Kutcher, Matthew E. MD; Ferguson, Adam R. PhD; Cohen, Mitchell J. MD
Hemorrhage remains the leading cause of potentially preventable death after trauma, complicated in up to a third of injured patients by coagulation abnormalities present on arrival to the emergency department.1 Although well-studied, the importance of specific clotting factor abnormalities to the complex phenomenon of acute traumatic coagulopathy is poorly understood. Strong correlations between clotting factor levels pose a significant challenge to identifying the isolated importance of any individual factor. This colinearity makes standard regression techniques prone to unstable results, difficult to generalize, and at risk of identifying spurious statistical significance. Several mathematical techniques exist to more clearly describe the patterns that exist in such complex, correlated data sets and to associate these patterns with binary outcomes; principal component analysis (PCA) is one such method.
PCA is a statistical pattern detection tool that distills a complex set of intercorrelations down to essential clusters of variables that move together as groups. To begin, a data set of correlated variables is decomposed into a smaller set of uncorrelated synthetic multivariables. A best-fit plane is described in this multivariate space, the axes of which are termed principal components (PCs); the location of each individual data point in multivariate space can then be specifically described relevant to these PCs. In this sense, PCA can be thought of as a multivariate form of the Pearson correlation, in which the best-fit line is replaced by a best-fit multivariate plane. The data set described here contains arrival clotting factor measurements in a panel of critically injured patients, in which many of the individual factor levels are highly correlated with each other. The application of PCA transforms each patient’s individual clotting factor measurements into a smaller set of PC “scores,” which can be interpreted as individual patient locations within multivariate outcome space. Furthermore, each PC can be broken down into “factor loadings” that describe both the contribution of each individual clotting factor into the calculation of that PC score as well as each factor’s relationships with other factors. A detailed explanation of the rationale and methodology of PCA is available in the Supplemental Digital Content (see Document, Supplemental Digital Content 1, http://links.lww.com/TA/A243).
As an example, a PC controlled entirely by a single predictor would have a loading coefficient of +1.0 for the predictor in question, with all other predictor coefficients equal to zero; in contrast, a PC determined by mixed contributions from multiple factors would have loading coefficients spanning values from −1.0 to +1.0 for each contributing factor, corresponding to positive and inverse Pearson correlations. Intuitively, PCs describe the internal structure of a set of correlated measurements allowing patients to be grouped by global patterns of clotting factor perturbations as defined by high or low PC scores, and the clotting factor interrelationships that define each group can be described. For a set of highly interrelated predictors such as clotting factor measurements, PCA accommodates degrees of colinearity that would make standard regression approaches unstable.
Therefore, we hypothesized that PCA would identify clinically predictive patterns in the complex clotting factor milieu after trauma. In this study, we apply PCA to identify and examine the clotting factor relationships underlying these clinical patterns and correlate these with outcomes in a panel of critically injured patients.
PATIENTS AND METHODS
Plasma was collected from 163 critically injured trauma patients on arrival to the emergency department of an urban Level I trauma center from February 2005 to October 2010 as part of an ongoing clinical study.2 Consecutive patients triggering activation of the highest level of a two-tiered triage system during the study period were prospectively enrolled under a waiver of consent; patients younger than 18 years, those with more than 5% body surface area burns, those who received more than 2 L of intravenous fluid before arrival, and those transferred from another institution were excluded. Patients were retrospectively excluded if informed consent was unobtainable or declined. Admission blood samples were collected via initial placement of a 16-gauge or larger peripheral intravenous line into 3.2% (0.109 M) sodium citrate and processed within 3 hours of being drawn; sample collection methodology is described in detail elsewhere.2 Activity levels of factors II, V, VII, VIII, IX, and X; antithrombin III and protein C; as well as antigen D-dimer levels were assayed using a Stago Compact Functional Coagulation Analyzer (Diagnostica Stago, Parsippany, NJ). Activated protein C was assayed using an established enzyme-linked immunosorbent assay method reported elsewhere.3 Demographics, laboratory and resuscitation data, and outcomes were collected in parallel. Injury was assessed by Injury Severity Score (ISS).4 Acute lung injury was based on the American-European consensus definition.5 Multiorgan failure was defined as a multiple organ dysfunction score of greater than 3 using established Denver criteria.6 The study was approved by the University of California Committee on Human Research.
