Background: The Coma Recovery Scale-Revised (CRS-R) is the most recommended instrument to examine the neurobehavioral condition of individuals with disorders of consciousness (DOCs). Different studies have investigated the prognostic value of the information provided by the conventional administration of the scale, while other measures derived from the scale have been proposed to improve the prognosis of DOCs. However, the heterogeneity of the data used in the different studies prevents a reliable comparison of the identified predictors and measures.

Objectives: This study investigates which information derived from the CRS-R provides the most reliable prediction of both the neurobehavioral state and recovery of consciousness at the discharge of a long-term neurorehabilitation program.

Methods: The clinical records of 171 individuals with DOCs admitted to an inpatient neurorehabilitation program for a minimum of 3 months were used to implement machine learning classifiers that were trained to predict the neurobehavioral state and recovery of consciousness at discharge.

Severe acquired brain injury, understood as any damage to the brain that causes coma for over 24 hours, can lead to a complex clinical condition characterized by impaired consciousness, commonly referred to as a disorder of consciousness (DOC). The concept of consciousness is multifaceted and complex and arises from the presence of both arousal, i.e. vigilance and wakefulness, and awareness, i.e. perception of the environment and self. Consequently, depending on the presence and nature of the behavioral responses to multisensory stimuli, individuals with DOC are diagnosed as either in an unresponsive wakefulness syndrome (UWS) or vegetative state (VS) if they only show reflexes, or in a minimally conscious state (MCS) if they show intentional responses. Two further subgroups have been proposed within this latter group that allow to categorize individuals in an MCS+ or in an MCS- according to the presence or absence of higher-level behaviors, respectively. Finally, individuals who show functional communication or functional use of objects are considered as emerging from the MCS.

Diagnosis of DOCs, therefore, poses a clinical challenge, as it requires the accurate analysis of behavioral signs that can be weak or inconsistent. The Coma Recovery Scale-Revised (CRS-R) is the most recommended instrument worldwide for assessing the neurobehavioral condition of individuals with DOC and features multiple cross-cultural adaptations. The CRS-R investigates the presence of 23 neurobehavioral responses, grouped in 6 different subscales, which evaluate auditory, visual, motor, oromotor, communication, and arousal functions. For each subscale, the responses are hierarchically ordered and are evaluated from higher responses (cognitively-mediated responses) to lower responses (reflexes). The diagnostic utility of the scale was first analyzed in 2004, but it was not until 2010 that its interrater reliability, internal consistency, and prognostic or diagnostic validity supported its use for diagnosis. Additionally, although this has been only proven for traumatic etiology, the scale has demonstrated strong construct validity, with confirmed evidence of monotonicity, mutual independence, and invariant item ordering. In this regard, the hierarchy of the CRS-R has also shown a lack of invariance across relevant group factors including age, sex, etiology, enrollment facility, time since injury, and time between assessments. However, in spite that all these properties of the CRS-R contribute to reduce its diagnostic error in comparison to consensus-based tools, it is recommended that the diagnosis of DOCs is based on the clinical findings from five consecutive assessments and combined with imaging or electrophysiological-derived measures. Interestingly, some authors have proposed alternative indices and measures derived from the CRS-R, such as the CRS-R Modified Score and the CRS-R Index, pursuing an increase in the accuracy of the original version.

The total score in the CRS-R has been also identified as an important predictor of recovery of responsiveness in non-traumatic individuals with UWS in Class I studies. In the case of mixed cohorts, however, the current guidelines neither confirm nor refuse the prognostic value of the CRS-R due to insufficient evidence. However, more recent studies highlight the relevance of the information provided by the CRS-R in the prediction of the recovery of consciousness. Measures derived from the CRS-R have been also proposed to improve prognosis of DOCs. Arnaldi and colleagues introduced the CRS+, a weighted score based on the CRS-R to investigate the prognostic value of sleep patterns in the recovery of consciousness. More recently, the Consciousness Domain Index was proposed, an unsupervised machine learning clustering technique based on information from the CRS-R sub-scales to improve the prediction of recovery of consciousness.

However, although the information provided by the CRS-R might be essential to predict the clinical progress of individuals with DOCs and many attempts exist to find alternative measures that improve the predictive value of the original instrument, the heterogeneity of the data used in the different studies prevent a reliable comparison of the identified predictors and measures. Consequently, the aim of this study was to determine which information derived from the CRS-R provides the most reliable prediction of both the neurobehavioral state and recovery of consciousness at discharge of a long-term neurorehabilitation program.

Inclusion Criteria:

• diagnosis with (Unresponsive Wakefulness Syndrome) UWS or Minimal Consciousness State (MCS) due to either a vascular, anoxic or traumatic origin

Exclusion Criteria:

• individuals who did not attended the program for at least three months

Patients with Disorder Of Consciousness (DOC) who had attended an inpatient neurorehabilitation program between February 2004 and December 2021 in three facilities of the same hospital network were retrospectively extracted from their medical records.