Anemia in the Setting of Traumatic Brain Injury: The Arguments For and Against Liberal Transfusion

Cerebral oxygen delivery

The main theoretical concern about anemia in the setting of TBI is that the injured brain may require normal (or possibly even supranormal) delivery of oxygen (DO₂) for optimal healing and clinical outcome. Cerebral DO₂ is the product of the oxygen content of arterial blood (C_aO₂) and cerebral blood flow (CBF), though it may be incorrect to assume that the construct of CBF at the global level is adequately representative of the spectrum of flow characteristics at the local level through the capillary network of the cerebrovasculature. C_aO₂ is largely dependent on the concentration of Hb and the extent to which Hb in arterial blood is bound with oxygen (oxygen saturation, or S_aO₂): \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland,xspace}\usepackage{amsmath,amsxtra}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6} \begin{document} \begin{align*} C_aO_2 ( mL / dL ) &= 1.39 \times [ Hb ] ( g / dL ) \times S_aO_2 ( \% ) \\&\quad+ 0.0031 \times P_aO_2 ( mm \ Hg ) \end{align*} \end{document}

Oxygen dissolved in plasma (0.0031 × P_aO₂) is a relatively small proportion of C_aO₂ at normal Hb levels but it can contribute to an appreciable proportion of oxygen consumption (VO₂) when 100% inspired fraction of oxygen (F_iO₂) is administered in the setting of severe anemia (47% of VO₂ at Hb = 7 g/dL and 74% at 3 g/dL) (Habler et al., 1998). By the Hagen-Poiseuille equation, CBF is proportional to cerebral perfusion pressure (CPP) and the radius (r) of the vessel to the fourth power and inversely proportional to the viscosity of blood (μ) and the length of the vessel (L): \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland,xspace}\usepackage{amsmath,amsxtra}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6} \begin{document} \begin{align*} CBF \propto \frac { CPP \times \pi \times r^4 } { 8 \times \mu \times L } \end{align*} \end{document}

Because blood behaves as a non-Newtonian fluid, the Hagen-Poiseuille equation does not precisely describe the relationship between CBF and viscosity, especially at low shear rates (Kee and Wood, 1984). Nonetheless, it holds that small changes in vascular tone generally have a large impact on CBF, and vascular tone mediates the influence of blood pressure (i.e., “pressure autoregulation”), carbon dioxide concentration (“CO₂ reactivity”), oxygen concentration, and local metabolites on CBF.

In the healthy state, several compensatory mechanisms offset the decreases in C_aO₂ that occur with anemia. Systemically, increases in heart rate and stroke volume result in increased cardiac output, while blood viscosity decreases. Changes in viscosity at the level of the microcirculation – though they are complex and involve factors such as erythrocyte aggregation and flexibility, platelet aggregation, and fibrinogen and α₂-macroglobulin levels – are predominantly a function of hematocrit (Kee and Wood, 1984) and thus reflect the restoration of lost blood volume with fluid shifts and administered crystalloid intravenous fluids. Mild anemia appears to increase cerebral vascular tone because decreased viscosity may paradoxically allow increased oxygen delivery (Hudak et al., 1989; Muizelaar et al., 1984), but severe anemia decreases vascular tone, augmenting CBF (Haggendal et al., 1966). The relative contributions of decreased viscosity on the one hand and vasodilation of the cerebrovasculature (resulting from decreased C_aO₂) on the other have been debated, but it appears that both factors help increase CBF at Hb levels <7 g/dL (Borgstrom et al., 1975; Fan et al., 1980). Vasodilation appears to be mediated by release of nitric oxide from perivascular cerebral neurons and sympathetic stimulation of β₂ adrenergic receptors (Hare et al., 2008). In rat models, selective inhibition of neuronal nitric oxide synthase (nNOS) resulted in attenuated increases in erythrocyte velocity with anemia (Hudetz et al., 2000), hemodilution increased nNOS expression (Hare et al., 2003), and treatment with a β₂ antagonist partially abrogated the increase in cortical CBF induced by anemia (Hare et al., 2006).

The net result of these compensatory mechanisms is that cerebral DO₂ and jugular venous oxygen saturation remain fairly constant despite increasing severity of anemia (Borgstrom et al., 1975; Fan et al., 1980). Perhaps because normal DO₂ in the absence of anemia is more than sufficient to meet the brain's oxygen consumption, moderate decrements in DO₂ with anemia appear to be well tolerated. With severe decreases in C_aO₂ from anemia, both CBF and oxygen extraction increase markedly (Brown et al., 1985; Todd et al., 1994a; van Bommel et al., 2002). However, at Hb levels <3.5 g/dL, these compensatory mechanisms probably can no longer stave off cerebral hypoxia (van Bommel et al., 2002).