Nonlinear PCA was performed using SPSS categories CAT-PCA (IBM, Chicago, IL); specific details are available in the Supplemental Digital Content Methods (see Document, Supplemental Digital Content 1, http://links.lww.com/TA/A243). Possible nonlinear correlations among coagulation factors were accounted for in the input stage, resulting in an output of continuous, linear PC scores and PC loadings;7 (see Figure, Supplemental Digital Content 2, http://links.lww.com/TA/A241). All measured factor levels were included as predictors: prothrombin; factors V, VII, VIII, IX, and X; D-dimer; activated protein C; protein C; and antithrombin III. PCs were considered significant for eigenvalues greater than or equal to 1.0 (see Figure, Supplemental Digital Content 3, http://links.lww.com/TA/A242); factor loadings were considered significant for coefficients greater than or equal to 0.3. Continuous PC scores were calculated for each patient along each PC axis, and differences between the highest and lowest quartile of patients for each significant PC were examined using standard univariate statistics. The predictive capacities of independent PC scores were evaluated using unadjusted logistic regression for the binary outcomes of mortality, multiorgan failure, acute lung injury, and ventilator-associated pneumonia (VAP) and for standard definitions of coagulopathy determined by admission international normalized ratio (INR) and partial thromboplastin time (PTT).
Data are presented as mean ± SD, median (interquartile range [IQR]), or percentage; univariate comparisons were made using Student’s t test for normally distributed data, Wilcoxon rank-sum testing for skewed data, and Fisher’s exact test for proportions. Missing predictor data were imputed using multiple imputation; results were similar to those obtained using only complete data as well as using a data set completed with population means (data not shown). An α of 0.05 was considered significant. All analysis was performed by the authors using SPSS categories (IBM) and Stata version 12 (Stata Corp; College Station, TX).
Our 163-patient study population had a mean age of 41.3 ± 19.3 years and a mean ISS of 23.2 ± 5.4; there was 31.0% penetrating and 61.2% brain injuries. Of 163 patients, 22 (19.0%) were coagulopathic on arrival as defined by INR of 1.3 or greater, and 56 (34.4%) were coagulopathic as defined by PTT of 30 or greater. Nonlinear PCA identified seven independent PCs, together accounting for 92.0% of the variance present in the data. PCs 1, 2, and 3 were considered significant (eigenvalues > 1.0); these together accounted for 67.5% of total variance. Eigenvalues, percentage of variance explained, and the factor loading matrix for all individual clotting factors for the three significant PCs are shown in Table 1. Factor loadings were considered significant for loading coefficients greater than or equal to 0.3.
PC1 accounted for 43.9% of overall variance, including significant negative factor loading on (analogous to inverse Pearson correlation with) prothrombin; factors V, VII, VIII, IX, and X; protein C; and antithrombin III (Table 1). To identify patient-level characteristics associated with high PC1 scores, patients with the highest quartile of PC1 score were compared with those in the lowest quartile (Table 2). Patients in the highest quartile of PC1 score had significantly more common penetrating injury (34.2% vs.12.5%, p = 0.035) and more severe injury (mean ISS, 32.2 vs. 23.8; p = 0.011) compared with those in the lowest quartile. High-PC1 patients also had significantly elevated admission INR (median, 1.2 vs. 1.0; p < 0.001) and PTT (median, 32.6 seconds vs. 27.0 seconds, p < 0.001), as well as lower admission platelet count (mean, 245 × 103/µL vs. 311 × 103/µL; p < 0.001). High-PC1 patients also had significantly higher transfusion requirements for red blood cells (median, 5 U vs. 0 U, p < 0.001), plasma (median, 2 U vs. 0 U; p < 0.001), and platelets (median [IQR], 0 U [0–2 U] vs. 0 U [0 U]; p = 0.003) and significantly higher mechanical ventilation requirements (median, 6.5 ventilator-free days vs. 17.5 ventilator-free days; p = 0.016). Expressed in odds ratios, each unit increase in PC1 score was associated with a 4.68-fold higher incidence of INR-based coagulopathy (p < 0.001), a 3.35-fold higher incidence of PTT-based coagulopathy (p < 0.001), and 1.48-fold higher mortality (p = 0.032).
