Association Between Quantitative Diffusion-Weighted Magnetic Resonance Neuroimaging and Outcome After Pediatric Cardiac Arrest

Abstract

Background and Objectives

Diffusion MRI can quantify the extent of hypoxic-ischemic brain injury after cardiac arrest. Our objective was to determine the association between the adult-derived threshold of apparent diffusion coefficient (ADC) <650 × 10⁻⁶ mm²/s in >10% of brain tissue and an unfavorable outcome after pediatric cardiac arrest. Since ADC decreases exponentially as a function of increasing age, we determined the association between (1) having >10% of brain tissue below a novel age-dependent ADC threshold, and (2) age-normalized whole-brain mean ADC and unfavorable outcome.

Methods

This was a retrospective study of patients aged ≤18 years who had cardiac arrest and a clinically obtained brain MRI within 7 days. The primary outcome was unfavorable neurologic status at hospital discharge based on the Pediatric Cerebral Performance Category score. ADC images were extracted from 3-direction diffusion imaging. We determined whether each patient had >10% of voxels with an ADC below prespecified thresholds. We computed the whole-brain mean ADC for each patient.

Results

One hundred thirty-four patients were analyzed. Patients with ADC <650 × 10⁻⁶ mm²/s in >10% of voxels had 15 times higher odds (95% CI 5–65) of an unfavorable outcome compared with patients with ADC <650 × 10⁻⁶ mm²/s (area under the receiver operating characteristic curve [AUROC] 0.72 [95% CI 0.63–0.80]). These ADC criteria had a sensitivity and specificity of 0.49 and 0.94, respectively, and positive and negative predictive values of 0.93 and 0.52, respectively, for an unfavorable outcome. The age-dependent ADC threshold that yielded optimal sensitivity and specificity for unfavorable outcomes was <300 × 10⁻⁶ mm²/s below each patient's predicted whole-brain mean ADC. The sensitivity, specificity, and positive and negative predictive values for this ADC threshold were 0.53, 0.96, 0.96, and 0.54, respectively (odds ratio [OR] 26.4 [95% CI 7.5–168.3]; AUROC 0.74 [95% CI 0.66–0.83]). Lower age-normalized whole-brain mean ADC was also associated with an unfavorable outcome (OR 0.42 [0.24–0.64], AUROC 0.76 [95% CI 0.66–0.82]).

Discussion

Quantitative diffusion thresholds on MRI within 7 days after cardiac arrest were associated with an unfavorable outcome in children. The age-independent ADC threshold was highly specific for predicting an unfavorable outcome. However, the specificity and sensitivity increased when using age-dependent ADC thresholds. Age-dependent ADC thresholds may improve prognostic accuracy and require further investigation in larger cohorts.

Classification of Evidence

This study provides Class III evidence that quantitative diffusion-weighted imaging within 7 days postarrest can predict an unfavorable clinical outcome in children.

Over 20,000 children experience cardiac arrest each year in the United States.^1-4 Hypoxic-ischemic brain injury (HIBI) is a major cause of mortality and the primary determinant of the degree of neurologic disability after cardiac arrest.^5-8 Neurocognitive and neurobehavioral sequelae of HIBI are prevalent in survivors and can have a significant effect on quality of life.^9,10 Neuroimaging and diffusion MRI, in particular, can quantify the degree of HIBI after cardiac arrest because it is highly sensitive to cytotoxic edema that occurs after cerebral ischemia.^11-13 Intensivists and neurologists routinely incorporate and prioritize the extent and severity of HIBI on diffusion MRI into their outcome predictions.^14,15 However, there are limited data on the association between MRI findings and outcomes after pediatric cardiac arrest.^7,16,17

