe but did not influence guideline compliance or clinical outcomes, perhaps due to low utilization.

Systems of Care Delivery and Optimization in the Postoperative Care Wards.

A third of all patients are at risk for a serious adverse event, including death, in the first month after undergoing a major surgery. Most of these events will occur within 24 hours of the operation but are unlikely to occur in the operating room or postanesthesia care unit. Most opioid-induced respiratory depression events in the postoperative period resulted in death (55%) or anoxic brain injury (22%). A future state of mature artificial intelligence and machine learning will improve situational awareness of acute clinical deterioration, minimize alert fatigue, and facilitate early intervention to minimize poor outcomes.

Code status at time of rapid response activation - Impact on escalation of care?

BACKGROUND: A code status documents the decision to receive or forgo cardiopulmonary resuscitation in the event of cardiac arrest. For patients who undergo a rapid response team activation (RRT) for possible escalation to an intensive care unit (ICU), the presence or absence of a code status represents a critical inflection point for guiding care decisions and resource utilization. This study characterizes the prevalence of code status at the time of RRT and how code status at RRT affects rates of intensive treatments in the ICU. METHODS: We conducted a single-center retrospective cohort study of 895 rapid response activations occurring over six months. The study included all rapid response team activations for non-obstetric adult inpatients documented in the patient chart. All data was obtained through retrospective chart review. STROBE reporting guidelines were followed. RESULTS: At the time of RRT activation, 56% of patients had a documented code status. Code status prevalence was much higher among medical rather than surgical services (74% vs. 13%). For patients escalated to the ICU, having a DNR code status at RRT was not associated with decreased odds of receiving cardioactive medications or advanced respiratory support. Before RRT activation, palliative care utilization was low (9%) but more than doubled after RRT (24% before discharge). CONCLUSIONS: Barely half of the patients had an active code status at the time of RRT activation. Similar rates of invasive ICU treatments among full code and DNR patients suggest that documented code statuses do not reflect in-depth goals of care discussions, nor does it guide medical teams caring for the patient at times of decompensation.

Incidence of Laryngotracheal Stenosis after Thermal Inhalation Airway Injury.

Inhalation injury is independently associated with burn mortality, yet little information is available on the incidence, risk factors, or functional outcomes of thermal injury to the airway. In patients with thermal inhalation injury, we sought to define the incidence of laryngotracheal stenosis (LTS), delineate risk factors associated with LTS development, and assess long-term tracheostomy dependence as a proxy for laryngeal function. Retrospective cohort study of adult patients treated for thermal inhalation injury at a single institution burn critical care unit from 2012 to 2017. Eligible patients' records were assessed for LTS (laryngeal, subglottic, or tracheal stenosis). Patient characteristics, burn injury characteristics, and treatment-specific covariates were assessed. Descriptive statistics, Mann-Whitney U-tests, odds ratio, and chi-square tests compared LTS versus non-LTS groups. Of 129 patients with thermal inhalation injury during the study period, 8 (6.2%) developed LTS. When compared with the non-LTS group, patients with LTS had greater mean TBSA (mean 30.3, Interquartile Range 7-57.5 vs 10.5, Interquartile Range 0-15.12, P = .01), higher grade of inhalation injury (mean 2.63 vs 1.80, P = .05), longer duration of intubation (12.63 vs 5.44; P < .001), and greater inflammatory response (mean white blood cell count on presentation 25.8 vs 14.9, P = .02, mean hyperglycemia on presentation 176.4 vs 136.9, P = .01). LTS patients had a significantly higher rate of tracheostomy dependence at last follow-up (50 vs 1.7%, P < .001). Six percent of patients with thermal inhalation injury develop LTS. LTS was associated with more severe thermal airway injury, longer duration of intubation, and more severe initial host inflammation. Patients with inhalation injury and LTS are at high risk for tracheostomy dependence. In burn patients with thermal inhalation injury, laryngeal evaluation and directed therapy should be incorporated early into multispecialty pathways of care.

Nursing attitudes towards continuous capnographic monitoring of floor patients.

