Cognitive Motor Dissociation in Disorders of Consciousness: An Individual Participant Data Meta‐Analysis

ABSTRACT

Introduction

In patients with disorders of consciousness (DoC), behavioral assessment often underestimates awareness when motor output is absent or unreliable. Cognitive motor dissociation (CMD) refers to patients who appear unresponsive but demonstrate volitional brain responses during neurodiagnostic tests. We characterized clinical and methodological factors associated with CMD detection.

Methods

Studies published between January 2000 and May 2026 were identified through MEDLINE, Embase, and Scopus. Individual participant data were analyzed using mixed‐effects logistic regression. Study quality was assessed using the Quality Assessment of Diagnostic Accuracy Studies‐2.

Results

Fifty‐six studies with 1248 patients (mean age 45.5 years; 66.1% male; 44.5% unresponsive wakefulness syndrome [UWS]) were included. In unadjusted analyses, CMD detection was lower in anoxic (OR 0.43, 95% CI 0.29–0.66) and cerebrovascular (OR 0.59, 95% CI 0.37–0.98) compared to traumatic brain injury, and higher in minimally conscious state minus (MCS−) compared to UWS (OR 1.64, 95% CI 1.14–2.35). In multivariable models, anoxic (OR 0.35, 95% CI 0.23–0.55) and cerebrovascular (OR 0.51, 95% CI 0.32–0.84) etiologies remained independently associated with lower CMD detection, while MCS− remained associated with higher detection (OR 1.50, 95% CI 1.02–2.23). Time since injury, diagnostic modality and task paradigms were not independently associated with CMD detection. Overall, the studies showed a relatively low risk of bias.

Conclusions

CMD is common but varies by etiology and clinical state. Its detection appears to be independent of the time of injury, the task paradigm, and the diagnostic modality, highlighting the importance of standardized protocols and longitudinal studies.

1. Introduction

Coma and other disorders of consciousness (DoC) affect millions globally each year [1]. Correct classification of their level of consciousness is crucial for prognosis, rehabilitation, and clinical decision‐making but detecting signs of consciousness in the low‐ or unresponsive patient is challenging [2, 3].

The introduction of the Coma Recovery Scale–Revised (CRS‐R) for bedside assessment of consciousness after severe acquired brain injury substantially improved the clinical classification of patients with DoC [4, 5], but misdiagnosis remains common [4]. It is important to note that even a structured behavioral assessment can underestimate the level of consciousness if voluntary motor reaction is absent, inconsistent, or too subtle [5]. This limitation has become increasingly apparent over the past two decades, as a growing body of literature indicates that some behaviorally unresponsive patients retain evidence of command‐following and volitional brain activity that can be detected using neurodiagnostic methods [6].

To assess this state of cognitive motor dissociation (CMD), a range of neurodiagnostic modalities has been developed. So‐called active paradigms require patients to modulate brain activity in response to a command, and a positive response is considered specific for CMD [7, 8]. Active paradigms are used in conjunction with fMRI [9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26] and EEG [11, 12, 13, 14, 15, 22, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50] [e51–e53], which are the most widely studied technologies to detect CMD [3][e54]. Furthermore, bedside‐compatible tools are being explored as cost‐effective alternatives for detecting CMD. These include automated pupillometry [e55, e56], functional near‐infrared spectroscopy (fNIRS) [e57–e59], and electromyography (for clinically invisible muscle activity) to capture volitional responses to motor commands or cognitive tasks [e60–e64].

Despite the substantial clinical and ethical implications related to CMD [e65], our understanding about the influence of test modalities, research settings, underlying brain mechanisms, and time courses on CMD detection is limited [5] [e54]. We therefore performed a meta‐analysis of all published CMD studies with available individual patient level data to evaluate and compare CMD detection rates across DoC groups, brain injury types, time courses, cognitive task paradigms, and technological modalities.

We hypothesized that CMD would be more frequent in non‐anoxic compared to anoxic brain injuries; more frequent with increased levels of clinical alertness; and more frequent in chronic brain injuries than in acute ones; and that different diagnostic modalities and cognitive tasks would identify CMD with comparable frequency.

