- Open Access
Altering physiological networks using drugs: steps towards personalized physiology
© Grossman et al.; licensee BioMed Central Ltd. 2013
- Published: 7 May 2013
The rise of personalized medicine has reminded us that each patient must be treated as an individual. One factor in making treatment decisions is the physiological state of each patient, but definitions of relevant states and methods to visualize state-related physiologic changes are scarce. We constructed correlation networks from physiologic data to demonstrate changes associated with pressor use in the intensive care unit.
We collected 29 physiological variables at one-minute intervals from nineteen trauma patients in the intensive care unit of an academic hospital and grouped each minute of data as receiving or not receiving pressors. For each group we constructed Spearman correlation networks of pairs of physiologic variables. To visualize drug-associated changes we split the networks into three components: an unchanging network, a network of connections with changing correlation sign, and a network of connections only present in one group.
Out of a possible 406 connections between the 29 physiological measures, 64, 39, and 48 were present in each of the three component networks. The static network confirms expected physiological relationships while the network of associations with changed correlation sign suggests putative changes due to the drugs. The network of associations present only with pressors suggests new relationships that could be worthy of study.
We demonstrated that visualizing physiological relationships using correlation networks provides insight into underlying physiologic states while also showing that many of these relationships change when the state is defined by the presence of drugs. This method applied to targeted experiments could change the way critical care patients are monitored and treated.
- Mean Arterial Pressure
- Intensive Care Unit Patient
- Injury Severity Score
- Clinical Decision Support System
- Correlation Network
A major goal of medical physiology has long been to fully specify a model for how molecules and systems in the body interact and react to changes in environment. Thus far the most widely known and complete model of this sort was produced by Guyton , systematically and mathematically detailing 354 independent elements that together regulate cardiac output. A major and necessary assumption inherent to this work and those that followed is the capability of a single set of equations - a "physiology" - to describe the ways in which systems and molecules interact. While the concept of a single unified set of relationships to define physiology has certainly served both science and medicine extremely well at the population level, we now know (and, some would argue, have always known) this view to be incorrect on the level of individuals . Regardless, the literature is full of population-based studies that associate single physiological measures with outcome. For example, many links have been made between the variability of such measures - mostly heart rate variability - to various adverse outcomes in a population of trauma patients [3–7]. While valuable in advancing medical knowledge the limits of population-based medicine are becoming clearer. The rise of personalized medicine via genomics, personalized dosing algorithms , and genome wide association studies showing individual risk of disease are two obvious examples of such limitations of population-based medicine [9, 10].
Applying the concept of patient specificity to physiology using multivariate data analysis and visualization tools has recently found some traction. Nelson et al. , using cerebral microdialysis, showed that patients exhibited individualistic physiological patterns that were also reflective of shock, but they could not find common patterns between patients. Rixen et al  devised a physiological state classifier based on seventeen variables collected once daily, enabling prediction of patients' deaths. Peleg et al  devised a biological simulator based on multivariate physiological measurements. Additionally, several patient-specific monitoring and learning algorithms have been developed to improve detection of severe and acute pathologies  and as triage assistance and clinical decision support systems [15, 16]. Our own previous work has shown the dependence of clinical assessment on data collection frequency  and the benefits of combining multivariate physiological data with continuous monitoring for managing secondary brain injury and gaining clinical insight not possible without such methods . We have also shown that physiological state as defined by agglomerative hierarchical clustering leads to identification of complex metabolic patterns naturally sorted for outcome measures including death and multiple organ failure .
One way to visualize and examine the interrelationships of physiological measurements and how they change is by using networks; this approach has become the basis for large efforts to understand the relationships between genetics and disease . Networks constructed using correlated pairs of variables have found uses in many areas, including integrative biology , metabolic networks , gene expression networks , and even the structure of the Internet . Furthermore, we have shown previously that physiological correlation networks changed based on whether patients contracted an infection during their stays in the intensive care unit (ICU) . Because physiology depends on disease state, we hypothesized that the effects of another common modifier of physiology, drugs, would be apparent when visualizing physiological correlation networks and comparing the cases of the drugs being given or not. Even if drugs are used that primarily affect a single molecule, many drugs potentially have a large set of off-target or side effects making broader changes in physiology that we believe could be visualized using clinical physiological networks.
