Association of nadir oxygen delivery on cardiopulmonary bypass with serum glial fibrillary acid protein levels in paediatric heart surgery patients

Abstract

OBJECTIVES

Protecting the brain during cardiac surgery is a major challenge. We evaluated associations between nadir oxygen delivery (DO2) during paediatric cardiac surgery and a biomarker of brain injury, glial fibrillary acidic protein (GFAP).

METHODS

Blood samples were obtained during a prospective, single-centre observational study of children undergoing congenital heart surgery with cardiopulmonary bypass (CPB) (2010–2011). Remnant blood samples, collected serially prior to cannulation for bypass and until incision closure, were analysed for GFAP levels. Perfusion records were reviewed to calculate nadir DO2. Linear regression analysis was used to assess the association between nadir DO2 and GFAP levels.

RESULTS

A total of 116 consecutive children were included, with the median age of 0.75 years (interquartile range: 0.42–8.00) and the median weight of 8.3 kg (5.8–20.0). Single-ventricle anatomy was present in 19 patients (16.4%). Deep hypothermic circulatory arrest (DHCA) was used in 14 patients (12.1%). On univariable analysis, nadir DO2 was significantly associated with GFAP values measured during rewarming on CPB (P = 0.005) and after CPB decannulation (P = 0.02). On multivariable analysis controlling for CPB time, DHCA and procedure risk category, a significant negative relationship remained between nadir DO2 and post-CPB GFAP (P = 0.03).

CONCLUSIONS

Lower nadir DO2 is associated with increased GFAP levels, suggesting that diminished DO2 during paediatric heart surgery may be contributing to neurological injury. The DO2–GFAP relationship may provide a useful measure for the implementation of neuroprotective strategies in paediatric heart surgery, including goal-directed perfusion.

INTRODUCTION

Despite dramatic advances in the field of congenital heart surgery (CHS) over the last 6 decades, abnormal neurodevelopmental outcomes continue to complicate a large proportion of cases—as many as a third, depending on the outcome evaluated [1, 2]. Some forms of neurological injury have characteristic radiological patterns, and some entail long-term sequelae, as they adversely affect neurodevelopmental outcomes few years later [2].

Brain injury after surgical repair of congenital heart disease is multifactorial with a variety of mechanisms shown to contribute. To begin with, infants with congenital heart disease appear to exhibit delayed brain development preoperatively [3]. Innate genetic characteristics, such as the presence of the apolipoprotein E ε2 allele, have been associated with worse neurodevelopmental outcomes following CHS [4]. The haemodynamic perturbations associated with heart surgery may also compromise cerebral oxygen delivery (DO2) and could compound maturational and genetic risks. A recent study proposed a role for diagnostic group and circulatory arrest in infants undergoing arch surgery as associated with new white matter brain injury in the CHS population [5]. Moreover, cardiopulmonary bypass (CPB) has been shown to be markedly pro-inflammatory across many populations; in paediatric patients, it clearly activates complement, stimulates leucocytes and induces capillary leak [6]. These factors are all thought to contribute to the white matter injury commonly observed on imaging following CHS [1].

Early detection (intra- and postoperative) of markers or correlates of CHS-related neurological injury could prompt intervention to reduce these insults. Blood-based brain biomarkers represent a potentially rapid means of diagnosing neurological injury. One such biomarker is glial fibrillary acidic protein (GFAP), an astrocyte intermediate filament protein, not normally present in blood and representing astrocyte injury or necrosis. GFAP has been linked to neurological outcomes in a variety of paediatric and adult populations [7–10]. Our group has previously shown that during CHS, GFAP levels increase significantly during CPB phases, with peak levels observed during rewarming and not affected by ultrafiltration [11]. In neonates treated with whole body cooling for hypoxic ischaemic encephalopathy, a similar pattern of peak GFAP levels with rewarming was shown to be associated with brain injury by MRI and worse functional outcomes [8]. Similarly, our group has shown in adults that both hypotension and the rewarming rate during cardiac surgery are associated with GFAP release [12, 13], and that higher GFAP levels are associated with cognitive decline in post-cardiac surgery patients [14].