We previously demonstrated a strong association between the extent of diffusion restriction on clinically obtained MRIs and neurologic outcome after out-of-hospital cardiac arrest using a pediatric MRI modification of the Alberta Stroke Program Early Computed Tomography Score.¹⁸ This scoring system requires manual scoring of diffusion MRIs and is subject to potential bias since involvement of a very small portion of a territory will result in the same score as involvement of the entire territory. Quantitative analysis of diffusion MRI can more precisely define the extent of HIBI. The apparent diffusion coefficient (ADC) is a measure of the magnitude of restricted diffusion within a voxel of brain tissue, with lower ADC values signifying more restricted diffusion and HIBI. In adults, having an ADC below 650 × 10⁻⁶ mm²/s in greater than 10% of brain tissue after cardiac arrest was predictive of poor outcome with a specificity of 0.96 arrest.^19,20 ADC decreases exponentially as a function of increasing age. This is due to a high water content in neonates and young infants that decreases as the brain undergoes maturation and myelination.^21-23 A few pediatric studies have shown an association between poor neurologic outcome and whole-brain ADC values.^24-26 However, these were limited by small sample sizes and did not account for patient age.

Our primary objective was to determine the association between the adult-derived threshold of having greater than 10% of brain tissue with an ADC below 650 × 10⁻⁶ mm²/s and an unfavorable outcome after pediatric cardiac arrest. We secondarily aimed to (1) determine the association between having greater than 10% of brain tissue below a novel age-based ADC threshold and an unfavorable outcome and (2) determine the association between whole-brain mean ADC and unfavorable outcome, accounting for patient age. The primary research question being addressed in the study is whether quantitative diffusion-weighted imaging (DWI) within 7 days postcardiac arrest can predict an unfavorable clinical outcome in children.

Methods

Standard Protocol Approvals, Registrations, and Patient Consents

The study was approved by the Children's Hospital of Philadelphia Institutional Review Board. Consent was not required, and this cohort represented a consecutive sample of patients.

Design and Setting

This was a retrospective observational study using an institutional cardiac arrest database at the Children's Hospital of Philadelphia between June 2005 and December 2019. Patients aged ≤18 years who experienced an in- or out-of-hospital cardiac arrest, received postcardiac arrest care in the pediatric intensive care unit (PICU), and had a clinically obtained brain MRI ≤7 days of their cardiac arrest were included. Only index MRIs were analyzed. We excluded patients with concomitant brain injury from a non-HIBI etiology, a baseline Pediatric Cerebral Performance Category (PCPC) score of 5 (i.e., coma or vegetative state), or evidence of severe brain atrophy or sequela of prior injury on MRI. Patients were excluded if diffusion data were missing or incomplete or if technical issues (e.g., artifact or diffusion sequence performed with incorrect parameters) precluded quantitative MRI diffusion analysis. Abstracted data included patient demographics, cardiac arrest characteristics, and details about postcardiac arrest care. Postcardiac arrest care was determined by the clinical team and guided by an institutional pathway.²⁷

Clinical Outcomes

The primary outcome was unfavorable neurologic status at hospital discharge based on changes in the PCPC score.^28-30 The secondary outcome was discharge mortality. PCPC is a 6-point scale of global neurologic function: (1) normal; (2) mild disability; (3) moderate disability; (4) severe disability; (5) coma or vegetative state; and (6) death. Baseline PCPC was defined as the patient's neurologic function before their cardiac arrest. PCPC scores at baseline and hospital discharge were retrospectively assigned by 2 trained nurse raters via medical record review. Discrepancies were resolved by consensus. Favorable outcome was defined as either no change in PCPC from baseline to hospital discharge or a change from PCPC 1 to 2. Unfavorable outcome was defined as a change in PCPC ≥1 from baseline that resulted in a hospital discharge PCPC of 3, 4, 5, or 6. PCPC raters were blinded to MRI data.

