- Email: [email protected]
Prediction of stable clinical conditions after elective coronary bypass surgery
346 died (1.97 ± 1.87 vs 3.46 ± 1.7, P = .03). Other Poincare plot parameters as well as standard vital signs for HR, SBP, SpO2, SI, and PP were not found to be significant between groups. Conclusions: Poincare plot analysis of the ECG is consistent with a high frequency parasympathetic dominance in our cohort of patients that died. This is reflected by significant differences in the Poincare plot descriptors SD1, SD2, and SD2/SD1. These data suggest that simple time-domain analyses of prehospital ECGs may provide additional information on the condition of the patient when other vital signs are normal and GCS scores are unattainable, such as in remote triage situations. doi:10.1016/j.jcrc.2007.10.026
Prediction of stable clinical conditions after elective coronary bypass surgery Kristien Van Loon a, Fabian Guiza b, Geert Meyfroidt c, Jean-Marie Aerts a, Hendrik Blockeel b, Greet Van den Berghe c, Daniel Berckmans a a Division Measure, Model &Manage Bioresponses, Katholieke Universiteit Leuven b Department of Computer Sciences, Katholieke Universiteit Leuven c Department of Intensive Care Medicine, University Hospital Gasthuisberg Objectives: In cardiac surgery, optimal use of intensive care unit (ICU) and operating room (OR) capacity requires the prediction of future availability of ICU beds. Minimal conditions required to start the weaning from mechanical ventilation in these patients are hemodynamic and respiratory stability, absence of bleeding and normothermia. This preliminary study is a test of 5 advanced data analysis and trend analysis tools, combined with a Gaussian Processes Classifier, to examine the influence of the trends of standard measured clinical variables in the first 4 hours after ICU admission, on the timeframe in which the minimal clinical conditions to start weaning of the mechanical ventilation are reached. Methods: For 103 patients, measurements of 14 physiologic variables obtained from a Patient Data Management System (Metavision, iMD-Soft) were used as inputs for developing the mathematical models in this study. Most variables (heart rate, systolic, diastolic and mean arterial blood pressure, systolic, diastolic and mean pulmonary artery pressure, PEEP, FiO2, blood and peripheral temperature) were stored every minute from the patient monitor or respirator. Blood loss was recorded manually by the nurses, approximately once every hour. Blood gas analysis was performed at least once every 4 hours (PO2, lactate). Four-hour time intervals of these measurements were used to build several timeseries models (multivariate autoregressive [MAR], multiple inputs/ multiple outputs [ARX], cepstral coefficients computed directly from the time-series [CEP] and from the ARX parameters [CEPARX], signal average). The actual prediction was done by using the parameters of these models as inputs for a Machine Learning algorithm (Gaussian Process Classifier). Including parameters of different types of time-series models as a representation of the time-varying signals, we incorporate knowledge of the dynamical behavior of the patients. We believe that this will lead to better predictive performances. The Gaussian Processes Classifier assigned for each patient a probability of belonging to each of the following classes: the patient will meet the clinical criteria within the first 8 hours after admission, between 8 and 16 hours, between
Abstracts Table 1
Results of the gaussian Process Classifier
Signal average MAR ARX CEP CEPARX
Before 8 h
8-16 h
16-24 h
After 24 h
0.814 0.637 0.601 0.662 0.729
0.565 0.586 0.566 0.724 0.635
0.619 0.545 0.667 0.617 0.847
0.671 0.671 0.620 0.658 0.700
16 and 24 hours and after 24 hours. These predictions were made from the analysis of data from the first 4 hours of ICU stay. Results: Each table entry in Table 1 corresponds to the aROC of the corresponding Gaussian Process Classifier through leave-one-out cross-validation. Conclusions: In general, it is clear that from analyzing the trend and variability of the selected parameters, it is hard to make predictions with good discrimination. For each prediction interval there was at least one model with an aROC above 0.7. When examining the results in detail, one can see that the further ahead the patient meets the minimal conditions to start weaning of the mechanical ventilation, different inputs will become more predictive for the task, for example the CEPARX-input classifier has the best performance for the 16 to 24h and theN 24-hour intervals, whereas the signal average-input results in a high performance for the first 8-hour interval. Future research will try to improve these results by integrating different combinations of model parameters or all model parameters at once, in the Gaussian Process Classifier. doi:10.1016/j.jcrc.2007.10.027
Spatial arrangements of drug-metabolizing enzymes influence metabolite formation Tai ning Lam a,b, C. Anthony Hunt a,b a University of California, San Francisco b The Program in Pharmaceutical Sciences and Pharmacogenomics Objectives: Xenobiotic transport is modeled traditionally using inductive, equation-based methods. New, discrete event, agentbased methods will enable exploration of emergent phenomena and the spatial and probabilistic nature of posited mechanisms at selected levels of detail. The same model can be used (reused) to study any number of compounds, real or hypothetical. The objective of these projects is to demonstrate an assembly of software components representing transporters, enzymes, spaces, and drugs for use in studying spatially influenced, metabolic phenomena. Our synthetic, object-oriented, computational analogue represents an in vitro, cellular, drug transport and metabolism system. Methods: We use the synthetic modeling method to construct an object-oriented, computational analogue of an in vitro drug transport system of interest. The synthetic modeling method offers advantages for understanding and unraveling the complexities of cellular drug metabolism (PMIDs: 17051440, 16435171, and 16135397). Mechanisms are encoded as algorithms. Software components representing in vitro counterparts are plugged together within the Swarm modeling and simulation framework, and operated in ways that represent drug transport across cellular barriers. The model's five spaces represent in vitro counterparts: apical extracellular space, apical membrane, cytoplasm, basal membrane, and basal extracellular space. Transporters are located