PC2 accounted for 13.4% of overall variance, including significant positive factor loading on (analogous to positive Pearson correlation with) factor VIII, D-dimer, and activated protein C levels (Table 1). Similarly to those previously mentioned, patients in the highest PC2 quartile were again compared with those in the lowest (Table 3). High PC2 patients had more severe injury (mean ISS, 37.3 vs. 23.0; p < 0.001), acidosis (mean pH, 7.27 vs. 7.23; p = 0.032), and base deficit (mean, −8.4 vs. −5.5, p = 0.029). High-PC2 patients received less prehospital intravenous fluid (median, 0 mL vs. 500 mL), consistent with expedited transport. Despite these differences in injury and shock severity, admission INR and PTT did not differ significantly by PC2 quartile (p = 0.440 and p = 0.756, respectively), although admission platelet count was higher in the high PC2 population (mean 288 × 103/µL vs. 238, p = 0.015). High-PC2 patients had significantly higher transfusion requirements for red blood cells (median, 6 U vs. 0 U; p < 0.001), plasma (median, 4 U vs. 0 U; p < 0.001), and platelets (median [IQR], 0 U [0–2] vs. 0 U [0 U]; p = 0.002). In outcomes, high-PC2 patients had prolonged intensive care unit (ICU) (median, 12 days vs. 4.5 days; p = 0.003) and total hospital stays (median 22.5 days vs. 8.5 days; p = 0.004), and significantly higher mechanical ventilation requirements (median, 2 ventilator-free days vs. 22 ventilator-free days; p = 0.002). The incidences of VAP (55.0% vs. 29.8%; p = 0.028), acute lung injury (73.5% vs. 26.7%; p < 0.001), multiorgan failure (32.5% vs. 4.2%; p < 0.001), and mortality (37.5% vs. 16.7%; p = 0.031) were all markedly higher in the high-PC2 population. Expressed in odds ratios, PC2 score was not associated with significant increases in the incidence of either INR-based or PTT-based coagulopathy (p = 0.220 and p = 0.340, respectively); however, each unit increase in PC2 score was associated with a 1.59-fold higher incidence of VAP, a 2.24-fold higher incidence of acute lung injury, a 1.83-fold higher incidence of multiorgan failure (p = 0.002), and 1.62-fold higher mortality (p = 0.006).
PC3 accounted for 10.1% of overall variance, including significant positive factor loading with factor VII and activated protein C and significant negative loading with factor VIII (Table 1). Patients in the highest PC3 quartile had significantly better Glasgow Coma Scale (GCS) score (median 9 vs. 6; p = 0.034) and lower admission PTT (median 26.6 seconds vs. 28.8 seconds; p < 0.001); high-PC3 patients also had a lower incidence of acute lung injury (40.0% vs. 67.6%; p = 0.033; Table 4). Unit increases in PC3 were associated with a 1.44-fold increase in the incidence of PTT-based coagulopathy (p = 0.041); however, PC3 scores were not significantly associated with other measured outcomes (all other p > 0.05).
To graphically assess interrelationships between PCs and their relationship to outcomes, scatter plots were generated representing each patient in multivariate space by PC1 and PC2 score (Fig. 1). Visually, the majority of patients with a prolonged INR and PTT are seen to have an elevated PC1 score; these are equally balanced between high and low PC2 scores (Fig. 1A and B). In comparison, most patients with multiorgan failure and mortality are seen to have an elevated PC2 score; these are equally balanced between high and low PC1 scores (Fig. 1C and D). Odds ratio data for all binary outcomes assessed are summarized in Table 5.
Here, we describe the use of PCA to interrogate the data structure of clotting factor levels in a panel of 163 critically injured trauma patients. Three significant uncorrelated multivariate components were identified, together explaining 67.5% of the total variance in observed clotting factor measurements. PC1 accounted for 43.2% of overall variance and consisted of significant negative loading coefficients for all procoagulant clotting factors as well as the anticoagulants protein C and antithrombin III. Intuitively, PC1 identifies global clotting factor depletion, and its patient-level values are associated with increased incidences of admission coagulopathy and mortality. PC2 accounted for 13.4% of variance, consisting principally of significant positive factor loading on D-dimer and activated protein C levels. Intuitively, PC2 identifies a fibrinolytic component to the clotting factor milieu; interestingly, this component is not significantly associated with admission coagulopathy by INR-based or PTT-based definitions but is significantly associated with VAP, acute lung injury, multiorgan failure, and mortality. PC3 accounted for 10.1% of variance, consisting principally of significant negative loading on factor VIII, with smaller contributions from factor VII and activated protein C. Intuitively, PC3 may account for an element of coagulopathy associated with consumption-driven depletion of factor VIII; PC3 was associated only with admission PTT-based coagulopathy.
