PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
 
Transfusion. Author manuscript; available in PMC 2012 September 1.
Published in final edited form as:
PMCID: PMC3138868
NIHMSID: NIHMS261237

BLOOD PRODUCT USE IN TRAUMA RESUSCITATION: Plasma deficit versus plasma ratio as predictors of mortality in trauma

Abstract

Introduction

Resuscitation of rapidly bleeding trauma patients with units of packed red blood cells (RBCs) and plasma given in a 1:1 ratio has been associated with improved outcome. However, demonstration of a benefit is confounded by survivor bias, and past work from our group has been unable to demonstrate a benefit.

Methods

We identified 438 adult direct primary trauma admissions at risk for massive transfusion who got ≥5 RBC units in the first 24 hours and had a probability of survival of 0.010 to 0.975. We correlated survival with RBC and plasma use by hour, both as a ratio (units of plasma/units of RBC) and a plasma deficit (units of RBC – units of plasma) in the group as a whole and among those using 5–9 and >9 units of RBC.

Results

Resuscitation was essentially complete in 58.3% by the end of the third hour and 77.9% by the end of the 6th hour. Mortality by hour was significantly associated with worse plasma deficit status in the first 2 hours of resuscitation (P<0.001 and 0.01) but not with plasma ratio. In a subgroup with TRISS 0.200–0.800, early plasma repletion was associated with less blood product use independently of injury severity (P<0.001).

Conclusions

1) The efficacy of plasma repletion plays out in the first few hours of resuscitation; 2) plasma deficit may be a more sensitive marker of efficacy in some populations; and 3) early plasma repletion appears to prevent some patients from going on to require massive transfusion.

Keywords: hemostatic resuscitation, damage control resuscitation, blood component therapy, coagulopathy, injury

Introduction

Resuscitation of rapidly bleeding trauma patients with units of packed red blood cells (RBC) and plasma given in a 1:1 ratio has been associated with improved outcome.(110) However, this finding is confounded by the speed at which massively hemorrhaging patients die and the rate at which type-specific plasma is thawed and delivered to the bedside.(11,12) These two events interact to create ‘survivor bias’ which accounts for some of the apparent association. In addition, blood product use in trauma patients is strongly associated with injury severity. Controlling for this association is difficult, particularly in retrospective studies dependent on registry data and the use of one or another of the injury scoring systems, and can lead to inappropriate attribution of either adverse or improved outcomes to the effects of transfusion.

Previous work from our institution has not demonstrated a survival advantage from the use of 1:1 ratio resuscitation (13) despite a large patient experience with massive transfusion (14) and our having been early proponents of this approach.(1517) One reason for this finding may be that when very large transfusions are given, the plasma:RBC ratio is not a good metric; a patient receiving 30 units of RBC and 20 units of plasma would have an ‘acceptable’ ratio of 2:3, but in reality have a substantial deficit of plasma. Calculation of overall ratio also fails to capture the time course of transfusion. The patient who receives 10 units of RBC followed an hour or so later by 10 units of plasma may not do as well as one who receives the same total number of blood products with RBC and plasma units alternating.

In an attempt to control for survivor bias; to provide insight into the scale, time-course and outcome of severely injured, rapidly bleeding civilian patients; and to characterize in detail our institutional experience with plasma resuscitation, we reviewed RBC and plasma usage and survival in a 5-year cohort. We had three hypotheses in this work. First, in rapidly bleeding trauma patients, the efficacy of plasma repletion in improving survival will be obvious in the first few hours of resuscitation, when most deaths from uncontrolled hemorrhage occur. Second, the deficit of plasma to RBC units may provide a more sensitive marker than does the ratio of plasma to RBC. Third, early plasma repletion will prevent some patients from requiring massive transfusion.

Methods

The University of Maryland R Adams Cowley Shock-Trauma Center is the primary adult trauma referral center for a catchment population of roughly 6 million. It admits 5,500 patients a year directly from the scene of injury and has maintained a trauma registry since the mid-1980s. Details of the scope, staffing and procedures of this registry have been published elsewhere. (12)

Using a database query process, we identified all primary trauma admissions 18 years or older admitted directly from the scene of injury from July 1, 2003, through June 30, 2008, who survived at least 15 minutes after admission and who received at least 1 unit of uncrossmatched Group O RBC in the trauma receiving unit (TRU). The use of uncrossmatched group O RBCs is a marker of the need for urgent transfusion.(18) We do not have a trauma transfusion protocol but do keep 10 units of uncrossmatched Group O RhD positive and 2 units of Group O RhD negative RBCs available in the TRU at all times. Since January 27, 2007, 4 units of thawed AB plasma have also been immediately available. Readmissions for prior trauma, non-trauma admissions, and admissions through the TRU for other services were excluded.

