SWIFT: A deep learning approach to prediction of hypoxemic events in critically-Ill patients using SpO2 waveform prediction.

Hypoxemia is a significant driver of mortality and poor clinical outcomes in conditions such as brain injury and cardiac arrest in critically ill patients, including COVID-19 patients. Given the host of negative clinical outcomes attributed to hypoxemia, identifying patients likely to experience hypoxemia would offer valuable opportunities for early and thus more effective intervention. We present SWIFT (SpO2 Waveform ICU Forecasting Technique), a deep learning model that predicts blood oxygen saturation (SpO2) waveforms 5 and 30 minutes in the future using only prior SpO2 values as inputs. When tested on novel data, SWIFT predicts more than 80% and 60% of hypoxemic events in critically ill and COVID-19 patients, respectively. SWIFT also predicts SpO2 waveforms with average MSE below .0007. SWIFT predicts both occurrence and magnitude of potential hypoxemic events 30 minutes in the future, allowing it to be used to inform clinical interventions, patient triaging, and optimal resource allocation. SWIFT may be used in clinical decision support systems to inform the management of critically ill patients during the COVID-19 pandemic and beyond.

Using Continuous Vital Sign Monitoring to Detect Early Deterioration in Adult Postoperative Inpatients.

BACKGROUND: Episodic vital sign collection (eVSC), as single data points, gives an incomplete picture of adult patients' postoperative physiologic status. LOCAL PROBLEM: Late detection of patient deterioration resulted in poor patient outcomes on a postsurgical unit. METHODS: Baseline demographic and outcome data were collected through retrospective chart review of all patients admitted to the surgical unit for 12 weeks prior to this quality improvement project. Data on the same outcomes were collected during the 12-week project. INTERVENTION: This project compared outcomes between the current standard of eVSC and the proposed standard of continuous vital sign monitoring (cVSM). RESULTS: Using cVSM demonstrated a statistically significant 27% decrease in the complication rate, and a clinically significant decrease in transfers to an intensive care unit and failure-to-rescue (FTR) events rate. CONCLUSIONS: cVSM demonstrated detection of early signs of patient deterioration to prevent FTR.

Technological Distractions (Part 2): A Summary of Approaches to Manage Clinical Alarms With Intent to Reduce Alarm Fatigue.

OBJECTIVE: Alarm fatigue is a widely recognized safety and quality problem where exposure to high rates of clinical alarms results in desensitization leading to dismissal of or slowed response to alarms. Nonactionable alarms are thought to be especially problematic. Despite these concerns, the number of clinical alarm signals has been increasing as an everincreasing number of medical technologies are added to the clinical care environment. DATA SOURCES: PubMed, SCOPUS, Embase, and CINAHL. STUDY SELECTION: We performed a systematic review of the literature focused on clinical alarms. We asked a primary key question; "what interventions have been attempted and resulted in the success of reducing alarm fatigue?" and 3-secondary key questions; "what are the negative effects on patients/families; what are the balancing outcomes (unintended consequences of interventions); and what human factor approaches apply to making an effective alarm?" DATA EXTRACTION: Articles relevant to the Key Questions were selected through an iterative review process and relevant data was extracted using a standardized tool. DATA SYNTHESIS: We found 62 articles that had relevant and usable data for at least one key question. We found that no study used/developed a clear definition of "alarm fatigue." For our primary key question 1, the relevant studies focused on three main areas: quality improvement/bundled activities; intervention comparisons; and analysis of algorithm-based false and total alarm suppression. All sought to reduce the number of total alarms and/or false alarms to improve the positive predictive value. Most studies were successful to varying degrees. None measured alarm fatigue directly. CONCLUSIONS: There is no agreed upon valid metric(s) for alarm fatigue, and the current methods are mostly indirect. Assuming that reducing the number of alarms and/or improving positive predictive value can reduce alarm fatigue, there are promising avenues to address patient safety and quality problem. Further investment is warranted not only in interventions that may reduce alarm fatigue but also in defining how to best measure it.

Review: Continuous Monitoring to Detect Failure to Rescue in Adult Postoperative Inpatients.

