Use of brain diffusion tensor imaging for the prediction of long-term neurological outcomes in patients after cardiac arrest: a multicentre, international, prospective, observational, cohort study

Summary

Background

Prediction of neurological outcome after cardiac arrest is a major challenge. The aim of this study was to assess whether quantitative whole-brain white matter fractional anisotropy (WWM-FA) measured by diffusion tensor imaging between day 7 and day 28 after cardiac arrest can predict long-term neurological outcome.

Methods

This prospective, observational, cohort study (part of the MRI-COMA study) was done in 14 centres in France, Italy, and Belgium. We enrolled patients aged 18 years or older who had been unconscious for at least 7 days after cardiac arrest into the derivation cohort. The following year, we recruited the validation cohort on the same basis. We also recruited a minimum of five healthy volunteers at each centre for the normalisation procedure. WWM-FA values were compared with standard criteria for unfavourable outcome, conventional MRI sequences (fluid-attenuated inversion recovery and diffusion-weighted imaging), and proton magnetic resonance spectroscopy. The primary outcome was the best achieved Glasgow-Pittsburgh Cerebral Performance Categories (CPC) at 6 months, dichotomised as favourable (CPC 1–2) and unfavourable outcome (CPC 3–5). Prognostication performance was assessed by the area under the receiver operating characteristic (ROC) curves and compared between groups. This study was registered with ClinicalTrials.gov, number NCT00577954.

Findings

Between Oct 1, 2006, and June 30, 2014, 185 patients were enrolled in the derivation cohort, of whom 150 had an interpretable multimodal MRI and were included in the analysis. 33 (22%) patients had a favourable neurological outcome at 6 months. Prognostic accuracy, as quantified by the area under the ROC curve, was significantly higher with the normalised WWM-FA value (area under the ROC curve 0·95, 95% CI 0·91–0·98) than with the standard criteria for unfavourable outcome or other MRI sequences. In a subsequent validation cohort of 50 patients (enrolled between April 1, 2015, and March 31, 2016), a normalised WWM-FA value lower than 0·91, set from the derivation cohort, had a negative predictive value of 71·4% (95% CI 41·9–91·6) and a positive predictive value of 100% (90·0–100), with 89·7% sensitivity (75·8–97·1) and 100% specificity (69·1–100) for the prediction of unfavourable outcome.

Interpretation

In patients who are unconscious 7 days after cardiac arrest, the normalised WWM-FA value, measured by diffusion tensor imaging, could be used to accurately predict neurological outcome at 6 months. This evidence requires confirmation from future large-scale trials with a strict protocol of withdrawal or limitation-of-care decisions and time window for MRI.

Funding

French Ministry of Health, French National Agency for Research, Italian Ministry of Health, and Regione Lombardia.

Introduction

Prognostication of comatose-resuscitated patients following cardiac arrest is challenging, particularly during the first week when lingering effects of sedatives and neuromuscular blocking agents,¹ hypothermia,² and unstable physiological status preclude detailed neurological examinations.³ Standard early predictors of unfavourable outcome after cardiac arrest include absence of brain-stem reflexes, absence of motor response other than extensor response, status myoclonus, high serum concentrations of neuron-specific enolase, and absence of cortical responses by somatosensory evoked potentials.⁴ All have substantial limitations in terms of reliability.² The publication of the Parisian Region Out of Hospital Cardiac Arrest (PROCAT) registry⁵ and the Save Hearts in Arizona Registry and Education (SHARE)⁶ have raised some concerns about the existence of late awakeners and the consecutive risk of inappropriate early prognostication. The latest published guidelines on cardiac arrest recommend delaying prognostication after therapeutic hypothermia, and basing it on multiple prognostic tools.7, 8 Physicians are, therefore, confronted with a difficult decision-making process, mainly in patients who do not recover consciousness by day 7.⁹