2. Methods

2.1. Objectives

This systematic review was conducted in accordance with the PRISMA 2020 guidelines [e66]. Using the PICO (Patients, Intervention, Comparison, Outcomes) framework, the primary objective was to review and meta‐analyze the frequency of CMD detection across different etiologies of severe acquired brain injury. Secondary objectives were to examine differences in CMD detection across DoCs, diagnostic modality, time since injury, and cognitive task paradigm.

2.2. Search Strategy

The search was done using MEDLINE (via PubMed), Embase, and Scopus for studies published between January 2000 and May 1, 2026. Two reviewers (P.P.L., M.H.) independently screened titles and abstracts. Full‐text review and data extraction were subsequently performed by F.W.A. and P.P.L. The full search strategy is provided in the Supporting Information.

2.3. Types of Studies: Inclusion and Exclusion Criteria

Included studies were original, peer‐reviewed studies investigating CMD in DoC patients. To reduce the influence of very small cohorts and convenience sampling, studies with fewer than five patients were excluded. Studies were required to report patient‐level data and to provide sufficient information on brain injury etiology and CMD assessment. In publications without single subject data, the corresponding author was contacted with a request to provide the necessary data. We excluded all non‐original publications, including reviews, meta‐analyses, editorials, commentaries, and conference papers, as well as studies published in languages other than English. Studies not specifically focusing on active paradigms were excluded, as were studies with fewer than five subjects.

2.4. Participants

The included patients were adults (≥ 18 years) treated in intensive care units, specialist hospital units (e.g., stroke units, neurology, or neurosurgery departments), step‐down units, rehabilitation facilities, or nursing homes, who met established criteria for DoC (Figure S1). Bedside diagnosis and classification of consciousness was denoted as stated by the authors. If patient‐specific data on consciousness was available, patients were grouped accordingly based on standardized clinical assessment using the CRS‐R, widely regarded as the most sensitive and specific instrument for differentiating patients into coma, unresponsive wakefulness syndrome (UWS), minimally conscious state (MCS; subdivided into MCS− and MCS+). While MCS− patients can have CMD and were included in our analysis, MCS+ patients cannot have CMD because they show clinical evidence of language processing [2], and these patients were therefore excluded.

2.5. Target Condition

The target condition was CMD, defined as evidence of preserved conscious or volitional cognitive processing detected by active neurodiagnostic paradigms despite absent or insufficient behavioral evidence of command following on bedside assessment.

2.6. Reference Standards

The reference standard was the bedside behavioral diagnosis of consciousness, preferably established using the CRS‐R. When CRS‐R data were unavailable, the diagnosis reported by the original study authors was used. Patients were classified as coma, UWS, and MCS−.

2.7. Index Tests

The index tests comprised active task‐based neurodiagnostic assessments used to detect CMD. Active paradigms were chosen, as they have greater specificity of CMD detection when compared to so‐called passive paradigms [5]. Paradigms were classified as active only when the authors found evidence that the patient actively followed and were consistent with the instructions provided. Passive instructions such as “pay attention” or “listen carefully” prior to auditory tests were not regarded as active paradigms. Included modalities were EEG, fMRI, pupillometry, fNIRS, EMG, and other validated neurodiagnostic tools.

fMRI‐based paradigms assessed task‐related blood‐oxygen‐level‐dependent (BOLD) signal changes during active cognitive tasks [e67, e68]. EEG‐based paradigms were included when they assessed higher‐order cognitive event‐related potentials, such as P300, N400, P600, or late positive complex [5]. Pupillometry‐based paradigms assessed task‐related changes in pupil diameter as a potential marker of preserved cognitive processing [e69]. fNIRS‐based paradigms assessed task‐related changes in cortical hemoglobin oxygenation as a proxy for neural activation during cognitive task performance [e58]. EMG‐based paradigms measured electromyographic signals, as an indicator of preserved motor function in patients with DoCs [e60].

