Introduction The Coma Recovery Scale-Revised (CRS-R) is the gold standard diagnostic tool for assessing patients with disorders of consciousness (DOC) after severe acquired brain injury (Giacino, Kalmar and Whyte, 2004; Seel et al., 2010). Differential diagnosis of DOC includes the unresponsive wakefulness syndrome (UWS;(Laureys et al., 2010)), characterized by the recovery of eye-opening but no behavioral evidence of self or environmental awareness, and the minimally conscious state (MCS; (Giacino et al., 2002)) defined by clearly discernible but inconsistent behavioral signs of conscious awareness. The CRS-R assesses reflexes and cognitively mediated behavior in six domains, namely auditory (4 items), visual (5 items), motor (6 items), oromotor (3 items), communication (2 items) and arousal (3 items). Items in every subscale are hierarchically ordered (i.e. reflexive to cognitively-mediated behaviors; higher level behaviors correspond to higher level of neurologic functioning and ability to demonstrate lower-level behaviors or disappearance of pathological behaviors as sign of recovery) and can be used to infer the patient’s level of consciousness (La Porta et al., 2013; Gerrard, Zafonte and Giacino, 2014). Several studies on DOC investigating markers of consciousness, recovery and treatment used the CRS-R total score (i.e. addition of the highest scores reached for each subscale) as regressor in neuroimaging analyses (Bruno et al., 2012; Thibaut et al., 2012; Margetis et al., 2014; Bagnato et al., 2015). However, ignoring the hierarchy of the subscales in the CRS-R total score reduces the sensitivity for the diagnosis of MCS patients (i.e., 100% specificity for UWS but false negative diagnostic error of 22%, with a cut-off CRS-R total score of 10 (Bodien et al., 2016)). In addition, the ordinal nature of the CRS-R total score make it limited to use with parametric statistical tests (e.g., requiring normal distribution). A solution to this problem has been proposed by Sattin and colleagues (2015) who computed a CRS-R modified score (CRS-R MS1), by considering reflexes and cognitively mediated behaviors separately, reliably distinguishing between UWS and MCS patients. These authors also argue that the interpretation of the total CRS-R scores is limited due to “the underlying assumption that if a patient is able to show higher-level behaviors, he/she is also able to show lower-level responses”. Sattin et al. (2015) propose to account for the number of presented responses in every subscale (i.e., every items in a subscale should be assessed and scored). One major drawback to this approach is that according to the CRS-R guidelines, the assessor should start assessing the highest item and move to the next subscale once an item is scored, in line with the hierarchical organization of the scale. This means that, if the CRS-R is performed according to the guidelines (for which the CRS-R has been validated), the CRS-R modified score cannot be calculated. Even if assessing all items might be valid, it is unlikely to be done in many clinical and research settings as it would increase assessment time and fatigue the patient. We here propose to adapt the CRS-R MS1 by considering only the highest score reached on every subscale, respecting the CRS-R guidelines. Methods One-hundred twenty-four patients admitted to the University Hospital of Liège were assessed multiple times with the CRS-R, at least once including the assessment of all items. Patients for whom the CRS-R assessment including all items provided the same diagnosis as the patient’s final diagnosis were selected. The study was approved by the ethics committee of the University Hospital of Liège and the legal guardians of patients gave written informed consent for participation in the study, in accordance with the Declaration of Helsinki. The CRS-R total score and two CRS-R MS were calculated for every patient. The CRS-R MS combines scores for reflexes and cognitive behaviors of every CRS-R subscale which can be used to obtain the CSR-R MS from a transposition matrix. The CRS-R MS1 was calculated as previously described (Sattin et al., 2015), and the CRS-R MS2 only used the highest score in every subscale (i.e., assuming that lower items were successful). Statistics were performed in R (R Core team, 2012). We assessed group differences in age (two sample t-test), time since injury (two sample t-test) and etiology (χ2 test). Receiver Operating Characteristic were calculated to obtain the sensitivity and specificity at several classification thresholds (package pROC (Robin et al., 2011)). We calculated the correlation between the CRSR MS1 and CRSR MS2 using Pearson correlation, and both scores with the CRS-R total score using Spearman correlation. Finally, we used a Kolmogorov-Smirnoff test to evaluate whether CRSR MS1 and CRSR MS2 come from different distributions (i.e., if one approach provides additional information over the other). Results Eighty-five MCS patients (26 females; mean age 40.4 (SD±17.4) years old; 43 traumatic; mean time since injury 2.7 (SD±4.0) years) and 39 UWS patients (14 females; mean age 50.6 (SD±16.5) years old; 29 traumatic; mean time since injury 1.2 (SD±1.8) years) were included in the study. MCS patients were older (t(77.6)-3.15, p<0.002 95%CI[-16.7, -3.7]), were in a more chronic stage (t(121.9)=2.9, p = 0.005, 95%CI[974,427]), and suffered more often from a traumatic brain injury (χ2=6.8, p = 0.01) than UWS patients. The ROC analysis for both MS showed an AUC of 1 (cut-off:8.315, 100% specificity and sensitivity). The ROC analysis for the CRS-R total score showed an AUC of 0.94 (cut-off:9, sensitivity = 100%, specificity = 67%). A correlation was found between the CRSR total score and both the CRSR MS1 (r = 0.94, p < 0.0001, figure 1A) and CRSR MS2 (r = 0.96, p < 0.0001, figure 1B). The two CRS-R MS correlated (r = 0.96, p = 0.0001, figure 1C). CRSR MS1 and CRSR MS2 were drawn from the same distribution (D(124)= 0.13, p = 0.25). Discussion CRSR MS2 correlated strongly with the CRSR MS1, and perfectly discriminated UWS from MCS patients. As for accurate diagnosis the CRS-R should be repeated (preferably five times (Wannez et al., 2018)) short assessments are preferred, and possibly also reduce effects of fatigue. Second, the CRSR MS2 can be calculated with CRS-R assessments performed according to the CRS-R guidelines, facilitating its use in clinical environments, and in research settings where CRSR MS2 can be used pro- and retrospectively for research protocols. Furthermore, the results indicate that the two modified scores share the same distribution. This suggests that assessing all CRS-R items as proposed previously does not significantly contribute to the stratification of patients. The CRSR MS2 code is available via: Github A remaining limitation of the proposed score is that it does not allow to distinguish MCS minus (i.e. showing language independent signs of awareness, like visual pursuit) from MCS plus (i.e. showing language dependent signs of awareness) patients, or emergence from MCS. However, a clear consensus about the diagnostic criteria is needed before an updated modified score can be provided. In conclusion, the current analyses show that the calculation of the CRS-R modified score using the highest item in every subscale is valid for clinical diagnosis, and provides perspective for its use for research. Figure Figure 1. Correlation between the CRS-R total score and the CRS-R MS1 (1A), CRSR MS2 (1B), and between the two modified CRS-R scores (1C). MCS plus patients are here characterized by command following, intelligible verbalization and/or intentional communication. Acknowledgements This project has received funding from the University and University Hospital of Liege, the Belgian National Funds for Scientific Research (FRS-FNRS), the European Union’s Horizon 2020 Framework Programme for Research and Innovation under the Specific Grant Agreement No. 785907 (Human Brain Project SGA2) the Luminous project (EU-H2020-fetopenga686764), the Center-TBI project (FP7-HEALTH- 602150), the Public Utility Foundation ‘Université Européenne du Travail’, “Fondazione Europea di Ricerca Biomedica”, the Bial Foundation, the Mind Science Foundation and the European Commission, the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No 778234, European Space Agency (ESA) and the Belgian Federal Science Policy Office (BELSPO) for their support in the framework of the PRODEX Programme. CC is a post-doctoral Marie Sklodowska-Curie fellow (H2020-MSCA-IF-2016-ADOC-752686), and SL is research director at FRS-FNRS. We are highly grateful to the members of the Liège Coma Science Group for their assistance in clinical evaluations, and we thank all the patients and their families and the Neurology department of the University hospital of Liège.