This preliminary cohort was described by age, sex, mechanism and type of injury, injury severity score (ISS) (19), probability of survival (using TRISS methodology) (20), hospital length of stay (LOS) and in hospital death; these data were then transferred to an Excel database and linked with the Quality Management database to capture timing and cause of death. Hourly blood product issue for each patient over the first 24 hours was determined from the blood bank data management system.

We then examined three subgroups of interest: those patients receiving >9 units of RBC in the first 24 hours of admission (‘massive transfusion’) and those who received 5 to 9 or 1 to 4 units in the same time period. In an attempt to control for the association between blood product use and injury severity but also to address the relative insensitivity of ISS in the assessment of penetrating injury and bleeding (20) and the age-diversity of our population, we examined plasma and RBC use as well as mortality in a subgroup of patients who had received at least 5 units of RBCs in the first 24 hours and whose probability of survival or TRISS was between 0.010 and 0.975. By eliminating from analysis patients who were either very likely to die (TRISS <0.010) or almost certain to survive (TRISS >0.975) we hoped to achieve a better focus on the impact of transfusion therapy.

Statistical methods

Plasma resuscitation was calculated as plasma ratio (units of plasma/units of RBC) and plasma deficit (RBC units – plasma units) at hourly intervals over the first 24 hours after admission. Overall mean plasma use among those receiving >9 units of RBCs in the first 24 hours was 4 units, so we defined ‘low deficit’ as 2 or fewer units and ‘high deficit’ as more than 6 units. Plasma ratio was calculated as a decimal and defined as >0.66 (‘better’ than 2:3), 0.66 to 0.34 (2:3 to ‘better than’ 1:3), and <0.34 (1:3 or less), in keeping with previously published work from other centers.(7,8)

Student’s t-test (Excel®; Microsoft; Redmond, WA) and analysis of variance (ANOVA) (21) were used to assess differences in means (age, ISS, TRISS, blood product use). Chi square analyses were used to assess differences in categorical variables (plasma deficit, mortality). (22) Probability values for results being due to chance (P) of 0.05 or less were considered significant. Corrections for repeat measures were not used in this exploratory study. P values of more than one decimal place less than 0.01 are shown as <0.001.

Results

Demographic and other summary information

Eight hundred forty-four primary trauma admission patients met the basic inclusion criterion of having received one unit of uncrossmatched group O RBC in the first hour of care. This group was 79% male, had a mean age of 39 years, a mean ISS of 32 and probability of survival of 0.639. They were in the trauma center for a mean of 12 days. Injuries were 53% vehicle related, 21% gun-shot wounds, 16% stabbings, 6% falls, 4% other. Three hundred one of these patients died, 49% from uncontrolled hemorrhage, 33% from brain injury, and 9% from multiple organ failure.

Table 1 summarizes RBC and plasma transfusion in the first 24 hours among those on whom complete information was available. Of those 835 individuals, 320 (38.3%) used 1 to 4 units in the first 24 hours, 208 (24.9%) used 5 to 9 units, and 307 (36.8%) used 10 units or more in that same period. Mean RBC and plasma use were 10 and 7 units, respectively, and mean product usage in the three subgroups increase proportionately. In each RBC-use group, some patients received no plasma at all (Table 2); in the 1 to 4 units group this proportion was almost 70% (217/320). Age and injury severity (reflected in mortality, LOS, time to death and both trauma scores) varied significantly but not always progressively through the three groups. ISS and LOS did vary progressively. The difference in age between the three groups was attributable to the relative youth of the 1 to 4 and >9 units groups and most of the difference in time-to-death was attributable to the relatively short time-to-death in the 1 to 4 units group. (The calculations regarding attribution are not shown.) The significant differences in mortality between the three groups are largely related to worse probability of survival/TRISS in the massive transfusion (>9 units) group.