Included cognitive task paradigms comprised of motor imagery tasks (e.g., imagining playing tennis), counting tasks based on auditory, visual, or tactile stimuli (e.g., counting sounds, light flashes, or tactile stimuli), motor command‐following tasks (e.g., “raise your arm”), gaze‐fixation tasks (e.g., directing gaze toward the correct answer), and task comprehension paradigms. Task comprehension paradigms were defined as active paradigms intended to assess whether patients understood and cognitively processed verbal instructions or task content, rather than engaging in mental imagery or motor command following (e.g., mental calculation).

2.8. Data Extraction

For each included study, the following variables were extracted: year of publication, diagnostic modality, time since injury, behavioral diagnosis at assessment, brain injury etiology, and cognitive task paradigm used to assess CMD. Brain injuries were categorized as acute (≤ 30 days from onset), subacute (31–89 days), or chronic (≥ 90 days). Brain injury etiology was categorized as TBI, anoxic brain injury, cerebrovascular event, meningitis/encephalitis, or other. Information on whether studies used serial behavioral assessments of consciousness was also extracted.

2.9. Statistical Analysis

The analysis was conducted using a one‐stage individual participant data meta‐analysis of CMD detection using mixed‐effects logistic regression models (generalized linear mixed models with binomial distribution and logit link). Results are reported as odds ratios (ORs) with 95% confidence intervals (CIs). p value of < 0.05 was considered statistically significant. The search was done using R version 4.5.2 (2025‐10‐31), with the packages readxl, dplyr, stringr, tidyr, and lme4.

Some studies used multiple modalities and/or cognitive tasks in the same patient. As individual patients could be evaluated with different modalities at different time points, each patient was assigned a unique identifier. This allowed summarization of CMD detection according to level of consciousness, brain injury etiology, and time since injury without duplicate counting. In secondary analyses of diagnostic modality and cognitive task paradigm, individual patients could contribute to more than one modality‐ or task‐specific estimate, but repeated recordings of the same modality or the same task were counted only once per patient. In studies with serial testing, patients were counted only once, that is, at the time CMD was identified.

UWS was used as the reference category for level of consciousness because it represented the clinical comparator for CMD analyses. TBI was used as the reference category for etiology as it was the most frequent etiology in the dataset. Acute injury was used as the reference category for time since injury because this timeframe was considered clinically most relevant for comparison. EEG was used as the reference category for diagnostic modality because it was the most frequently represented modality, and motor imagery was selected as the reference task because it represents the historically established paradigm in CMD research and was used in the first reported case of CMD [e70].

2.10. Assessment of Methodological Quality

Using the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS‐2) [e71], each study was assessed for their methodological quality. QUADAS‐2 comprises four domains: Participant selection, index test, reference standard, and flow of participants through the study and timing of the index tests and reference standard (flow and timing). Each domain is assessed for risk of bias, and the first three domains are also assessed for concerns regarding applicability. Risk of bias and concerns about applicability are judged as “low,” “high,” or “unclear.”

3. Results

The initial search yielded 9542 studies, and after title and abstract screening, 56 studies were included for data extraction in this review (Figure 1). Most of the studies were prospective observational cohort studies or case–control studies with patients being compared to healthy subjects. Eligible data was identified for 1248 unique patients (mean age 45.46 years, 18–91 years, median age 46; 66.14% males). Table 1 shows patient characteristics, and Table 2 and Figure 2 and provide group‐ and modality‐specific data.

FIGURE 1.

PRISMA flowchart. Of 9542 studies retrieved from the initial literature search, 56 had individual participant level data and were included. See Section 2 for details.

TABLE 1.

Patient‐level predictors of CMD (univariate analysis).