The global procoagulant depletion phenotype described by PC1 matches the clinical intuition that patients with overwhelming tissue injury and acute hemorrhage frequently present with coagulopathy resulting from clotting factor consumption. The early recognition of this “vicious cycle” of self-perpetuating consumptive coagulopathy is a mainstay of trauma resuscitation,8 with recent studies extending these clinical observations to confirm the association of specific clotting factor deficits to poor outcomes after injury.9 The global procoagulant depletion phenotype reflected by PC1 and its association with both coagulopathy and mortality reflect the clinical intuition that clotting factor depletion must be anticipated and treated early to minimize its deleterious effects; precisely this insight has driven recent trends in plasma-based hemostatic resuscitation therapy for critically injured patients.10,11
The fibrinolytic phenotype described by PC2 is significantly associated with multiple functional outcomes but is interestingly not associated with standard admission laboratory values such as INR or PTT. This subtlety supports the hypothesis that injury-induced activation of endogenous anticoagulants and other systemic effectors, as opposed to consumptive depletion of procoagulant factors alone, plays a critical role in the pathophysiology of acute traumatic coagulopathy. Recent work suggests several candidate biochemical pathways that may mediate dysfunctional coagulation after trauma independently of clotting factor depletion, including catecholamine-mediated degradation of the endothelial glycocalyx12,13 and activation of the protein C system.2,3 While activated protein C mediates receptor-independent proteolysis of activated factors Va and VIIIa as well as derepression of fibrinolysis,2 it may also play an additional receptor-dependent role in potentiating the systemic inflammatory response.3 Kerschen et al.14 recently showed that a recombinant form of activated protein C with targeted mutation leading to less than 10% anticoagulant activity is equivalent to native protein in reducing mortality in sepsis models in mice. A prospective study of clotting factor levels in 71 injured patients identified a significant negative correlation between severity of systemic hypoperfusion after injury and the activity of several procoagulant factors but found that decreased factor V activity may occur via a consumption-independent mechanism such as protein C–mediated cleavage.15
This identification of PC2 as a fibrinolytic component is also consistent with a growing recognition of the importance of hyperfibrinolysis to acute traumatic coagulopathy. Hyperfibrinolysis is estimate to occur from 3% to 20% of significantly injured patients, with mortality spanning 38.5% to 100%.16 At the biochemical level, the presence of hyperfibrinolysis is associated with significantly elevated levels of D-dimer and activated protein C.17 Recent intriguing data suggest that the use of plasminogen-targeted antifibrinolytics such as tranexamic acid may provide the missing pharmacologic treatment for the hyperfibrinolytic component of acute traumatic coagulopathy.18,19 Taken together, these results suggest that aggressive clotting factor repletion by empiric plasma-based therapies may inadequately treat the fibrinolytic component of acute traumatic coagulopathy and that targeted therapies are a promising area of active investigation.
As with other single-center prospective studies examining the relationship between admission clotting factor studies and outcomes, several limitations are important to interpretation of these data. Although our sample size is modest, highly cited work in PCA suggests that 5 to 10 samples for each predictor included is adequate for robust results,20,21 which our 163-patient panel well exceeds for our 10-predictor model. Importantly, this analysis does not provide for strict causal interpretation; neither PCs themselves nor the individual patient PC scores provided by PCA are directly clinically interpretable. Furthermore, the PCA model itself is not generalizable or portable but is only applicable to the data set from which it is derived; PC scores cannot be derived for novel patients outside of the data set presented here. For this exploratory analysis, we included all classical clotting factor measurements available; however, other relevant physiologic measures (such as pH or temperature) and clotting cascade elements (such as calcium and fibrinogen) were not uniformly available. Thus, the final PCA model may be sensitive to predictor selection. Cognates to common predictor selection strategies for regression analysis (such as forward/backward and information criteria-based selection) are not well-established for the construction of PCA models and will require detailed sensitivity analyses to validate. Overall, however, the purpose of the current study was not to construct a comprehensive predictive model but instead to interrogate the clotting cascade for clinically compelling patterns. We suggest that these patterns identify unique groups of patients that would otherwise go undetected based on standard clinical characteristics alone and that the further analysis of these groups may identify novel molecular markers and potential therapeutic targets.