Table 1
Patterns of red blood cell (RBC) usage in primary trauma admission patients who received at least 1 unit of uncrossmatched RBC in the first 24 hours of resuscitation and among whom complete data was available. Values are presented as mean and standard ...
Table 2
Red blood cell (RBC) and plasma transfusion in the first 24 hours among 438 primary trauma patients who received at least 5 units of RBC in the first 24 hours and whose probability of survival score (TRISS) was 0.010 to 0.975.

Blood product use

The above data suggested that the group of patients who received 1 to 4 units of RBC in the first 24 hours, despite having been identified by the transfusion of at least 1 unit of uncrossmatched RBC in the TRU, represented too heterogeneous a population and had too low a total RBC and plasma use to support realistic analysis of the efficacy of plasma resuscitation. Therefore, this group was not analyzed further.

The next stage of analysis focused on those who got 5 or more units of RBCs in the first 24 hours and whose predicted probability of survival/TRISS was 0.010 to 0.975.

Timing of plasma repletion

Figure 1 and Tables 2 and and33 summarize RBC and plasma use in this cohort. Of the 393 patients who got any plasma, 130 got 5 to 9 units of RBCs in the first 24 hours with 32 deaths (24.6%) and 263 got >9 RBC units with 125 deaths (47.5%, P<0.001). More than half (58.3%) of these patients got more than 2/3 of their total plasma replacement before the end of the third hour of resuscitation Almost 80% (77.9%) got more than two thirds of their total plasma replacement before the end of the sixth hour. In the group as a whole, and in the two subgroups, proportional mortality was less among those who got more plasma after three or six hours, but the differences were not statistically significant. Probability of survival/TRISS also appeared more favorable among those who got additional plasma after three hours and in the massive transfusion group – this difference was significant (P=0.02). Median times to death were shorter in all the groups who achieved high ratios of plasma early.

Figure 1
Mean hourly usage of units of RBC and plasma among patients with probability of survival scores (TRISS) between 0.010 to 0.975 and categorized as A) all patients receiving at least 5 units of RBC in the first 24 hours, B) patients receiving 5 to 9 units ...
Table 3
Timing of plasma transfusion, probability of survival, mortality, and time to death.

Plasma deficit versus plasma ratio as predictors of mortality

The three panels of Figure 2 show mortality (number of deaths in the stated interval divided by number of those surviving to that interval) in three plasma deficit status categories at intervals over the first 24 hours. Panel A includes all 438 patients; Panel B, the 169 patients who got 5 to 9 units of RBC in the first 24 hours and Panel C, the 269 massive transfusion patients, those who got >9 units in that period. Overall and in the massive transfusion group (Panels A and C), high, moderate, and low deficit status are associated respectively with high, mid-range, and lower mortality in the first hour of resuscitation (P<0.001). By the end of the second hour of resuscitation, this association is fading in the overall group (Panel A, P=0.01) and has disappeared in the massive transfusion group (Panel C, P=0.3).

Figure 2
Figure 2a. Proportional mortality among patients who had probability of survival scores (TRISS) between 0.010 to 0.975, and who survived to specific time points, grouped by deficit status at that time point as ‘Low’ deficit = 0–2 ...

The 5 to 9 RBC units group (Panel B) presents quite a distinct picture. As noted above, overall mortality is significantly less in this group, and this difference is clearly visible by the end of the first hour. Mortality in the high deficit status category varies markedly in the first three hours but after three hours, no more deaths are recorded in this deficit category. Mortality in the moderate deficit category is elevated in the first hour, approaching significance (0.06), and drops thereafter, reprising the gradient appearance of the plasma deficit categories across time in the overall and >9 RBC groups. Mortality in the low deficit category in this 5 to 9 RBC units group remains essentially unchanged at about 20% throughout the 24 hour period but by 12 and 24 hours is significantly greater than in either the mid-range or high-deficit categories (P=0.03 and 0.02, respectively.)