Acute isovolemic anemia and brain function in healthy human subjects

Studies of acute isovolemic anemia in healthy volunteer human subjects suggest CNS function becomes impaired at much more moderate levels of anemia than those that cause cerebral hypoxia (Hb = 3.5 g/dL). This laboratory technique involves phlebotomy and isovolemic replacement of fluid losses with plasma and albumin. In this setting, anemia results in cognitive impairment when the Hb level drops to <7 g/dL (Weiskopf et al., 2000). Deficits include slower responses in simple addition tests, increased reaction time for digit–symbol substitution, and poorer immediate and delayed memory, all of which can be reversed by increasing Hb levels (Weiskopf et al., 2000) or administering supplemental oxygen (Weiskopf et al., 2002). Subjects with Hb = 5 g/dL breathing room air have increased latencies of P300 potentials with auditory testing, suggesting that the observed decrement of cognitive function is mediated by central processing rather than a non-specific effect of anemia on attention (Weiskopf et al., 2005). At Hb = 5 g/dL, only a small proportion of healthy subjects have electrocardiographic changes possibly representing myocardial ischemia (Leung et al., 2000), and reduction of systemic DO₂ to very low levels (7.3 mL O₂/kg·min) with beta-blocker medication in subjects with Hb = 5 g/dL results in virtually no change in arterial lactate levels (Lieberman et al., 2000). These observations suggest that, for healthy subjects, the brain may be the most sensitive organ in the body to anemia – though only at levels as severe as 5–6 g Hb/dL. Because the injured brain may have different oxygen requirements and impaired autoregulation compared with the uninjured brain, the relevance of these findings to the clinical management of TBI remains unclear.

Anemia in the setting of traumatic brain injury

Anemic patients with TBI may be at risk for cerebral hypoxia for two additional reasons besides merely the anemia. First, by virtue of hemorrhage or arterial hypoxia from associated extracranial injuries, patients with TBI are prone to unexpected further decrements in cerebral DO₂. Second, the injured brain may be particularly sensitive to harm from such insults. Neuropathological evidence of brain ischemia is evident in approximately 90% of deaths from TBI (Graham et al., 1989). Also consistent with this notion of occult ischemia is the observation that increased F_iO₂ allows the metabolic rate of at-risk brain tissue in patients with TBI to increase (Nortje et al., 2008). Episodic and/or sustained decrements in DO₂ could lead to infarction of this at-risk tissue without being severe enough to affect other organ systems. Because the injured brain may not reflect obvious clinical manifestations of secondary injury in real time, the fear is that such an insult during the early post-injury period would remain occult until after the damage becomes irreversible.

Furthermore, several studies suggest that the protective mechanisms that compensate for decreased DO₂ are disrupted with TBI. Xenon-enhanced computed tomography has shown that global or regional decrements in CBF are common within hours following TBI (Bouma et al., 1992), and CO₂ reactivity is also abnormal during this time (Marion et al., 1991). In animal models, TBI followed by blood loss reduces CBF, DO₂, and electroencephalographic activity (DeWitt et al., 1992). Consistent with this notion, experimental TBI has been shown to attenuate markedly the increase in CBF that naturally occurs with isovolemic hemodilution, dramatically reducing DO₂ (Todd et al., 1994b). In humans with TBI, CBF does not increase as anemia becomes more severe, as one would expect under normal circumstances (Cruz et al., 1993). Clinical studies suggest that anemia may contribute to episodes of low jugular venous saturation (S_jO₂) (Robertson, 1993; Schoon et al., 2002; Sheinberg et al., 1992), which have been associated with worse neurologic outcome (Gopinath et al., 1994; Robertson, 1993; Schoon et al., 2002).

Anemia and traumatic brain injury in animal models

Experimental manipulation of the severity of anemia in the setting of brain injury or ischemia might allow identification of an optimal hematocrit or a threshold below which anemia is clearly harmful. One study of a dog ischemic brain model examined the influence of isovolemic hemodilution on infarct size and found that a hematocrit of 30% resulted in smaller infarcts than hematocrits of 25%, 35%, 40%, or no hemodilution (50%) (Lee et al., 1994). The authors postulated that the optimal balance of viscosity and C_aO₂ occurs at a hematocrit of 30%. A similar study in rabbits found larger infarcts with Hb = 6 g/dL versus 11 g/dL (Reasoner et al., 1996). Hemodilution to a hematocrit of 31% (as opposed to 26% or 36%) has also been associated with maximal DO₂ and cerebral metabolic rate following global brain ischemia (Tu et al., 1997). In a rat model of TBI, dilutional anemia to Hb = 5–7 g/dL resulted in increased cerebral contusion area and evidence of increased neuronal apoptosis (Hare et al., 2007). Considered together, the implication of these studies is that TBI may be aggravated by levels of Hb <10 g/dL.