Recent data suggest that nadir DO2 during CPB, rather than nadir haematocrit or hypotension alone, may drive postoperative complications following cardiac surgery such as acute kidney injury [15–17]. On the basis of the above findings, we sought to determine whether GFAP levels, as a marker of brain injury, were associated with hypoperfusion as measured by nadir DO2 on CPB within a population of paediatric heart surgery patients. Our hypothesis was that nadir DO2 was independently negatively associated with higher GFAP levels in children undergoing CHS.

METHODS

Patient population

After obtaining Institutional Review Board approval (NA00001068), we undertook a prospective, observational study of children <18 years of age undergoing repair of congenital heart disease on CPB at the Johns Hopkins Hospital between October 2010 and December 2011. Patients undergoing surgery not on CPB were excluded.

Conduct of cardiopulmonary bypass and procedures

All patients were treated by the same group of anaesthesiologists, surgeons, perfusionists and intensivists at our institution. Patients were centrally cannulated for CPB. Among patients who underwent ultrafiltration, zero-balance ultrafiltration (ZBUF) was the modality used with a 65 kDa hemoconcentrator. Of note, we have previously shown that our institution's use of ZBUF in this manner does not affect GFAP levels, and that GFAP is undetectable in the ultrafiltrate [11]. Following surgery, all patients who were transported intubated to the same paediatric intensive care unit to recover. GFAP assay results were not available to any clinicians caring for patients enrolled in this study.

To provide a means of risk-stratifying the varied CHS procedures, we utilized procedure risk categories as defined by the Risk Adjustment for Congenital Heart Surgery (RACHS) score [18]. By this method, CHS procedures were grouped into one of six categories according to a consensus-based process, assessing each procedure's perceived risk of in-hospital mortality. When combined procedures were performed, we assigned the highest RACHS 1–6 risk score of any individual component procedure.

Blood collection and biomarker assays

Starting after the induction of anaesthesia, waste blood from samples collected to measure activated clotting times was collected at baseline prior to cannulation, and then during the cooling, rewarming and post-CPB phases of each operation. Blood was centrifuged for 8 min at 3000 r.p.m. to separate out the plasma layer, which was then stored at −80°C.

An electrochemiluminescent sandwich immunoassay (Meso Scale Discovery, Gaithersburg, MD, USA) was used to detect and quantify GFAP levels [7, 8, 19–21]. The assay's detection range is 0.001–40.0 ng/ml. GFAP levels below the lower limit of detection were recorded as 0. Blood samples were taken from four separate time periods: pre-bypass, cooling on bypass, rewarming on bypass and post-bypass. If multiple samples were available during the same CPB phase, the GFAP results were averaged. Samples were assayed in duplicate, with an interassay coefficient of variation threshold of <10%.

Statistical analysis and outcomes

Clinical data collected on chart review included patient demographic data, presence of single-ventricle anatomy, any chromosomal abnormalities diagnosed, CPB times, cross-clamp times, deep hypothermic circulatory arrest (DHCA) times, nadir temperature in the operating room, postoperative complication, length of stay and survival. Additionally, from perfusion records, we collected data on nadir haemoglobin and nadir DO2 while on CPB. The latter was calculated according to the equation: DO2 in ml/min/m² BSA = pump flow in litres × [10 × haemoglobin in grams per decilitre × 1.36 × haemoglobin saturation % + 0.003 × arterial partial pressure of oxygen (PaO2)]/m² BSA [17], where BSA is the body surface area. Values were recorded each 15 min on bypass, with the nadir DO2 taken as the lowest of all calculated values. Of note, patients who had a nadir DO2 which occurred following one of our outcome GFAP levels (i.e. after the cooling or rewarming phases of the operation) were excluded from analyses of the GFAP level it followed. Thus, we focused our analyses on our primary outcomes of GFAP levels during rewarming and after separation from CPB.