The two sections of Table 4 examine plasma deficit and plasma ratio as predictors of mortality. At 24 hours in the overall group, plasma ratio did not distinguish mortality (P=0.5). In contrast, plasma deficit category in these patients at 24 hours was clearly associated with mortality (P=<0.001). However, it was also clearly associated with probability of survival/TRISS (P=0.001); that is, those with a high deficit had significantly higher mortality but also significantly worse probability of survival/TRISS. At 24 hours into resuscitation in the 5 to 9 RBC units group, neither deficit status nor ratio predicted mortality. In the massive transfusion group at 24 hours, as in the overall group, worse plasma deficit status was associated with mortality (P=0.007) and with TRISS (P=0.01), but plasma ratio was not associated with either mortality or TRISS (P=0.4 for both). Likewise, at three hours into resuscitation, ratio was not associated with mortality in the combined, the 5 to 9 RBC unit, or the >9 RBC unit groups (P=0.2, 0.5, and 0.4, respectively), whereas worsening plasma deficit was clearly associated with mortality in these same groupings (P<0.001, P =0.03 and 0.003, respectively).

Table 4
Plasma deficit (units of RBC transfused minus units of plasma transfused) and ratio of plasma repletion (expressed as units of plasma divided by units of RBC) as predictors of mortality in rapidly bleeding trauma patients.

Early plasma repletion status as a predictor of RBC use at 24 hours

Low deficit status at three hours did not independently predict use of less than 10 units of RBC at 24 hours (data not shown). However, when we repeated this analysis among a further subgroup with a probability of survival/TRISS between 0.200 and 0.800 (SD: <0.200; mortality 35.1% to 63.6%) an effect emerged. In these patients, low deficit status at three hours did predict the use of less than 10 units of RBC at 24 hours (Table 5) as well as lower mortality (P=0.02), independent of probability of survival (P=0.3 for likelihood of true difference between TRISS among the three deficit status categories).

Table 5
Plasma deficit (units of RBC transfused minus units of plasma transfused) at 3 hours as a predictor of RBC use by 24 hours into resuscitation in a subgroup of 154 rapidly bleeding patients with probability of survival (TRISS) between 0.200 and 0.800.

Discussion

Our trauma patient cohort was selected for receiving at least one unit of uncrossmatched Group O RBC in the first hour of care and broken down into groups and categories based on specific subsequent patterns of RBC and plasma use. The cohort was drawn from a civilian population of trauma patients who often require massive transfusion: young, predominantly male, and with predominantly blunt injuries but with a slightly higher proportion of penetrating injuries than would be anticipated. (12) However, the cohort also contains a sizeable group of older female patients admitted after motor-vehicle-associated injury (including pedestrians struck by motor vehicles) or after falls. These “new trauma demographics” are the probable explanation for the significantly older mean age of the subgroup who received 5 to 9 units of RBCs in the first 24 hours. They also require consideration when choosing scoring systems to compare injury severity.

Overall mortality among those undergoing massive transfusion at our center has improved somewhat since 2000 (14) but remains very high. Likewise, our patterns of plasma use have changed somewhat with the availability of thawed AB plasma in our TRU, but overall, our center has been in the vanguard of advocating 1:1 plasma to RBC repletion in rapidly bleeding trauma patients for more than a decade, which we assume is part of the explanation for our past inability to demonstrate strong differences in mortality between massively transfused patients from more recent and historical control groups. (11,13)

Because we developed our study groups from retrospective trauma registry data derived routinely as an institutional process rather than from the dedicated activity of an independent prospective cohort or case-control study, we chose to limit analyses to those descriptive statistics and univariate analytic procedures that we felt best represented the degree of inference legitimately possible from these data. We specifically chose not to apply Cox regression analysis to our data because the usefulness of the resulting inferences is highly dependent on the independence of the variables. Given both the general nature of registries as data repositories and the intense correlation of blood use with injury severity, we felt that a more cautious approach to analysis was justified.

Our decision to use probability of survival or TRISS, rather than the more conventional ISS, as our principle means of attempting to control for injury severity emerges from this same dependence on registry data. Our registry routinely provides Revised Trauma Score (RTS), ISS, and TRISS. The RTS is a physiologic score derived from the Glasgow Coma Scale (GCS), systolic blood pressure and respiratory rate at admission to the trauma center but does not include any anatomic assessment or factors for age or mechanism of injury. The ISS is a sum-of-squares weighting of injury severity in the three worst injured of six body regions but allows only one score per body region and incorporates only anatomic injury assessment. The TRISS methodology for assessing probability of survival combines the RTS and ISS with additional factors for age and blunt versus penetrating injury and thus incorporates the limitations and problems associated with both RTS and ISS.(23) With that acknowledged, among the scoring systems available to us, we felt that probability of survival/TRISS best reflected the diversity of anatomic and physiologic injury in our population and would be the best tool to tease out the confounding effects of injury severity on our data.