Failure to rescue, or the unexpected death of a patient due to a preventable complication, is a nationally documented problem with numerous and multifaceted contributing factors. These factors include the frequency and method of collecting vital sign data, response to abnormal vital signs, and delays in the escalation of care for general ward patients who are showing signs of clinical deterioration. Patients' clinical deterioration can be complicated by concurrent secondary factors, including opioid abuse/dependence, being uninsured, or having sleep-disordered breathing. Using the Johns Hopkins Nursing Evidence-Based Practice Model, this integrative review synthesizes 43 research and nonresearch sources of evidence. Published between 2001 and 2017, these sources of evidence focus on failure to rescue, the multifaceted contributing factors to failure to rescue, and how continuous vital sign monitoring could ameliorate failure to rescue and its causes. Recommendations from the sources of evidence have been divided into system, structural, or technological categories.

Two-State Collaborative Study of a Multifaceted Intervention to Decrease Ventilator-Associated Events.

OBJECTIVES: Ventilator-associated events are associated with increased mortality, prolonged mechanical ventilation, and longer ICU stay. Given strong national interest in improving ventilated patient care, the National Institute of Health and Agency for Healthcare Research and Quality funded a two-state collaborative to reduce ventilator-associated events. We describe the collaborative's impact on ventilator-associated event rates in 56 ICUs. DESIGN: Longitudinal quasi-experimental study. SETTING: Fifty-six ICUs at 38 hospitals in Maryland and Pennsylvania from October 2012 to March 2015. INTERVENTIONS: We organized a multifaceted intervention to improve adherence with evidence-based practices, unit teamwork, and safety culture. Evidence-based interventions promoted by the collaborative included head-of-bed elevation, use of subglottic secretion drainage endotracheal tubes, oral care, chlorhexidine mouth care, and daily spontaneous awakening and breathing trials. Each unit established a multidisciplinary quality improvement team. We coached teams to establish comprehensive unit-based safety programs through monthly teleconferences. Data were collected on rounds using a common tool and entered into a Web-based portal. MEASUREMENTS AND RESULTS: ICUs reported 69,417 ventilated patient-days of intervention compliance observations and 1,022 unit-months of ventilator-associated event data. Compliance with all evidence-based interventions improved over the course of the collaborative. The quarterly mean ventilator-associated event rate significantly decreased from 7.34 to 4.58 cases per 1,000 ventilator-days after 24 months of implementation (p = 0.007). During the same time period, infection-related ventilator-associated complication and possible and probable ventilator-associated pneumonia rates decreased from 3.15 to 1.56 and 1.41 to 0.31 cases per 1,000 ventilator-days (p = 0.018, p = 0.012), respectively. CONCLUSIONS: A multifaceted intervention was associated with improved compliance with evidence-based interventions and decreases in ventilator-associated event, infection-related ventilator-associated complication, and probable ventilator-associated pneumonia. Our study is the largest to date affirming that best practices can prevent ventilator-associated events.

Technologic Distractions (Part 1): Summary of Approaches to Manage Alert Quantity With Intent to Reduce Alert Fatigue and Suggestions for Alert Fatigue Metrics.

OBJECTIVE: To provide ICU clinicians with evidence-based guidance on tested interventions that reduce or prevent alert fatigue within clinical decision support systems. DESIGN: Systematic review of PubMed, Embase, SCOPUS, and CINAHL for relevant literature from 1966 to February 2017. PATIENTS: Focus on critically ill patients and included evaluations in other patient care settings, as well. INTERVENTIONS: Identified interventions designed to reduce or prevent alert fatigue within clinical decision support systems. MEASUREMENTS AND MAIN RESULTS: Study selection was based on one primary key question to identify effective interventions that attempted to reduce alert fatigue and three secondary key questions that covered the negative effects of alert fatigue, potential unintended consequences of efforts to reduce alert fatigue, and ideal alert quantity. Data were abstracted by two reviewers independently using a standardized abstraction tool. Surveys, meeting abstracts, "gray" literature, studies not available in English, and studies with non-original data were excluded. For the primary key question, articles were excluded if they did not provide a comparator as key question 1 was designed as a problem, intervention, comparison, and outcome question. We anticipated that reduction in alert fatigue, including the concept of desensitization may not be directly measured and thus considered interventions that reduced alert quantity as a surrogate marker for alert fatigue. Twenty-six articles met the inclusion criteria. CONCLUSION: Approaches for managing alert fatigue in the ICU are provided as a result of reviewing tested interventions that reduced alert quantity with the anticipated effect of reducing fatigue. Suggested alert management strategies include prioritizing alerts, developing sophisticated alerts, customizing commercially available alerts, and including end user opinion in alert selection. Alert fatigue itself is studied less frequently, as an outcome, and there is a need for more precise evaluation. Standardized metrics for alert fatigue is needed to advance the field. Suggestions for standardized metrics are provided in this document.

