Eliezer Bose

Using Supervised Machine Learning to Classify Real Alerts and Artifact in Online Multisignal Vital Sign Monitoring Data.

Objective: The use of machine-learning algorithms to classify alerts as real or artifacts in online noninvasive vital sign data streams to reduce alarm fatigue and missed true instability. Design: Observational cohort study. Setting: Twenty-four–bed trauma step-down unit. Patients: Two thousand one hundred fifty-three patients. Intervention: Noninvasive vital sign monitoring data (heart rate, respiratory rate, peripheral oximetry) recorded on all admissions at 1/20 Hz, and noninvasive blood pressure less frequently, and partitioned data into training/validation (294 admissions; 22,980 monitoring hours) and test sets (2,057 admissions; 156,177 monitoring hours). Alerts were vital sign deviations beyond stability thresholds. A four-member expert committee annotated a subset of alerts (576 in training/validation set, 397 in test set) as real or artifact selected by active learning, upon which we trained machine-learning algorithms. The best model was evaluated on test set alerts to enact online alert classification over time. Measurements and Main Results: The Random Forest model discriminated between real and artifact as the alerts evolved online in the test set with area under the curve performance of 0.79 (95% CI, 0.67–0.93) for peripheral oximetry at the instant the vital sign first crossed threshold and increased to 0.87 (95% CI, 0.71–0.95) at 3 minutes into the alerting period. Blood pressure area under the curve started at 0.77 (95% CI, 0.64–0.95) and increased to 0.87 (95% CI, 0.71–0.98), whereas respiratory rate area under the curve started at 0.85 (95% CI, 0.77–0.95) and increased to 0.97 (95% CI, 0.94–1.00). Heart rate alerts were too few for model development. Conclusions: Machine-learning models can discern clinically relevant peripheral oximetry, blood pressure, and respiratory rate alerts from artifacts in an online monitoring dataset (area under the curve > 0.87).

The Interface Between Monitoring and Physiology at the Bedside

Hemodynamic instability as a clinical state represents either a perfusion failure with clinical manifestations of circulatory shock or heart failure or 1 or more out-of-threshold hemodynamic monitoring values, which may not necessarily be pathologic. Different types of causes of circulatory shock require different types of treatment modalities, making these distinctions important. Diagnostic approaches or therapies based on data derived from hemodynamic monitoring assume that specific patterns of derangements reflect specific disease processes, which respond to appropriate interventions. Hemodynamic monitoring at the bedside improves patient outcomes when used to make treatment decisions at the right time for patients experiencing hemodynamic instability.

Monitoring cardiorespiratory instability: Current approaches and implications for nursing practice

Unrecognised in-hospital cardiorespiratory instability (CRI) risks adverse patient outcomes. Although step down unit (SDU) patients have continuous non-invasive physiologic monitoring of vital signs and a ratio of one nurse to four to six patients, detection of CRI is still suboptimal. Telemedicine provides additional surveillance but, due to high costs and unclear investment returns, is not routinely used in SDUs. Rapid response teams have been tested as possible approaches to support CRI patients outside the intensive care unit with mixed outcomes. Technology-enabled early warning scores, though rigorously studied, may not detect subtle instability. Efforts to utilise nursing intuition as a means to promote early identification of CRI have been explored, but the problem still persists. Monitoring systems hold promise, but nursing surveillance remains the key to reliable early detection and recognition. Research directed towards improving nursing surveillance and facilitating decision-making is needed to ensure safe patient outcomes and prevent CRI.

Temporal distribution of instability events in continuously monitored step-down unit patients: Implications for Rapid Response Systems

AIM Medical Emergency Teams (MET) activations are more frequent during daytime and weekdays, but whether due to greater patient instability, proximity from admission time, or caregiver concentration is unclear. We sought to determine if instability events, when they occurred, varied in their temporal distribution. METHODS Monitoring data were recorded (frequency 1/20Hz) in 634 SDU patients (41,635 monitoring hours). Vital sign excursion beyond our MET trigger thresholds defined alerts. The resultant 1399 alerts from 216 patients were tallied according to clock hour and time elapsed since admission. We fit patient ID (n=216), clock hour, time since SDU admission, and alert present into a null model and three mixed effect logistic regression models: clock hour, hours elapsed since admission, and both clock hour and time elapsed since admission as fixed effect covariates. We performed likelihood ratio tests on these models to assess if, among all alerts, there were proportionally more alerts for any given clock hour, or proximity to admission time. RESULTS Only time elapsed since admission (p<0.001), and not clock hour adjusting for time elapsed since admission (p=0.885), was significant for temporal disproportion. Results were unchanged if the first 24h following admission were excluded from the models. CONCLUSION Although instability alerts are distributed most frequently within 24h after SDU admission in unstable patients, they are otherwise not more likely to distribute proportionally more frequently during certain clock hours. If MET utilization peaks do not coincide with admission time peaks, other variables contributing to unrecognized instability should be explored.