Taken together, these results suggest that PCA accurately identifies patterns embedded in the complex milieu of the coagulation cascade in injured patients. The independent consumptive and fibrinolytic components identified here show robust correlation with patient-level outcomes and match prevailing clinical intuition regarding drivers of acute traumatic coagulopathy. In particular, the fibrinolytic phenotype is associated with markedly poor outcomes but not with abnormalities in INR or PTT, highlighting the inadequacy of these measures in describing the complexity of traumatic coagulopathy and the need for validated markers of abnormal fibrinolysis after trauma. Whereas standard regression techniques are problematic for investigating the role of clotting factors in isolation, PCA is well suited to address just such highly collinear systems of predictors as patterns. The pattern finding capability demonstrated here holds promise for elucidating critical mechanisms underlying the pathophysiology of acute traumatic coagulopathy. Further development of data-driven analytical methods such as PCA may ultimately provide critical insights to drive advances in clinical care.
M.E.K., A.R.F., and M.J.C. prepared the article, performed all data analyses, and take full responsibility for the data as presented.
This study was supported by NIH T32 GM-08258-20 (M.E.K.), NIH NS-067092 (A.R.F.), and NIH GM-085689 (M.J.C.).
1. Hess JR, Brohi K, Dutton RP, et al. The coagulopathy of trauma: a review of mechanisms. J Trauma. 2008; 65: 748–754.
2. Brohi K, Cohen MJ, Ganter MT, Matthay MA, Mackersie RC, Pittet JF. Acute traumatic coagulopathy: initiated by hypoperfusion: modulated through the protein C pathway? Ann Surg. 2007; 245: 812–818.
3. Cohen MJ, Call M, Nelson M, et al. Critical role of activated protein C in early coagulopathy and later organ failure, infection and death in trauma patients. Ann Surg. 2011.
4. Baker SP, O’Neill B, Haddon W Jr, Long WB. The injury severity score: a method for describing patients with multiple injuries and evaluating emergency care. J Trauma. 1974; 14: 187–196.
5. Bernard GR, Artigas A, Brigham KL, et al. The American-European Consensus Conference on ARDS. Definitions, mechanisms, relevant outcomes, and clinical trial coordination. Am J Respir Crit Care Med. 1994; 149 (Pt 1): 818–824.
6. Sauaia A, Moore EE, Johnson JL, Ciesla DJ, Biffl WL, Banerjee A. Validation of postinjury multiple organ failure scores. Shock. 2009; 31: 438–447.
7. Linting M, Meulman JJ, Groenen PJ, van der Koojj AJ. Nonlinear principal components analysis: introduction and application. Psychol Methods. 2007; 12: 336–358.
8. Kashuk JL, Moore EE, Millikan JS, Moore JB. Major abdominal vascular trauma—a unified approach. J Trauma. 1982; 22: 672–679.
9. Rizoli SB, Scarpelini S, Callum J, et al. Clotting factor deficiency in early trauma-associated coagulopathy. J Trauma. 2011; 71 (Suppl 1): S427–S434.
10. Duchesne JC, Holcomb JB. Damage control resuscitation: addressing trauma-induced coagulopathy. Br J Hosp Med (Lond). 2009; 70: 22–25.
11. Holcomb JB, Wade CE, Michalek JE, et al. Increased plasma and platelet to red blood cell ratios improves outcome in 466 massively transfused civilian trauma patients. Ann Surg. 2008; 248: 447–458.
12. Johansson PI, Sorensen AM, Perner A, et al. High sCD40L levels early after trauma are associated with enhanced shock, sympathoadrenal activation, tissue and endothelial damage, coagulopathy and mortality. J Thromb Haemost. 2011.
13. Johansson PI, Stensballe J, Rasmussen LS, Ostrowski SR. A high admission syndecan-1 level, a marker of endothelial glycocalyx degradation, is associated with inflammation, protein C depletion, fibrinolysis, and increased mortality in trauma patients. Ann Surg. 2011; 254: 194–200.
14. Kerschen EJ, Fernandez JA, Cooley BC, et al. Endotoxemia and sepsis mortality reduction by non-anticoagulant activated protein C. J Exp Med. 2007; 204: 2439–2448.
15. Jansen JO, Scarpelini S, Pinto R, Tien HC, Callum J, Rizoli SB. Hypoperfusion in severely injured trauma patients is associated with reduced coagulation factor activity. J Trauma. 2011; 71 (Suppl 1): S435–S440.