Anesthesiologists and Disaster Medicine: A Needs Assessment for Education and Training and Reported Willingness to Respond.

BACKGROUND: Anesthesiologists provide comprehensive health care across the emergency department, operating room, and intensive care unit. To date, anesthesiologists' perspectives regarding disaster medicine and public health preparedness have not been described. METHODS: Anesthesiologists' thoughts and attitudes were assessed via a Web-based survey at 3 major academic institutions. Frequencies, percentages, and odds ratios (ORs) were used to assess self-reported perceptions of knowledge and skills, as well as attitudes and beliefs regarding education and training, employee development, professional obligation, safety, psychological readiness, efficacy, personal preparedness, and willingness to respond (WTR). Three representative disaster scenarios (natural disaster [ND], radiological event [RE], and pandemic influenza [PI]) were investigated. Results are reported as percent or OR (95% confidence interval). RESULTS: Participants included 175 anesthesiology attendings (attendings) and 95 anesthesiology residents (residents) representing a 47% and 51% response rate, respectively. A minority of attendings indicated that their hospital provides adequate pre-event preparation and training (31% [23-38] ND, 14% [9-21] RE, and 40% [31-49] PI). Few residents felt that their residency program provided them with adequate preparation and training (22% [14-33] ND, 16% [8-27] RE, and 17% [9-29] PI). Greater than 85% of attendings (89% [84-94] ND, 88% [81-92] RE, and 87% [80-92] PI) and 70% of residents (81% [71-89] ND, 71% [58-81] RE, and 82% [70-90] PI) believe that their hospital or residency program, respectively, should provide them with preparation and training. Approximately one-half of attendings and residents are confident that they would be safe at work during response to a ND or PI (55% [47-64] and 58% [49-67] of attendings; 59% [48-70] and 48% [35-61] of residents, respectively), whereas approximately one-third responded the same regarding a RE (31% [24-40] of attendings and 28% [18-41] of residents). Fewer than 40% of attendings (34% [26-43]) and residents (38% [27-51]) designated who would take care of their family obligations in the event they were called into work during a disaster. Regardless of severity, 79% (71-85) of attendings and 73% (62-82) of residents indicated WTR to a ND, whereas 81% (73-87) of attendings and 70% (58-81) of residents indicated WTR to PI. Fewer were willing to respond to a RE (63% [55-71] of attendings and 52% [39-64] of residents). In adjusted logistic regression analyses, those anesthesiologists who reported knowing one's role in response to a ND (OR, 15.8 [4.5-55.3]) or feeling psychologically prepared to respond to a ND (OR, 6.9 [2.5-19.0]) were found to be more willing to respond. Similar results were found for RE and PI constructs. Both attendings and residents were willing to respond in whatever capacity needed, not specifically to provide anesthesia. CONCLUSIONS: Few anesthesiologists reported receiving sufficient education and training in disaster medicine and public health preparedness. Providing education and training and enhancing related employee services may further bolster WTR and help to build a more capable and effective medical workforce for disaster response.

Evaluation of Noninvasive Hemoglobin Monitoring in Surgical Critical Care Patients.

