Effects of Extreme Hemodilution during Cardiac Surgery on Cognitive Function in the Elderly

Subsequent to approval by the Duke University Health System Institutional Review Board (Durham, NC) and informed consent, patients older than 65 yr undergoing coronary artery bypass graft surgery with CPB were enrolled into this clinical trial, designed as a prospective, randomized, blinded, interventional trial. Patients were randomized to two treatment groups: (1) moderate hemodilution (MH), wherein the hematocrit on CPB was maintained at 27% or greater; and (2) profound hemodilution (PH), wherein the hematocrit on CPB was maintained at 15–18%. A group assignment schedule was prepared using a randomization function in SAS® version 9.1 (SAS, Cary, NC) and stored in consecutively numbered sealed envelopes until allocation. Assignments were stratified by two sex groups and five age groups (65–69, 70–74, 75–79, 80–84, and 85+ yr). Within strata, the randomized assignments were balanced in varying block sizes of 6, 8, and 10. Although operating room personnel were aware of the patient’s treatment assignment, both the patient and the investigators responsible for evaluating the neurologic/neurocognitive status of the patient were blinded to the treatment assignment. Patients were excluded from participation if they had a history of symptomatic cerebrovascular disease (e.g. , stroke with a residual deficit), psychiatric illness (any clinical diagnoses requiring therapy), renal failure (serum creatinine >2 mg/dl), active liver disease (liver function tests >1.5 times the upper limit of normal), alcoholism (>2 drinks/day), or chronic anemia (hematocrit <30%); were unable to read; or had less than a seventh grade education.

Anesthesia was induced and maintained with midazolam, fentanyl, and isoflurane. All patients underwent nonpulsatile hypothermic (30°–32°C) CPB. The perfusion apparatus consisted of a Cobe CML membrane oxygenator (COBE Chem Labs, Lakewood, CO), Sarns 7000 MDX pump (3M Inc., Ann Arbor, MI), and Pall SP3840 arterial line filter (Pall Biomedical Products Co., Glen Cove, NY). Perfusion was maintained at pump flow rates of 2–2.4 l · min⁻¹· m²throughout CPB to maintain a mean arterial pressure at 50–80 mm Hg. The pump was primed with crystalloid and arterial blood gases were measured every 15–30 min to maintain arterial carbon dioxide partial pressures of 35–40 mm Hg, unadjusted for temperature (α-stat), and oxygen partial pressures of 150–250 mm Hg.

Before the initiation of CPB, the estimated blood volume for each patient was calculated as described in table 1. For patients randomized to MH, the volume of packed erythrocytes to be added to achieve a hematocrit of 27% or greater was calculated as

For patients randomized to PH, the volume of blood to be removed to achieve a hematocrit of 15–18% was calculated as

where volume added to the pump prime was 500–1,000 ml hetastarch solution, 6%.

In the MH group, packed erythrocytes were added to the pump prime (if needed) before the onset of CPB. In the PH group, the calculated volume of heparinized blood was drained via  the venous circuit into storage bags upon initiation of CPB. Harvested blood was stored at room temperature and added back to the venous reservoir immediately before separation from CPB. Blood was also returned to the circuit during CPB to maintain a hematocrit of 15% or greater or if the patient’s clinical condition dictated the transfusion of blood (e.g. , profound hypotension unresponsive to phenylephrine or mixed venous saturation <50% for more than 10 min).

Trained psychometricians blinded to the treatment group individually examined patients with a battery of five cognitive tests on the day before surgery and again at 6 weeks after CPB. Instruments included the Short Story module of the Randt Memory Test, Digit Span subtest of the Wechsler Adult Intelligence Scale–Revised Test, Modified Visual Reproduction Test from the Wechsler Memory Scale, Digit Symbol subtest of the Wechsler Adult Intelligence Scale–Revised Test, and Trail Making Test (Part B).

To characterize cognitive function over time while minimizing potential redundancy in the cognitive measures, a factor analysis was performed on the 10 cognitive test scores from baseline. The 10 scores were incorporated into a principal components analysis using SAS Proc Factor, with orthogonal rotation (a linear transformation of the data) to produce uncorrelated factors. The factor analysis was conducted on the enrolled patients in this study, and scoring coefficients for all time points were determined using this sample’s baseline rotated factor scores; thus, cognitive domains remained consistent over time. We chose a four-factor solution, which accounts for 84% of the variability in the original 10 test scores and represents four cognitive domains: (1) verbal memory and language comprehension, short-term and delayed; (2) attention, psychomotor processing speed, and concentration; (3) abstraction and visuospatial orientation; and (4) figural memory. Two summary measures were calculated to represent cognitive function: (1) “Cognitive deficit” (the binary outcome) was defined as a decline of 1 SD or more in performance on at least one of the four domains. (2) To quantify overall cognitive function and the degree of learning (i.e. , practice effect from repeated exposure to the testing procedures), a “baseline cognitive index” was first calculated as the sum of the four preoperative domain scores. A continuous change score (the continuous outcome) was then calculated by subtracting the baseline from the follow-up cognitive index.