Anemia and outcomes from traumatic brain injury

Some of the first studies to recognize secondary insults to the injured brain focused on the potential harm from anemia (Table 1). Over 30 years ago, Miller and associates (1978) described anemia (defined as hematocrit <30%) among other factors that, in composite, were associated with increased mortality from TBI. However, no attempt was made to account for confounding by extracranial injuries and the influence of anemia was not evaluated independent of other secondary insults, such as hypotension and hypoxia. A subsequent larger study from the same group did not confirm anemia by itself to predict poor neurologic outcome (Miller et al., 1981).

The several retrospective cohort studies that have examined clinical outcomes have not consistently demonstrated harm from anemia. Ariza and associates (2004) found no association between having a hematocrit <30% during the first 3 days post injury and 6-month functional neurological outcomes as assessed by a series of neuropsychological tests. However, this study involved a relatively small number of subjects (n = 57), the investigators did not account for patients who were too impaired to participate in testing, and they did not describe any attempt to adjust for confounding factors. Salim and associates (2008) conducted a retrospective analysis of 1150 patients with severe TBI to examine the influence of anemia on mortality. They defined anemia as Hb <9.0 g/dL for three consecutive measurements within the first 7 days of hospitalization and found that, without adjusting for transfusion, anemia was predictive of 1.6 times higher odds of death. Van Beek and associates (2007) examined the association between initial Hb and different cutoff values of the Glasgow Outcome Scale (GOS) score using data from several prior randomized trials involving patients with TBI. They described a consistent pattern of worse outcome with lower Hb, with no particular threshold that resulted in better prediction of poor outcomes. Sánchez-Olmedo and associates (2005) found that Hb <10.0 g/dL within the first 24 h of ICU admission was not predictive of brain death among patients with severe TBI. Duane and associates (2008) defined anemia as Hb <8.0 g/dL and found it to be associated with increased mortality. In an analysis of 169 patients with severe TBI, Carlson and associates (2006) concluded that each additional day with a hematocrit <30% was associated with a favorable increment in the GOS and Rancho Los Amigos scores at hospital discharge. However, they included both the number of RBC units transfused and lowest hematocrit in their model and did not present associations as a summary risk measure, making interpretation of their findings difficult. Schirmer-Mikalsen and associates (2007) did not identify an association between having a Hb <8 g/dL and worse GOS.

Case reports exist of good neurological outcomes in patients with TBI who developed severe anemia but refused transfusion (Cothren et al., 2002; Logemann et al., 1997; Victorino and Wisner, 1997). Jehovah's Witness patients undergoing neurosurgical procedures do not seem to experience adverse outcomes (Suess et al., 2001); however, this study evaluated few patients with TBI and anemia was not severe (mean hematocrit decrease to 33%).

The challenges in applying findings from these studies to clinical practice include inconsistent definitions of TBI and anemia, inadequate consideration of the time frame during which anemia may be important, failure to account for variation in the degree of anemia over time, residual confounding from unmeasured or unknown factors that are correlated with anemia but that independently affect outcome, and lack of agreement on the most meaningful outcome measures. Furthermore, it remains unknown whether the relationship between anemia and outcome may be influenced by other factors such as cerebral perfusion pressure, pH, P_aCO₂, and cerebral metabolic rate (i.e., temperature, sedative medications).

Transfusion and outcomes from traumatic brain injury

Measures to curtail blood loss from injuries and phlebotomy are obviously helpful, and recent advances in more aggressively preventing and treating trauma-induced coagulopathy – such as limiting crystalloid fluids, replacing clotting factors and fibrinogen prior to the development of laboratory abnormalities (Holcomb et al., 2008; Karlsson et al., 2009), and administering antifibrinolytic agents (Shakur et al., 2010) – provide more tools to limit blood loss. However, the development of anemia in the setting of TBI is not always preventable, and, once all measures to limit blood loss have been exhausted, the most relevant question from a therapeutic standpoint then becomes what criteria should be used to prompt transfusion of such patients. Presently, there is little direct information from randomized trials to evaluate this question for euvolemic patients with TBI.