MRI Acquisition and Processing

Three-direction diffusion imaging with b-values of 0, 500, and 1,000 s/mm², performed on either a 1.5 or 3 T Siemens scanner (Erlangen, Germany), was a standard clinical sequence for all brain MRIs. Scan parameters varied as patients were imaged on 8 different MRI scanners over the study period. All diffusion scans were performed with a slice thickness of 4 mm. For 1.5 T scans, time to echo (TE) ranged from 72 to 122 milliseconds, and for 3 T scans, TE ranged from 54 to 103 milliseconds. The trace diffusion-weighted image for each b-value and the corresponding ADC image maps were computed by the scanner and extracted for analysis. MRI analysis was performed using FSL (FMRIB software library).³¹ The trace diffusion-weighted image corresponding to b = 1,000 s/mm² was intensity thresholded and binarized to create a brain tissue mask, which excluded soft tissue, bone, and CSF surrounding the brain and ventricles. The tissue mask was applied to the ADC map to restrict measurements to brain tissue. The whole-brain mean ADC was computed for each patient. A lower ADC indicated worse HIBI. For each patient, we then computed a 1,000 bin histogram spanning their ADC range. From this individualized histogram, we determined the percent of voxels within brain tissue below various prespecified ADC thresholds. MRI processing was conducted by investigators blinded to clinical data including outcome assessment.

Evaluation of an Adult-Derived, Age-Independent ADC Threshold

We assessed the sensitivity and specificity of the adult-derived threshold of ADC <650 × 10⁻⁶ mm²/s in >10% of voxels for an unfavorable outcome.^19,20 To ensure that this was the most appropriate for children, we also computed the sensitivity and specificity using ADC thresholds ranging from 400 to 1,000 × 10⁻⁶ mm²/s in 50 × 10⁻⁶ mm²/s increments in >5%, 10%, and 15% of voxels.

Evaluation of a Novel Age-Dependent ADC Threshold

Because the brain matures and myelinates with age, ADC values exponentially decrease with age.^21-23 Therefore, to account for age, we generated an exponential regression model fitting whole-brain mean ADC by age from 118 healthy controls (Figure 1, solid regression line).²¹ All healthy controls had a baseline PCPC of 1, no neurodevelopmental concerns, and MRIs that were anatomically normal. This data set was chosen because we had access to the raw data to perform ADC calculations. From this exponential regression model, we determined the predicted whole-brain mean ADC for each postcardiac arrest patient based on their age, rounded to the nearest 50 × 10⁻⁶ mm²/s. We then computed the sensitivity and specificity for unfavorable outcomes using thresholds created by subtracting values ranging from 100 to 400 × 10⁻⁶ mm²/s in 50 × 10⁻⁶ mm²/s increments from each patient's predicted whole-brain mean ADC for age. For example, if the predicted whole-brain mean ADC for age for a patient was 1,100 × 10⁻⁶ mm²/s, the sensitivity and specificity were calculated using thresholds from ADC <1,000 to <700 × 10⁻⁶ mm²/s in >10% of voxels.

Figure 1. ADC Values by Age.

Exponential regression (black line) with 95% CI (dashed line) of the whole-brain mean ADC in healthy controls compared with the whole-brain mean ADC in favorable (open circles) and unfavorable (closed circles) patients. ADC = apparent diffusion coefficient.

Evaluation of Age-Normalized Whole-Brain Mean ADC

To evaluate whether whole-brain mean ADC for cardiac arrest patients was associated with an unfavorable outcome accounting for patient age, we assigned ADC Z-scores to patients using the exponential regression generated from healthy controls.

Statistical Analysis

We report descriptive statistics as frequencies and percentages for categorical variables and median with interquartile range (IQR) for continuous variables. We used χ² and Fisher exact tests to evaluate the associations between categorical variables and dichotomous outcomes (unfavorable vs favorable outcome) and Wilcoxon rank-sum for continuous variables between outcome groups. We used univariable and multivariable logistic regression models to assess the relationship between patients who had >10% of voxels below an ADC threshold of <650 × 10⁻⁶ mm²/s and an unfavorable outcome. Multivariable models were adjusted for time from cardiac arrest and MRI, cardiopulmonary resuscitation (CPR) duration, and initial pH. The same models were applied with an age-dependent ADC threshold that was calculated based on >10% of voxels with ADC <300 below each patient's predicted whole-brain mean ADC for age. We performed sensitivity analyses examining whether scanner strength and postcardiac arrest targeted temperature management to 33°C significantly affected outcomes. The sensitivity analysis for targeted temperature management to 33°C was unable to be performed for the age-independent ADC threshold (>10% of voxels below an ADC threshold of <650 × 10⁻⁶ mm²/s) due to outcome distribution when these patients were removed. We performed univariable logistic regression to assess the relationship between age-normalized ADC Z-scores and unfavorable outcomes and multivariable logistic regression adjusting for time between cardiac arrest and MRI, CPR duration, and initial postcardiac arrest pH. Statistics were analyzed using JMP (SAS Institute, Cary, NC), SAS version 9.4 (SAS Institute), and broom packages of R version 3.4 (R Core Team 2019).