To demonstrate this we examined the effects of administering pressors to patients in the ICU by collecting high-frequency clinical physiological measurements and using them to create correlation networks during times when pressors were given during the normal course of treatment and separately for times when they were not. We then show that the structure of these networks changed when pressors were administered, with changes falling into two categories: the sign of the correlation changed and correlations that appeared or disappeared when pressors were given. We also provide an example demonstrating that both patient individuality and drug differences can visibly affect physiological relationships. These methods could open the door to improved treatment strategies for critical care patients, and may even lead to clinically-relevant models of personalized physiology gleaned from the data that are normally collected in every ICU.
Physiological data collection
This study was approved by and conducted under supervision of the Committee on Human Research at the University of California San Francisco. We collected physiological data from nineteen severely injured poly-trauma patients from the intensive care unit at San Francisco General Hospital between May 2004 and June 2005, as previously described [19, 25]. Briefly, patients were included based on admission to the ICU and a requirement for ongoing resuscitation.
Types of data collected and their abbreviations.
Arterial O2/CO2 partial pressure
Mean arterial blood pressure
Muscle oxygen temperature
Muscle oxygen tension
Oxygen saturation percentage
Fraction of inspired oxygen
Muscle lactate concentration
Muscle glucose concentration
Muscle glutamate concentration
Muscle pyruvate concentration
Muscle lactate/pyruvate ratio
Bedside finger-stick glucose reading
Mechanical lung compliance
Positive end expiratory pressure
Volume of air per minute
Central venous pressure
Blood urea nitrogen
Additional data preparation has been described elsewhere . Briefly, data were collected, de-identified, integrated into a single database, and hand-curated. Muscle microdialysis data were interpolated, while all other measures were retained unaltered. All collected variables and their abbreviations are summarized in Table 1.
Univariate data analysis
We grouped our minute-by-minute data from all patients according to whether pressors were being given and used Wilcoxon-Mann-Whitney (Mann-Whitney U) tests to determine whether the univariate physiological measures came from the same underlying population in those two cases. Bonferroni correction was used to correct for multiple comparisons with significance set at p < 0.05.
Correlation network construction
We grouped all of our minute-by-minute data, regardless of patient, according to whether pressors were being administered during that minute, and calculated the Spearman correlation coefficient for each possible pairing of physiological measurements. We chose rank instead of linear correlation to directly show the monotonic relationships between variables without the need for them to be linear. To maximize data utility, we calculated the correlation for each pair of measurements using all those time points containing data for both variables of that pair. While this can yield a correlation matrix that is not positive definite, we are concerned with the coefficients themselves rather than manipulating the resulting matrix. We calculated the p-value for each correlation using Student's t-test on the standard z-transform and used Bonferroni correction for multiple comparisons. We considered a connection (edge) between two physiological measures (nodes) to exist if the correlation between them was statistically significant.
To assist in comparing the physiological correlation network when pressors were given ("pressors on") to the one generated when they were not ("pressors off"), we visualized the union of these two networks into three interesting and distinct cases. The first, "static", network contains those connections present in both the "pressors on" and "pressors off" networks, and with the same sign of directionality in both cases. The second, "either-or", network consists of connections present in either of the two networks, but not in both. The third, "sign-change", network consists of connections present both with and without pressors being given, but with a different sign of directionality in the two cases (i.e. positive correlation in one network and negative correlation in the other). Rather than simply display these networks as tables of correlation values we visualize them as a network to lend clarity. Each physiological variable becomes a node and each significant correlation becomes a line connecting the two relevant nodes. This way the differences in correlation direction and strength can easily be seen across the different conditions.
Individual patient comparison
where β i is the slope of the linear regression of the sample Yi on Xi, r i 2 is the Pearson correlation coefficient associated with that same regression, and σ Xi indicates the sample standard deviation of X i . The resulting statistic follows a t distribution with N1 + N2 - 4 degrees of freedom, where N1 and N2 are the numbers of samples in the first and second groups, respectively. We compared the regression slopes across the cases of receiving and not receiving pressors within two individual patients. We also compared the regression slopes of two individuals at times when they were being given pressors to times they were not.
Gun shot wound
Pedestrians vs. auto
Mean ± s.d.