	Studies, n	CMD patients, n	Non‐CMD patients, n	Total patients, n	CMD%	Odds ratio (95% CI)	p
Type of injury (reference: TBI)
Traumatic brain injury	51	131	215	346	37.9	Reference	—
Anoxic brain injury	38	63	196	259	24.3	0.43 (0.29–0.66)	p < 0.0001
Cerebrovascular event	31	57	121	178	32.0	0.59 (0.37–0.98)	p = 0.02
Meningitis/encephalitis	15	6	13	19	31.6	0.79 (0.27–2.33)	p = 0.66
Other	12	12	44	56	21.4	0.56 (0.26–1.19)	p = 0.13
Nontraumatic UNS	6	17	31	48	35.4	1.04 (0.46–2.35)	p = 0.92
Traumatic/anoxic	4	1	5	6	16.7	0.38 (0.04–3.75)	p = 0.40
Level of consciousness (reference: UWS)
UWS	52	162	409	571	28.4	Reference	—
Coma	10	39	80	119	32.8	0.78 (0.42–1.46)	p = 0.44
MCS−	37	91	144	235	38.7	1.64 (1.14–2.35)	p = 0.008
Time since injury (reference: acute)
Acute	17	105	228	333	31.5	Reference	—
Subacute	28	48	103	151	31.8	1.05 (0.59–1.87)	p = 0.89
Chronic	43	131	290	421	31.1	1.00 (0.59–1.70)	p = 0.99

Note: Patient‐level analyses of 1248 individual patients across 56 studies. Analyses of brain injury type and time since injury were restricted to patients in coma, UWS, or MCS−. Significant results are shown in bold.

Abbreviations: CMD, cognitive motor dissociation; MCS, minimally conscious state; TBI, traumatic brain injury; UWS, unresponsive wakefulness syndrome.

TABLE 2.

CMD detection according to diagnostic modality and task paradigm (univariate analysis).

	Studies, n	CMD patients, n	Non‐CMD patients, n	Total patients, n	CMD%	Odds ratio (95% CI)	p
Diagnostic method used (reference: EEG)
EEG	34	157	353	510	30.8	Reference	—
fMRI	17	48	145	193	24.9	0.90 (0.51–1.59)	p = 0.73
fNIRS	3	25	40	65	38.5	1.50 (0.51–4.42)	p = 0.46
EMG	2	10	16	26	38.5	1.24 (0.32–4.83)	p = 0.75
Pupillometry	2	42	88	130	32.3	0.81 (0.26–2.49)	p = 0.71
Other	3	12	26	38	31.6	0.97 (0.31–3.10)	p = 0.97
Task paradigm (reference: motor imagery)
Motor imagery	18	65	208	273	23.8	Reference	—
Counting auditory stimuli	13	70	164	234	29.9	1.26 (0.64–2.50)	p = 0.50
Counting visual stimuli	4	39	55	94	41.5	1.76 (0.61–5.12)	p = 0.30
Counting tactile stimuli	4	18	27	45	40.0	1.48 (0.53–4.17)	p = 0.46
Task comprehension	5	48	118	166	28.9	0.69 (0.27–1.77)	p = 0.44
Motor commands	10	44	93	137	32.1	1.09 (0.50–2.37)	p = 0.84
Spatial navigation	5	11	33	44	28.0	1.16 (0.46–2.95)	p = 0.75
Gaze fixation	3	11	11	22	50.0	2.78 (0.74–10.48)	p = 0.13

Note: In analyses of diagnostic modality and task paradigm, individual patients could contribute to more than one modality‐ or task‐specific estimate; totals therefore do not represent unique patients and should not be summed across rows. Repeated recordings of the same modality or same task were counted only once per patient. Odds ratios (ORs), 95% confidence intervals (CIs), and p values were estimated using mixed‐effects logistic regression models.

Abbreviations: EEG, electroencephalography; EMG, electromyography; fMRI, functional magnetic resonance imaging; fNIRS, functional near‐infrared spectroscopy.

FIGURE 2.

