Multicenter prospective study on predictors of short-term outcome in disorders of consciousness

Design of the study

The present prospective, observational study was conducted in the framework of the IBIA DoC-SIG and involved 12 medical institutions with expertise in diagnosis and care of adults after severe acquired brain injuries. The participating units (intensive care units, n = 2; intensive specialized rehabilitation units for postacute patients, n = 8; neurology departments, n = 2) were located in Europe (n = 10), North America (n = 1), and Asia (n = 1).

The study was conducted in 3 phases. In the first phase, a team of multidisciplinary investigators from all participating sites identified clinical and neurophysiologic measures that are feasible to collect in most clinical settings^{10,12,13,15,–,17} and potentially informative for the purposes of the study, as suggested by recent recommendation statements for prognosis in adults with prolonged DoC.⁴ All participating centers had well-established expertise in using the selected clinical and neurophysiologic tools. In the second phase, the skilled examiners, selected among the staff of each participating center for their long-standing experience in the evaluation of patients with DoC, participated in several conference calls to ensure the consistency of administration and scoring of the clinical scales, as well as of acquisition and analysis of the neurophysiologic tests selected for the study. In the third phase, all participating centers enrolled a convenience sample of patients with prolonged DoC from January to December 2017 and followed them clinically until 6 months after brain injury.

Study population

Inclusion criteria were (1) age 18–80 years; (2) clinical diagnosis of VS/UWS or MCS, according to standard diagnostic criteria^1,3; (3) anoxic, traumatic, or vascular (i.e., hemorrhagic or ischemic) etiology; (4) time postinjury ranging from 1 to 3 months. Exclusion criteria were (1) previous history of acquired brain injury or psychiatric or neurodegenerative diseases; (2) coexisting neoplasms, severe organ dysfunction, or unstable clinical condition (e.g., hemodynamic instability or severe respiratory failure).

Clinical assessment

At study entry, each center collected patient demographic data (e.g., age, sex) and information about medical history (e.g., etiology, time postinjury). Within 1 week from study entry, all patients were assessed by skilled investigators by means of repeated (at least 3 in a week) CRS-R evaluations¹⁵ in order to confirm the patients' clinical diagnosis¹⁸ and to determine the best CRS-R total score. Moreover, hospital staff assessed patients' behavioral response to nociceptive stimuli and functional disability level by means of the Nociception Coma Scale–Revised (NCS-R)¹⁹ and Disability Rating Scale (DRS)²⁰ total scores, respectively. Patients were clinically followed up at 6 months after brain injury by repeated CRS-R conducted by the same clinical staff from their hospital stay or, if after discharge, in an outpatient follow-up clinic, at home, or in chronic care facilities.

Neurophysiologic examination

Skilled neurophysiologists blind to the patient's clinical diagnosis recorded and analyzed multimodal neurophysiologic data including standard EEG, SEPs, and event-related potentials (ERPs) within 15 days from study entry, according to the shared protocols.

The standard EEG was recorded by 19 electrodes placed on the patients' scalp according to the international 10–20 system (Fp1, Fp2, F7, F8, F3, F4, C3, C4, T3, T4, P3, P4, T5, T6, O1, O2, Fz, Cz, Pz). EEG was recorded for a minimum of 35 minutes according to standard procedure of eye-closed recording during a waking rest with filter settings 0.53–70 Hz and notch filter on. Eyes closure for the analysis of predominant activity was obtained with forced eye closing by a cotton wool applied by paper patch in wakefulness condition (spontaneous eye opening). A morning EEG was acquired bedside after customary nursing procedures and (at least) 10 hours after administration of drugs acting on the CNS, such as myorelaxants and sedative drugs (e.g., benzodiazepines, neuroleptics) in order to optimize vigilance. Long-term antiepileptic treatment was continued, if clinically indicated.

EEG background activity was classified into 5 severity categories according to recently proposed criteria for patients with DoC²¹: (1) normal EEG activity, in presence of predominant posterior alpha rhythm and of the anterior-posterior gradient (APG), without focal or hemispheric slowing or epileptiform abnormalities; (2) mildly abnormal (MiA) EEG, characterized by predominant posterior theta activity (>20 µV), symmetric or not, with frequent (10%–49% of recording) posterior alpha rhythms; (3) moderately abnormal (MoA) EEG, characterized by predominant posterior theta activity (>20 µV), symmetric or not, poorly organized APG, even with rare (<1% of recording) or occasional (1%–9% of recording) posterior alpha rhythms; (4) diffuse slowing (DS), defined as an EEG background activity with predominant diffuse theta or theta/delta rhythms at amplitude >20 µV, without APG; (5) low voltage (LV) EEG, defined as a predominant EEG activity (theta or delta) <20 µV over most brain regions.