Categorical and continuous demographic characteristics were compared between treatment groups with Pearson chi-square and t  tests, respectively. Continuous variables are reported as mean ± SD or as median with interquartile range (IQR). The effect of the hemodilution group on postoperative cognition was tested using variable linear and logistic regression modeling accounting for age, years of education, and baseline cognition; interactions with age were also examined. Secondary post hoc  analyses were similarly conducted using the area under the curve for hematocrit below the pre-CPB value as a predictor variable. This variable was chosen because of an abundance of more recent (after the majority of study enrollment) literature indicating that nadir hematocrit was a predictor of adverse outcome after cardiac surgery.7–11 P < 0.05 was considered significant; all analyses were performed with SAS® version 9.1. No adjustment was made for multiple comparisons because all post hoc  analyses were considered exploratory. Because primary outcome comparisons were conducted only after study termination and not during the yearly Data Safety and Monitoring Board review, statistical penalties were not applied for interim analysis.

We expected that the incidence of cognitive deficit in patients older than 65 yr would be approximately 35%. We hypothesized that the profound hemodilution strategy would increase this incidence to 50%, and a sample size of 170 per group would yield power of 80% at a significance level of 0.05 to detect this difference. To allow for a 10% loss to follow-up, we intended to recruit a total of 374 patients.

From June 14, 1999 to February 12, 2002, a total of 121 patients were consented to participate in the study (fig. 1,table 2). Thirteen of these patients were not subsequently enrolled (refused neurologic testing = 6, changed mind = 4, exclusion criteria developed = 1, surgical decision = 1, change in surgical schedule = 1). At the third scheduled review of the Data Safety and Monitoring Board, termination of the trial was recommended because of an increase in the incidence of adverse events. At the point of termination, 56 patients were randomized to the MH group and 52 were randomized to the PH group (n = 108).

Demographic characteristics of the randomized patients are listed in table 2; as expected from randomization, no significant differences were seen. The mean hematocrits during CPB for the PH and MH groups were 18.0 ± 1.7% and 26.9 ± 2.8%, respectively. Mean arterial pressure during CPB in the MH group was 55.6 ± 6.9 mm Hg compared with 52.4 ± 5.2 mm Hg in the PH group (P = 0.11). The mean volume of blood removed in the PH group before the onset of CPB was 1578 ml (IQR, 640–2,000 ml). Eighty-nine percent of the MH patients were transfused with homologous blood products compared with 88% of the PH patients (P = 0.86). Intraoperatively and postoperatively, the MH group received a median of 900 ml (IQR, 600–1,500 ml) packed erythrocytes compared with 900 ml (IQR, 600–1,900 ml) (P = 0.78); similarly, there were no differences in the transfusion of fresh frozen plasma or platelets.

Cognitive deficits, defined as a decline of 1 SD or more in performance on at least one of the four domains, were present at 6 weeks after surgery in 37.5% of patients randomized to MH and in 42.5% of patients randomized to PH (P = 0.65). The continuous cognitive score was also not significantly different between the treatment groups. Multivariable analysis accounting for the covariable effects of age, baseline level of cognition, and years of education, however, revealed a significant treatment group–by–age interaction, such that older patients in the PH group were more likely to experience cognitive decline (binary outcome: P = 0.03, table 3; continuous outcome: P = 0.02).

