Description of Clinicians' Predictive Abilities Before and After a Diffusion Tensor Imaging (DTI)

With a mortality rate of over 13% in the neuro-intensive care unit, 86% of which are related to the withdrawal of active treatment, predicting a reliable neurological prognosis is of paramount importance in neuro-intensive care to help clinicians inform family members appropriately and tailor care to the patient's presumed wishes. This allows for the justification of intensive resuscitation treatments in patients with a presumed good prognosis and, conversely, avoids keeping patients with an unfavorable neurological prognosis alive against their presumed wishes.

Current guidelines advocate a multimodal neuroprognostic approach, where the clinician synthesizes prognostic factors from clinical examination, EEG, plasma or CSF biomarkers, and imaging to collaboratively develop an individualized prognosis, even though no study has evaluated the quality of this multimodal prognosis. Furthermore, the existing variability depending on the nosological context makes this approach complex and likely difficult to reproduce. Brain Magnetic Resonance Imaging (MRI) is an important tool for neuroprognostication in patients with neurological injuries, and for this reason, it is commonly used in clinical practice as part of a multimodal approach.

More specifically, brain MRI with diffusion tensor imaging (DTI) sequences is validated prognostic tool in post anoxic coma, traumatic brain injury, and more recently in spontaneous hemorrhagic stroke, subarachnoid hemorrhage, and hypoglycemia. It allows for relatively precise classification of patients after cerebral anoxia or traumatic brain injury into 3 categories: those with an unfavorable outcome (specificity greater than 95%), those with an indeterminate outcome, and those with a favorable outcome (specificity greater than 95%).

It is currently being evaluated in the COMABASE study for its prognostic value in other cerebral pathologies (spontaneous hemorrhagic stroke, subarachnoid aneurysmal hemorrhage, and hypoglycemia).

Despite its excellent discriminatory values (AUC 0.93), it has not been compared to the gold standard, namely multimodal neuroprognosis performed by a panel of clinicians.

The investigators wish to study the informational contribution of MRI with DTI to the clinician's prognostic development. To this end, the investigators will study the degree of prognostic uncertainty experienced by clinicians before DTI scans are performed to determine whether MRI helps reduce this uncertainty.

This research aims to evaluate the prognostic ability of clinicians prescribing diffusion tensor imaging (DTI), specifically the fractional anisotropy and mean diffusivity in the white matter, associated with patient age, and the time between the onset of symptoms and the MRI scan. It also aims to evaluate the predictive power of clinicians integrating multimodal signals (clinical finfings, EEG, and CT scan) regarding neurological prognosis at the time of the MRI scan. The clinicians' predictions will be assessed on the day of the MRI scan and a few hours prior to it, to ensure the closest possible comparison with the prognosis provided by MRI markers.

The investigators need to recruit 220 patients and 70 clinicians across 8 intensives care units in Paris, France.

The planned duration of the research will be 3 years and 2 months, with an inclusion period of 2 years associated with a follow-up period of 14 months (12 months +/- 2 months) in order to include a total of 220 patients.

The primary objective of this study is to compare the prognostic performance of two prescribing clinicians using MRI and Diffusion Tensor Imaging (DTI) before and after the MRI results.

Secondary objectives are:

• to study the informational gain of MRI in improving clinicians' predictions after obtaining the MRI results.
• To study predicitive factors of the quality of the clinicians' prognosis, including: the time between the discontinuation of sedation and the MRI scan, the clinicians' experience, and the level of altered consciousness at the time of the MRI scan.
• To study the concordance between the two clinicians' prognoses before the MRI scan and the concordance between the two clinicians' prognoses after receiving the MRI results.

The primary outcome measure (detailed above) is the clinicians' ROC curve for predicting unfavorable patient outcomes at 12 months (Glasgow Outcome Scale Extended (GOSE <5).

Detailed protocol:

The investigators will screen every MRI prescribed to establish a prognosis for an intensive care patient.

The day before the MRI scan, the investigators will request the participation in the study of patient's relative and two clinicians from the same Intensive Care Unit (ICU), who are responsible for the patient awaiting prognosis.