To analyze EEG reactivity to external stimulation, 5 kinds of stimuli were randomly administered (each stimulus 3 times) during EEG recording: (1) eye opening and (forced) eye closing; (2) tactile stimuli (wiping on the back of right and left forearm with cotton wool); (3) noxious stimulation (pressing fingernail beds on each hand); (4) acoustic stimulation (hand clapping); (5) intermittent photic stimulation (IPS) by 1, 3, 6, 9, 12, 15, 18, and 21 Hz flashes in 10-second trains presented through closed eyelids with 5-second interval between 2 trains. We analyzed 3-second EEG activity after stimulus onset to detect any clear and reproducible change in frequency or amplitude in cerebral activity that could be ascribed to multisensory stimulation.²² EEG reactivity to each kind of stimulus was defined as present if it was recorded in at least one out of 3 trials. The effect of IPS was considered as present if the physiologic photic driving response was elicited over posterior region. Not clearly detectable EEG responses to stimuli were considered as absent. If artifacts were present in more than 50% of EEG recording time, EEG acquisition was repeated within 1 week. When no EEG of sufficient quality could be obtained, EEG data were excluded from analysis.

SEPs were recorded during electrical stimulation of the patients' median nerve, delivered by means of a bipolar surface electrode placed on the patients' right and left wrists. Stimulation intensity was set to 4/3 of individual motor threshold. Each pulse lasted 0.2 ms and was delivered with a 3 Hz rate (time base 50 ms; bandwidth 5 Hz–3 KHz). Recording stainless steel needle electrodes or silver cups were placed at Erb's point (referred to contralateral Erb's point), spinous process Cv7 (referred to anterior neck), C3′ and C4′ (referred to each other and to Fz). At least 2 averages of 300 responses were repeated and superimposed. The presence or absence of N20/P25 cortical components was evaluated.²³

ERPs were obtained by means of a simple oddball paradigm using auditory stimulation.²⁴ Two randomly intermixed tones (not-target: 1,000 Hz, overall probability 80%; target: 2,000 Hz, overall probability 20%) were delivered binaurally through earphones at a rate of 1 tone (90 dB SPL, 50 ms plateau time, 2 ms rise and fall slope) every 1.1 seconds. Each patient was asked to keep a mental count of the rare (target) tones while ignoring the frequent (not-target) tones. This procedure was conducted independently of patients' language comprehension abilities on the behavioral examination.²⁵ EEG activity was continuously recorded (band pass filtered at 0.1–100 Hz; digitized at a sampling rate of 1,024 Hz) from at least 3 (Fz, Cz, and Pz) scalp electrodes (10–20 International System) all referred to linked earlobes, with the ground electrode positioned on Fpz. In the presence of a N100, detection of late cortical component P300 was evaluated by visual identification by a trained neurophysiologist based on latency (280–500 ms) and scalp topography central to parietal (80–150 ms).²⁴

Outcome definition

The primary outcome was a patient's clinical diagnosis at 6 months postinjury. Clinical diagnosis was established according to the standardized clinical criteria for VS/UWS,^1,2 MCS, and for emergence from MCS,³ confirmed by CRS-R.⁴ For the purpose of statistical analysis, we classified a patient as improved if there was an improvement in clinical diagnosis with respect to study entry (i.e., patient in MCS at baseline who emerged from MCS, and patient in VS/UWS at baseline who progressed to MCS or eventually regained full consciousness). We classified a patient as not improved if the clinical diagnosis worsened (i.e., patient in MCS at baseline who worsened to VS/UWS), or did not improve, or if the patient died.

Data management and statistical analyses

Referring staff from all participating sites entered, through a forced-choice format, patient data into a secure, password-protected, electronic database uploaded onto a shared folder. Data entry fields for each patient were referred to by an anonymous code that did not allow the identity of the patient to be traced, in order to protect confidentiality and to safeguard privacy. Regular checks on data entry were periodically carried out by the coordinator research group (A.E., S.F., A.M.). If any missing data or inconsistencies were detected, the center in question was encouraged to check it thoroughly. When completed, the database was removed from the online shared folder and kept on file safely stored by the principal investigator (A.E.).