16. Schochl H, Voelckel W, Maegele M, Solomon C. Trauma-associated hyperfibrinolysis. Hamostaseologie. 2012; 32: 22–27.
17. Kutcher ME, Cripps MW, McCreery RC, et al. Criteria for empiric treatment of hyperfibrinolysis after trauma. J Trauma Acute Care Surg. 2012; 73: 87–93.
18. Morrison JJ, Dubose JJ, Rasmussen TE, Midwinter MJ. Military Application of Tranexamic Acid in Trauma Emergency Resuscitation (MATTERs) Study. Arch Surg. 2011.
19. Shakur H, Roberts I, Bautista R, et al. Effects of tranexamic acid on death, vascular occlusive events, and blood transfusion in trauma patients with significant haemorrhage (CRASH-2): a randomised, placebo-controlled trial. Lancet. 2010; 376: 23–32.
20. Nunnally JO. Psychometric Theory. New York, NY: McGraw-Hill; 1978.
21. Tabachnik BG, Fidell LS. Using Multivariate Statistics. 5th ed. Boston, MA: Pearson Education; 2007.
Dr. Sandro Rizoli (Toronto, Canada): This study, as I understand it, has two components. It explores the mechanisms underlying acute coagulopathy of trauma and, second, it explores the use of a mathematical statistical tool to identify patterns in a complex system such as coagulation.
The mathematical tool is called Principal Component Analysis and even after the brilliant presentation I still am not sure that I could explain to anyone what it really is.
Very little is known about coagulation and trauma. We know it is complex, multifactorial, it changes over time and we know it changes in response to surgical interventions.
We also know it very difficult to determine the importance of each clotting element in this complex process. So a mathematical tool that can make it easier to understand is welcome and worthwhile testing.
So over a five-year period, Dr. Cohen and his group drew blood from 163 severely-injured patients on arrival to hospital and performed many coagulation assays. Of the 163 patients included, only 22 were coagulopathic, defined by an INR of 1.3, which could be a soft definition considering that 1,5 is more commonly used.
So my first question is considering that the goal of this study was to understand coagulopathy as a disease, shouldn’t the statistical analysis be done only on the 22 patients that were in fact coagulopathic?
Otherwise this study was done on non-coagulopathic patients and might simply demonstrate the variations in clotting factors after trauma, which do not necessarily lead to coagulopathy.
The second question is, many of the differences found between patients within the same PC were statistically significant but not clinically relevant. A platelet count of 250 versus 300 or a PTT of 32 versus 27 doesn’t seem clinically relevant to me. So how can this statistically but not clinically relevant difference be taken to explain coagulopathy?
And, finally, concerning your interpretation of the role of hyperfibrinolysis in trauma, could we extrapolate your conclusions and say that every trauma has a component of hyperfibrinolysis and thus should receive Tranexamic acid?
So, once again, I want to congratulate the authors for their brilliant work and I thank the ASAT for the privilege.
Dr. Ronald V. Maier (Seattle, Washington): Very nice study, attempting to better elucidate the complex coagulopathic patterns in the injured patient.
One of the issues is to follow the process serially over time in these patients with different treatments. But using the early data you have demonstrating differential treatments prehospital, can you assess how much crystalloid is required to increase bleeding significantly?
You should have had enough variability in these patients as far as prehospital volume of resuscitation that you can interrogate the impact of a large volume of crystalloid on coagulopathy or type of coagulopathy at the time of hospital presentation. Thank you.
Dr. Susan Rowell (Portland, Oregon): Given the worse outcomes associated with PC2 that are not detectible by INR and PTT, I am curious as to whether you have done any platelet function analysis or have correlated this with TEG data which could be more easily performed at the bedside? Thank you.
Dr. Scott G. Thomas (South Bend, Indiana): I’m interested in your thoughts on recent discussions with regard to TEG replacing INR. And, at your institution, are you looking at that as a primary modality of looking at coagulopathy and trauma? As we know, Dr. Holcomb just recently published a paper saying that this may be the new trend. And is this your plan in the future for looking at coagulopathic patients in trauma? Thank you.
Dr. Andre Cap (Houston, Texas): I’m curious about why we didn’t see fibrinogen results in your table of coagulation factors. I was wondering if you had measured fibrinogen and how those values align with the various patterns you’re observing.