OBJECTIVE: To assess the clinical utility of noninvasive hemoglobin monitoring based on pulse cooximetry in the ICU setting. DESIGN AND SETTING: A total of 358 surgical patients from a large urban, academic hospital had the noninvasive hemoglobin monitoring pulse cooximeter placed at admission to the ICU. Core and stat laboratory hemoglobin measurements were taken at the discretion of the clinicians, who were blinded to noninvasive hemoglobin monitoring values. MEASUREMENT AND MAIN RESULTS: There was a poor correlation between the 2,465 time-matched noninvasive hemoglobin monitoring and laboratory hemoglobin measurements (r = 0.29). Bland-Altman analysis showed a positive bias of 1.0 g/dL and limits of agreement of -2.5 to 4.6 g/dL. Accuracy was best at laboratory values of 10.5-14.5 g/dL and least at laboratory values of 6.5-8 g/dL. At hemoglobin values that would ordinarily identify a patient as requiring a transfusion (< 8 g/dL), noninvasive hemoglobin monitoring consistently overestimated the patient's true hemoglobin. When sequential laboratory values declined below 8 g/dL (n = 102) and 7 g/dL (n = 13), the sensitivity and specificity of noninvasive hemoglobin monitoring at identifying these events were 27% and 7%, respectively. At a threshold of 8 g/dL, continuous noninvasive hemoglobin monitoring values reached the threshold before the labs in 45 of 102 instances (44%) and at 7 g/dL, noninvasive hemoglobin monitoring did so in three of 13 instances (23%). Noninvasive hemoglobin monitoring minus laboratory hemoglobin differences showed an intraclass correlation coefficient of 0.47 within individual patients. Longer length of stay and higher All Patient Refined Diagnostic-Related Groups severity of illness were associated with poor noninvasive hemoglobin monitoring accuracy. CONCLUSIONS: Although noninvasive hemoglobin monitoring technology holds promise, it is not yet an acceptable substitute for laboratory hemoglobin measurements. Noninvasive hemoglobin monitoring performs most poorly in the lower hemoglobin ranges that include commonly used transfusion trigger thresholds and is not consistent within individual patients. Further refinement of the signal acquisition and analysis algorithms and clinical reevaluation are needed.

Validity of the Agency for Health Care Research and Quality Patient Safety Indicators and the Centers for Medicare and Medicaid Hospital-acquired Conditions: A Systematic Review and Meta-Analysis.

BACKGROUND: The Agency for Health Care Research and Quality Patient Safety Indicators (PSIs) and Centers for Medicare and Medicaid Services Hospital-acquired Conditions (HACs) are increasingly being used for pay-for-performance and public reporting despite concerns over their validity. Given the potential for these measures to misinform patients, misclassify hospitals, and misapply financial and reputational harm to hospitals, these need to be rigorously evaluated. We performed a systematic review and meta-analysis to assess PSI and HAC measure validity. METHODS: We searched MEDLINE and the gray literature from January 1, 1990 through January 14, 2015 for studies that addressed the validity of the HAC measures and PSIs. Secondary outcomes included the effects of present on admission (POA) modifiers, and the most common reasons for discrepancies. We developed pooled results for measures evaluated by ≥3 studies. We propose a threshold of 80% for positive predictive value or sensitivity for pay-for-performance and public reporting suitability. RESULTS: Only 5 measures, Iatrogenic Pneumothorax (PSI 6/HAC 17), Central Line-associated Bloodstream Infections (PSI 7), Postoperative hemorrhage/hematoma (PSI 9), Postoperative deep vein thrombosis/pulmonary embolus (PSI 12), and Accidental Puncture/Laceration (PSI 15), had sufficient data for pooled meta-analysis. Only PSI 15 (Accidental Puncture and Laceration) met our proposed threshold for validity (positive predictive value only) but this result was weakened by considerable heterogeneity. Coding errors were the most common reasons for discrepancies between medical record review and administrative databases. POA modifiers may improve the validity of some measures. CONCLUSION: This systematic review finds that there is limited validity for the PSI and HAC measures when measured against the reference standard of a medical chart review. Their use, as they currently exist, for public reporting and pay-for-performance, should be publicly reevaluated in light of these findings.

The prevalence of long QT interval in post-operative intensive care unit patients.