Descriptive CMD detection rates across clinical and methodological factors. Bar plots showing the proportion of patients identified with cognitive motor dissociation (CMD) across categories of etiology, level of consciousness, time since injury, and diagnostic modality. Percentages represent the proportion of patients within each subgroup demonstrating CMD based on available assessments. Statistically significant differences from univariate analyses are shown as *p < 0.05; **p < 0.01; ***p < 0.001. CMD detection rates were highest in patients with traumatic brain injury and lowest in anoxic injury. Detection increased with higher behavioral levels of consciousness, with the lowest rates observed in unresponsive wakefulness syndrome (UWS) and higher rates in minimally conscious state (MCS− and MCS+). Rates were comparable across acute, subacute, and chronic time points. Among diagnostic modalities, higher detection rates were observed with functional neuroimaging and pupillometry compared to EEG‐based paradigms. Dagger (†) denotes groups in which overt behavioral evidence of consciousness is present (MCS+ and locked‐in syndrome [LIS]); detection of command‐following in these groups does not imply CMD but reflects task‐related responses in patients with clinically observable awareness.

3.1. CMD Detection Across Acquired Brain Injury Type

TBI was the most frequent type of brain injury (37.9%), followed by anoxic (28.4%) and cerebrovascular (19.52%) brain injuries; the remainder included other or unspecified etiologies. Using TBI as the reference category, CMD detection was less often in anoxic (24.3% vs. 37.9%; OR 0.43, 95% CI 0.29–0.66; p < 0.0001), and cerebrovascular (32.0% vs. 37.9%; OR 0.59, 95% CI 0.37–0.98; p = 0.02) brain injuries.

3.2. CMD Detection Across Levels of Consciousness

Among included patients, 44.5% were classified as UWS, 18.3% as MCS−, and 9.3% as coma, with the remainder DoC patients showing clinical signs of consciousness and being classified as MCS+, MCS, or locked‐in syndrome. Compared to UWS, the likelihood of CMD detection increased in patients with MCS− (38.7% vs. 28.4%; OR 1.64, 95% CI 1.14–2.35; p = 0.008). In contrast, CMD detection did not differ significantly between coma and UWS (32.8% vs. 28.4%; OR 0.78, 95% CI 0.42–1.46; p = 0.44).

3.3. CMD Detection by Time Since Injury

At the time of CMD assessment, 31.5% of patients had acute, 16.7% subacute, and 46.5% chronic brain injuries. Compared with the acute phase, CMD detection did not differ significantly in the subacute phase (31.8% vs. 31.5%; OR 1.05, 95% CI 0.59–1.87; p = 0.89) or the chronic phase (31.1% vs. 31.8%; OR 1.00, 95% CI 0.59–1.70; p = 0.99).

3.4. CMD Detection Across Diagnostic Modalities

EEG was the most frequently used modality, contributing 510 recordings across 34 studies. fMRI contributed 193 recordings across 17 studies. Pupillometry was used in 130 patient‐level assessments across two studies, and fNIRS was reported in three studies including 65 patients. Compared with EEG, CMD detection did not differ significantly with fMRI (24.9% vs. 30.8%; OR 0.90, 95% CI 0.51–1.59; p = 0.73), pupillometry (32.3% vs. 30.8%; OR 0.81, 95% CI 0.26–2.49; p = 0.71), EMG (38.5% vs. 30.8%; OR 1.24, 95% CI 0.32–4.83; p = 0.75), or fNIRS (38.5 vs. 30.8%; OR 1.50, 95% CI 0.51–4.42; p = 0.46).

3.5. CMD Detection Across Cognitive Tasks Paradigms

Motor imagery was the most frequently used cognitive task paradigm, reported in 18 studies, followed by auditory counting in 13 studies, motor command following in 10 studies, and task comprehension in 5 studies. Using motor imagery as the reference task paradigm, no significant differences in CMD detection were observed for counting auditory stimuli (29.9% vs. 23.8%; OR 1.26, 95% CI 0.64–2.50; p = 0.50) or counting visual stimuli (41.5% vs. 23.8%; OR 1.76, 95% CI 0.61–5.12; p = 0.30) or counting tactile stimuli (40.0% vs. 23.8%; OR 1.48, 95% CI 0.53–4.17; p = 0.46) or motor command following (32.1% vs. 23.8%; OR 1.09, 95% CI 0.50–2.37; p = 0.84) or task comprehension (28.9% vs. 23.8%; OR 0.69, 95% CI 0.27–1.77; p = 0.44) or spatial navigation (28.0% vs. 23.8%; OR 1.16, 95% CI 0.46–2.95; p = 0.75) or gaze fixation (50.0% vs. 23.8%; OR 2.78, 95% CI 0.74–10.48; p = 0.13).