Data Availability

Anonymized data will be shared by request from any qualified investigator.

Results

Two hundred thirty-nine patients aged ≤18 years had cardiac arrest, survived PICU admission, and had an MRI obtained for clinical indications (Figure 2). Patients were excluded because MRI was obtained >7 days postcardiac arrest (n = 59), MRI was not analyzable for technical reasons or missing/incomplete data (n = 22), presence of concomitant brain injury from a non-HIBI etiology (n = 19), or preexisting severe neurologic injury with a baseline PCPC of 5 or MRI evidence of preexisting severe brain atrophy or injury (n = 5). One hundred thirty-four patients were included in analyses. Fifty-three percent (n = 71) of patients were scanned on a 1.5 T, and 47% (n = 63) were scanned on a 3 T MRI. There were no differences in scanner field strength between patients with ADC <650 × 10⁻⁶ mm²/s in >10% and ≤10% of voxels (p = 0.76).

Figure 2. Consort Diagram.

HIBI = hypoxic-ischemic brain injury; PCPC = pediatric cerebral performance category; PICU = pediatric intensive care unit.

Association of an Adult-Derived, Age-Independent ADC Threshold With an Unfavorable Outcome

We determined that the adult-derived ADC threshold of <650 × 10⁻⁶ mm²/s in >10% of voxels prioritized specificity for unfavorable outcomes across all pediatric patients without overly sacrificing sensitivity (Table 1). Thirty-four percent (n = 45) of patients had ADC <650 × 10⁻⁶ mm²/s in >10% of voxels. Patients with ADC <650 × 10⁻⁶ mm²/s in >10% of voxels more commonly had pulseless electrical activity or asystole, received longer durations of CPR and more epinephrine doses, and had lower initial pH and higher initial lactate values than patients with ADC <650 × 10⁻⁶ mm²/s in ≤10% of voxels (each p < 0.05, Table 2), and had a longer time from cardiac arrest to MRI (median 4 IQR [3–6] vs 3 [2–5] days, p = 0.03).

Table 1.

Sensitivity and Specificity of Unfavorable Outcomes for >5%, 10%, and 15% of Voxels With an ADC Less Than Prespecified Thresholds From 400 to 1,000 × 10⁻⁶ mm²/s

Table 2.

Patient Demographics and Cardiac Arrest Characteristics by Whether ADC <650 × 10⁻⁶ mm²/s in ≤10% or >10% of MRI Voxels

Sixty-three percent (n = 85) of patients had an unfavorable outcome, and 26% (n = 35) died (Table 3). Patients with ADC <650 × 10⁻⁶ mm²/s in >10% of voxels were more likely to have an unfavorable outcome than patients with ≤10% of voxels (p < 0.001). Sixty-four percent (29/45) of patients with ADC <650 × 10⁻⁶ mm²/s in >10% of voxels died, whereas 7% (6/89) of patients with ≤10% of voxels died (Table 2). Modes of death for patients with ADC <650 × 10⁻⁶ mm²/s in >10% of voxels were withdrawal due to neurologic prognosis (55%, 16/29), death by neurologic criteria (41%, 12/29), and rearrest without return of spontaneous circulation (3% 1/29). All patients with ADC <650 × 10⁻⁶ mm²/s in ≤10% of voxels who died (n = 6) had withdrawal due to poor neurologic prognosis.

Table 3.