Vector Autoregressive Models and Granger Causality in Time Series Analysis in Nursing Research: Dynamic Changes Among Vital Signs Prior to Cardiorespiratory Instability Events as an Example

Background Patients undergoing continuous vital sign monitoring (heart rate [HR], respiratory rate [RR], pulse oximetry [SpO2]) in real time display interrelated vital sign changes during situations of physiological stress. Patterns in this physiological cross-talk could portend impending cardiorespiratory instability (CRI). Vector autoregressive (VAR) modeling with Granger causality tests is one of the most flexible ways to elucidate underlying causal mechanisms in time series data. Purpose The purpose of this article is to illustrate the development of patient-specific VAR models using vital sign time series data in a sample of acutely ill, monitored, step-down unit patients and determine their Granger causal dynamics prior to onset of an incident CRI. Approach CRI was defined as vital signs beyond stipulated normality thresholds (HR = 40–140/minute, RR = 8–36/minute, SpO2 < 85%) and persisting for 3 minutes within a 5-minute moving window (60% of the duration of the window). A 6-hour time segment prior to onset of first CRI was chosen for time series modeling in 20 patients using a six-step procedure: (a) the uniform time series for each vital sign was assessed for stationarity, (b) appropriate lag was determined using a lag-length selection criteria, (c) the VAR model was constructed, (d) residual autocorrelation was assessed with the Lagrange Multiplier test, (e) stability of the VAR system was checked, and (f) Granger causality was evaluated in the final stable model. Results The primary cause of incident CRI was low SpO2 (60% of cases), followed by out-of-range RR (30%) and HR (10%). Granger causality testing revealed that change in RR caused change in HR (21%; i.e., RR changed before HR changed) more often than change in HR causing change in RR (15%). Similarly, changes in RR caused changes in SpO2 (15%) more often than changes in SpO2 caused changes in RR (9%). For HR and SpO2, changes in HR causing changes in SpO2 and changes in SpO2 causing changes in HR occurred with equal frequency (18%). Discussion Within this sample of acutely ill patients who experienced a CRI event, VAR modeling indicated that RR changes tend to occur before changes in HR and SpO2. These findings suggest that contextual assessment of RR changes as the earliest sign of CRI is warranted. Use of VAR modeling may be helpful in other nursing research applications based on time series data.

Pulmonary Artery Catheterization

Pulmonary artery catheterization has been available as a clinical tool since 1964 and quickly became the gold standard for hemodynamic monitoring of critically ill patients. The pulmonary artery catheter allows simultaneous monitoring of continuously mixed venous O2 saturation (SvO2), cardiac output (CO), right atrial pressure (Pra), pulmonary arterial pressure (Ppa), right ventricular ejection fraction (RVef), and, by mathematical derivation, right ventricular end-diastolic volume (EDV) and, by intermittent distal balloon occlusion, pulmonary artery occlusion pressure (Ppao). No other cardiovascular monitoring device shares this pluripotential presence. While there is no debate about the measurements a pulmonary artery catheter can offer, there is controversy surrounding the benefits of this device. Nonetheless, it is clearly a useful hemodynamic monitoring tool in selected patients. However, the decision to place this invasive device must be based on the specific information sought.

Risk for Cardiorespiratory Instability Following Transfer to a Monitored Step-Down Unit

BACKGROUND: Hospitalized patients who develop at least one instance of cardiorespiratory instability (CRI) have poorer outcomes. We sought to describe the admission characteristics, drivers, and time to onset of initial CRI events in monitored step-down unit (SDU) patients. METHODS: Admission characteristics and continuous monitoring data (frequency 1/20 Hz) were recorded in 307 subjects. Vital sign deviations beyond local instability trigger threshold criteria, with a tolerance of 40 s and cumulative duration of 4 of 5 min, were classified as CRI events. The CRI driver was defined as the first vital sign to cross a threshold and meet persistence criteria. Time to onset of initial CRI was the number of days from SDU admission to initial CRI, and duration was length of the initial CRI epoch. RESULTS: Subjects transferred to the SDU from units with higher monitoring capability were more likely to develop CRI (CRI n = 133 [44%] vs no CRI n = 174 [31%] P = .042). Time to onset varied according to the CRI driver. Subjects with at least one CRI event had a longer hospital stay (CRI 11.3 ± 10.2 d vs no CRI 7.8 ± 9.2 d, P < .001) and SDU stay (CRI 6.1 ± 4.9 d vs no CRI 3.5 ± 2.9 d, P < .001). First events were more often due to SpO2, whereas breathing frequency was the most common driver of all CRI. CONCLUSIONS: Initial CRI most commonly occurred due to SpO2 and was associated with prolonged SDU and hospital stay. Findings suggest the need for clinicians to more closely monitor SDU patients transferred from an ICU and parameters (SpO2, breathing frequency) that more commonly precede CRI events.

Semi automated adjudication of vital sign alerts in step-down units

Machine Learning (ML) has shown predictive utility in analyzing vital sign (VS) data collected from physiologically unstable monitored patients. Training an ML model usually requires sizable amounts of labeled ground-truth data typically obtained via laborious manual chart reviews by expert clinicians.

Machine learning can classify vital sign alerts as real or artifact in online continuous monitoring data

Alarm hazards continue to be the top patient safety concern of 2015. Machine learning (ML) can be used to classify patterns in monitoring data to differentiate real alerts from artifact.

150: CHARACTERIZING INITIAL CARDIORESPIRATORY INSTABILITY PATTERNS IN MONITORED STEP-DOWN UNIT PATIENTS.

Crit Care Med 2015 • Volume 43 • Number 12 (Suppl.) Median lowest arterial oxygen saturation was 92% with apneic oxygenation versus 90% with usual care (95% confidence interval for the difference -1.6% to 7.4%; P = .16). There was no difference between apneic oxygenation and usual care in incidence of oxygen saturation < 90% (44.7% versus 47.2%; P = .87), oxygen saturation < 80% (15.8% versus 25.0%; P = .22), or decrease in oxygen saturation > 3% (53.9% versus 55.6%; P = .87). Duration of mechanical ventilation, ICU length of stay, and in-hospital mortality were similar between study groups. Conclusions: Apneic oxygenation does not significantly increase lowest arterial oxygen saturation during endotracheal intubation of critically ill patients compared to usual care. These findings do not support the routine use of apneic oxygenation during endotracheal intubation of critically ill adults. (clinicaltrials. gov Identifier: NCT02051816).