3.6. Multivariable Analysis

In multivariable mixed‐effects logistic regression adjusting for etiology, level of consciousness, time since injury, diagnostic modality, and task paradigm, CMD detection remained significantly associated with etiology and behavioral diagnosis (Table 3, Figure 3). Compared to traumatic brain injury, CMD was less frequently detected in anoxic brain injury (OR 0.35, 95% CI 0.23–0.55; p < 0.001) and cerebrovascular event (OR 0.51, 95% CI 0.32–0.84; p = 0.007). CMD was more likely to be detected in patients with MCS− compared to UWS (OR 1.50, 95% CI 1.02–2.23; p = 0.04).

TABLE 3.

Multivariable mixed‐effects logistic regression for CMD detection.

Variables	Comparison (reference)	Odds ratio (95% CI)	p
Etiology	Anoxic vs. TBI	0.35 (0.23–0.55)	< 0.001
	Cerebrovascular event vs. TBI	0.51 (0.32–0.84)	0.007
	Others	NS
Level of consciousness	MCS− vs. UWS	1.50 (1.02–2.23)	0.04
	Coma vs. UWS	0.78 (0.39–1.57)	0.49

Note: See Table S1 for all model coefficients. Significant results are shown in bold.

FIGURE 3.

Factors associated with cognitive motor dissociation (CMD) detection. Forest plot showing odds ratios (ORs) with 95% confidence intervals (CIs) for clinical and methodological factors associated with CMD detection. Gray markers represent unadjusted (univariable) estimates, and green markers represent adjusted estimates from a multivariable mixed‐effects logistic regression model including etiology, level of consciousness, time since injury, diagnostic modality, and task paradigm. Models account for clustering at the study and patient level. The dashed vertical line indicates no association (OR = 1). Statistically significant differences from univariate analyses are shown as *p < 0.05; **p < 0.01; ***p < 0.001. Associations with etiology and level of consciousness remained significant after adjustment, whereas differences between diagnostic modalities were attenuated.

There was no significant association between time since injury and CMD detection or between the task paradigm and CMD detection. Similarly, no diagnostic modality demonstrated a statistically significant advantage after adjustment, although pupillometry and fNIRS showed a non‐significant trend toward higher detection rates. Model coefficients from all multivariable analyses are available online (Table S1).

3.7. QUADAS‐2 Results

Using QUADAS‐2, concerns were often evident in the domains of patient selection, index test, and flow and timing. Applicability concerns were limited overall, with the greatest concern observed for the index tests domain and flow and time domain, whereas patient selection and reference standard applicability concerns were limited (Figure 4).

FIGURE 4.

Assessment of study quality. Graphic overview of the systematic evaluation with respect to risk of bias and concerns of applicability of 56 original studies on CMD detection using QUADAS‐2, a revised tool for the Quality Assessment of Diagnostic Accuracy Studies. On average, risk of bias and concerns of applicability were low in three of four studies.

Some bias was noted for the majority of studies, concerning the selection of index test and the flow and timing of their evaluations, whereas few or no concerns were seen with respect to the reference standard and patient selection applicability.

4. Discussion

In this individual participant data meta‐analysis, we evaluated clinical and methodological factors associated with the detection of CMD across a large cohort of DoC patients. Using multivariable mixed‐effects models that accounted for both study‐level and patient‐level clustering, we found that CMD detection was primarily associated with the etiology of brain injury and clinical level of consciousness, whereas time since injury and diagnostic modality were not independently associated with CMD detection.