I agree with the previous comments about TEG analysis and how it correlates with other measures of coagulation function. Thanks very much.
Dr. Matthew Kutcher (San Francisco, California): Thank you for your insightful questions. In terms of Dr. Rizoli’s question—whether these analyses should be performed on coagulopathic versus normal patients—interestingly, it turns out that the PC scores in coagulopathic versus normal patients were remarkably different.
In receiver-operator characteristic analyses, we found that a PC1 score is a very strong predictor of having an elevated admission INR with an area under the curve of about 800—so these scores really are clinically relevant.
I think the focus of this paper, however, was really as an exploratory analysis to understand whether these principal components would independently recapitulate these clinical patterns.
And so I think the question is: how do you identify patients who are at risk of bleeding to death, versus those who don’t hemorrhage but who are at risk of having an uncontrolled inflammatory response to trauma.
That’s where we’re really looking to go with this work, and that’s our interest in a closer exploration of the real meaning of PC2. This is biological work that remains to be done.
In terms of the question of clinical relevance, we simply use mean values to try to isolate and understand a subset of patients with high PC scores to see what the differences in these populations are. The difference in the mean platelet counts, for example, is not particularly large—but we already know what a low or high platelet count by itself means.
So: are these patterns differentiating patients who were or were not coagulopathic early in their clinical course, or, better yet, are not going to bleed to death but who will have later end-organ damage from their early inflammatory response to trauma? I think that most of our standard laboratory measurements are poor measures of that. That’s how our interest in this work began in the first place.
In terms of Dr. Rizoli’s comments on hyperfibrinolysis, we’re very interested in the use of agents like aminocaproic and tranexamic acid for the treatment of hyperfibrinolysis. And I think the TEG is a great way of looking that.
TEG, in a way, is a little bit like Principal Components Analysis in that it distills the clotting cascade into a visual set of patterns from which more useful numbers like an R or an MA can be extracted. So a hyperfibrinolytic TEG has a collection of different numbers that you can use to describe it, but it has a particular shape. And in a sense that shape is that pattern that we look for.
So if you had a patient who came in with a very high PC1 score and a low to normal PC2 score—basically someone with depletion coagulopathy—you could treat that patient very well with fresh frozen plasma.
But if you had a similar patient with a high PC1 score but also a high PC2 score, plasma might fix their depletion coagulopathy but really wouldn’t address the problem of their hyperfibrinolysis. This is why we’ve been interested in identifying guidelines for empiric and data-based anti-fibrinolytic therapies.
In terms of Dr. Maier’s question about prehospital crystalloid, in San Francisco the average transport time is less than fifteen minutes and we’re lucky if they have an IV by the time they get in the door, so our prehospital crystalloid volumes tend to be in the 0 to 500 range.
And so we don’t really have the variance to determine many differences based on prehospital data but, as you point out, I think looking longitudinally during the course of in-hospital resuscitation we will have the ability to find out what perturbs this system over time, based on responses to crystalloid, red cells, FFP, anti-fibrinolytics and other resuscitative measures.
And we look forward to doing that work—it’s just statistically more complex and is going to take more data and time.
I think platelet function and TEG would be excellent datasets to use the PCA technique on. We actually do have longitudinal platelet function as well as TEG data on some of these patients in addition to their clotting factor profiles. We look forward to doing the same sort of exploratory analyses in those arenas as well.
In response to Dr. Thomas’ question about looking at TEG-based resuscitation, I think that this is exactly where this sense that we need a better understanding of patterns comes from.
And, as I alluded to, I think hyperfibrinolysis is one of the critically important patterns to identify early. The difference between PC1 and PC2 is exactly like one might see in a TEG, where you can identify hyperfibrinolysis, or you can identify a separate platelet-based deficit, or a clotting factor-based deficit, or any number of these things occurring at the same time. And all of these are treated differently. Both the PCA data here and TEG used as a resuscitation guide clinically make sense, since it parallels the way we think. Patterns, not flow-charts.
And finally, in terms of fibrinogen as well as some other clotting cascade elements and adjuncts, we don’t have as complete a set of measurements for these as we do from our standard factor level profiles, but we are in the process of expanding our standard battery of measurements as well as going back to re-measure some of these in our stored samples as well.
Thank you very much for the privilege of presenting our work today.
© 2013 Lippincott Williams & Wilkins, Inc.