Patient Demographics and Cardiac Arrest Characteristics by Outcome

Patients with ADC <650 × 10⁻⁶ mm²/s in >10% of voxels were much more likely to have an unfavorable outcome compared with patients with ADC <650 × 10⁻⁶ mm²/s in ≤10% of voxels (odds ratio [OR] 15.0; 95% CI 5–65); area under the receiver operating characteristic curve (AUROC) 0.72 (95% CI 0.63–0.80) with a sensitivity of 0.49, a specificity of 0.94, and positive and negative predictive values of 0.93 and 0.52, respectively (Table 4). After accounting for time from cardiac arrest to MRI, CPR duration, and initial pH, patients with ADC <650 × 10⁻⁶ mm²/s in >10% of voxels had higher odds of an unfavorable outcome (adjusted OR 22 [95% CI 4–409]; AUROC 0.86 [95% CI 0.79–0.93]). Sensitivity analysis including scanner strength (1.5 vs 3 T) into the model yielded results similar to the original model.

Table 4.

Association Between ADC and Unfavorable Outcome

Association of a Novel Age-Dependent ADC Threshold With an Unfavorable Outcome

The ADC threshold that yielded optimal sensitivity and specificity for unfavorable outcomes was <300 × 10⁻⁶ mm²/s below each patient's predicted whole-brain mean ADC in >10% of voxels (Table 5). The sensitivity, specificity, positive predictive value, and negative predictive value for this threshold were 0.53, 0.96, 0.96, and 0.54, respectively, with an AUROC of 0.74 (95% CI 0.66–0.83; Table 4). After accounting for time from cardiac arrest to MRI, CPR duration, and initial pH, patients with ADC <300 × 10⁻⁶ mm²/s below their predicted whole-brain mean ADC in >10% voxels had higher odds of an unfavorable outcome (adjusted OR 21 [95% CI 4–392]; AUROC 0.86 [95% CI 0.78–0.93]). Sensitivity analyses including scanner strength (1.5 vs 3 T) and targeted temperature management (33 vs 36°C) into the models yielded results similar to the original models.

Table 5.

Sensitivity and Specificity of Unfavorable Outcomes With >10% of Voxels With an ADC Less Than Thresholds Ranging From 100 to 400 × 10⁻⁶ mm²/s Less Than the ADC Predicted for Age by the Exponential Regression Generated From Heathy Controls

Association of an Age-Normalized Whole-Brain Mean ADC With an Unfavorable Outcome

The whole-brain mean ADC for patients (favorable outcome: open circles; unfavorable outcome: black circles) and healthy controls (solid regression line with 95% CIs [dashed line]) is shown in Figure 1. Patients with an unfavorable outcome had a lower whole-brain mean ADC compared with patients with a favorable outcome (876 [727–930] vs 969 [898–1,042] × 10⁻⁶ mm²/s, p < 0.001). A unit decrease in the whole-brain mean ADC Z-score to account for patient age was associated with higher odds of having an unfavorable outcome (OR 0.42 [0.24–0.64], AUROC 0.76 [95% CI 0.66–0.82]; Table 4).

Classification of Evidence

This study provides Class III evidence that quantitative DWI within 7 days postarrest can predict an unfavorable clinical outcome in children.

Discussion

These data demonstrate that quantitative diffusion thresholds on clinically obtained MRIs in children within 7 days after cardiac arrest were associated with unfavorable neurologic outcomes. The odds of unfavorable outcomes were significantly higher among patients with ADC <650 × 10⁻⁶ mm²/s in >10% of brain tissue compared with patients with ADC <650 × 10⁻⁶ mm²/s in ≤10% of brain tissue. In addition, these adult-derived ADC criteria had a high specificity (0.94) and positive predictive value (0.93) for unfavorable outcomes. The odds of unfavorable outcomes were significantly higher in the children with >10% of brain tissue below this fixed, age-independent ADC threshold, after adjusting for time between cardiac arrest and MRI, CPR duration, and initial pH. The specificity and sensitivity slightly improved when using either age-dependent ADC thresholds or age-normalized whole-brain mean ADC.

