Artificial Intelligence Outcome Prediction in Neonates with Encephalopathy (AI-OPiNE).

"Just Accepted" papers have undergone full peer review and have been accepted for publication in Radiology: Artificial Intelligence. This article will undergo copyediting, layout, and proof review before it is published in its final version. Please note that during production of the final copyedited article, errors may be discovered which could affect the content. Purpose To develop a deep learning algorithm to predict 2-year neurodevelopmental outcomes in neonates with hypoxic-ischemic encephalopathy (HIE) using MRI and basic clinical data. Materials and Methods In this study, MRI data of term neonates with encephalopathy in the High Dose Erythropoietin for Asphyxia (HEAL) trial (ClinicalTrials.gov: NCT02811263), who were enrolled from 17 institutions between January 25th, 2017 and October ninth, 2019, were retrospectively analyzed. The harmonized MRI protocol included T1-weighted, T2-weighted, and diffusion tensor imaging. Deep learning classifiers were trained to predict the primary outcome of the HEAL trial (death or any neurodevelopmental impairment [NDI] at 2 years) using multisequence MRI and basic clinical variables, including sex and gestational age at birth. Model performance was evaluated on a test sets comprising 10% of cases from 15 institutions (in-distribution test set, n = 41) and 100% of cases from 2 institutions (out-of-distribution test set, n = 41). Model performance in predicting additional secondary outcomes, including death alone, was also assessed. Results For the 414 neonates (mean gestational age, 39 weeks ± 1.4, 232 males, 182 females), in the study cohort, 198 (48%) died or had any NDI at 2 years. The deep learning model achieved an area under the receiver operating characteristic curve (AUC) of 0.74 (95% CI: 0.60-0.86) and 63% accuracy on the in-distribution test set and an AUC of 0.77 (95% CI: 0.63-0.90) and 78% accuracy on the out-of-distribution test set. Performance was similar or better for predicting secondary outcomes. Conclusion Deep learning analysis of neonatal brain MRI yielded high performance for predicting 2-year neurodevelopmental outcomes. ©RSNA, 2024.

Duration of cardiopulmonary resuscitation and phenotype of post-cardiac arrest brain injury.

BACKGROUND: Patients resuscitated from cardiac arrest have variable severity of primary hypoxic ischemic brain injury (HIBI). Signatures of primary HIBI on brain imaging and electroencephalography (EEG) include diffuse cerebral edema and burst suppression with identical bursts (BSIB). We hypothesize distinct phenotypes of primary HIBI are associated with increasing cardiopulmonary resuscitation (CPR) duration. METHODS: We identified from our prospective registry of both in-and out-of-hospital CA patients treated between January 2010 to January 2020 for this cohort study. We abstracted CPR duration, neurological examination, initial brain computed tomography gray to white ratio (GWR), and initial EEG pattern. We considered four phenotypes on presentation: awake; comatose with neither BSIB nor cerebral edema (non-malignant coma); BSIB; and cerebral edema (GWR ≤ 1.20). BSIB and cerebral edema were considered as non-mutually exclusive outcomes. We generated predicted probabilities of brain injury phenotype using localized regression. RESULTS: We included 2,440 patients, of whom 545 (23%) were awake, 1,065 (44%) had non-malignant coma, 548 (23%) had BSIB and 438 (18%) had cerebral edema. Only 92 (4%) had both BSIB and edema. Median CPR duration was 16 [IQR 8-28] minutes. Median CPR duration increased in a stepwise manner across groups: awake 6 [3-13] minutes; non-malignant coma 15 [8-25] minutes; BSIB 21 [13-31] minutes; cerebral edema 32 [22-46] minutes. Predicted probability of phenotype changes over time. CONCLUSIONS: Brain injury phenotype is related to CPR duration, which is a surrogate for severity of HIBI. The sequence of most likely primary HIBI phenotype with progressively longer CPR duration is awake, coma without BSIB or edema, BSIB, and finally cerebral edema.

Early risk stratification for progression to death by neurological criteria following out-of-hospital cardiac arrest.

