Asli Ozdas

Investigation of vocal jitter and glottal flow spectrum as possible cues for depression and near-term suicidal risk

Among the many clinical decisions that psychiatrists must make, assessment of a patients risk of committing suicide is definitely among the most important, complex, and demanding. When reviewing his clinical experience, one of the authors observed that successful predictions of suicidality were often based on the patients voice independent of content. The voices of suicidal patients judged to be high-risk near-term exhibited unique qualities, which distinguished them from nonsuicidal patients. We investigated the discriminating power of two excitation-based speech parameters, vocal jitter and glottal flow spectrum, for distinguishing among high-risk near-term suicidal, major depressed, and nonsuicidal patients. Our sample consisted of ten high-risk near-term suicidal patients, ten major depressed patients, and ten nondepressed control subjects. As a result of two sample statistical analyses, mean vocal jitter was found to be a significant discriminator only between suicidal and nondepressed control groups (p<0.05). The slope of the glottal flow spectrum, on the other hand, was a significant discriminator between all three groups (p<0.05). A maximum likelihood classifier, developed by combining the a posteriori probabilities of these two features, yielded correct classification scores of 85% between near-term suicidal patients and nondepressed controls, 90% between depressed patients and nondepressed controls, and 75% between near-term suicidal patients and depressed patients. These preliminary classification results support the hypothesized link between phonation and near-term suicidal risk. However, validation of the proposed measures on a larger sample size is necessary.

The impact of a normoglycemic management protocol on clinical outcomes in the trauma intensive care unit.

BACKGROUND The purpose of this study was to determine if protocol-driven normoglycemic management in trauma patients affected glucose control, ventilator-associated pneumonia, surgical-site infection, and inpatient mortality. METHODS A prospective, consecutive-series, historically controlled study design evaluated protocol-driven normoglycemic management among trauma patients at Vanderbilt University Medical Center. Those mechanically ventilated > or =24 hours and > or =15 years of age were included. A glycemic-control protocol required insulin infusion therapy for glucose >110 mg/dL. Control patients included those who met criteria, were admitted the year preceding protocol implementation, and had hyperglycemia treated at the physicians discretion. RESULTS Eight hundred eighteen patients met study criteria; 383 were managed without protocol; 435 underwent protocol. The protocol group had lower glucose levels 7 of 14 days measured. After admission, both groups had mean daily glucose levels <150 mg/dL. No difference in pneumonia (31.6% vs 34.5%; p = .413), surgical infection (5.0% vs 5.7%; p = .645) or mortality (12.3% vs 13.1%; p = .722) occurred between groups. If one episode of blood glucose level was > or =150 mg/dL (n = 638; 78.0%), outcomes were worse: higher daily glucose levels for 14 days after admission (p < .001), pneumonia rates (35.9% vs 23.3%; p = .002), and mortality (14.6% vs 6.1%; p = .002). One or more days of glucose > or =150 mg/dL had a 2- to 3-fold increase in the odds of death. Protocol use in these patients was not associated with outcome improvement. CONCLUSIONS Protocol-driven management decreased glucose levels 7 of 14 days after admission without outcome change. One or more glucose levels > or =150 mg/dL were associated with worse outcome.

A Computerized Insulin Infusion Titration Protocol Improves Glucose Control With Less Hypoglycemia Compared to a Manual Titration Protocol in a Trauma Intensive Care Unit