Research in context

Evidence before this study

We searched MEDLINE for reports on the use of diffusion measures on MRI in cardiac arrest published in English between database inception and Nov 24, 2017, with the following search terms for the patient category of cardiac arrest: “heart arrest”, “cardiac arrest”, “cardiopulmonary resuscitation”, “ischemic-hypoxic encephalopathy”, “hypoxia-ischemia”, or “post anoxic coma'; the following search terms for the diffusion category: “magnetic resonance”, “MRI”, “MR”, “neuroimaging”, “apparent diffusion coefficient”, “diffusion-weighted imaging”, or “diffusion tensor imaging”; and the following search terms for outcome: “prediction”, “predictors”, “prognosis”, “prediction model”, or “outcome”. The search yielded 21 observational cohorts, of which 17 reported outcomes. Eight studies predominantly examined apparent diffusion coefficient (ADC), seven diffusion-weighted imaging (DWI), and two diffusion tensor imaging (DTI).

Except one, all the studies were based on MRI acquired during the first week after cardiac arrest. Diffuse DWI abnormalities were highly specific but only modestly sensitive of poor outcome. Being a qualitative technique, DWI is prone to interobserver variability, but it can be standardised using semiquantitative methods such as ADC. Patients with poor outcomes exhibited a nadir in ADC values at 3–5 days after cardiac arrest, which therefore seemed to be the optimum time window for prognostication using ADC. Thresholds have been determined either regionally or globally. One study reported prognostic thresholds for percentage of brain volume below specific values of ADC. However, this study was done in a small cohort (51 patients) and was monocentric. Indeed, all published multicentre studies on ADC had to use semiquantitative analysis because ADC metrics highly depend on diffusion acquisition parameters. Two particular bicentric studies reported the use of DTI derived parameters. Outcome was not the primary endpoint in the one including 97 patients in which data were not normalised. The second one, from our group, proposed a predictive composite score derived on 57 patients that was not externally validated and did not show direct clinical transferability.

Added value of this study

To our knowledge, this is the largest, prospectively followed up cohort of patients who were still comatose 7 days after cardiac arrest reporting the predictive value of fractional anisotropy, derived from DTI. Our findings provide evidence that normalised whole-brain white matter fractional anisotropy can be useful for the prediction of poor outcome. This study also resolved two methodological issues: the requirement of measurement calibration in each centre by acquiring healthy controls to establish measurements of reference; and the great need for extensive quality checks of MRI acquisitions that led to 25 (14%) of 185 patients being excluded from our derivation cohort.

Implications of all the available evidence

Our results are relevant in the clinical setting because they might provide reliable outcome predictors and could possibly improve diagnosis of late awakeners in survivors after cardiac arrest who were still unresponsive to simple orders after 7 days. The findings of our study support the use of quantitative MRI (DTI) for proxy information and management of care withdrawal decisions in this selected population of patients with cardiac arrest.

Diffusion MRI is emerging as a promising prognostic tool.¹⁰ One such measure is diffusion-weighted imaging (DWI) with the whole-brain apparent diffusion coefficient, but the ideal time window for this technique is short (between days 3 and 5 after cardiac arrest)11, 12 and it has a low sensitivity (25–30%) for unfavourable prognosis, despite high specificity (95–100%).11, 12, 13 An extension of DWI, diffusion tensor imaging (DTI), especially the calculation of fractional anisotropy (FA), allows in-vivo quantification of white matter injuries that occur in an acute, subacute, or delayed manner after global anoxia.14, 15 Another technique, proton magnetic resonance spectroscopy (¹H-MRS), allows the in-vivo quantification of the concentrations of brain metabolites characterising cellular dysfunction and neuronal loss. However, little evidence exists as to how these techniques might contribute to an improved prognostication for outcome after cardiac arrest.¹⁶

The goal of this study was to assess and validate the prognostic performance of whole-brain white matter FA (WWM-FA) in comatose patients 7 days after cardiac arrest, as compared with clinical symptoms, morphological MRI, and ¹H-MRS. Optimum cutoff was defined from a derivation cohort and assessed in an independent validation cohort.