Summary statistics were calculated for all patients. Differences in categorical variables were computed by the χ² or Fisher's exact test, while continuous variables were analysed with Student's t-test or rank-sum tests depending on distribution. A significance level of P < 0.05 was used for all statistical tests, which was computed using STATA 12.0 (STATA Corporation, College Station, TX, USA). To assess the relationship of our biomarkers with nadir DO2, we utilized linear regression analysis. A power calculation for linear regression suggested that, to detect an improvement of 0.05 in R² by adding nadir DO2 to a model incorporation controlling variables, we would need to accrue ∼100–110 patients, assuming low R² values in the range of 0.1–0.2 and only 70% power. Sensitivity analyses were conducted excluding outlier biomarker levels (defined as >2 standard deviations from the mean) to confirm the results of regression analysis. Multivariable models were built forwards using likelihood-ratio tests and the Akaike information criterion to assess model strength while building the most parsimonious model possible.

RESULTS

The study cohort included 116 patients between the ages of 2 days and 18 years (Table 1). Of note, the median age was 9 months, the median weight was 7.6 kg and 25% of patients presented with single-ventricle anatomy. Chromosomal abnormalities were present in 12.9% of patients, and the majority of patients (73.3%) had already had a prior CHS. Procedures involving atrial or ventricular septal defect repairs, Norwood/Glenn/Fontan procedures and tetralogy of Fallot repairs accounted for about three-quarters of all operations. Fourteen patients underwent surgery utilizing DHCA. Median nadir DO2 on CPB was 255 ml O₂/min/m² BSA [interquartile range (IQR): 198–310]. The median time at which nadir DO2 occurred was 45 min following CPB initiation (IQR: 30–60); by comparison, the median total bypass time for the entire cohort was 120 min (IQR: 92–164).

Table 1:

Summary statistics for study patients

N	116
Median age	9 mo (3 mo–8 years)
Median weight (kg)	7.6 (5–20)
Male gender	57.8 (67)
Single-ventricle anatomy	25.0 (29)
Chromosomal abnormality	12.9 (15)
Prior heart surgery	36.2 (42)
Procedure types
ASD/VSD repair	30.2 (35)
Norwood/Glenn/Fontan (single-ventricle patients)	25.0 (29)
Tetralogy of Fallot repair	16.4 (19)
Ross procedure	5.2 (6)
Pulmonary valve replacement	4.3 (5)
AV canal repair	2.6 (3)
Hypoplastic aortic arch repair	2.6 (3)
Other procedure	13.8 (16)
Median bypass time (min)	120 (92–164)
Median cross-clamp time (min)	65 (39–94)
Required DHCA	12.1 (14%)
DHCA with SACP	42.9 (6%)
Median DHCA time (min)	11 (7–28)
Median RACHS risk category	3 (2–3)
Median nadir temperature in OR	29.2 (27.1–32.2)
Median nadir oxygen delivery on bypass (DO2)	255 (198–310)
Median ICU LOS (days)	3 (2–5.5)
Median hospital LOS (days)	5 (3–10)
Postoperative seizures	1.9 (2)
Postoperative cerebrovascular accident	0.9 (1)
Survival to hospital discharge	97.4 (113)

All data shown as median (interquartile range), or % (n).

ASD: atrial septal defect; AV: atrioventricular; mo: months; DHCA: deep hypothermic circulatory arrest; OR: operating room; DO2: oxygen delivery on cardiopulmonary bypass in ml O₂/min/m² body surface area; ICU: intensive care unit; LOS: length of stay in days; SACP: selective antegrade cerebral perfusion; RACHS: Risk Adjustment in Congenital Heart Surgery procedure risk category (1–6); VSD: ventricular septal defect.

Postoperatively, patients spent a median of 3 days in the ICU and 5 days in the hospital, with only 2 patients (1.9%) experiencing postoperative seizures. Survival to hospital discharge was observed in 97.4% of patients.

As previously described, GFAP levels varied throughout our four phases of operations (Fig. 1). Levels measured prior to the initiation of CPB (median: 0.001 ng/ml, IQR: 0.001–0.012, 95th percentile: 0.130 ng/ml) differed significantly from levels while cooling on CPB (median: 0.040, IQR: 0.001–0.128, 95th percentile: 0.588 ng/ml), rewarming on CPB (median: 0.197, IQR: 0.047–0.438, 95th percentile 1.900 ng/ml) and after separation from CPB (median: 0.104, IQR: 0.02–0.188, 95th percentile: 0.532 ng/ml) (all P < 0.01 for comparisons).