A series of observational studies have examined the association of transfusion per se with clinical outcomes (Table 2). George and associates (2008) retrospectively analyzed 82 patients with severe TBI whose lowest Hb value was 8.0–10.0 g/dL, thus excluding patients who were either highly likely to receive blood (lowest Hb <8.0 g/dL) or highly unlikely to do so (lowest Hb >10.0 g/dL). With adjustment, transfusion was associated with a greater hazard of death (Cox proportional hazards regression) but no difference in the likelihood of in-hospital death (logistic regression). The aforementioned study by Salim and associates (2008) also examined the association between transfusion and mortality, and found that, with adjustment for the presence of anemia as a binary predictor (and inclusion of an interaction term between anemia and transfusion in the model), transfusion was associated with a 2.2 times higher odds of death. Duane and associates (2008) identified that the number of units of all blood products (RBCs, plasma, platelets, and cryoprecipitate) transfused in patients with isolated TBI was an independent predictor of mortality, even after adjustment for the minimum Hb value; however, they did not find such a relationship when they focused on RBC units only. The studies by Carlson and associates (2006) and Warner and associates (2010) both identified an association between any blood transfusion and poorer functional outcome.

Studies using surrogate outcomes suggest that transfusion may improve some surrogate measures. Transfusion increases brain tissue oxygen levels (P_btO₂) in neurosurgical patients (Figaji et al., 2010; Smith et al., 2005; Zygun et al., 2009), yet the effect may have been mediated by expansion of intravascular volume instead of increased red cell mass per se (Griesdale and Chittock, 2005). Decreases in P_btO₂ have been correlated with worse outcome (Maloney-Wilensky et al., 2009; Stiefel et al., 2006), but this relationship has not been observed universally (Meixensberger et al., 2003; Vath et al., 2001) (perhaps because the relationship may not be causal or local measurement may not reliably reflect global cerebral oxygenation). Transfusion of patients with subarachnoid hemorrhage to Hb = 10 g/dL has been associated with greater risk of vasospasm (Smith et al., 2004).

Although the above studies appear to demonstrate harm from transfusion, as with the analyses of anemia as a predictor of outcome, they suffer from unmeasured or unknown characteristics related to the severity of the illness that potentially confound any relationship between transfusion and outcomes. They also generally failed to consider the potential for a dose-dependent effect from transfusion. Those studies that include terms for both anemia and transfusion in statistical models might also have introduced so much collinearity into the models that interpretation of the results becomes difficult. Also, testing multiple different versions of the same hypothesis without correcting the alpha level (i.e., setting it to less than 0.05) might have led to identification of spurious associations.

The design advantages of randomized trials can potentially address some of these weaknesses. The only information from a randomized trial to address this topic comes from the Transfusion Requirements in Critical Care (TRICC) investigators. The TRICC trial (Hebert et al., 1999) (see more detail below) evaluated a liberal (transfusion to keep Hb = 10.0–12.0 g/dL) versus a restrictive (transfusion to keep Hb = 7.0–9.0 g/dL) transfusion policy. The investigators presented hypothesis-generating post hoc subgroup analyses of both the 203 trauma patients enrolled in the trial (McIntyre et al., 2004) and the 67 such patients with moderate to severe TBI (McIntyre et al., 2006), which did not show a difference between treatment arms in 30-day mortality (Table 2).

Transfusion of the critically ill or injured patient

Many observational studies of critically ill patients also suggest that blood transfusion is associated with worse outcomes, even with attempts to control for shock and the severity of the underlying illness. Two large prospective cohort studies in Europe and the United States found transfusion to be an independent risk factor for death in ICU patients in general (Corwin et al., 2004; Vincent et al., 2002). These studies both used propensity score techniques to adjust for baseline imbalances associated with the indication for transfusion, which possibly increases the ability to isolate the impact of transfusion from confounding factors. Studies specifically of patients with traumatic injuries describe transfusion to be associated with increased risk of infection (Agarwal et al., 1993; Carson et al., 1999; Edna and Bjerkeset, 1992; Graves et al., 1989), multiple organ failure (Moore et al., 1997; Sauaia et al., 1994; Tran et al., 1993), and death (Dunne et al., 2004; Malone et al., 2003). Increased risk for infection has also been associated with blood transfusion following elective spine procedures (Triulzi et al., 1992), cardiac valve or revascularization procedures (Miholic et al., 1985), colorectal cancer resection (Tartter, 1988), and bowel resection for Crohn's disease (Tartter et al., 1988). Transfusion has long been linked to acute lung injury (Silliman et al., 2009; Triulzi, 2009), and it may increase the risk for repeat hemorrhage in the setting of acute gastrointestinal bleeding (Hearnshaw et al., 2010). Even in patients with cardiac risk factors, who presumably would be at greater risk for ill effects of anemia, transfusion to a higher level of Hb does not seem to be associated with improved survival (Hebert et al., 2001). One evaluation of patients with acute coronary syndromes enrolled in major cardiology trials found, after adjustment for confounding factors, increased mortality associated with transfusion (Rao et al., 2004). In practice, physicians generally accept relatively low transfusion thresholds (Hb = 7.9 g/dL) for ICU patients whose primary diagnosis is a cardiac problem (Walsh et al., 2005).