Our finding that quantitative diffusion MRI metrics are highly predictive of unfavorable outcomes after pediatric cardiac arrest supports the incorporation of MRI into neuroprognostic considerations used to direct patient care and counsel families. ADC metrics were also associated with other factors known to be associated with an unfavorable outcome including CPR duration, initial rhythm, number of epinephrine doses, and initial pH and lactate.⁷ Our cohort included patients after both in- and out-of-hospital cardiac arrest, as well as those who were neurologically normal (PCPC 1) and had developmental disabilities (PCPC 2–4) before their cardiac arrest, consistent with contemporary demographics of patients cared for in pediatric ICUs after cardiac arrest.^7,32,33

We analyzed quantitative diffusion MRI in a cohort of 134 children after cardiac arrest. Several smaller studies have similarly shown an association between quantitative diffusion MRI injury and outcomes, but they included only whole-brain ADC or ADC in select regions and did not account for patient age.^24-26 ADC is exponentially related to age, with younger infants and children having a higher ADC in the whole brain and individual structures due to higher water content in the immature brain.^21-23 Qualitative accounting for age when neuroradiologists assess the extent or severity of diffusion restriction relies primarily on clinical expertise,^18,34,35 whereas quantitative MRI analysis requires specification of an ADC threshold below which a brain region is considered to have sustained hypoxic-ischemic injury.

In this study, we examined the predictive ability of both age-independent and age-dependent ADC thresholds. We first used an ADC threshold of <650 × 10⁻⁶ mm²/s in >10% of brain tissue derived from a multicenter study of diffusion MRI in comatose adults after cardiac arrest.^19,20,36 We demonstrated that this threshold has similar specificity (0.94 vs 0.96) and positive predictive value (0.93 vs 0.94) between children and adults. This implies that brain tissue with an ADC <650 × 10⁻⁶ mm²/s has sustained hypoxic-ischemic injury independent of the ADC value in the patient's brain prearrest. We explored an ADC threshold for ischemia that accounted for patient age using ADC values generated from healthy controls. This approach yielded improved specificity and sensitivity, likely driven by the youngest patients who are in a highly nonlinear period of ADC change. It suggests that generating normal ADC values from a larger data set of healthy controls may facilitate identifying more precise, age-specific ADC thresholds that more accurately predict a neurologic outcome.

We prioritized specificity because it represented the proportion of patients with a favorable outcome with ADC <650 × 10⁻⁶ mm²/s in ≤10% of brain tissue. Thus, if a patient only had a small amount of HIBI on MRI using these ADC criteria, their likelihood of a favorable outcome is very high. This minimizes the chances of a patient receiving an outcome prediction of a poor prognosis and having withdrawal of life-sustaining therapy (WLST) when they may have had a favorable outcome. Our lower sensitivity suggests that more factors can contribute to an unfavorable outcome beyond the degree of HIBI on MRI, including the presence of comorbid conditions and the severity of injury to other organs. In addition, 35% of patients had prearrest PCPC scores of 2–4. In these patients, it is possible that a relatively small amount of HIBI could contribute to a more impaired neurologic status and a higher PCPC score at discharge.

In our study, there were 3 patients whose diffusion injury exceeded the primary ADC threshold, but had favorable outcomes. These patients were not infants. Two had prearrest PCPC scores of 1 and hospital discharge PCPC scores of 2, whereas 1 had both a prearrest and hospital discharge PCPC scores of 3. This reinforces the need for multimodal neuroprognostication that combines quantitative MRI with EEG background, neurologic examination, and other biomarkers to improve prognostic accuracy; however, prospective studies should evaluate the prognostic value of these modalities alone and in combination.^7,37 Although the quantitative ADC data were not available to the clinicians, MRIs and reports generated by pediatric neuroradiologists were available to treating clinicians and were likely incorporated into discussions regarding care goals including decisions about WLST. This potential manifestation of the self-fulfilling prophecy bias may be a partial explanation for why the sensitivity of the primary ADC threshold for an unfavorable outcome was low in our study.³⁸ Quantitative diffusion data only provide the relative amount of brain tissue with ischemia but do not address the anatomical distribution of the ischemia and whether the ischemia is localized to structures associated with worse neurologic disability.¹⁸ Future studies should assess whether combining these quantitative diffusion data with conventional MRI sequences or other imaging modalities could provide additional prognostic value. In addition, 35% of patients in this study had prearrest PCPC scores of >1. A smaller amount of hypoxic-ischemic injury in these patients with preexisting brain dysfunction may have disproportionally resulted in further functional impairment.