Figure 1:

Glial fibrillary acidic protein (GFAP) levels by phase of operation. Asterisk denotes significant differences from pre-CPB levels. Of note, GFAP levels differed significantly from each other among cooling, rewarming and post-CPB phases. CPB: cardiopulmonary bypass.

Rewarming GFAP levels also differed when stratifying by several of the largest basic anatomic patient categories, though post-CPB GFAP levels did not. Median rewarming GFAP levels for ASD/VSD patients were 0.201 ng/ml (0.057—0.270), 0.124 (0.350–0.516) for single-ventricle patients, and 0.308 (0.240–1.310) for tetralogy of Fallot patients (P = 0.006). Post-CPB GFAP levels were 0.118 (0.045–0.188) for ASD/VSD patients, 0.080 (0.040–0.283) for single-ventricle patients and 0.127 (0.057–0.324) for tetralogy of Fallot patients (P = 0.44).

We assessed the relationship of these levels to several operative characteristics—procedure risk category, DHCA use and ZBUF use. Median GFAP levels, particularly during rewarming, were positively correlated with procedures assigned higher RACHS risk categories (Fig. 2). Rewarming GFAP levels for RACHS categories 5 and 6 differed significantly from those of RACHS category 3 and below (P = 0.04, 0.04 and 0.004 for RACHS levels 3 vs 4, 5 and 5, respectively). Post-CPB GFAP levels for RACHS category 6 differed significantly from those of RACHS category 3 and below (P = 0.04). Positive correlations between increasing RACHS risk category and GFAP were borderline significant for cooling GFAP levels (coefficient 0.041, P = 0.06), and significant for rewarming GFAP (coefficient 0.335, P < 0.001). Patients requiring DHCA were more likely to have elevated GFAP levels at rewarming (1.06 vs 0.33 ng/ml, P < 0.001), but not after CPB (0.23 vs 0.15, P = 0.16); among the 14 DHCA patients, there were no significant differences in GFAP levels between those who received selective antegrade cerebral perfusion versus those who did not, either at rewarming (1.69 vs 0.67 ng/ml, P = 0.10) or post-CPB (0.27 vs 0.21 ng/ml, P = 0.65). Similarly, ZBUF in 97 children (83.6%) did not affect rewarming (no ZBUF, 0.51 vs ZBUF recipients, 0.41, P = 0.65) or post-CPB GFAP levels (0.11 vs 0.17 ng/ml, P = 0.27).

Figure 2:

Median GFAP levels at the four operation phases, stratified by RACHS procedure risk categories. Note that median pre-CPB GFAP levels (green bars) were ≤0.005 for all RACHS categories. Asterisk denotes significant differences (between RACHS categories) from the same phases of GFAP measurement. Rewarming GFAP levels for RACHS 4/5/6 differ significantly from RACHS 3 and below; post-CPB GFAP levels for RACHS 6 differ significantly from RACHS 3 and below. GFAP: glial fibrillary acidic protein; RACHS: Risk Adjustment for Congenital Heart Surgery method; CPB: cardiopulmonary bypass.

On univariable regression analysis, we found significant associations between nadir DO2 and GFAP during the rewarming and post-CPB phases of operations (Tables 2 and 3). On sensitivity analysis, these univariable relationships remained significant even after excluding all outlying GFAP values above 2.5 ng/ml, or even above 1 ng/ml. Moreover, these associations were all negative, implying that higher nadir DO2 on CPB was associated with a lower GFAP level during the three phases. We then constructed parsimonious multivariable regression models for GFAP levels during both phases to control for confounding. Only CPB time (continuous variable, in minutes) was independently associated with rewarming GFAP levels (Table 2). However, on multivariable analysis controlling for CPB time, DHCA use and RACHS procedure risk category, we found that nadir DO2 was independently associated with post-CPB GFAP levels (Table 3; Fig. 3).