The aforementioned Transfusion Requirements in Critical Care (TRICC) trial (Hebert et al., 1999) that evaluated a liberal (Hb = 10.0–12.0 g/dL) versus a restrictive (Hb = 7.0–9.0 g/dL) transfusion policy suggests that adult ICU patients who have adequate intravascular volume and moderate anemia do not benefit from blood transfusion. In this non-inferiority comparison of 838 patients, the primary outcome of 30-day mortality was similar in the two groups: 18.7% in the restrictive group versus 23.3% in the liberal group (absolute difference 4.7%; 95% C.I. − 0.8 to 10.2%). Subgroup analyses that were planned a priori showed that younger patients (<55 years of age) and those who were less severely ill (APACHE II score ≤20) had greater mortality with liberal transfusion. An analysis of the “number needed to harm” suggested that, for every 40 additional units of blood transfused to patients in these two subgroups in order to keep the Hb at at least 10 g/dL, one additional death would occur (Ely and Bernard, 1999).

A similar non-inferiority trial in anemic pediatric ICU patients yielded results consistent with no difference between approaches. The Transfusion Requirements in the Pediatric Intensive Care Unit (TRIPICU) Study compared liberal (transfusion threshold of Hb = 9.5 g/dL) versus restrictive (threshold of Hb = 7.0 g/dL) transfusion strategies among 637 children up to 14 years old. The primary outcome, a composite of death or new or progressive organ dysfunction at 28 days, occurred in 11.9% of patients in the restrictive arm and 12.3% in the liberal arm (absolute difference 0.4%; 95% C.I. − 4.6 to 5.5%).

A meta-analysis of restrictive versus liberal transfusion policies that incorporated other much smaller trials did not reveal any substantial heterogeneity in outcomes across trials (Hill et al., 2002).

Putative mechanisms of harm from transfusion

If transfusion harms patients, whether young, less sick, or otherwise, the mechanism by which it does so remains under speculation. The TRICC trial did not reveal a specific mechanism though it raised the possibility that liberal transfusion causes more cardiac complications (Hebert et al., 1999). Blood transfusion has long been recognized to suppress the immune system, possibly from allogeneic erythrocytes, allogeneic leukocytes, or soluble factors (Blajchman, 2002; Raghavan and Marik, 2005). Another possible mechanism is that transfused erythrocytes are not as deformable as native erythrocytes and thus may block microcirculatory flow (Marik and Sibbald, 1993; Morisaki and Sibbald, 2004). Increased age of blood has also been postulated to be harmful (Zallen et al., 1999), and a small randomized trial has recently been conducted to evaluate this possibility specifically in patients with TBI (ClinicalTrials.gov, 2010b).

Threshold for blood transfusion

Although clinical guidelines have recommended that anemia not be the sole consideration in decisions regarding transfusion (NIH Consensus Conference, 1988), large prospective observational studies, as well as recent surveys, continue to suggest that physicians frequently use Hb or hematocrit values as the main determinant in this decision (Corwin et al., 2004; Hebert et al., 1998; Palmieri and Greenhalgh, 2004; Sena et al., 2009). Although actual transfusion practices involving TBI patients are not well described, neurosurgeons appear to prefer higher Hb values as a transfusion threshold. Among physicians at U.S. Level I trauma centers, neurosurgeons favored transfusion at a higher mean Hb threshold than trauma surgeons or non-surgeon intensivists in clinical scenarios whether the intracranial pressure was presented as normal (8.3 vs. 7.5 and 7.5 g/dL respectively) or elevated (8.9 vs. 8.0 and 8.4 g/dL respectively) (Sena et al., 2009). Although all three groups of specialists recognized secondary brain injury as an important problem, neurosurgeons much less frequently expressed concern for potentially harmful effects of transfusion.

Since TBI is often sustained by young, otherwise healthy people (Kalsbeek et al., 1980) and the TRICC trial suggests that this particular subgroup of critically ill patients is at risk for harm from liberal transfusion, there seems to be genuine disagreement and clinical equipoise as to whether brain-injured patients would benefit from more liberal (to maintain Hb ≥10 g/dL) or more restrictive transfusion (to maintain Hb ≥7 g/dL).