Assessment of Brain Magnetic Resonance and Spectroscopy Imaging Findings and Outcomes After Pediatric Cardiac Arrest

This cohort study analyzes the association of brain lesions seen on magnetic resonance imaging and concentrations of lactate and N-acetylaspartate seen on magnetic resonance spectroscopy with 1-year outcomes after pediatric cardiac arrest.


Introduction
Children resuscitated from cardiac arrest have high morbidity and mortality rates, mostly due to global cerebral hypoxic-ischemic injury. 1,2 Accurate tools to assess patient outcomes are essential for shared clinical decision-making. Consensus statements for postarrest care of children include brain computed tomography or magnetic resonance imaging (MRI) to assess outcomes, but high-quality evidence is lacking. [3][4][5] Conventional T2-weighted imaging and diffusion-weighted imaging (DWI) sequences can detect injury within hours to days after a hypoxic-ischemic event. Single-center studies 6-8 of pediatric cardiac arrest have found an association of regional brain injury seen on T2-weighted imaging and DWI sequences and outcomes. Furthermore, studies 8,9 using magnetic resonance spectroscopy (MRS), which analyzes brain biochemistry, show that increased lactate and decreased N-acetylaspartate (NAA) are promising biomarkers associated with outcomes after pediatric cardiac arrest. A reduction of NAA is associated with neuronal and axonal injury, whereas lactate accumulates when pyruvate oxidation in the citric acid cycle is impaired. 10 However, multicenter, prospective studies of children with cardiac arrest are lacking.
The objective of this prospective multicenter cohort study was to analyze the association of brain features seen on MRI and MRS performed within 2 weeks among children who experienced cardiac arrest with the composite outcome of death or unfavorable adaptive behavior at 1 year. We hypothesized that brain MRI and MRS features would be associated with the composite outcome.

Study Design and Setting
The Personalizing Outcomes After Child Cardiac Arrest (POCCA) prospective cohort study 12 was performed at 14 pediatric intensive care units in the US between May 16, 2017, and August 19, 2020, including follow-up through 1 year. The University of Pittsburgh institutional review board approved the study centrally. Independent institutional review board approval was obtained by 2 sites (Children's Healthcare of Atlanta and Children's Wisconsin), because they could not participate in the central review board. Written informed consent from a parent or guardian was required for participation and written and/or verbal assent was obtained from the children when appropriate according to local center guidelines. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cohort studies. 11

Participants
Eligible children included those aged 48 hours to 17 years who were enrolled in the parent study 12 and underwent a brain MRI or MRS as part of routine clinical care within the first 14 days after being resuscitated from in-hospital or out-of-hospital cardiac arrest. See eFigure 1 in Supplement 1 for more information on participant eligibility. race and ethnicity, were extracted from the medical record using National Institutes of Healthrecommended categories, and included African American or Black, Asian, White, and unknown race, and Hispanic ethnicity. 14 Race and ethnicity were recorded to evaluate for outcome differences. 15 The MRI protocol included axial and coronal T2-weighted sequences. Whole-brain DWI was performed at b = 1000 seconds/mm 2 with 6 directions. In-plane resolution was 1.1 × 1.1 × 6.0 mm 3 with no gap. Scan time was echo time to repetition time of 87.4 per 10 000 milliseconds.

Brain MRI and MRS
Brain MRI images were scored and analyzed by 2 pediatric neuroradiologists (S.N. and S.S.) who were blinded to both patient outcome and each other's scores. T2-weighted and DWI images were examined for signal intensity abnormalities and scored using a system adapted from Hirsch et al 16 (eAppendix 2 in Supplement 1). Findings in white and gray matter were reported separately when appropriate. Lesions seen on T2-weighted imaging and DWI were also scored by severity (0 = none, 1 = mild, 2 = moderate, 3 = severe). MRI Injury Score was a sum of T2-weighted imaging and DWI lesions in gray and white matter (maximum score, 34). The Cortex Score was a sum of gray and white matter scores from 4 cortical lobes. The Deep Gray Matter Score was a sum of basal ganglia (lenticular and caudate) and thalamus scores. Interobserver variability was described, and disagreement was resolved by consensus.
MRS data were acquired using a single-voxel point-resolved spectroscopy sequence with an echo time of 35 milliseconds and repetition time of 2 seconds (3-Tesla). MRS data were obtained from 4 regions of interest (ROIs): (1) basal ganglia, (2) thalamus, (3) parietal white matter, and (4) parietooccipital gray matter (eAppendix 3 in Supplement 1). These ROIs correspond to those regions most vulnerable to global hypoxic-ischemic injury and were standardized in the protocol. 8,17,18 Axial, sagittal, and coronal orientations were assessed for accurate voxel placement. Volumes of ROIs were typically 3 to 6 cm 3 , and acquisition times (128 averages) were 3 to 5 minutes per ROI (eAppendix 1 in Supplement 1).
Raw MRS data files were transferred offline for fully automated postprocessing using LCModel software version 6.3-1P (Stephen Provencher Inc). 16,19 LCModel processing provides objective measures for the signal-to-noise ratio and the spectral line width (full width at half maximum) for objective quality assessment. Spectra with a low signal-to-noise ratio of more than 2 SDs below the mean were discarded. We also excluded spectra with poor line width more than 2 SDs above the mean. From the remaining spectra, absolute concentrations (institutional units) of NAA and lactate were determined using the unsuppressed water signal as an internal reference, with a water content of 75% set as an approximation for our patient population. We also computed and analyzed the lactate-to-NAA ratio. Pediatric neuroradiologists verified MRS ROI placements at each center. MRS measurements were performed on 1 hemisphere only because cardiac arrest produces global hypoxia-ischemia.