Table 2:

Univariable and multivariable regression results for rewarming GFAP levels

Rewarming GFAP	Univariable				Multivariable
	Coefficient	95% CI		P-value	Coefficient	95% CI		P-value
Age (years)	−0.035	−0.061	−0.009	0.01	−0.027	−0.056	0.002	0.06
Weight (kg)	−0.008	−0.015	−0.002	0.02
Male gender	0.110	−0.181	0.400	0.46
Single-ventricle anatomy	0.139	−0.197	0.476	0.41
Chromosomal abnormality	0.305	−0.095	0.704	0.13
Prior cardiac surgery	−0.254	−0.552	0.044	0.09
CPB time (min)	0.006	0.004	0.009	<0.001	0.048	0.002	0.008	0.001
Cross-clamp time (min)	0.008	0.005	0.012	<0.001
DHCA used	0.734	0.330	1.137	<0.001	−0.052	−0.609	0.505	0.85
DHCA time (min)	−0.001	−0.055	0.052	0.95
RACHS risk category (1–6)	0.336	0.207	0.464	<0.001	0.154	−0.053	0.361	0.14
Nadir temperature (°C)	−0.066	−0.093	−0.038	<0.001
Nadir DO2	−0.002	−0.004	−0.001	0.005	−0.001	−0.002	0.001	0.54

Multivariable model R² = 0.30. Bold type denotes significant associations on multivariable analysis.

CI: confidence interval; CPB: cardiopulmonary bypass; DHCA: deep hypothermic circulatory arrest; DO2: oxygen delivery in ml O₂/min/m² BSA; RACHS: Risk Adjustment in Congenital Heart Surgery procedure risk category.

Table 3:

Univariable and multivariable regression results for Post-CPB GFAP levels

Post-CPB GFAP	Univariable				Multivariable
	Coefficient	95% CI		P-value	Coefficient	95% CI		P-value
Age (years)	−0.009	−0.016	−0.002	0.02
Weight (kg)	−0.002	−0.004	0.000	0.03
Male gender	−0.013	−0.097	0.071	0.76
Single-ventricle anatomy	0.032	−0.064	0.129	0.51
Chromosomal abnormality	0.064	−0.054	0.183	0.29
Prior cardiac surgery	−0.073	−0.159	0.013	0.09
CPB time (min)	0.000	0.000	0.001	0.20	0.000	−0.001	0.001	0.72
Cross-clamp time (min)	0.001	0.000	0.002	0.05
DHCA used	0.085	−0.033	0.204	0.16	−0.001	−0.192	0.191	1.00
DHCA time (min)	−0.002	−0.013	0.009	0.70
RACHS risk category (1–6)	0.024	−0.015	0.064	0.20	−0.022	−0.094	0.050	0.54
Nadir temperature (°C)	−0.010	−0.018	−0.002	0.02
Nadir DO2	−0.001	−0.001	0.000	0.02	−0.001	−0.001	0.000	0.03

Multivariable model R² = 0.07. Bold type denotes significant associations on multivariable analysis.

CI: confidence interval; CPB: cardiopulmonary bypass; DHCA: deep hypothermic circulatory arrest; DO2: oxygen delivery in ml O₂/min/m² BSA; RACHS: Risk Adjustment in Congenital Heart Surgery procedure risk category.

Figure 3:

Univariable (i.e. unadjusted) correlation between nadir oxygen delivery on cardiopulmonary bypass (DO2) and post-CPB glial fibrillary acidic protein (GFAP) levels. High outlying values excluded (>2 standard deviations). CPB: cardiopulmonary bypass; BSA: body surface area in m².