Post hoc  analyses conducted using the area under the curve for hematocrit below the pre-CPB value as a predictor variable and adjusting for the effects of age, baseline level of cognition, and years of education also revealed a significant interaction between hematocrit-area below baseline and age (P = 0.02). To determine whether the maximum decrease in hematocrit from baseline was as important as the area under the curve (decrease + duration), additional modeling was conducted using only the maximum decrease in hematocrit from baseline as the predictor variable. The mean decrease in the hematocrit from baseline in the MH group was 11.7 ± 4.5% versus  19.9 ± 4.6% in the PH group (P < 0.001). Again, a significant interaction with age was detected; a greater decrease in the cognitive change score was present in older patients with a greater decrease from baseline hematocrit (table 4and fig. 2). To investigate the possibility of a nonlinear association between hematocrit decrease and cognitive change, an analysis using restricted cubic splines was performed in the subset of patients who were aged 70 yr or older (n = 51). Restricted cubic splines, which are smooth at the joint points, or knots (slope is allowed to vary at these points) and which are constrained to be linear in the tails, can greatly improve the fit of the model.12Based on the size of the data set, four knots were placed at the 10th, 25th, 75th, and 90th percentiles. Figure 3shows the resulting fitted line with 95% confidence intervals indicating that cognitive decline was relatively unchanged until the decline in hematocrit from baseline exceeded approximately 12% points.

At the third Data Safety and Monitoring Board review, it was noted that the MH group was experiencing a greater number of pulmonary complications as well as a trend to a greater number of serious adverse events (table 5). At the request of the Data Safety and Monitoring Board, we also compared the occurrence of adverse events in the study groups with events in a group of patients (n = 323) enrolled in a contemporaneous noninterventional trial. The greater occurrence of pulmonary complications was even more significant (P = 0.002) when the MH group was compared with this nonintervened group. This difference persisted (P = 0.05) when the MH group was compared with a subset of the nonintervened patients who were matched to the hemodilution patients on study date, age, and CPB time (cross clamp time, Hannan score, Parsonnet score, and Charlson comorbidity index were also similar; mean hematocrit during CPB = 23.5 ± 3.7%). The pulmonary complications that varied the most between the MH and nonintervened groups were the occurrence of postoperative pneumonia (13.3% vs.  5.9%) and pulmonary edema (26.7% vs.  8.8%). MH patients with postoperative pneumonia received a median of 4,500 ml (IQR, 3,600–5,400 ml) packed erythrocytes, and those with pulmonary edema received 3,300 ml (IQR, 1,050–10,299 ml) packed erythrocytes. The overall transfusion rate in the MH (89%) and PH (88%) groups was also significantly higher than the matched control group (69%; P = 0.02).

To our knowledge, this is the first prospective, randomized study of hemodilution during cardiac surgery with CPB in adults. This trial was terminated early because of an increase in adverse events in the moderate hemodilution group. Despite the incomplete enrollment, we found that older patients experiencing profound hemodilution were more likely to experience cognitive decline. In secondary analyses, an association between the maximum decrease in hematocrit from baseline and neurocognitive decline in the elderly was also detected; a decrease in hematocrit of 12% points or greater from baseline approximated the threshold for this cognitive decline.

The effects of hemodilution during CPB on ischemic neurologic injury have been studied in both animals and humans. Although early animal studies indicated a potential benefit to hemodilution, strengthening the belief that hemodilution during cardiac surgery was without consequence,13,14most of these studies were limited in that they did not mimic CPB or reduce hematocrit below 30%. In a landmark study in healthy dogs, Cook et al. ,2attempting to define the “critical hematocrit,” reported that increases in cerebral blood flow (CBF) compensated for the decreased arterial oxygen content from hemodilution and that cerebral oxygen delivery was maintained to a hematocrit of approximately 14% during normothermic CPB. Subsequent study by these same investigators demonstrated that cerebral oxygen demand was maintained to a hematocrit of 11% when hypothermia was applied. However, with progressive temperature reduction, a progressively smaller increase in CBF was seen, and “physiologically important changes in cerebral oxygen supply” were reported at hematocrits of 18, 15, and 12% with temperatures of 38°, 28°, and 18°C, respectively.1Limits to the extent of hemodilution were also described by Lee et al. ,4who reported that hemodilution to a hematocrit of 30% reduced cerebral infarct volumes in a dog model, but the benefit was reversed when the hematocrit was further reduced to 25%. Similarly, Reasoner et al.  3found an increase in hemispheric infarct size after middle cerebral artery occlusion in rabbits when marked hemodilution (hematocrit = 18%) was used. More recently, Homi et al.  15also hemodiluted rats surgically prepared for CPB to a hematocrit of 18% and reported both worsened functional neurologic performance and greater cerebral infarct volumes 24 h after middle cerebral artery occlusion, when compared with control animals maintained at a hematocrit of 33%.