Outcomes
The primary outcome was the association of brain MRI Injury Score and MRS features with an unfavorable outcome (either death or survival with a Vineland Adaptive Behavior Scales, Third Edition 20  score <70) at 1 year. Patient follow-up was standardized and performed at each center as per the parent study. 12,20

Statistical Analysis
Frequencies and percentages were reported for categorical variables. Results were nonparametric and thus presented as median (IQR). Outcome group comparisons were made using Kruskal-Wallis, Mann-Whitney, Fisher exact, and χ 2 tests, as appropriate. Simple and weighted κ statistics are reported for MRI scores (eTable 2 in Supplement 1). P < .05 was considered significant for unfavorable vs favorable outcomes at 1 year.
Multivariable logistic regression modeling was performed to evaluate the association of MRI Injury Score with 1-year outcomes. Variables initially tested in univariate models included MRI Injury Score, cause of arrest, arrest location, total number of epinephrine doses, first monitored rhythm, age, sex, and witnessed event status. [21][22][23] Covariates significant at P < .20 were then evaluated using multivariable logistic regression models with stepwise selection and entry and removal levels of .20.
A multivariable regression for MRS results was not possible due to limited sample size. Multivariable analysis of the area under the receiver operator characteristic curve (AUROC) was used to assess the accuracy of clinical variables alone and with MRI Injury Score to assess 1-year outcomes. Missing data were not imputed. Only patients with primary outcome data available were analyzed. Analyses were conducted using SAS statistical software version 9.2 (SAS Institute, Inc). Data analysis was conducted from January 2022 to February 2023.  Analyzing these same regions by the severity of lesions, temporal lobe gray matter, parietal lobe white matter, basal ganglia, thalamus, and centrum semiovale scores were higher (ie, indicating worse outcomes) for children with an unfavorable outcome than for children with a favorable outcome ( Table 3). DWI lesions in all 4 lobes, in both gray and white matter, basal ganglia, thalamus, cerebellar, brain stem white matter, and centrum semiovale were associated with an unfavorable outcome.     All summary MRI Injury Scores except the T2-weighted imaging cortex score were significantly higher in children with an unfavorable outcome compared to children with a favorable outcome (      Unfavorable outcome (n = 23) P value

Case Studies
In Supplement 1, eFigure 2 and eTable 7 highlight 3 patient cases. These cases are representative of more severe injury, moderately severe injury, and no injury at 1 year.