38 ± 18
18 - 72
ICU length of stay (days)
24 ± 21
1 - 78
Hospital length of stay (days)
40 ± 42
1 - 172
28 ± 10
16 - 50
Injury sites (maximum AIS)
# of patients
Effect of pressors on individual measures
The core physiological network response to pressors
We next looked at the 406 possible pairings of the 29 physiological variables. We comprehensively computed Spearman correlation coefficients for each pair of variables, using all data points when both variables were simultaneously measured. We computed pair-wise correlations separately using those data points associated with both the absence and presence of pressors. To generate a more complete picture of physiological changes concurrent with administering pressors, we used a correlation network-based analysis to study the interrelationships between our 29 clinical measurements. We visualized the differences in these correlational networks in three ways to highlight the similarities and differences in the physiological relationships when pressors were given. The "static" network is composed of those relationships that exist and have the same sign of directionality in both conditions and contains 64 out of a possible 406 edges (Additional file 2). These relationships should therefore include known physiology and serve as useful controls. For example, hemoglobin and hematocrit are strongly positively correlated both with and without pressors (ρ = 0.98, p < 10-20 in both cases). The three independent glucose measurements are also all positively correlated (five of six ρ > 0.62 with p < 10-8, exception: muscle and lab glucose correlation was not significant with pressors on due to lack of sufficient concurrent samples), as is base excess with pH (ρ = 0.66 and 0.89, both p < 10-16). Muscle oxygen tension is negatively correlated with muscle lactate concentration (ρ = -0.28 and -0.65, p < 10-20) and the lactate:pyruvate ratio (mLP) (ρ = -0.17 and -0.25, p < 10-20) both with and without pressors, as would be expected since a reduction in oxygen available to muscle should lead to an increase the relative amount of anaerobic respiration and the resulting lactate. Ventilation parameters are also correlated as expected; arterial oxygen, oxygen saturation, and arterial oxygen:inspired oxygen (PF) ratio are all positively correlated, while PF ratio is anti-correlated with inspired oxygen fraction (all |ρ| > 0.46, all p < 10-7 with one p = 10-4). These findings reassured us that our methodology was specific enough to re-establish known, and intuitive physiological relations not associated with pressors.
The second stage of analysis, the "either-or" network, (Additional file 3) contains the 48 edges that exist either while pressors were given or not, but not in both cases. Twenty-nine edges are present only when pressors are not. For example, blood urea nitrogen (BUN) and creatinine concentrations are strongly correlated without pressors (ρ = 0.60, p << 10-60) but with pressors have only a statistically insignificant negative correlation (ρ = -0.06, p = 1). Nineteen edges are present only when pressors are being given. As we have fewer measurements with pressors than without, these correlations may be more likely be indicative of physiological changes rather than being anomalous due to large numbers of data points. Interpreting these emerging relationships can be difficult, as some may be reflective of the reason for giving pressors while others could be a consequence of pressors. For example, the coupling between arterial and muscle oxygen is associated with pressors (ρ = 0.55, p = 10-4). Since under normal conditions arterial oxygen remains nearly saturated, one could hypothesize (and eventually test) that oxygen transport is impaired at times when pressors are given. While receiving pressors pH becomes strongly negatively correlated with muscle lactate (ρ = -0.76, p = 8 × 10-8) and with muscle pyruvate (ρ = -0.86, p = 2 × 10-11), possibly indicating an increased susceptibility to acidosis. Perhaps the most interesting of our three analysis cases is what we term the "sign change" network, consisting of those edges that are oppositely correlated depending on whether pressors are being administered. Thirty-nine edges fall into this case (Additional file 4). While pressors were being given, the relationships between mean arterial pressure (MAP) and 8 other measures significantly change direction, in stark contrast to the edges connecting MAP to three other measures in both the static and either-or networks. The relationships between oxygen saturation (SpO2) and nearly all of the muscle microdialysis measures changed from positive to negative correlations while pressors were being given.
We set out to show that giving drugs to ICU patients can modify their physiology on a wider scale than expected from the main effect of the drugs, and that correlation networks can be used to visualize these changes. We have shown that changes in both univariate measures and correlation networks describing physiological states are associated with pressors being given to critical care patients and demonstrated a method to visualize these differences as three distinct cases. The "static" case includes a core physiology that is not associated with pressor use, while the "sign change" and "either-or" networks show two different types of physiological change associated with pressors. After narrowing our attention to only the strongest correlations from the original correlation networks, we examined a set of 58 edges to demonstrate the subtlety with which these changes can manifest in individual patients, even in a relationship that is part of the "static" network.
Defining physiological state is a difficult yet important endeavor that could change the way critical care patients receive treatment. In previous studies by our group and others [2, 12, 18, 19] the definition of physiologic state was based on instantaneous values or time histories of single physiological measurements. This approach has been successful, giving rise to scoring systems such as APACHE-III, the Injury Severity Score, and many systems for outcome prediction [12, 16, 28, 29]. Those studies mainly focused either on algorithm development or outcome scores. Here, we focused more on developing a technique for a deeper understanding of physiology, using a novel application of building correlation networks from temporally dense data.