A limitation of this study was that patients and healthy controls were imaged on different scanners with varied diffusion protocols, which can introduce interscanner ADC variance. Although diffusion MRI measurement may vary with MRI hardware, field strength, and pulse sequence parameters, data demonstrate that ADC has good reproducibility across different scanner platforms and different field strengths.^39,40 Grech-Sollars et al.³⁹ reported ADC interscanner coefficient of variation to be less than 4% for brain ADC measures and recommended an effect size greater than 8% be considered beyond the interscanner variation. In this study, the ADC threshold of <650 × 10⁻⁶ mm²/s represented an approximate 30% decrease from typical ADC levels. Because of limited published data, the healthy control regression line used in this study for children aged 0–18 years was based on data from children aged 0–8 years.²¹ Nevertheless, our results suggest that these estimates were useful overall, presumably because changes in the whole-brain mean ADC is slow after age 2 years. Future studies are warranted to fully characterize the developmental trajectory of whole-brain mean ADC across the pediatric age range.

Quantifying diffusion restriction is currently limited to the research domain and is not available clinically. These data provide a proof of concept that these calculations are feasible to perform in pediatrics with retrospective data. With the development of a more comprehensive set of normative data across the pediatric age spectrum, it may be possible to develop an automated pipeline to perform these computational analyses and provide results to bedside clinicians in a reasonable and clinically meaningful time frame to be used for clinical care and neuroprognostic discussions with families. Input from MRI researchers will be crucial to implementing these types of analyses across centers with different scanner manufacturers and data formats. Specific prognostic ADC thresholds could then be prospectively validated.

The time between cardiac arrest and MRI statistically differed between patients with ADC <650 × 10⁻⁶ mm²/s in >10% and ≤10% of voxels (median 4 vs 3 days, respectively). However, this difference is probably not clinically meaningful as pseudonormalization of the diffusion signal does not occur until 5–7 days after cardiac arrest.^12,41 We postulate that patients with more severe HIBI may have been less stable for transport to MRI on postarrest day 3 and thus more likely to have had MRI delayed. Another potential limitation is that the perceived severity of HIBI on MRI by the clinical team may have contributed to issuing a poor prognosis and, ultimately, WLST. This would have biased our analyses toward the null as patients with an unfavorable outcome would have had less severe HIBI by quantitative ADC. In addition, it is unlikely in our cohort that patients would have transitioned from an unfavorable to a favorable outcome if WLST had not occurred, as prior data suggest that the amount of diffusion restriction is similar between patients who survived cardiac arrest with an unfavorable outcome and those who died from WLST due to poor neurologic prognosis.¹⁸ We only assessed the outcome at hospital discharge, and the patients' functional status may evolve over time with rehabilitation services.

An age-independent quantitative diffusion threshold of ADC <650 × 10⁻⁶ mm²/s in >10% of brain tissue is highly specific for unfavorable outcome in children after cardiac arrest. Age-dependent ADC thresholds may improve prognostic accuracy and require further study in larger cohorts.

Glossary

ADC

apparent diffusion coefficient

AUROC

area under the receiver operating characteristic curve

CPR

cardiopulmonary resuscitation

DWI

diffusion-weighted imaging

HIBI

hypoxic-ischemic brain injury

IQR

interquartile range

OR

odds ratio

PCPC

Pediatric Cerebral Performance Category

PICU

pediatric intensive care unit

TE

time to echo

WLST

withdrawal of life-sustaining therapy

Appendix. Authors

Study Funding

This study was funded by the Department of Anesthesiology and Critical Care, Children's Hospital of Philadelphia, University of Pennsylvania Perelman School of Medicine.

Disclosure

M.P. Kirschen's institution has received research support from the NIH. J.I. Berman has served as a consultant for McGowan Associates. The other authors report no relevant disclosures. Go to Neurology.org/N for full disclosures.