INTRODUCTION: Nurses' perceptions of the utility of capnography monitoring are inconsistent in previous studies. We sought to outline the limitations of a uniform education effort in bringing about consistent views of capnography among nurses. METHODS: A survey was administered to 22 nurses in three subacute care floors participating in a pragmatic clinical trial employing capnography monitoring in a large, urban tertiary care hospital. A 5-point Likert scale was used to assess the value and acceptance nurses ascribed to the practice. Means and SD were calculated for each response. RESULTS: Survey results indicated inconsistency in the valuation of capnography, coupled with varying degrees of acceptance of its use. The mean for the level of perceived impact of capnography use on patient safety was 3.86, yet the perceived risk of removing capnography was represented by a mean of 2.57. The levels of urgency attached to apnoea alarms (mean 3.57, SD 1.57) were lower than those for alarms for oxygen saturation violations (mean 3.67, SD 1.32). The necessity for pulse oximetry monitoring was perceived as much higher than that for capnography monitoring (mean 1.76, SD 1.34), where '1' represented pulse oximetry as more necessary and '5' represented capnography as more necessary. CONCLUSIONS: Nursing acceptance of capnography monitoring is a difficult endpoint to achieve. There is a need for better accounting for the external and internal influences on nurse perceptions and values to have greater success with the implementation of similar monitoring.

Cryoamputation as a temporizing measure in severe burn injury.

Cryoamputation, or physiologic amputation, is a well-described procedure typically used to amputate gangrenous lower extremities. In such cases the patient is too unstable for transport to the operating room, so cryoamputation using dry ice or other refrigerant allows for immediate bedside intervention and later operative amputation when the patient is more stable. In this study the authors describe the use of cryoamputation to stabilize a burn patient with a nonviable upper extremity considered to be contributing significantly to his metabolic acidosis. This experience suggests that cryoamputation may be a reasonable technique to consider when a burn patient presents with a nonviable extremity but is too unstable for immediate operative amputation.

The science and economics of improving clinical communication.

This article presents a complex clinical scenario based on actual communication breakdowns that led to a sentinel event. Basic communication theory that underlies clinical interactions and the tenets of health care economic evaluation are reviewed. The process of the handoff as it relates to clinical interactions is discussed and the weaknesses in communication arising from handoff failures in the operative and critical care environments are examined. The discussion follows by looking at the influences of current medical culture, emerging technology, and changing care environments and their impact on communication behaviors and resultant effect on patient outcomes. A detailed cost analysis of the charges incurred for both standard and escalated care required for the case is followed by a discussion of the economic basis for improving clinical communication and patient safety using the SBAR tool.

Workflow in intensive care unit remote monitoring: A time-and-motion study.

OBJECTIVE: To investigate workflow in intensive care unit remote monitoring, a technology-driven practice that allows critical care specialists to perform proactive and continuous patient care from a remote site. DESIGN: A time-and-motion study. SETTING: Facility that remotely monitored 132 beds in nine intensive care units. PARTICIPANTS: Six physicians and seven registered nurses. INTERVENTIONS: Participants were observed for 47 and 39 hrs, respectively. MEASUREMENTS AND MAIN RESULTS: Clinicians' workflow was analyzed as goal-oriented tasks and activities. Major variables of interest included the times spent on different types of tasks and activities, the frequencies of accessing various information resources, and the occurrence and management of interruptions in workflow. Physicians spent 70%, 3%, 3%, and 24% of their time on patient monitoring, collaboration, system maintenance, and administrative/social/personal tasks, respectively. For nurses, the time allocations were 46%, 3%, 4%, and 17%, respectively. Nurses spent another 30% of their time maintaining health records. In monitoring patients, physicians spent more percentage times communicating with others than the nurses (13% vs. 7%, p = .026) and accessed the in-unit clinical information system more frequently (42 vs. 14 times per hour, p = .027), while nurses spent more percentage times monitoring real-time vitals (16% vs. 2%, p = .012). Physicians' and nurses' workflows were interrupted at a rate of 2.2 and 7.5 times per hour (p < .001), with an average duration of 101 and 45 secs, respectively (p = .006). The sources of interruptions were significantly different for physicians and nurses (p < .001). CONCLUSIONS: Physicians' and nurses' task performance and information utilization reflect the distributed nature of work organization in intensive care unit remote monitoring. Workflow interruption, clinical information system usability, and collaboration with bedside caregivers are the major issues that may affect the quality and efficiency of clinicians' work in this particular critical care setting.

A time-motion study of registered nurses' workflow in intensive care unit remote monitoring.

Utilizing advanced information technology, Intensive Care Unit (ICU) remote monitoring allows highly trained specialists to oversee a large number of patients at multiple sites on a continuous basis. In the current research, we conducted a time-motion study of registered nurses' work in an ICU remote monitoring facility. Data were collected on seven nurses through 40 hours of observation. The results showed that nurses' essential tasks were centered on three themes: monitoring patients, maintaining patients' health records, and managing technology use. In monitoring patients, nurses spent 52% of the time assimilating information embedded in a clinical information system and 15% on monitoring live vitals. System-generated alerts frequently interrupted nurses in their task performance and redirected them to manage suddenly appearing events. These findings provide insight into nurses' workflow in a new, technology-driven critical care setting and have important implications for system design, work engineering, and personnel selection and training.