The severity of patient illnesses and medication complexity in post-operative critically ill patients increase the risk for a prolonged QT interval. We determined the prevalence of prolonged QTc in surgical intensive care unit (SICU) patients. We performed a prospective cross-sectional study over a 15-month period at a major academic center. SICU pre-admission and admission EKGs, patient demographics, and laboratory values were analyzed. QTc was evaluated as both a continuous and dichotomous outcome (prolonged QTc > 440 ms). 281 patients were included in the study: 92 % (n = 257) post-operative and 8 % (n = 24) non-operative. On pre-admission EKGs, 32 % of the post-operative group and 42 % of the non-operative group had prolonged QTc (p = 0.25); on post-admission EKGs, 67 % of the post-operative group but only 33 % of the non-operative group had prolonged QTc (p < 0.01). The average change in QTc in the post-operative group was +30.7 ms, as compared to +2 ms in the non-operative group (p < 0.01). On multivariable adjustment for long QTc as a dichotomous outcome, pre-admission prolonged QTc (OR 3.93, CI 1.93-8.00) and having had an operative procedure (OR 4.04, CI 1.67-9.83) were associated with developing prolonged QTc. For QTc as a continuous outcome, intra-operative beta-blocker use was associated with a statistically-significant decrease in QTc duration. None of the patients developed a lethal arrhythmia in the ICU. Prolonged QTc is common among post-operative SICU patients (67 %), however lethal arrhythmias are uncommon. The operative experience increases the risk for long QTc.

Diagnostic errors in the pediatric and neonatal ICU: a systematic review.

OBJECTIVE: Diagnostic errors lead to preventable hospital morbidity and mortality. ICU patients may be at particularly high risk for misdiagnosis. Little is known about misdiagnosis in pediatrics, including PICU and neonatal ICU. We sought to assess diagnostic errors in PICU and neonatal ICU settings by systematic review. DATA SOURCES: We searched PubMed, Embase, CINAHL, and Cochrane. STUDY SELECTION: We identified observational studies reporting autopsy-confirmed diagnostic errors in PICU or neonatal ICU using standard Goldman criteria. DATA EXTRACTION: We abstracted patient characteristics, diagnostic error description, rates and error classes using standard Goldman criteria for autopsy misdiagnoses and calculated descriptive statistics. DATA SYNTHESIS: We screened 329 citations, examined 79 full-text articles, and included 13 studies (seven PICU; six neonatal ICU). The PICU studies examined a total of 1,063 deaths and 498 autopsies. Neonatal ICU studies examined a total of 2,124 neonatal deaths and 1,259 autopsies. Major diagnostic errors were found in 19.6% of autopsied PICU and neonatal ICU deaths (class I, 4.5%; class II, 15.1%). Class I (potentially lethal) misdiagnoses in the PICU (43% infections, 37% vascular) and neonatal ICU (62% infections, 21% congenital/metabolic) differed slightly. Although missed infections were most common in both settings, missed vascular events were more common in the PICU and missed congenital conditions in the neonatal ICU. CONCLUSION: Diagnostic errors in PICU/neonatal ICU populations are most commonly due to infection. Further research is needed to better quantify pediatric intensive care-related misdiagnosis and to define potential strategies to reduce their frequency or mitigate misdiagnosis-related harm.

Rapid-response systems as a patient safety strategy: a systematic review.

Rapid-response systems (RRSs) are a popular intervention in U.S. hospitals and are supported by accreditors and quality improvement organizations. The purpose of this review is to evaluate the effectiveness and implementation of these systems in acute care settings. A literature search was performed between 1 January 2000 through 30 October 2012 using PubMed, PsycINFO, CINAHL, and the Cochrane Central Register of Controlled Trials. Studies published in any language evaluating outcome changes that occurred after implementing an RRS and differences between groups using and not using an RRS (effectiveness) or describing methods used by RRSs (implementation) were reviewed. A single reviewer (checked by a second reviewer) abstracted data and rated study quality and strength of evidence. Moderate-strength evidence from a high-quality meta-analysis of 18 studies and 26 lower-quality before-and-after studies published after that meta-analysis showed that RRSs are associated with reduced rates of cardiorespiratory arrest outside of the intensive care unit and reduced mortality. Eighteen studies examining facilitators of and barriers to implementation suggested that the rate of use of RRSs could be improved.

Rapid Response Systems.