We also assessed for a possible relationship between GFAP levels and outcomes among patients who experienced postoperative seizures (n = 2) or who died prior to discharge (n = 3). The two children who seized postoperatively had similar mean GFAP levels to the rest of the cohort (rewarming GFAP, 0.15 in seizing children vs 0.43 ng/ml for rest of cohort, P = 0.59; post-CPB GFAP, 0.11 vs 0.16 ng/ml, P = 0.75). One 5-year old experienced a 5-min seizure on postoperative day 3 and was found to have extra-axial subdural blood near the left occipital lobe and cerebellum; she recovered well otherwise and suffered no permanent disability on follow-up. The other patient experienced a postoperative tear in a right ventricular outflow patch which resulted in an air embolism, seizing and cerebrovascular accident on POD3; her intraoperative GFAP levels were all below the lower limit of detection. Her neurological status recovered to baseline prior to hospital discharge. Regarding mortality, causes of death were: post-cardiotomy failure following tetralogy of Fallot repair at an outside hospital (and failure to recover despite mechanical circulatory support); fixed pulmonary hypertension and right ventricular failure following truncus arteriosus repair in a patient with DiGeorge syndrome; and post-cardiotomy failure following complex repair in a patient with numerous defects including double outlet right ventricle, hypoplastic left ventricle and interrupted aortic arch type B. Overall, non-survivors appeared to have higher mean rewarming GFAP levels than survivors (1.49 vs 0.39 ng/ml, P = 0.009). This difference could not be assessed in post-GFAP levels, as two of three post-CPB GFAP values were unavailable secondary to postoperative demise.

DISCUSSION

In the present study, we found that GFAP was strongly associated with nadir DO2 on CPB during rewarming, and post-CPB phases of operations. Moreover, on multivariable analysis controlling for operation length (i.e. bypass times), DHCA use and procedure risk category, this association remained significant for post-CPB GFAP levels. However, we were unable to document any relationship between GFAP levels and clinical outcomes in our small, retrospective sample.

Data on the specificity and predictive ability of GFAP as a marker of neurological injury in a variety of clinical scenarios continue to accumulate. GFAP expression has been associated with histopathological evidence of neurological damage in patients succumbing to brain injury [12, 13, 22]. Within the cardiac surgery patient population, GFAP release has been associated with hypotension as well as rapid rewarming rates, and with the outcome of neurocognitive decline [12–14]. In other populations, GFAP has been shown to have not only predictive ability following traumatic brain injury in adults, but also utility as an early diagnostic tool even on the first day following TBI [9]. Excitingly, the recent article by Papa et al. [23] demonstrated in a large cohort of trauma patients that GFAP has strong diagnostic accuracy in predicting diagnosis of mild-to-moderate traumatic brain injury. In adult cardiac arrest victims, serum GFAP levels were higher in patients with poor early and late neurological outcomes [10]. In the paediatric population, GFAP has been shown to be a biomarker of brain injury in children placed on extracorporeal membrane oxygenation when compared with healthy controls, with significant differences in GFAP levels seen between survivors and non-survivors [7]. Neonates with hypoxic–ischaemic encephalopathy have also been shown to have higher serum GFAP levels in the first week of life when compared with controls and significantly associated with neurodevelopmental outcomes at 12 months of age [8, 24]. Table 4 summarizes recent literature regarding GFAP as a potential biomarker with proposed thresholds for various types of neurological injury in different populations.

Table 4:

Review of recent publications examining the relationship between glial fibrillary acidic protein (GFAP) levels and neurological injury

Study	Patient population	Primary outcome	GFAP threshold (ng/ml) for outcome
Nylen et al. [25]	Adults with severe traumatic brain injury (n = 59)	Severe neurological disability, vegetative state or death	Lowest 0.38; median 2.72
Honda et al. [9]	Adults with severe trauma (n = 34)	Traumatic brain injury	About >1.0 (?) (means/medians not reported)
Kaneko et al. [10]	Adults post-cardiac arrest with return of spontaneous circulation (n = 44)	Severe neurological disability, vegetative state or death	>0.1
Bembea et al. [7]	Paediatric patients on extracorporeal membrane oxygenation (n = 22)	Brain injury (haemorrhage, oedema or death)	>0.436
Ennen et al. [8]	Neonates with hypoxic–ischaemic encephalopathy (n = 23) versus controls (n = 23)	Hypoxic–ischaemic encephalopathy	>0.08
Savage et al. [20]	Paediatric patients w/ sickle cell (n = 295)	Cerebral infarct on MRI	>0.227
Stewart et al. [19]	Low birthweight (<2500 g) infants	Periventricular white matter injury	>0.04
McKenney et al. [21]	Neonates with CHD (n = 56) and controls (healthy, n = 23; w/ HIE, n = 23)	CHD, HIE	Median 0.042 in CHD patients; 0.108 in HIE patients, 0.03 in healthy controls

CHD: congenital heart disease; HIE: hypoxic ischaemic encephalopathy; MRI: magnetic resonance imaging.