BACKGROUND Previous studies reflect reduced morbidity and mortality with intensive blood glucose control in critically ill patients. Unfortunately, implementation of such protocols has proved challenging. This study evaluated the degree of glucose control using manual paper-based vs computer-based insulin protocols in a trauma intensive care unit. METHODS Of 1455 trauma admissions from May 31 to December 31, 2005, a cohort of 552 critically ill patients met study entry criteria. The patients received intensive blood glucose management with IV insulin infusions. Using Fishers exact test, the authors compared patients managed with a computerized protocol vs a paper-based insulin protocol with respect to the portion of glucose values in a target range of 80-110 mg/dL, the incidence of hyperglycemia (> or =150 mg/dL), and the incidence of hypoglycemia (< or =40 mg/dL). RESULTS Three hundred nine patients were managed with a manual paper-based protocol and 243 were managed with a computerized protocol. The total number of blood glucose values across both groups was 21,178. Mean admission glucose was higher in the computer-based protocol group (170 vs 152 mg/dL; p < .001, t-test). Despite this finding by Fishers exact test, glucose control was superior in the computerized group; a higher portion of glucose values was in range 80-110 mg/dL (41.8% vs 34.0%; p < .001), less hyperglycemia occurred (12.8% vs 15.1%; p < .001), and less hypoglycemia occurred (0.2% vs 0.5%; p < .001). CONCLUSIONS A computerized insulin titration protocol improves glucose control by (1) increasing the percentage of glucose values in range, (2) reducing hyperglycemia, and (3) reducing severe hypoglycemia.

Cardiac uncoupling and heart rate variability stratify ICU patients by mortality : A study of 2088 trauma patients

Objective:We have previously shown that cardiac uncoupling (reduced heart rate variability) in the first 24 hours of trauma ICU stay is a robust predictor of mortality. We hypothesize that cardiac uncoupling over the entire ICU stay independently predicts mortality, reveals patterns of injury, and heralds complications. Methods:A total of 2088 trauma ICU patients satisfied the inclusion criteria for this study. Cardiac uncoupling by outcome was compared using the Wilcoxon rank sum test. Risk of death from cardiac uncoupling and covariates (age, ISS, AIS Head Score, total transfusion requirements) was assessed using multivariate logistic regression models at each ICU day. Univariate logistic regression was used to assess risk of death from uncoupling irrespective of covariates at each ICU day. Results:A total of 1325 (63.5%) patients displayed some degree of uncoupling over their ICU stay. The difference in uncoupling between survivors and nonsurvivors is both dramatic and consistent across the entire ICU stay, indicating that the presence of uncoupling is unrelated to the cause of death. However, the magnitude of uncoupling varies by day when data is stratified by cause of death. Conclusions:Cardiac uncoupling: 1) is an independent predictor of death throughout the ICU stay, 2) has a predictive window of 2 to 4 days, and 3) appears to increase in response to inflammation, infection, and multiple organ failure.

Reduced heart rate volatility: An early predictor of death in trauma patients

Objective:To determine if using dense data capture to measure heart rate volatility (standard deviation) measured in 5-minute intervals predicts death. Background:Fundamental approaches to assessing vital signs in the critically ill have changed little since the early 1900s. Our prior work in this area has demonstrated the utility of densely sampled data and, in particular, heart rate volatility over the entire patient stay, for predicting death and prolonged ventilation. Methods:Approximately 120 million heart rate data points were prospectively collected and archived from 1316 trauma ICU patients over 30 months. Data were sampled every 1 to 4 seconds, stored in a relational database, linked to outcome data, and de-identified. HR standard deviation was continuously computed over 5-minute intervals (CVRD, cardiac volatility–related dysfunction). Logistic regression models incorporating age and injury severity score were developed on a test set of patients (N = 923), and prospectively analyzed in a distinct validation set (N = 393) for the first 24 hours of ICU data. Results:Distribution of CVRD varied by survival in the test set. Prospective evaluation of the model in the validation set gave an area in the receiver operating curve of 0.81 with a sensitivity and specificity of 70.1 and 80.0, respectively. CVRD predict death as early as 24 hours in the validation set. Conclusions:CVRD identifies a subgroup of patients with a high probability of dying. Death is predicted within first 24 hours of stay. We hypothesize CVRD is a surrogate for autonomic nervous system dysfunction.