The power of network-based analysis to discover relationships and improve clinical insight is gaining prominence . Constructing correlation networks out of readily measured clinical physiological variables can be the first step towards moving these innovative research tools to the clinic. One primary use for these networks might be in customization of drug therapies. For example, Adourian, et al., have used correlation networks to identify biomarkers for hepatic drug toxicity , but also make the point that empirical approaches using easily acquired data are likely to shed more light on complex physiology than normal univariate procedures.
One potential criticism of this work is that the use of a correlation network analysis on metabolites and variables yields results that are already known and generally well understood. However, it has been shown that similarly constructed correlation networks do not match up with those expected from interaction networks and that both approaches contribute useful information . Our work provides more evidence in favor of visualizing physiology via correlation networks, as ours contain unexpected associations that do not conform to standard physiological thinking. Establishing causation is a difficulty in this work, as in all correlation network analyses. In this case, some of the correlations associated only with pressors are likely to be reflective of the reasons for giving the drugs while others are likely to be due to the drugs. It is likely, for example, that the negative correlation that appears between muscle oxygen tension and heart rate when pressors are given reflects impairment in peripheral oxygen transport that drives the cardiovascular system to work harder , resulting in increased heart rate that is among the diagnostic criteria for shock . At the same time, the uncoupling of BUN from creatinine during pressor administration while associating with heart rate and muscle glucose does not have any clear connection to shock or pressors. Notwithstanding this inherent ambiguity, our method is capable of suggesting putative physiological relationships associated with the physiologic state of a patient that could shed light on poorly understood aspects of physiology. Future work can look at how relationships between measurements change over time, and can provide temporal evidence for causation by examining network changes immediately preceding and following the start of pressors.
To draw clinically useful conclusions about this "new" physiology, one must also improve on our main limitation of having 19 patients receive five different pressors in various combinations. Consequently, we limit our claims and say that the application of these methods to high frequency multivariate physiological data provides an excellent way to discover changes in physiology associated with varying physiologic state and we recommend further studies to delve into the specific ways in which this information can be put to practical use. These studies could involve more patients given well-defined doses of individual pressors and would enable the use of these methods to deeply understand their effects. As additional comprehensive data sets become available - and indeed they are currently being generated - these methods can easily be applied in a variety of ways to provide insight into the rapidly changing physiology of ICU patients. Ideally, future studies could also produce concrete treatment recommendations based on individual patients' physiologic state. At a minimum, though, this sort of analysis could provide an additional source of data for physicians to incorporate into their normal diagnostic procedure. Furthermore, because our methods are computationally simple they lend themselves well to eventual implementation in clinical decision support systems and improved patient monitors that incorporate state information.
We have shown that correlation networks can be built from physiological associations and that drugs modify these networks. Using correlation network analysis we determined that the strength and directionality of associations between clinical physiological variables changed when pressors were given, both in aggregate and in individual patients. In fact, the majority of these associations changed when pressors were given, reinforcing the notion that complex physiology cannot easily be modeled as a single set of mathematical relationships between metabolic variables. We also presented a proof of concept showing that patients have inherent differences in their physiology that could be exploited by a "personalized physiology" approach to medicine. We also demonstrated that different drugs have differential effects on physiological networks and that applying network analysis to clinical physiology provides information that could be useful in treating patients. Further studies in personalized physiology that open up the middle level of personalized medicine, a level between genes and demographics, have the potential to change the way drugs are evaluated during clinical trials and prescribed in the hospital or clinic.
Supported by the Hewlett Packard Foundation and Stanford Bio-X Graduate Fellowship (ADG), Lucile Packard Foundation for Children's Health and NIH R01 GM079719 (AJB), NIH K08 GM085689 (MJC) R01 NS050173 (GTM), UC Discovery Grant GCP06 10226 (GTM, MJC), CDC R49/CCR903697 (MJC, GTM). The authors wish to thank Diane Morabito for her expertise in manually curating and cleaning these data, and Shai Shen-Orr and Fraser Cameron for providing helpful comments.
The publication costs for this article were funded by AJB using NIH R01 GM079719.
This article has been published as part of BMC Medical Genomics Volume 6 Supplement 2, 2013: Selected articles from the Second Annual Translational Bioinformatics Conference (TBC 2012). The full contents of the supplement are available online at http://www.biomedcentral.com/bmcmedgenomics/supplements/6/S2.