If the relatives and the 2 clinicians agree to participate in the study, the 2 clinicians will complete a questionnaire, the day before or the same day, few hours before MRI scan, assessing their knowledge about patient's prognostic factors and allowing them to rate between 0 and 100, the probability of an unfavorable outcome. Note that a clinician may participate in the study for multiple patients.

Then the patient will undergo the MRI scan with DTI. The MRI sequences will contain at minimum, the FLuid Attenuated Inversion Recovery images (FLAIR), Diffusion-Weighted Imaging (DWI), braimstem spectroscopy, Susceptibility-Weighted Imaging (SWI) and Diffusion Tensor Imaging (DTI).

The DTI is aimed to obtain the standardized fractional anisotropy and the standardized mean diffusivity in the 17 regions of white matter defined by the Johns Hopkins Hospital Atlas. Thanks to the fractional anisotropy, the mean diffusivity, the patient age and the time between the brain injury and the MRI scan, the calculator published in Intensive care Medecine (Puybasset et al 2026 PMID : 41782003), will provide the probability of the unfavorable outcome for each patient.

As soon as the MRI results are obtained (on the same day or one day after if possible):

The same 2 clinicians will review the images from MRI and the results from the calculator and will complete once again the questionnaire about prognostic factors and their estimation of the unfavorable outcome.

The investigator will collect data from the patient in intensive care retrospectively as demographic data, the type of brain lesion, the date of the brain lesion, the date sedation was discontinued, the detailed Glasgow Coma Scale score on the day or the day before the MRI scan, eye movement tracking and fluctuations in level of consciousness, duration of ventilation, the limitation of active therapy, the date of discharge from intensive care or the date of death. Others data collections will include the time between the discontinuation of sedation and the MRI scan, the clinician's experience level, and the patient's level of consciousness at the time of the MRI to study their influence on the clinician's prognosis accuracy.

Follow-up of the population at 1 year (+/- 2 months). If the patient is alive and not under legal guardianship, a telephone call will be made to inform the patient and ask for their consent to continue the research.

Then, a Glasgow Outcomes Scale Extended (GOSE) will be administered to the patient or a relative, and a Medical Ouctomes study Short Form General Health Survey (SF-12) will be completed, blinded to the prognostic assessment established by diffusion tensor MRI and the clinicians.

The investigators will analyze clinicians' preMRI prognostic result, clinicians' post MRI prognosic results and the DTI calculator to establish the ROC curve for these 3 prognosis assessments.

The inclusion and non-inclusion criteria are detailed above.

Duration of participant participation and duration of the research:

Inclusion period: 2 years Duration of patient participation: maximum 14 months Duration of the research: 3 years and 2 months

The sample size calculation was performed with the secondary objective of comparing clinician prognosis to that of MRI. It was based on the sample size calculation described in the 2024 publication by Jung et al. for comparing two diagnostic tolls using paired data.

The investigators calculated a sample size of 209 patients based on parameters collected from two previously published studies on this topic:

• an AUC of 0.80 for clinicians (Amzallag et al.)
• an AUC of 0.90 for DTI MRI (Puybasset et al.)
• an estimated prevalence of unfavorable outcomes in this population requiring multimodal assessment of 70 %
• a correlation between the two diagnostic test of 0.5
• an alpha risk of 0.05
• a power of 90 %
• estimating the number of missing data points on MRI or clinician predictions at 5%, the investigators aim to include 220 patients.

Comparison using the DeLong method for paired studies between the Receiver Operating Characteristic (ROC) curve predicting an unfavorable outcome at 12 months, established by the predictive model based on fractional anisotropy values, mean white matter diffusivity, patient age, and the time between the brain lesion and the mean prediction established by a clinician in charge of the patient the day before the MRI.

This comparison will be performed on all patients and then on patients who survived the intensive care unit.

The agreement between the clinicians' predictions will be studied using intraclass correlation coefficients for paired measures.

All tests were two-sided, and p values < 0.05 were considered statistically significant.

Patients with missing data on MRI prognostic information clinician-pre-MRI prognosis, or GOSE score at one year will be excluded from the study. These patients will not be replaced, but the investigators are accounting for this loss of of information in the sample size calculation, estimating it at approximately 5%