The Unsteady Return of Command-Following: Recovery and Instability of Bedside Motor Command-Following After Acute Brain Injury

Technical Summary: The Unsteady Return of Command-Following

Problem Statement
Following a verbal command is the primary bedside indicator marking the transition from unresponsiveness to overt recovery of consciousness after acute brain injury. While this milestone is critical for clinical decision-making regarding prognosis and goals of care, its behavior in the acute intensive care unit (ICU) setting is poorly characterized at scale. Specifically, three aspects remain unquantified: the timing of the first return of command-following across different injury phenotypes, the stability of this state once it appears (i.e., whether it is sustained or fluctuating), and the extent to which the examination is suppressed by sedation. Current understanding relies heavily on rehabilitation settings or single-phenotype studies that treat command-following as a predictor rather than a dynamic outcome, often failing to account for competing risks like death or withdrawal of care.

Methodology
This study employed a retrospective cohort design using the MIMIC-IV database (Beth Israel Deaconess Medical Center, 2008–2019), involving 13,900 adults with acute brain injury who had at least three charted motor examinations. The cohort was stratified into six phenotypes: acute ischemic stroke, traumatic brain injury (TBI), intracerebral hemorrhage, subarachnoid hemorrhage, subdural hemorrhage, and anoxic injury.

• Outcome Definition: The primary endpoint was the first occurrence of the Glasgow Coma Scale (GCS) motor response "Obeys Commands" (score 6).
• Competing Risks: The analysis utilized a competing-risks framework with three absorbing states: (1) first command-following, (2) in-hospital death or discharge to hospice, and (3) alive ICU discharge without ever following commands.
• Instability Metrics: Instability was quantified by measuring transient first recoveries (falling below threshold at the next exam) and the frequency of threshold crossings.
• Sedation Analysis: Examinations were tagged for concurrent sedation (hypnotics, opioids), neuromuscular blockade, and mechanical ventilation. Sedation depth was assessed using the Richmond Agitation-Sedation Scale (RASS) where available.
• Validation & Modeling: Findings were externally validated in the multicenter eICU Collaborative Research Database. Additionally, predictive probes (gradient boosting, deep temporal models, and a local large language model) were used to assess how much information about sustained recovery is encoded in the first 24 hours of bedside behavior.

Key Results

• Recovery Trajectory: Among 5,498 patients who were not following commands at admission, 43.5% recovered by 24 hours and 65.0% by 14 days. Recovery timing was significantly phenotype-dependent: ischemic stroke patients showed the highest recovery rates (77.2% by 14 days), while anoxic injury patients showed the lowest (36.9%).
• Instability: Recovery was frequently unsteady. Of the 3,573 patients who recovered at least once, 22.2% experienced a transient first recovery (failing the next exam), and 62.4% crossed the command-following threshold in and out at least once. Only 37.6% of recoveries were sustained transitions that held until discharge.
• Sedation Dependence: Non-following was strongly associated with sedation depth. When patients were deeply sedated (RASS -4 or -5), 98.0% of examinations showed no command-following, compared to 28.1% otherwise. The risk of non-following was highest with hypnotic sedatives and during invasive ventilation.
• Predictability: Machine learning models using only first-day features could discriminate sustained recovery with an AUC of 0.82, significantly better than a baseline of age and phenotype (0.64), though many patients remained undetermined on day one. Complex temporal models did not significantly outperform simple first-day summaries.
• External Validation: The eICU cohort replicated the primary findings, showing a 14-day incidence of 64.8% and a transient recovery rate of 22.7%, confirming the generalizability of the instability and sedation dependence across 208 hospitals.

Significance and Claims
The paper argues that overt bedside command-following is an early but inherently unstable and sedation-sensitive signal. The central claim is that a single charted observation of "Obeys Commands" is an unreliable index of the underlying neurological state. The study quantifies that the gap between "ever following commands" and "sustained command-following" is substantial, with frequent threshold fluctuations occurring even in the acute phase.

The authors conclude that clinical interpretation of this milestone must account for:

1. Phenotype-specific timing: Recovery rates vary more than twofold across injury types.
2. Fluctuation: The state is dynamic, with repeated threshold crossings being the norm rather than the exception.
3. Sedation: The examination is heavily suppressed by pharmacological agents, meaning an absent response often reflects drug effect rather than injury severity.

The study posits that these findings extend the known instability of consciousness from rehabilitation settings to the acute ICU, suggesting that prognosis should not be anchored on a single observation but rather on serial, sedation-aware assessments. The authors emphasize that while the overt signal is useful, it cannot detect covert consciousness or cognitive-motor dissociation, and future research should link these fluctuation patterns to validated functional outcomes and advanced neurophysiological markers.