- Guyton AC, Coleman TG, Granger HJ: Circulation: Overall Regulation. Annual Review of Physiology. 1972, 34: 13-44. 10.1146/annurev.ph.34.030172.000305.View ArticlePubMedGoogle Scholar
- Buchman T: Physiologic stability and physiologic state. J Trauma. 1996, 41 (4): 599-605. 10.1097/00005373-199610000-00002.View ArticlePubMedGoogle Scholar
- Norris PR, Canter JA, Jenkins JM, Moore JH, Williams AE, Morris JA: Personalized medicine: genetic variation and loss of physiologic complexity are associated with mortality in 644 trauma patients. Ann Surg. 2009, 250 (4): 524-530.PubMedGoogle Scholar
- Riordan WP, Norris PR, Jenkins JM, Morris JA: Early loss of heart rate complexity predicts mortality regardless of mechanism, anatomic location, or severity of injury in 2178 trauma patients. J Surg Res. 2009, 156 (2): 283-289. 10.1016/j.jss.2009.03.086.View ArticlePubMedGoogle Scholar
- Dossett LA, Cao H, Mowery NT, Dortch MJ, Morris JM, May AK: Blood glucose variability is associated with mortality in the surgical intensive care unit. Am Surg. 2008, 74 (8): 679-685. discussion 685PubMedGoogle Scholar
- Morris JA, Norris PR, Waitman LR, Ozdas A, Guillamondegui OD, Jenkins JM: Adrenal insufficiency, heart rate variability, and complex biologic systems: a study of 1,871 critically ill trauma patients. J Am Coll Surg. 2007, 204 (5): 885-892. 10.1016/j.jamcollsurg.2007.01.019. discussion 892-883View ArticlePubMedGoogle Scholar
- Norris PR, Ozdas A, Cao H, Williams AE, Harrell FE, Jenkins JM, Morris JA: Cardiac uncoupling and heart rate variability stratify ICU patients by mortality: a study of 2088 trauma patients. Ann Surg. 2006, 243 (6): 804-812. 10.1097/01.sla.0000219642.92637.fd. discussion 812-804PubMed CentralView ArticlePubMedGoogle Scholar
- Consortium IWP, Klein TE, Altman RB, Eriksson N, Gage BF, Kimmel SE, Lee M-TM, Limdi NA, Page D, Roden DM, et al: Estimation of the warfarin dose with clinical and pharmacogenetic data. N Engl J Med. 2009, 360 (8): 753-764.View ArticleGoogle Scholar
- Butte AJ: The ultimate model organism. Science. 2008, 320 (5874): 325-327. 10.1126/science.1158343.PubMed CentralView ArticlePubMedGoogle Scholar
- Stoll M, Cowley A, Tonellato P, Greene A, Kaldunski M, Roman R, Dumas P, Schork N, Wang Z, Jacob H: A genomic-systems biology map for cardiovascular function. Science. 2001, 294 (5547): 1723-1726. 10.1126/science.1062117.View ArticlePubMedGoogle Scholar
- Nelson D, Bellander B, Maccallum R, Axelsson J, Alm M, Wallin M, Weitzberg E, Rudehill A: Cerebral microdialysis of patients with severe traumatic brain injury exhibits highly individualistic patterns as visualized by cluster analysis with self-organizing maps. Crit Care Med. 2004, 32 (12): 2428-2436. 10.1097/01.CCM.0000147688.08813.9C.View ArticlePubMedGoogle Scholar
- Rixen D, Siegel J, Abu-Salih A, Bertolini M, Panagakos F, Espina N: Physiologic state severity classification as an indicator of posttrauma cytokine response. Shock. 1995, 4 (1): 27-10.1097/00024382-199507000-00004.View ArticlePubMedGoogle Scholar
- Peleg M, Rubin D, Altman RB: Using Petri Net tools to study properties and dynamics of biological systems. J Am Med Inform Assoc. 2005, 12 (2): 181-199.PubMed CentralView ArticlePubMedGoogle Scholar
- Zhang Y, Szolovits P: Patient-specific learning in real time for adaptive monitoring in critical care. J Biomed Inform. 2008, 41 (3): 452-460. 10.1016/j.jbi.2008.03.011.PubMed CentralView ArticlePubMedGoogle Scholar
- Patel V, Zhang J, Yoskowitz N, Green R, Sayan O: Translational cognition for decision support in critical care environments: a review. J Biomed Inform. 2008, 41 (3): 413-431. 10.1016/j.jbi.2008.01.013.PubMed CentralView ArticlePubMedGoogle Scholar