For this study, we chose to examine nadir DO2 because of its demonstrated relationship to outcomes following cardiac surgery, particularly acute kidney injury [15–17]. Unfortunately, nadir DO2 has garnered less attention in regard to neurological injury thus far—an association that is difficult to study due to the relative paucity of neurological events observed in adult cardiac surgery patients and the complexity of neurocognitive assessments. However, we note that the kidneys are likely second to the brain in overall organ sensitivity to perturbations in DO2 and/or haemodynamics. We previously noted significant associations between GFAP levels and patient age, CPB time and temperature nadir in the operation room [11]. In the present study, we documented univariable associations between nadir DO2 and serum GFAP levels during several operative phases. Most importantly, even after controlling for CPB times and perceived in-hospital mortality risks associated with procedures, nadir DO2 was significantly associated with post-CPB serum GFAP levels. These findings are consistent with the evolving literature on nadir DO2 described above and its end-organ effects.

In the absence of common, easily measurable outcomes related to neurological injury and function, especially in children, we believe that our findings tentatively supporting a relationship between nadir DO2 and serum GFAP are significant in and of themselves. We would note that our reported GFAP values comport with published data by our group and others in terms of relative levels; unfortunately, we cannot relate our data to clinical neurological outcomes in any meaningful way, as only 2 patients had seizures and 3 patients died in this cohort of CHS patients. The finding, for example, that 27 of our children had post-CPB GFAP levels >0.4 ng/ml is interesting as GFAP values above this approximate level have been associated with injury in certain series [7–10, 19–21, 25]. None of these children had any documented obvious neurological injury on discharge following surgery, precluding our ability to draw conclusions about the clinical significance of elevated GFAP levels in these patients. In the future, given the accumulating literature on GFAP's value as a potential biomarker of neurological injury, we speculate that GFAP may not only aid in outcome prediction but also perhaps play a role in neurological monitoring of patients during or early after heart surgery. A rapid assay could allow for the detection of vulnerable patients and prompt measures to improve perfusion—if GFAP's association with outcomes can be demonstrated in this population.

Several limitations influence the inferences that can be drawn from our data. Though we attempted to control for differences in patient and operative complexity by adjusting our multivariable models for CPB time and RACHS procedure risk category, the dramatic heterogeneity of congenital heart defect lesions and procedures to correct them is a major challenge. Though other patient- and procedure-specific variables fell out of our multivariable model on its construction, one cannot assume we have therefore controlled for all possible confounders. Our data cannot truly clarify, for example, whether nadir DO2 is causative of GFAP release, or whether it may act as a proxy for operative difficulty as with bypass time and nadir temperature, for example. Additionally, our ability to perform a granular analysis was limited by the perfusion data available, which at the time was only recorded at 15-min intervals. A more useful metric of nadir DO2 would be a computation of area under the curve below a given threshold of DO2, but in the absence of such electronic data, we elected instead to take nadir DO2 at a single time point. Undoubtedly, this method fails to capture many potential episodes of hypoperfusion on CPB.

In conclusion, we documented significant associations between nadir DO2 and GFAP during CHS on CPB. This relationship persisted for post-CPB serum GFAP levels on multivariable analysis controlling for procedure duration and risk as well as DHCA use. We also emphasize that our documentation of this relationship cannot be construed to imply any association with neurological outcomes, which we were unable to effectively measure. Further studies across a far larger multicentre sample set and more granular neurocognitive measures are required to determine whether bypass-related variables like nadir DO2 are causally related to GFAP release, and whether GFAP is truly predictive of neurological outcome following CHS.