Morbid Obesity is Not a Risk Factor for Mortality in Critically Ill Trauma Patients

BACKGROUND Age, Injury severity score (ISS), hyperglycemia (HGL) at admission, and morbid obesity are known risk factors of poor outcome in trauma patients. Our aim was to which risk factors had the highest risk of death in the critically ill trauma patient. METHODS A Trauma Registry of the American College of Surgeons database retrospective study was performed at our Level I trauma center from January 2000 to October 2004. Inclusion criteria were age >15 years and >or=3 days hospital stay. Data collected included age, gender, and ISS. Groups were divided into nonobese and morbidly obese (MO) (body mass index, BMI >or=40 kg/m2) and into HGL (mean >or=150 mg/dL on initial hospital day) and non-HGL. Primary outcome was 30-day mortality. Differences in mortality and demographic variables between groups were compared using Fishers exact and Wilcoxons rank sum tests. Univariate and multivariate logistic regression was used to assess the relationship of HGL, morbid obesity, age, and injury severity to risk of death. Relationships were assessed using odds ratios (OR) and area under the receiver operator characteristic curve (AUC). RESULTS A total of 1,334 patients met study criteria and 70.5% were male. Demographic means were age 40.3, ISS 25.7, length of stay 13.4, and BMI 27.5. The most common mechanism of injury was motor vehicle collision 55.1%. Overall mortality was 4.7%. Mortality was higher in HGL versus non-HGL (8.7% vs. 3.5%; p < 0.001). Mortality was higher in MO versus nonobese, but not significantly (7.8 vs. 4.6%; not significant [NS] p = 0.222). Univariate logistic regression relationships of death to age OR: 1.031, p < 0.001, AUC +/- SE: 0.639 +/- 0.042; ISS OR: 1.044, p < 0.001, AUC +/- SE: 0.649 +/- 0.039; HGL OR: 2.765, p < 0.001; MO: OR: NS, p = NS, AUC +/- SE: NS. Relationships were similar in a combined multivariate model. CONCLUSION HGL >150 mg/dL on the day of admission is associated with twofold increase in mortality, and an outcome measure should be followed. Morbid obesity (BMI >or=40) is not an independent risk factor for mortality in the critically ill trauma patient.

Volatility: a new vital sign identified using a novel bedside monitoring strategy.

BACKGROUND SIMON (Signal Interpretation and Monitoring) monitors and archives continuous physiologic data in the ICU (HR, BP, CPP, ICP, CI, EDVI, SVO2, SPO2, SVRI, PAP, and CVP). We hypothesized: heart rate (HR) volatility predicts outcome better than measures of central tendency (mean and median). METHODS More than 600 million physiologic data points were archived from 923 patients over 2 years in a level one trauma center. Data were collected every 1 to 4 seconds, stored in a MS-SQL 7.0 relational database, linked to TRACS, and de-identified. Age, gender, race, Injury Severity Score (ISS), and HR statistics were analyzed with respect to outcome (death and ventilator days) using logistic and Poisson regression. RESULTS We analyzed 85 million HR data points, which represent more than 71,000 hours of continuous data capture. Mean HR varied by age, gender and ISS, but did not correlate with death or ventilator days. Measures of volatility (SD, % HR >120) correlated with death and prolonged ventilation. CONCLUSIONS 1) Volatility predicts death better than measures of central tendency. 2) Volatility is a new vital sign that we will apply to other physiologic parameters, and that can only be fully explored using techniques of dense data capture like SIMON. 3) Densely sampled aggregated physiologic data may identify sub-groups of patients requiring new treatment strategies.