- Abad-Grau M, Ierache J, Cervino C, Sebastiani P: Evolution and challenges in the design of computational systems for triage assistance. J Biomed Inform. 2008, 41 (3): 432-441. 10.1016/j.jbi.2008.01.007.PubMed CentralView ArticlePubMedGoogle Scholar
- Hemphill JC, Barton C, Morabito DJ, Manley GT: Influence of data resolution and interpolation method on assessment of secondary brain insults in neurocritical care. Physiol Meas. 2005, 26 (4): 373-386. 10.1088/0967-3334/26/4/004.View ArticlePubMedGoogle Scholar
- Sorani M, Hemphill JC, Morabito DJ, Rosenthal G, Manley GT: New approaches to physiological informatics in neurocritical care. Neurocrit Care. 2007, 7 (1): 45-52. 10.1007/s12028-007-0043-7.View ArticlePubMedGoogle Scholar
- Cohen MJ, Grossman AD, Morabito D, Knudson MM, Butte AJ, Manley GT: Identification of complex metabolic states in critically injured patients using bioinformatic cluster analysis. Critical care (London, England). 2010, 14 (1): R10-10.1186/cc8864.View ArticleGoogle Scholar
- Schadt E, Friend S, Shaywitz D: OPINION A network view of disease and compound screening. Nature Reviews Drug Discovery. 2009, 8 (4): 286-295. 10.1038/nrd2826.View ArticlePubMedGoogle Scholar
- Adourian A, Jennings E, Balasubramanian R, Hines WM, Damian D, Plasterer TN, Clish CB, Stroobant P, McBurney R, Verheij ER, et al: Correlation network analysis for data integration and biomarker selection. Molecular Biosystems. 2008, 4 (3): 249-259. 10.1039/b708489g.View ArticlePubMedGoogle Scholar
- Müller-Linow M, Weckwerth W, Hütt M: Consistency analysis of metabolic correlation networks. BMC Systems Biology. 2007, 1: 44-10.1186/1752-0509-1-44.PubMed CentralView ArticlePubMedGoogle Scholar
- Butte AJ, Tamayo P, Slonim D, Golub TR, Kohane IS: Discovering functional relationships between RNA expression and chemotherapeutic susceptibility using relevance networks. Proc Natl Acad Sci USA. 2000, 97 (22): 12182-12186. 10.1073/pnas.220392197.PubMed CentralView ArticlePubMedGoogle Scholar
- Maslov S, Sneppen K, Zaliznyak A: Detection of topological patterns in complex networks: correlation profile of the internet. Physica A. 2004, 333: 529-540.View ArticleGoogle Scholar
- Grossman AD, Cohen MJ, Manley GT, Butte AJ: Infection in the intensive care unit alters physiological networks. BMC Bioinformatics. 2009, 10 (Suppl 9): S4-10.1186/1471-2105-10-S9-S4.PubMed CentralView ArticlePubMedGoogle Scholar
- Ikossi D, Knudson M, Morabito DJ, Cohen MJ, Wan J, Khaw L, Stewart C, Hemphill JC, Manley GT: Continuous muscle tissue oxygenation in critically injured patients: a prospective observational study. J Trauma. 2006, 61 (4): 780-10.1097/01.ta.0000239500.71419.58.View ArticlePubMedGoogle Scholar
- Sheskin D: Handbook of parametric and nonparametric statistical procedures. 2007, 1736-4Google Scholar
- Chen L, McKenna T, Reisner A, Gribok A, Reifman J: Decision tool for the early diagnosis of trauma patient hypovolemia. J Biomed Inform. 2008, 41 (3): 469-478. 10.1016/j.jbi.2007.12.002.View ArticlePubMedGoogle Scholar
- Dagum P, Galper A, Horvitz E, Seiver A: Uncertain reasoning and forecasting. International Journal of Forecasting. 1995, 11 (1): 73-87. 10.1016/0169-2070(94)02009-E.View ArticleGoogle Scholar
- Despopoulos A, Silbernagl S: Color Atlas of Physiology. 2003Google Scholar
- Bridges E, Dukes S: Cardiovascular aspects of septic shock: pathophysiology, monitoring, and treatment. Crit Care Nurse. 2005, 25 (2): 14-PubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.