Prediction of stable clinical conditions after elective coronary bypass surgery Kristien Van Loon a, Fabian Guiza b, Geert Meyfroidt c, Jean-Marie Aerts a, Hendrik Blockeel b, Greet Van den Berghe c, Daniel Berckmans a a Division Measure, Model &Manage Bioresponses, Katholieke Universiteit Leuven b Department of Computer Sciences, Katholieke Universiteit Leuven c Department of Intensive Care Medicine, University Hospital Gasthuisberg Objectives: In cardiac surgery, optimal use of intensive care unit (ICU) and operating room (OR) capacity requires the prediction of future availability of ICU beds. Minimal conditions required to start the weaning from mechanical ventilation in these patients are hemodynamic and respiratory stability, absence of bleeding and normothermia. This preliminary study is a test of 5 advanced data analysis and trend analysis tools, combined with a Gaussian Processes Classifier, to examine the influence of the trends of standard measured clinical variables in the first 4 hours after ICU admission, on the timeframe in which the minimal clinical conditions to start weaning of the mechanical ventilation are reached. Methods: For 103 patients, measurements of 14 physiologic variables obtained from a Patient Data Management System (Metavision, iMD-Soft) were used as inputs for developing the mathematical models in this study. Most variables (heart rate, systolic, diastolic and mean arterial blood pressure, systolic, diastolic and mean pulmonary artery pressure, PEEP, FiO2, blood and peripheral temperature) were stored every minute from the patient monitor or respirator. Blood loss was recorded manually by the nurses, approximately once every hour. Blood gas analysis was performed at least once every 4 hours (PO2, lactate). Four-hour time intervals of these measurements were used to build several timeseries models (multivariate autoregressive [MAR], multiple inputs/ multiple outputs [ARX], cepstral coefficients computed directly from the time-series [CEP] and from the ARX parameters [CEPARX], signal average). The actual prediction was done by using the parameters of these models as inputs for a Machine Learning algorithm (Gaussian Process Classifier). Including parameters of different types of time-series models as a representation of the time-varying signals, we incorporate knowledge of the dynamical behavior of the patients. We believe that this will lead to better predictive performances. The Gaussian Processes Classifier assigned for each patient a probability of belonging to each of the following classes: the patient will meet the clinical criteria within the first 8 hours after admission, between 8 and 16 hours, between
Abstracts Table 1
Results of the gaussian Process Classifier
Signal average MAR ARX CEP CEPARX
Before 8 h
8-16 h
16-24 h
After 24 h
0.814 0.637 0.601 0.662 0.729
0.565 0.586 0.566 0.724 0.635
0.619 0.545 0.667 0.617 0.847
0.671 0.671 0.620 0.658 0.700
16 and 24 hours and after 24 hours. These predictions were made from the analysis of data from the first 4 hours of ICU stay. Results: Each table entry in Table 1 corresponds to the aROC of the corresponding Gaussian Process Classifier through leave-one-out cross-validation. Conclusions: In general, it is clear that from analyzing the trend and variability of the selected parameters, it is hard to make predictions with good discrimination. For each prediction interval there was at least one model with an aROC above 0.7. When examining the results in detail, one can see that the further ahead the patient meets the minimal conditions to start weaning of the mechanical ventilation, different inputs will become more predictive for the task, for example the CEPARX-input classifier has the best performance for the 16 to 24h and theN 24-hour intervals, whereas the signal average-input results in a high performance for the first 8-hour interval. Future research will try to improve these results by integrating different combinations of model parameters or all model parameters at once, in the Gaussian Process Classifier. doi:10.1016/j.jcrc.2007.10.027
Spatial arrangements of drug-metabolizing enzymes influence metabolite formation Tai ning Lam a,b, C. Anthony Hunt a,b a University of California, San Francisco b The Program in Pharmaceutical Sciences and Pharmacogenomics Objectives: Xenobiotic transport is modeled traditionally using inductive, equation-based methods. New, discrete event, agentbased methods will enable exploration of emergent phenomena and the spatial and probabilistic nature of posited mechanisms at selected levels of detail. The same model can be used (reused) to study any number of compounds, real or hypothetical. The objective of these projects is to demonstrate an assembly of software components representing transporters, enzymes, spaces, and drugs for use in studying spatially influenced, metabolic phenomena. Our synthetic, object-oriented, computational analogue represents an in vitro, cellular, drug transport and metabolism system. Methods: We use the synthetic modeling method to construct an object-oriented, computational analogue of an in vitro drug transport system of interest. The synthetic modeling method offers advantages for understanding and unraveling the complexities of cellular drug metabolism (PMIDs: 17051440, 16435171, and 16135397). Mechanisms are encoded as algorithms. Software components representing in vitro counterparts are plugged together within the Swarm modeling and simulation framework, and operated in ways that represent drug transport across cellular barriers. The model's five spaces represent in vitro counterparts: apical extracellular space, apical membrane, cytoplasm, basal membrane, and basal extracellular space. Transporters are located