Medical decision support using machine learning for early detection of late-onset neonatal sepsis

OBJECTIVE The objective was to develop non-invasive predictive models for late-onset neonatal sepsis from off-the-shelf medical data and electronic medical records (EMR). DESIGN The data used in this study are from 299 infants admitted to the neonatal intensive care unit in the Monroe Carell Jr. Childrens Hospital at Vanderbilt and evaluated for late-onset sepsis. Gold standard diagnostic labels (sepsis negative, culture positive sepsis, culture negative/clinical sepsis) were assigned based on all the laboratory, clinical and microbiology data available in EMR. Only data that were available up to 12 h after phlebotomy for blood culture testing were used to build predictive models using machine learning (ML) algorithms. MEASUREMENT We compared sensitivity, specificity, positive predictive value and negative predictive value of sepsis treatment of physicians with the predictions of models generated by ML algorithms. RESULTS The treatment sensitivity of all the nine ML algorithms and specificity of eight out of the nine ML algorithms tested exceeded that of the physician when culture-negative sepsis was included. When culture-negative sepsis was excluded both sensitivity and specificity exceeded that of the physician for all the ML algorithms. The top three predictive variables were the hematocrit or packed cell volume, chorioamnionitis and respiratory rate. CONCLUSIONS Predictive models developed from off-the-shelf and EMR data using ML algorithms exceeded the treatment sensitivity and treatment specificity of clinicians. A prospective study is warranted to assess the clinical utility of the ML algorithms in improving the accuracy of antibiotic use in the management of neonatal sepsis.

Social, organizational, and contextual characteristics of clinical decision support systems for intensive insulin therapy: A literature review and case study

INTRODUCTION Evaluations of computerized clinical decision support systems (CDSS) typically focus on clinical performance changes and do not include social, organizational, and contextual characteristics explaining use and effectiveness. Studies of CDSS for intensive insulin therapy (IIT) are no exception, and the literature lacks an understanding of effective computer-based IIT implementation and operation. RESULTS This paper presents (1) a literature review of computer-based IIT evaluations through the lens of institutional theory, a discipline from sociology and organization studies, to demonstrate the inconsistent reporting of workflow and care process execution and (2) a single-site case study to illustrate how computer-based IIT requires substantial organizational change and creates additional complexity with unintended consequences including error. DISCUSSION Computer-based IIT requires organizational commitment and attention to site-specific technology, workflow, and care processes to achieve intensive insulin therapy goals. The complex interaction between clinicians, blood glucose testing devices, and CDSS may contribute to workflow inefficiency and error. Evaluations rarely focus on the perspective of nurses, the primary users of computer-based IIT whose knowledge can potentially lead to process and care improvements. CONCLUSION This paper addresses a gap in the literature concerning the social, organizational, and contextual characteristics of CDSS in general and for intensive insulin therapy specifically. Additionally, this paper identifies areas for future research to define optimal computer-based IIT process execution: the frequency and effect of manual data entry error of blood glucose values, the frequency and effect of nurse overrides of CDSS insulin dosing recommendations, and comprehensive ethnographic study of CDSS for IIT.

A Simple Non-physiological Artifact Filter for Invasive Arterial Blood Pressure Monitoring: a Study of 1852 Trauma ICU Patients

Invasive arterial blood pressure (BP) is a vital sign in hemodynamic monitoring of trauma intensive care unit (ICU) patients. Continuous BP analysis can potentially provide additional information about patient status, predict morbidity and mortality, and automatically populate electronic nurse charting systems than intermittent monitoring. Challenges to routine application in the ICU include integration of complex physiological data collection systems, artifacts, missing data, and the various clinical interventions that may temporarily corrupt the BP signal. We have developed and previously described SIMON (signal interpretation and monitoring), a physiological data collection system in the Trauma ICU at Vanderbilt University. In order to extract useful information from continuous arterial line BP monitoring, it is necessary to remove non-physiological artifacts. In this setting, potential artifacts appear to be caused by resonance, over-damping, and data transmission. We designed a simple filter to identify various sources of non-physiological artifacts using statistical signal processing techniques. We implemented the filter to arterial invasive BP signals of 1852 trauma patients throughout their length of ICU stay. After filtering, the power of BP measures to predict hospital death was enhanced. Therefore, we concluded that our strategy of removing non-physiological artifact was simple, fast and useful for an accurate assessment of BP measures in trauma patients