Statistical analyses were performed by the coordinator researcher group (A.E., A.M., L.T.). Descriptive statistics were presented as means and 95% confidence intervals (CIs) for continuous variables and as frequencies for categorical variables. For between-group comparisons, continuous variables were submitted to Shapiro-Wilk test for normality. As scores of several variables departed significantly from the norm, we compared baseline findings as a function of diagnosis (VS/UWS vs MCS), of etiology, and of outcome (improved vs not improved) by nonparametric Mann-Whitney U test for continuous variables and χ² for categorical variables. The level of significance was adjusted for multiple comparisons using the false discovery rate correction; corrected p was set at 0.05 (2-tailed). Effect size for nonparametric statistics was expressed as η² (95% CI), and for χ² statistics as Cramer V (95% CI), computed by means of a bootstrapping procedure (1,000 samples; CI = BCa [bias corrected and accelerated]). The results were considered statistically significant when corrected p was <0.05 and 0 <η²/V (95% CI) < 1.

To investigate the relationships between predictors at baseline and outcome at 6 months postinjury, we applied 2 regression models. Prior to any regression analysis, we tested multicollinearity among the predictors by checking for the variance inflation factor (VIF) and tolerance. A VIF value less than 10 or tolerance (that is, the inverse of the VIF) less than 0.1 should not involve any issues with multicollinearity; in general, the lower the VIF value, the smaller the correlation among variables.²⁶

Then, we used a standard logistic regression model in which we included the selected variables as independent predictors and the outcome (improved = 1; not improved = 0) as the dependent variable. However, since standard regression models might suffer from overfitting, we also used a penalized regression model (least absolute shrinkage and selection operator [LASSO]) as an additional analysis. In the LASSO model, coefficients are biased from a penalty term in order to increase the predictive accuracy. The applied penalty could shrink to zero and then pull out from the final model factors whose original predictive value was low, thus including only the most important variables. The degree of shrinkage was determined by a 10-fold cross-validation. No p values are reported for the LASSO method, which does not provide tests of significance, whereas we reported model fit accuracy for both logistic and LASSO regression models. Moreover, model fit for the logistic regression was assessed by means of the Hosmer-Lemeshow goodness of fit test, for which a p value less than 0.05 suggests a poor fit. The Nagelkerke pseudo-R², which indicates the amount of variability accounted by all predictors in the model, was also computed. For evaluating the discrimination ability of the logistic regression model, the area under the curve (AUC) of the receiver operating characteristic curve was obtained. Accuracy of predictors of the logistic regression model was also assessed by a bootstrapping procedure (1,000 samples; CI = 95%; BCa).

All continuous variables were standardized prior to regression analyses. Moreover, as the LASSO model only deals with dichotomic categorical variables, and for the sake of consistency, in both the standard and the LASSO regressions multichotomic variables categorized as follows. For etiology, we contrasted traumatic vs nontraumatic (i.e., anoxic and vascular) brain injury, as often found in the literature.^1,6 As regards background EEG patterns and consistent with recent literature,²¹ we contrasted patterns with residual alpha rhythm (normal, MiA, MoA) vs patterns lacking alpha rhythm (DS, LV). Possible missing data were handled by listwise deletion. As this conservative procedure model limited the analysis to patients for whom all relevant data were available, we had to perform the same regression analyses twice, including or excluding ERP as an independent variable (ERPs were available for a subset of patients; see below). The level of significance was set at 0.05 (2-tailed).

In presenting the results, we focused on the logistic regression model and added the LASSO results as auxiliary; moreover, we also presented the main concordances and discordances between the logistic and the LASSO model. Statistical analyses were performed by IBM SPSS Statistics v. 25 (IBM Corp., Armonk, NY) and R version 3.5.3 (R Foundation for Statistical Computing, Vienna, Austria), including the R packages glmnet and caret for the LASSO model.

Standard protocol approvals and patient consents

The institutional review board of the coordinating center and of each center involved in the study reviewed and approved the same outline of the project, shared by all centers and translated in the respective languages. This study was then approved by the ethical standards committee of the coordinator center and of each center involved in the study, and performed according to the ethical standards of the Declaration of Helsinki (1964) and its later amendments. The surrogate decision-makers of the patients enrolled in the study provided their written informed consent after a semiformalized interview in which the purposes, procedures, and time points of the longitudinal study were clearly explained. The original forms were collected and stored at each participant center.

Data availability

Any anonymized data not published within the article will be shared upon request from any qualified investigator.