BACKGROUND: Some patients resuscitated from out-of-hospital cardiac arrest (OHCA) progress to death by neurological criteria (DNC). We hypothesized that initial brain imaging, electroencephalography (EEG), and arrest characteristics predict progression to DNC. METHODS: We identified comatose OHCA patients from January 2010 to February 2020 treated at a single quaternary care facility in Western Pennsylvania. We abstracted demographics and arrest characteristics; Pittsburgh Cardiac Arrest Category, initial motor exam and pupillary light reflex; initial brain computed tomography (CT) grey-to-white ratio (GWR), sulcal or basal cistern effacement; initial EEG background and suppression ratio. We used two modeling approaches: fast and frugal tree (FFT) analysis to create an interpretable clinical risk stratification tool and ridge regression for comparison. We used bootstrapping to randomly partition cases into 80% training and 20% test sets and evaluated test set sensitivity and specificity. RESULTS: We included 1,569 patients, of whom 147 (9%) had diagnosed DNC. Across bootstrap samples, >99% of FFTs included three predictors: sulcal effacement, and in cases without sulcal effacement, the combination of EEG background suppression and GWR ≤ 1.23. This tree had mean sensitivity and specificity of 87% and 81%. Ridge regression with all available predictors had mean sensitivity 91 % and mean specificity 83%. Subjects falsely predicted as likely to progress to DNC generally died of rearrest or withdrawal of life sustaining therapies due to poor neurological prognosis. Two of these cases awakened from coma during the index hospitalization. CONCLUSIONS: Sulcal effacement on presenting brain CT or EEG suppression with GWR ≤ 1.23 predict progression to DNC after OHCA.

Recovery among post-arrest patients with mild-to-moderate cerebral edema.

BACKGROUND: Cerebral edema after cardiac arrest may be a modifiable cause of secondary brain injury. We aimed to identify processes of care associated with recovery in a cohort of patients with mild to moderate edema. METHODS: We conducted a retrospective cohort study of adults resuscitated from out-of-hospital arrest (OHCA) at a single center from 2010 to 2018. We included those with cerebral edema ranging from mild to moderate (gray to white matter attenuation ratio (GWR) 1.2 to 1.3 on initial brain computerized tomography (CT). We used Pittsburgh Cardiac Arrest Category (PCAC) to adjust for illness severity and considered the following values in the first 24 h of admission as additional predictors: GWR, lab values affecting serum osmolality (sodium, glucose, blood urea nitrogen (BUN)), total osmolality, change in osmolality from 0 to 24 h, cardiac etiology of arrest, targeted temperature to 33 °C (vs 36 °C), time-weighted mean arterial pressure (MAP), partial pressures of arterial oxygen and carbon dioxide and select medications. Our primary outcome was discharge with cerebral performance category 1-3. We used unadjusted and adjusted logistic regression for analysis. RESULTS: We included 214 patients for whom CT was performed median 3.8 [IQR 2.4-5.2] hours after collapse. Median age was 57 [IQR 48-67] years, 82 (38%) were female, and 68 (32%) arrested from ventricular tachycardia or fibrillation. In adjusted models, modifiable processes of care were not associated with outcome. CONCLUSIONS: Illness severity, but not modifiable processes of care, were associated with recovery among post-arrest patients with mild-to-moderate cerebral edema.

Association of Initial Illness Severity and Outcomes After Cardiac Arrest With Targeted Temperature Management at 36 °C or 33 °C.