The deleterious consequences of extreme hemodilution during CPB in humans have also been recently (subsequent to study inception and the majority of enrollment) highlighted by a series of retrospective database studies. In virtually every outcome examined, an independent, direct association between the degree of hemodilution during CPB and the adverse outcome of interest was identified. For example, Karkouti et al. ,11studying 10,949 patients undergoing cardiac surgery with CPB, reported a 10% increase in the odds of experiencing a perioperative stroke with each percent decrease in hematocrit. When acute renal failure was examined, these same investigators reported a 230% increase in the odds of developing acute postoperative renal failure for those with a CPB nadir hematocrit less than 21%.10Interestingly, the odds of developing renal failure was also increased in those with a hematocrit greater than 25%, suggesting that an “optimal” hematocrit to manage this outcome might be somewhere between 21% and 25%. The lowest hematocrit on CPB has also been associated with greater in-hospital mortality7,8and reduced survival up to 6 yr after surgery.9A single prospective randomized trial in infants confirms the deleterious effects of extreme hemodilution. In that study, 147 infants were randomized to a hematocrit of 20% or 30% at the onset of low-flow CPB using a pH-stat strategy.16The lower hematocrit group had lower nadirs of cardiac index, higher serum lactate levels 60 min after CPB, and at age 1 yr, worse scores on a psychomotor development index. In contrast, a single retrospective study found no correlation between cognitive performance and hematocrit levels preoperatively, 30 min after CPB, 10 min after the end of CPB, or on the first postoperative day.17However, only 1 of 111 patients in that study had a hematocrit on CPB less than 20%.

Acute isovolumic anemia to a hematocrit of 15–18% (the same range as that of the PH group in our study) in healthy volunteers has been reported to increase reaction time and degrade immediate and delayed memory during cognitive testing.18These slowed responses are thought to result not from a nonspecific effect on attention but from impaired central processing as detected by an increase in the P300 evoked potential latency.19Although the reported effects on cognition were transient and reversible by the administration of oxygen or erythrocytes, subjects were severely anemic for only brief periods of time, and it was uncertain whether protracted periods of anemia (as seen with CPB) would have produced greater impairment. In the setting of cardiac surgery with CPB, it is likely that any deleterious effect of severe anemia is compounded by CPB-related alterations in cerebral physiology. Based on the work of Cook et al. ,20it is widely believed that during CPB, an increased CBF as a consequence of hemodilution, maintains cerebral oxygen delivery. Furthermore, a close coupling of oxygen delivery and demand is seen in normothermic CPB, but this coupling is lost during hypothermic CPB; CBF is unchanged with the addition of hypothermia despite the large decrease in cerebral metabolic rate.20,21Because CBF is uniformly increased with severe anemia during CPB, the worsening of cognitive function seen in these patients could then be a consequence of an increased delivery of cerebral emboli. Pathologic and Doppler studies22,23have long supported the association between embolic load and neurocognitive injury after cardiac surgery, but the clinical relevance of this association has been questioned by others.24Although data clearly demonstrating an increase in embolic load with severe anemia during CPB are lacking, several studies suggest that such an association is plausible. In a dog study examining the relation between CPB flow rate and cerebral embolization, it was noted that tissues with high blood flow received more emboli than tissues with lower blood flow.25Similarly, in a study evaluating the safety of a perfluorocarbon emulsion administered during CPB, an increase in CBF seen after hemodilution and emulsion administration was accompanied by a greater number of transcranial Doppler–detected cerebral emboli.26 

The impact of aging upon CBF and oxygenation in the setting of severe anemia or CPB has not been widely studied, so we can only speculate as to the reasons for our finding that extreme hemodilution was detrimental only in the elderly. A study in rats has shown that the cerebral hyperemic response to anemia is preserved in the aged, whereas a study of 12 subjects undergoing CPB revealed that advancing age further increased the magnitude of this hyperemia, albeit in the postoperative period.27,28The greater increases in CBF seen in the elderly may again be simply associated with a larger embolic load and, therefore, greater neurocognitive injury.

