The triple variable index combines information generated over time from common monitoring variables to identify patients expressing distinct patterns of intraoperative physiology.

BACKGROUND: Mean arterial pressure (MAP), bispectral index (BIS), and minimum alveolar concentration (MAC) represent valuable, yet dynamic intraoperative monitoring variables. They provide information related to poor outcomes when considered together, however their collective behavior across time has not been characterized. METHODS: We have developed the Triple Variable Index (TVI), a composite variable representing the sum of z-scores from MAP, BIS, and MAC values that occur together during surgery. We generated a TVI expression profile, defined as the sequential TVI values expressed across time, for each surgery where concurrent MAP, BIS, and MAC monitoring occurred in an adult patient (≥18 years) at the University of Pittsburgh Medical Center between January and July 2014 (n = 5296). Patterns of TVI expression were identified using k-means clustering and compared across numerous patient, procedure, and outcome characteristics. TVI and the triple low state were compared as prediction models for 30-day postoperative mortality. RESULTS: The median frequency MAP, BIS, and MAC were recorded was one measurement every 3, 5, and 5 min. Three expression patterns were identified: elevated, mixed, and depressed. The elevated pattern displayed the highest average MAP, BIS, and MAC values (86.5 mmHg, 45.3, and 0.98, respectively), while the depressed pattern displayed the lowest values (76.6 mmHg, 38.0, 0.66). Patterns (elevated, mixed, depressed) were distinct across the following characteristics: average patient age (52, 53, 54 years), American Society of Anesthesiologists Physical Status 4 (6.7, 16.1, 27.3%) and 5 (0.1, 0.6, 1.6%) categories, cardiac (2.2, 6.5, 16.1%) and emergent (5.8, 10.5, 12.8%) surgery, cardiopulmonary bypass use (0.3, 2.6, 9.8%), intraoperative medication administration including etomidate (3.0, 7.3, 12.6%), hydromorphone (47.6, 26.3, 25.2%), ketamine (11.2, 4.6, 3.0%), dexmedetomidine (18.4, 16.6, 13.6%), phenylephrine (74.0, 74.8, 83.0), epinephrine (2.0, 6.0, 18.0%), norepinephrine (2.4, 7.5, 21.2%), vasopressin (3.4, 7.6, 21.0%), succinylcholine (74.0, 69.0, 61.9%), intraoperative hypotension (28.8, 33.0, 52.3%) and the triple low state (9.4, 30.3, 80.0%) exposure, and 30-day postoperative mortality (0.8, 2.7, 5.6%). TVI was a better predictor of patients that died or survived in the 30 days following surgery compared to cumulative triple low state exposure (AUC 0.68 versus 0.62, p < 0.05). CONCLUSIONS: Surgeries that share similar patterns of TVI expression display distinct patient, procedure, and outcome characteristics.

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).