The hospital rapid response system (RRS) is a patient safety and quality intervention that responds quickly to clinical deteriorations on general wards with the goal of preventing cardiopulmonary arrests, reducing hospital mortality, and facilitating triage and level of care escalations. The RRS is one of the first organized, and systematic, elements of the "ICU without walls" model. RRSs have been shown to be effective in preventing deterioration to cardiopulmonary arrest on general hospital wards and reducing total and unexpected hospital mortality. Recent studies have demonstrated that this benefit can be enhanced through targeted improvements and modifications of existing RRSs.

SWIFT: A Deep Learning Approach to Prediction of Hypoxemic Events in Critically-Ill Patients Using SpO 2 Waveform Prediction.

Hypoxemia is a significant driver of mortality and poor clinical outcomes in conditions such as brain injury and cardiac arrest in critically ill patients, including COVID-19 patients. Given the host of negative clinical outcomes attributed to hypoxemia, identifying patients likely to experience hypoxemia would offer valuable opportunities for early and thus more effective intervention. We present SWIFT (SpO 2 W aveform I CU F orecasting T echnique), a deep learning model that predicts blood oxygen saturation (SpO 2 ) waveforms 5 and 30 minutes in the future using only prior SpO 2 values as inputs. When tested on novel data, SWIFT predicts more than 80% and 60% of hypoxemic events in critically ill and COVID-19 patients, respectively. SWIFT also predicts SpO 2 waveforms with average MSE below .0007. SWIFT provides information on both occurrence and magnitude of potential hypoxemic events 30 minutes in advance, allowing it to be used to inform clinical interventions, patient triaging, and optimal resource allocation. SWIFT may be used in clinical decision support systems to inform the management of critically ill patients during the COVID-19 pandemic and beyond.

Long-term mortality and quality of life in sepsis: a systematic review.

BACKGROUND: Long-term outcomes from sepsis are poorly understood, and sepsis in patients may have different long-term effects on mortality and quality of life. Long-term outcome studies of other critical illnesses such as acute lung injury have demonstrated incremental health effects that persist after hospital discharge. Whether patients with sepsis have similar long-term mortality and quality-of-life effects is unclear. OBJECTIVE: We performed a systematic review of studies reporting long-term mortality and quality-of-life data (>3 months) in patients with sepsis, severe sepsis, and septic shock using defined search criteria. DESIGN: Systematic review of the literature. INTERVENTIONS: None. MAIN RESULTS: Patients with sepsis showed ongoing mortality up to 2 yrs and beyond after the standard 28-day inhospital mortality end point. Patients with sepsis also had decrements in quality-of-life measures after hospital discharge. Results were consistent across varying severity of illness and different patient populations in different countries, including large and small studies. In addition, these results were consistent within observational and randomized, controlled trials. Study quality was limited by inadequate control groups and poor adjustment for confounding variables. CONCLUSIONS: Patients with sepsis have ongoing mortality beyond short-term end points, and survivors consistently demonstrate impaired quality of life. The use of 28-day mortality as an end point for clinical studies may lead to inaccurate inferences. Both observational and interventional future studies should include longer-term end points to better-understand the natural history of sepsis and the effect of interventions on patient morbidities.

Using evidence, rigorous measurement, and collaboration to eliminate central catheter-associated bloodstream infections.

Healthcare-associated infections are common, costly, and often lethal. Although there is growing pressure to reduce these infections, one project thus far has unprecedented collaboration among many groups at every level of health care. After this project produced a 66% reduction in central catheter-associated bloodstream infections and a median central catheter-associated bloodstream infection rate of zero across >100 intensive care units in Michigan, the Agency for Healthcare Research and Quality awarded a grant to spread this project to ten additional states. A program, called On the CUSP: Stop BSI, was formulated from the Michigan project, and additional funding from the Agency for Healthcare Research and Quality and private philanthropy has positioned the program for implementation state by state across the United States. Furthermore, the program is being implemented throughout Spain and England and is undergoing pilot testing in several hospitals in Peru. The model in this program balances the tension between being scientifically rigorous and feasible. The three main components of the model include translating evidence into practice at the bedside to prevent central catheter-associated bloodstream infections, improving culture and teamwork, and having a data collection system to monitor central catheter-associated bloodstream infections and other variables. If successful, this program will be the first national quality improvement program in the United States with quantifiable and measurable goals.