Overall, our study illustrates very clearly the problems associated with trying to draw conclusions based on retrospective data and inclusion criteria and outcomes at 24 hours in massively bleeding patients. The problem of survivor bias in such studies was elegantly displayed by Snyder and colleagues from the University of Alabama.(24) In that study, the authors tracked massively transfused patients through low to high plasma repletion groups over the course of the first 24 hours of treatment and demonstrated that what looks like decreased mortality among those with better plasma-to-RBC replacement ratios is just a matter of surviving long enough to move from a low-repletion group (the majority of patients in the early hours after admission) into higher-repletion groups.

The Alabama study is also interesting in that it provides the key both to our inability to demonstrate decreased mortality on the basis of ratio of plasma repletion and on our success in demonstrating decreased mortality on the basis of deficit. In the Alabama study, the median time to first plasma replacement was 93 minutes. Our data, which also follows a group of massively bleeding patients through the early hours of resuscitation, shows a different aspect of the survivor bias issue in a situation where median time to first plasma replacement was less than half that observed by Snyder and colleagues. In a situation where massively bleeding patients begin aggressive RBC and plasma resuscitation within moments of admission, as they do at our center, the movement of patients from a low ratio (or high deficit) group into a mid- or high ratio/low deficit group happens quickly. The people who remain in the low ratio/high deficit group at the end of the second and third hours are to a large extent not those who got left behind but rather are those who could survive without aggressive plasma replacement. The striking disappearance, in Figure 2 Panel B, after the 3rd hour of resuscitation, of mortality associated with high deficit status is a clear illustration of this.

On the other hand, at our center, massively bleeding patients are not only recognized immediately and correctly (as judged by the close association with the subsequently derived TRISS scores), but their plasma use status is immediately differentiated by the transfusion of 10 to 12 units of uncrossmatched group O RBC and 4 units of thawed AB plasma, thereby automatically thrusting them into a moderate to high deficit status. The results shown in Table 5 suggest that at least in some severely but not overwhelmingly injured patients, early 1:1 repletion of plasma may have decreased mortality and prevented the need for massive transfusion. Acknowledging their questionable epidemiologic legitimacy as a retro-fitted analysis, the results in Table 5 nevertheless support the advocacy of some major trauma centers, including Vanderbilt and the University of Copenhagen, for protocol-based preemptive trauma transfusion packages at roughly 1:1 repletion levels that, in the experience of those groups, have not only improved mortality but appear to decrease RBC and plasma use. (8,9)

Summary and conclusions

The recent explosion of clinical studies and expert commentary on plasma repletion in trauma patients with massive transfusion has tended to fall out into two general patterns. Most clinical studies show benefit from early repletion approaching 1:1 and are confounded by survival bias, and most expert opinion seeks to define, one way or another, an optimal replacement ratio. Here we present a large a 5-year retrospective review of blood products issued and mortality from a large center and draw somewhat different inferences. First, the effects of plasma repletion play out in the first two to three hours of care for massively bleeding individuals and that plasma deficit rather than unit ratios may be a more indicative measure. Second, the achievable differences in outcome are real but far less dramatic than suggested in most earlier reports on plasma repletion. Finally, although we would like to see randomized controlled trials, what is probably more important is a re-focus of research into the clinical predictors that will allow clinicians on the front line to know who is going to need early plasma repletion and who will do well without it.

Acknowledgments

Drs. Dutton, Scalea, and Hess have received support from NHLBI grant 5U01HL072359-07

Footnotes

Conflict of Interest: The authors report no financial conflict of interest with the content of this manuscript.