Importance: It is uncertain what the optimal target temperature is for targeted temperature management (TTM) in patients who are comatose following cardiac arrest. Objective: To examine whether illness severity is associated with changes in the association between target temperature and patient outcome. Design, Setting, and Participants: This cohort study compared outcomes for 1319 patients who were comatose after cardiac arrest at a single center in Pittsburgh, Pennsylvania, from January 2010 to December 2018. Initial illness severity was based on coma and organ failure scores, presence of severe cerebral edema, and presence of highly malignant electroencephalogram (EEG) after resuscitation. Exposure: TTM at 36 °C or 33 °C. Main Outcomes and Measures: Primary outcome was survival to hospital discharge, and secondary outcomes were modified Rankin Scale and cerebral performance category. Results: Among 1319 patients, 728 (55.2%) had TTM at 33 °C (451 [62.0%] men; median [interquartile range] age, 61 [50-72] years) and 591 (44.8%) had TTM at 36 °C (353 [59.7%] men; median [interquartile range] age, 59 [48-69] years). Overall, 184 of 187 patients (98.4%) with severe cerebral edema died and 234 of 243 patients (96.3%) with highly malignant EEG died regardless of TTM strategy. Comparing TTM at 33 °C with TTM at 36 °C in 911 patients (69.1%) with neither severe cerebral edema nor highly malignant EEG, survival was lower in patients with mild to moderate coma and no shock (risk difference, -13.8%; 95% CI, -24.4% to -3.2%) but higher in patients with mild to moderate coma and cardiopulmonary failure (risk difference, 21.8%; 95% CI, 5.4% to 38.2%) or with severe coma (risk difference, 9.7%; 95% CI, 4.0% to 15.3%). Interactions were similar for functional outcomes. Most deaths (633 of 968 [65.4%]) resulted after withdrawal of life-sustaining therapies. Conclusions and Relevance: In this study, TTM at 33 °C was associated with better survival than TTM at 36 °C among patients with the most severe post-cardiac arrest illness but without severe cerebral edema or malignant EEG. However, TTM at 36 °C was associated with better survival among patients with mild- to moderate-severity illness.

Duration and clinical features of cardiac arrest predict early severe cerebral edema.

BACKGROUND: Severe brain edema appears early after cardiopulmonary resuscitation (CPR) in a subset of patients and portends a poor prognosis. We tested whether clinical features of patients or resuscitation during out-of-hospital cardiac arrest (OHCA) are associated with early, severe cerebral edema. METHOD/RESEARCH DESIGN: We reviewed pre-hospital and hospital records for comatose patients surviving to hospital admission after OHCA who had computed tomography (CT) of brain at the time of hospital admission available for inspection. We measured the gray-white ratio (GWR) of X-ray attenuation between the caudate nucleus and posterior limb of the internal capsule, defining severe cerebral edema as GWR < 1.20. We calculated associations between severe cerebral edema and patient or resuscitation variables. RESULTS: Between 2010 and 2019, 1340 subjects were admitted of whom 296 (22%) showed severe cerebral edema on initial CT. Subjects with severe edema had lower survival (5/296, 2% vs. 377/1044, 36%). Severe edema was independently associated with total CPR duration, total dose of epinephrine, younger age, non-shockable arrest rhythms, fewer total number of rescue shocks, rearrest after initial return of pulses, and non-cardiac arrest etiology. Prevalence of severe cerebral edema increased from 2% among subjects with 0-10 min of CPR to 31% among subjects with >40 min of CPR. CONCLUSION: CPR duration along with easily measurable clinical and resuscitation characteristics predict early severe cerebral edema after OHCA. Future interventional trials should consider targeting or preventing cerebral edema after prolonged hypoxic-ischemic brain injury especially in patients with high risk clinical features.

Unsupervised learning of early post-arrest brain injury phenotypes.