The principal limitation to our study is the failure to reach the targeted enrollment as a consequence of the safety concerns. Nevertheless, the early termination of our study because of an increase in pulmonary complications, notably pneumonia and pulmonary edema, suggests that aggressive transfusion to maintain a hematocrit of 27% or greater during CPB may not be prudent. The association between packed erythrocyte transfusion and nosocomial pneumonia has been reported previously, with most studies suggesting a transfusion threshold of 3–4 units before the increased risk of infection is evident.29–31In the patients with postoperative pulmonary edema who also received large volumes of packed erythrocytes, transfusion-related acute lung injury is a consideration because coronary artery bypass graft surgery and massive transfusion have been implicated as risk factors for transfusion-related acute lung injury.32,33Another limitation to our study is the lack of brain imaging or transcranial Doppler data in the enrolled patients; therefore, we are left to speculate that a higher embolic load may have contributed to the cognitive decline. Finally, 21% of our baseline population did not return for follow-up testing. As expected, nonreturning patients were sicker (worse Charlson comorbidity score) than returnees; however, the other demographic characteristics listed in table 2were not different, and the rate of loss to follow-up was not different between the two treatment groups.

In summary, this trial is the first to evaluate the effect of hemodilution on neurocognitive outcomes after adult cardiac surgery in a prospective randomized manner. Because of an increase in adverse events in patients randomized to a higher hematocrit level, this study was terminated early. Despite the incomplete enrollment, we report that older patients experiencing profound hemodilution were more likely to experience cognitive decline. Our results suggest that both extreme hemodilution in the elderly and aggressive transfusion should be used with caution in the management of CPB during cardiac surgery.

Director:  Joseph P. Mathew, M.D., Professor, Department of Anesthesiology, Duke University Medical Center, Durham, North Carolina.

Co-Director:  James A. Blumenthal, Ph.D., Professor, Department of Psychology, Duke University Medical Center.

Anesthesiology:  Hilary P. Grocott, M.D., Professor, G. Burkhard Mackensen, M.D., Ph.D., Associate Professor, Joseph P. Mathew, M.D., Professor, David McDonagh, M.D., Assistant Professor, Terri Monk, M.D., Professor, Mark F. Newman, M.D., Professor, Mihai V. Podgoreanu, M.D., Assistant Professor, Debra A. Schwinn, M.D., Professor, Andrew D. Shaw, M.D., Associate Professor, Mark Stafford-Smith, M.D., Professor, Madhav Swaminathan, M.D., Assistant Professor, David Warner, M.D., Professor, Bonita L. Funk, R.N., Clinical Research Coordinator, Narai Balajonda, M.D., Clinical Research Coordinator, Maria Celerian, M.D., Clinical Research Coordinator, Roger L. Hall, A.A.S., Clinical Research Coordinator, Gladwell Mbochi, A.A.S., Clinical Trials Assistant, Richard Morris, Ph.D., Assistant Professor, Charles R. Peters, M.A., Clinical Trials Specialist, Barbara Phillips-Bute, Ph.D., Assistant Professor, Elizabeth Perez, R.N., Clinical Research Coordinator, Prometheus T. Solon, M.D., Clinical Trials Assistant, Ashley Western, R.N., Clinical Research Coordinator, Peter Waweru, Clinical Trials Specialist, William D. White, M.P.H., Statistician, Department of Anesthesiology, Duke University Medical Center.

Behavioral Medicine:  Michael A. Babyak, Ph.D., Associate Professor, Department of Psychiatry, James A. Blumenthal, Ph.D., Professor, Department of Psychology, Duke University Medical Center.

Cardiology:  Daniel B. Mark, M.D., M.P.H., Professor, Michael H. Sketch, Jr., M.D., Professor, Department of Medicine, Duke University Medical Center.

Neurology:  Ellen R. Bennett, Ph.D., Assistant Professor, Carmelo Graffagnino, M.D., Assistant Professor, Daniel T. Laskowitz, M.D., Associate Professor, Warren J. Strittmatter, M.D., Professor, Kathleen A. Welsh-Bohmer, Ph.D., Assistant Professor, Department of Medicine, Duke University Medical Center.

Perfusion Services:  Greg Smigla, B.S., C.C.P., Perfusionist, Ian Shearer, B.S., C.C.P., Perfusionist, Department of Perfusion Services, Duke University Medical Center.

Surgery:  Thomas A. D’Amico, M.D., Associate Professor, R. Duane Davis, M.D., Professor, Donald D. Glower, M.D., Professor, R. David Harpole, M.D., Professor, G. Chad Hughes, M.D., Assistant Professor, James Jaggers, M.D., Associate Professor, Shu Lin, M.D., Assistant Professor, Andrew Lodge, M.D., Assistant Professor, James E. Lowe, M.D., Professor, Carmelo Milano, M.D., Associate Professor, Peter K. Smith, M.D., Professor, Eric M. Toloza, M.D., Ph.D., Assistant Professor, Walter G. Wolfe, M.D., Professor, Department of Surgery, Duke University Medical Center.