References

1. Borgman MA, Spinella PC, Perkins JG, et al. The ratio of blood products transfused affects mortality in patients receiving massive transfusions at a combat support hospital. J Trauma. 2007;63:805–13. [PubMed]
2. Duchesne JC, Hunt JP, Wahl G, et al. Review of current blood transfusions strategies in a mature level I trauma center: were we wrong for the last 60 years? J Trauma. 2008;65:272–6. [PubMed]
3. Maegele M, Lefering R, Paffrath T, et al. Red-blood-cell to plasma ratios transfused during massive transfusion are associated with mortality in severe multiple injury: a retrospective analysis from the Trauma Registry of the Deutsche Gesellschaft fur Unfallchirurgie. Vox Sang. 2008;95:112–9. [PubMed]
4. 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;248:447–58. [PubMed]
5. Kashuk JL, Moore EE, Johnson JL, et al. Postinjury life threatening coagulopathy: is 1:1 fresh frozen plasma: packed red blood cells the answer? J Trauma. 2008;65(2):261–70. [PubMed]
6. Sperry JL, Ochoa JB, Gunn SR, et al. An FFP:PRBC transfusion ratio >/=1:1.5 is associated with a lower risk of mortality after massive transfusion. J Trauma. 2008;65:986–93. [PubMed]
7. Johansson PI, Stensballe J, Rosenberg I, et al. Proactive administration of platelets and plasma for patients with a ruptured abdominal aortic aneurysm: evaluating a change in transfusion practice. Transfusion. 2007;47:593–8. [PubMed]
8. Johansson PI, Stensballe J. Effect of Haemostatic Control Resuscitation on mortality in massively bleeding patients: a before and after study. Vox Sang. 2009;96:111–8. [PMC free article] [PubMed]
9. Cotton BA, Gunter OL, Isbell J, et al. Damage control hematology: impact of a defined exsanguination protocol on mortality and blood utilization. J Trauma. 2008;64:1177–1183. [PubMed]
10. Gunter OL, Jr, Au BK, Isbell JM, et al. Optimizing outcomes in damage control resuscitation: identifying blood product ratios associated with improved survival. J Trauma. 2008;65:527–34. [PubMed]
11. Stansbury LG, Dutton DP, Stein DM, et al. Controversy in trauma resuscitation: Do ratios of plasma to red cells matter? Transfus Med Rev. 2009;23:255–65. [PubMed]
12. Dutton DP, Stansbury LG, Leone S, et al. Trauma mortality in mature trauma systems: are we doing better? An analysis of trauma mortality patterns, 1997–2008. J Trauma. 2010 January 10; E-pub. [PubMed]
13. Scalea TM, Bochicchio KM, Lumpkins K, et al. Early aggressive use of fresh frozen plasma does not improve outcome in critically injured trauma patients. Ann Surg. 2008;248:578–84. [PubMed]
14. Como JJ, Dutton RP, Scalea TM, Edelman BB, Hess JR. Blood transfusion rates in the care of acute trauma. Transfusion. 2004;44:809–13. [PubMed]
15. Hess JR, Hiippala S. Optimizing the use of blood products in trauma care. Crit Care. 2005;9(Suppl 5):S10–4. Epub 2005 Oct 7. [PMC free article] [PubMed]
16. Ketchum L, Hess JR, Hiippala S. Indications for early fresh frozen plasma, cryoprecipitate, and platelet transfusion in trauma. J Trauma. 2006 Jun;60(6 Suppl):S51–8. [PubMed]
17. Hess JR, Dutton RB, Holcomb JB, Scalea TM. Giving plasma at a 1:1 ratio with red cells in resuscitation: who might benefit? Transfusion. 2008 Aug;48(8):1763–5. Epub 2008 May 14. [PubMed]
18. Nunez TC, Dutton WD, May AK, Holcomb JB, Young PP, Cotton BA. Emergency department blood transfusion predicts early massive transfusion and early blood component requirement. Transfusion. 2010 May 7; [Epub ahead of print] [PubMed]
19. Baker SP, O’Neill B. The injury severity score: an update. J Trauma. 1976;16:882–5. [PubMed]
20. Boyd CR, Tolson MA, Copes WS. Evaluating trauma care: the TRISS method. Trauma Score and the Injury Severity Score. J Trauma. 1987;27:370–8. [PubMed]
21. Gerstman B. Chapter 12 “Analysis of Variance” StatPrimer (Version 6.4) accessed as: http://www.sjsu.edu/faculty/gerstman/StatPrimer/anova-a.pdf.
22. Fleiss JL. Statistical Methods for Rates and Proportions. New York: John Wiley and Sons; 1973. p. 17.
23. Pohlman TH, Bjerko HS, Offner P. Trauma scoring systems. [Accessed 17 August 2010]. eMedicine.medscape.com/article/434076.
24. Snyder CW, Weinberg JA, McGwin G, Jr, et al. The relationship of blood product ratio to mortality: survival benefit or survival bias? J Trauma. 2009;66:358–62. [PubMed]