INTRODUCTION: Trials may be neutral when they do not appropriately target the experimental intervention. We speculated multimodality assessment of early hypoxic-ischemic brain injury would identify phenotypes likely to benefit from therapeutic interventions. METHODS: We performed a retrospective study including comatose patients resuscitated from out-of-hospital cardiac arrest (OHCA) by one of 126 emergency medical services or in-hospital arrest at one of 26 hospitals from 2011 to 2019. All patients were ultimately transported to a single tertiary center for care including standardized initial neurological examination, brain imaging and electroencephalography; targeted temperature management (TTM); hemodynamic optimization targeting mean arterial pressure (MAP) >80 mmHg; and, coronary angiography for clinical suspicion for acute coronary syndrome. We used unsupervised learning to identify brain injury phenotypes defined by admission neurodiagnostics. We tested for interactions between phenotype and TTM, hemodynamic management and cardiac catheterization in models predicting recovery. RESULTS: We included 1086 patients with mean (SD) age 58 (17) years of whom 955 (88%) were resuscitated from OHCA. Survival to hospital discharge was 27%, and 248 (23%) were discharged with Cerebral Performance Category (CPC) 1-3. We identified 5 clusters defining distinct brain injury phenotypes, each comprising 14% to 30% of the cohort with discharge CPC 1-3 in 59% to <1%. We found significant interactions between cluster and TTM strategy (P = 0.01), MAP (P < 0.001) and coronary angiography (P = 0.04) in models predicting outcomes. CONCLUSIONS: We identified patterns of early hypoxic-ischemic injury based on multiple diagnostic modalities that predict responsiveness to several therapeutic interventions recently tested in neutral clinical trials.

Differential association of subtypes of epileptiform activity with outcome after cardiac arrest.

BACKGROUND: Epileptiform activity is common after cardiac arrest, although intensity of electroencephalographic (EEG) monitoring may affect detection rates. Prior work has grouped these patterns together as "malignant," without considering discrete subtypes. We describe the incidence of distinct patterns in the ictal-interictal spectrum at two centers and their association with outcomes. METHODS: We analyzed a retrospective cohort of comatose post-arrest patients admitted at two academic centers from January 2011 to October 2014. One center uses routine continuous EEG, the other acquires "spot" EEG at the treating physicians' discretion. We reviewed all available EEG data and classified epileptiform patterns. We abstracted antiepileptic drugs (AEDs) administrations from the electronic medical record. We compared apparent incidence of each pattern between centers, and compared outcomes (awakening from coma, survival to discharge, discharge modified Rankin Scale (mRS) 0-2) across EEG patterns and number of AEDs administered. RESULTS: We included 818 patients. Routine continuous EEG was associated with a higher apparent incidence of polyspike burst-suppression (25% vs 13% P < 0.001). Frequency of other epileptiform findings did not differ. Among patients with any epileptiform pattern, only 2/258 (1%, 95%CI 0-3%) were discharged with mRS 0-2, although 24/258 (9%, 95%CI 6-14%) awakened and 36/258 (14%, 95%CI 10-19%) survived. The proportions that awakened and survived decreased in a stepwise manner with progressively worse EEG patterns (range 38% to 2% and 32% to 7%, respectively). Among patients receiving ≥3 AEDs, only 5/80 (6%, 95%CI 2-14%) awakened and 1/80 (1%, 95%CI 0-7%) had a mRS 0-2. CONCLUSION: We found high rates of epileptiform EEG findings, regardless of intensity of EEG monitoring. The association of distinct ictal-interictal EEG findings with outcome was variable.

Neurostimulant use is associated with improved survival in comatose patients after cardiac arrest regardless of electroencephalographic substrate.

AIM: Identify EEG patterns that predict or preclude favorable response in comatose post-arrest patients receiving neurostimulants. METHODS: We examined a retrospective cohort of consecutive electroencephalography (EEG)-monitored comatose post-arrest patients. We classified the last day of EEG recording before neurostimulant administration based on continuity (continuous/discontinuous), reactivity (yes/no) and malignant patterns (periodic discharges, suppression burst, myoclonic status epilepticus or seizures; yes/no). In subjects who did not receive neurostimulants, we examined the last 24h of available recording. For our primary analysis, we used logistic regression to identify EEG predictors of favorable response to treatment (awakening). RESULTS: In 585 subjects, mean (SD) age was 57 (17) years and 227 (39%) were female. Forty-seven patients (8%) received a neurostimulant. Neurostimulant administration independently predicted improved survival to hospital discharge in the overall cohort (adjusted odds ratio (aOR) 4.00, 95% CI 1.68-9.52) although functionally favorable survival did not differ. No EEG characteristic predicted favorable response to neurostimulants. In each subgroup of unfavorable EEG characteristics, neurostimulants were associated with increased survival to hospital discharge (discontinuous background: 44% vs 7%, P=0.004; non-reactive background: 56% vs 6%, P<0.001; malignant patterns: 63% vs 5%, P<0.001). CONCLUSION: EEG patterns described as ominous after cardiac arrest did not preclude survival or awakening after neurostimulant administration. These data are limited by their observational nature and potential for selection bias, but suggest that EEG patterns alone should not affect consideration of neurostimulant use.