Discussion
In this cohort study of children with cardiac arrest, brain MRI and MRS features from sequences obtained within the first 2 weeks after resuscitation were associated with 1-year outcomes. Our main findings were that (1) lesions observed in a majority of brain regions using DWI and T2-weighted MRI (to a lesser degree) were associated with 1-year outcomes; (2) higher MRI injury and DWI deep gray matter injury scores were associated with an unfavorable 1-year outcome in multivariable models; and (3) decreased NAA, increased lactate, and the lactate-to-NAA ratio were associated with an unfavorable 1-year outcome in all ROIs in univariate analyses.
Brain MRI is commonly used to evaluate hypoxic-ischemic injury following pediatric cardiac arrest, but evidence regarding best practices in terms of sequences and timing postarrest is lacking. 1,4,24 Lesions on DWI sequences may become apparent as early as a few hours after arrest, peak at 3 to 7 days after arrest, and then may recede. 25 MRS metabolites may offer better stability over time. 26 In practice, however, children are not always clinically stable (eg, vasoactive requirement) or safe (eg, extracorporeal membrane oxygenation) to undergo brain MRI or MRS in the first few days after arrest. Conversely, for a stable patient, a clinician may decide the benefit is not worth the risks and costs (eg, intrahospital transport, need for multiple intensive care unit clinicians' time off the unit, and/or additional medications necessary for procedural facilitation). 27,28 Thus, our study analyzed the earliest clinically performed test occurring within the first 2 weeks after arrest.
Prior studies 6,7,[29][30][31] have shown that cytotoxic injury occurs most frequently in regions with increased vulnerability to hypoxia-ischemia, especially the occipital and parietal lobes and basal ganglia. Our MRI findings, conducted, to our knowledge, for the first time in a multicenter study, confirm these injury patterns. Furthermore, our study provides a more detailed accounting of regional gray and white matter injury location and severity, adapting a clinically relevant scoring system from the adult cardiac arrest literature. 16 Prior single-center studies 6,31,32 examining brain MRI and MRS as prognostic tools varied in terms of patient population (eg, mix of children with cardiac arrest along with other neurologic conditions or only drowning patients), imaging details (eg, timing, magnet strength, and scoring), and outcome assessment (eg, outcome measures and outcome timing). Those foundational studies identified specific brain regions (watershed regions, 31 generalized and occipital cortex edema, 32 and basal ganglia 6 ) associated with outcomes using T2-weighted imaging. DWI is another sequence that is performed regularly as the standard of care in this population and may be a more robust factor associated with outcomes. In an earlier study 6 at our center, restricted diffusion in cortical gray and white matter and basal ganglia was also associated with outcomes after pediatric cardiac arrest.
Another study 33 that evaluated DWI in children with cardiac arrest found that lesions in the basal ganglia and cortex were associated with an unfavorable outcome, and that measured global apparent diffusion coefficient values of less than 31% of uninjured white matter also were associated with a poor outcome. Finally, in another recent single-center study 7 , investigators adapted a validated stroke imaging tool to quantitate injury with DWI using clinically indicated MRI performed within the first 2 weeks of the event. They found an AUROC of 0.96 for DWI and 0.92 for T2-weighted imaging to differentiate favorable and unfavorable outcomes in children with out-of-hospital cardiac arrest.
Our study affirmed the importance of imaging findings across a multicenter cohort of children who had a cardiac arrest and further suggests that MRS may provide additional key information.
There are no MRI scoring tools validated for outcomes after pediatric cardiac arrest. We adapted an adult cardiac arrest MRI scoring tool for pediatric use given its use of both T2-weighted and DWI sequences, elements specific to cardiac arrest, incorporation of lesion severity of both white and gray matter, and validity in adults. 16 In this study, we were able to show an association of MRI Injury Score with outcomes, which can be used to support shared decision-making. In POCCA, 12 the MRI Injury and DWI Deep Gray Matter Scores were associated with our study's primary outcome at 1 year after adjustment for a limited number of clinical variables due to sampling size, with excellent accuracy.
External validation of these scores in a new cohort is essential to confirm clinical capacity.
Finally, MRS has potential to improve diagnostic accuracy for children with cardiac arrest.
Decreased NAA-to-creatine ratios and increased lactate in the occipital gray and parietal white cortical matter on brain MRS were associated with worse outcome in studies of children with cardiac arrest and other acute neurologic conditions in seminal studies in the 1990s. 17 Lesions on MRI were correlated with altered metabolites on MRS in children who had cardiac arrest due to drowning. 32 Our pilot work confirmed these findings in the basal ganglia and thalamus. 8  and specificity (95%) and was associated with neurodevelopmental outcomes.

Strengths and Limitations
Strengths of our study include the multicenter design, individual site training, and harmonizing of MRI and MRS protocols for each center's platform. Clinically, brain MRI is commonly used to evaluate postcardiac arrest as described in an international survey 24 . Nearly one-half of the respondents reported performing a brain MRI for any surviving child and one-third of respondents reported performing a brain MRI for children with newly diagnosed impairment. Almost all responding center personnel reported performing conventional and DWI primary MRI sequences with fewer than one-third performing MRS. Once validated, our data may help inform future evidence-based guidelines for using and interpreting of MRI for prognostication after pediatric cardiac arrest. Future directions also include determining whether brain MRI and MRS improves diagnostic accuracy when in combination with key clinical variables, blood-based biomarkers, and other clinically relevant testing (eg, electroencephalography).
This study has some limitations. Patient and arrest variables in the overall parent study population 12 for patients who had or did not undergo brain MRI is in eTable 1 in Supplement 1.
Notably, patients with longer arrest duration, who were more deeply comatose after arrest, and who had more hospital disposition to rehabilitation had MRI performed. Patients with an unfavorable outcome had MRI or MRS assessed earlier than those with a favorable outcome, potentially biasing MRI and MRS findings. Our analysis may thus not represent all lesions given their natural evolution over time. 27 Indeed, given that imaging occurred earlier in patients with unfavorable outcomes suggests that in some cases, the clinical decision to image is not based solely on patient stability.
Intensive care unit clinicians determined whether a brain MRI or MRS was clinically indicated. We did not collect information on why a brain MRI or MRS was not performed. The use of clinically indicated brain imaging may have introduced bias. Clinicians were aware of imaging findings, potentially introducing bias in clinical care that could be associated with outcomes. Sample size remains a limitation, specifically for high-quality MRS data. In this study, MRS ROIs were selected according to the availability of substantial control data regarding normal age-dependent metabolic changes. It is conceivable that the metabolic evaluation of brain injury can be improved by selecting a different set of ROIs placed in brain regions that are more associated with long-term outcomes. Outcome analysis was not adjusted for prearrest baseline function, but entry criteria excluded children with severe coma at baseline. The POCCA protocol 12 did not record who died due to withdrawal from lifesustaining therapy.

Conclusions
In this cohort study of pediatric patients with cardiac arrest who underwent MRI and MRS within 2 weeks, higher MRI Injury Score, increased lactate, and decreased NAA in regions of interest were associated with an unfavorable outcome at 1 year. External validation is needed to confirm these findings for clinical implementation.

ARTICLE INFORMATION
Accepted for Publication: April 28, 2023.