Clinically distinct electroencephalographic phenotypes of early myoclonus after cardiac arrest.

OBJECTIVE: We tested the hypothesis that there are readily classifiable electroencephalographic (EEG) phenotypes of early postanoxic multifocal myoclonus (PAMM) that develop after cardiac arrest. METHODS: We studied a cohort of consecutive comatose patients treated after cardiac arrest from January 2012 to February 2015. For patients with clinically evident myoclonus before awakening, 2 expert physicians reviewed and classified all EEG recordings. Major categories included: Pattern 1, suppression-burst background with high-amplitude polyspikes in lockstep with myoclonic jerks; and Pattern 2, continuous background with narrow, vertex spike-wave discharges in lockstep with myoclonic jerks. Other patterns were subcortical myoclonus and unclassifiable. We compared population characteristics and outcomes across these EEG subtypes. RESULTS: Overall, 401 patients were included, of whom 69 (16%) had early myoclonus. Among these patients, Pattern 1 was the most common, occurring in 48 patients (74%), whereas Pattern 2 occurred in 8 patients (12%). The remaining patients had subcortical myoclonus (n = 2, 3%) or other patterns (n = 7, 11%). No patients with Pattern 1, subcortical myoclonus, or other patterns survived with favorable outcome. By contrast, 4 of 8 patients (50%) with Pattern 2 on EEG survived, and 4 of 4 (100%) survivors had favorable outcomes despite remaining comatose for 1 to 2 weeks postarrest. INTERPRETATION: Early PAMM is common after cardiac arrest. We describe 2 distinct patterns with distinct prognostic significances. For patients with Pattern 1 EEGs, it may be appropriate to abandon our current clinical standard of aggressive therapy with conventional antiepileptic therapy in favor of early limitation of care or novel neuroprotective strategies. Ann Neurol 2016;80:175-184.

Group-Based Trajectory Modeling of Suppression Ratio After Cardiac Arrest.

BACKGROUND: Existing studies of quantitative electroencephalography (qEEG) as a prognostic tool after cardiac arrest (CA) use methods that ignore the longitudinal pattern of qEEG data, resulting in significant information loss and precluding analysis of clinically important temporal trends. We tested the utility of group-based trajectory modeling (GBTM) for qEEG classification, focusing on the specific example of suppression ratio (SR). METHODS: We included comatose CA patients hospitalized from April 2010 to October 2014, excluding CA from trauma or neurological catastrophe. We used Persyst®v12 to generate SR trends and used semi-quantitative methods to choose appropriate sampling and averaging strategies. We used GBTM to partition SR data into different trajectories and regression associate trajectories with outcome. We derived a multivariate logistic model using clinical variables without qEEG to predict survival, then added trajectories and/or non-longitudinal SR estimates, and assessed changes in model performance. RESULTS: Overall, 289 CA patients had ≥36 h of EEG yielding 10,404 h of data (mean age 57 years, 81 % arrested out-of-hospital, 33 % shockable rhythms, 31 % overall survival, 17 % discharged to home or acute rehabilitation). We identified 4 distinct SR trajectories associated with survival (62, 26, 12, and 0 %, P < 0.0001 across groups) and CPC (35, 10, 4, and 0 %, P < 0.0001 across groups). Adding trajectories significantly improved model performance compared to adding non-longitudinal data. CONCLUSIONS: Longitudinal analysis of continuous qEEG data using GBTM provides more predictive information than analysis of qEEG at single time-points after CA.