4.1. Etiology and Level of Consciousness as Primary Determinants

The most robust finding of this study is the strong association between etiology and CMD detection. In line with our hypothesis, CMD was significantly less likely in patients with anoxic and cerebrovascular etiologies compared to traumatic brain injury. This finding is consistent with poorer functional recovery and reduced network integrity after hypoxic–ischemic injury and supports the notion that preserved large‐scale cortical connectivity is a key substrate for covert command‐following capacity [e72–e80]. Reassuringly, the data were internally consistent, with CMD being rarest in patients with mixed injuries, that is, the combination of anoxic and traumatic injuries had additive detrimental effects.

In parallel, CMD was more frequently detected in patients with MCS− compared to UWS, independent of other factors. This aligns with current neurobiological models of consciousness, in which MCS− is characterized by partially preserved frontoparietal network function [e75, e81]. Importantly, this result also supports the construct validity of CMD as a marker of residual consciousness and, possibly, improved clinical outcome [7, 8] [e82].

4.2. Lack of Independent Effect of Time Since Injury

In contrast to our hypothesis and some prior cohort studies reporting higher CMD detection rates in patients with longer time since injury—likely reflecting survival and case‐mix effects rather than a true temporal relationship [6]—we did not observe an independent association after multivariable adjustment. Consistent with prospective ICU data demonstrating that CMD can already be detected early after severe acquired brain injury [7, 8], the present findings suggest that apparent differences between acute and chronic populations are largely explained by differences in etiology and clinical level of consciousness rather than time per se. The clinically important message is that CMD should be actively sought already early after injury in the ICU.

4.3. Diagnostic Modality and Task Paradigms: No Clear Superiority After Adjustment

A key finding of this study is the absence of a statistically significant difference in CMD detection across diagnostic modalities after multivariable adjustment, which indicates that these differences are largely explained by underlying patient characteristics. For example, modalities such as fMRI are logistically challenging and therefore more commonly applied in stable chronic patients [e79], whereas EEG is frequently used in acute and more severely impaired ICU populations [7, 8].

Although pupillometry [e55, e56] and fNIRS [e57–e59] showed a trend toward higher CMD detection, these effects did not reach statistical significance and were associated with wide confidence intervals, reflecting limited sample sizes. Taken together, these findings suggest that for the time being, no single modality can be considered superior in isolation and support the use of multimodal approaches [e83, e84] to maximize sensitivity for detecting CMD in severe acquired brain injury.

Among task paradigms, gaze‐based tasks were associated with higher CMD detection, although this finding should again be interpreted cautiously due to the small sample size. Nevertheless, it may be the case that differences between paradigms are genuine and reflect varying cognitive demands and neural substrates. Tasks such as motor imagery require sustained attention and intact premotor and frontoparietal networks, which may be particularly vulnerable in acute brain injury [e77]. In contrast, lower‐demand paradigms or those engaging alternative networks may be more sensitive across acute and chronic patient groups [5], highlighting the rationale for longitudinal, multimodal, and multiparadigm testing and suggesting the possibility that in the future individualized task selection may improve CMD detection.

4.4. Systematic Literature Quality Assessment

Using the QUADAS‐2, we found that, overall, the studies showed a relatively low risk of bias in the reference standard and patient selection domains. Some concerns remained in the index test and flow and timing domains. Concerns in the index test domain were mainly related to lack of blinding to the reference standard, meaning that interpretation of neurodiagnostic findings may not always have been independent of the clinical evaluation. This is particularly relevant in CMD research, where borderline findings may be susceptible to observer expectations [2]. In addition, the index test domain was inherently heterogeneous, encompassing multiple diagnostic modalities and cognitive task paradigms. A positive index test result was not defined consistently across studies, with differing thresholds for positivity and methodological inconsistency in CMD detection across modalities or paradigms. Bias could therefore arise from variations in interpretation, rather than true differences in diagnostic performance.

4.5. Methodological Considerations

A key strength of this study is the comprehensive inclusion of all published studies since January 2000 with available individual patient‐level data, allowing for a more granular and clinically meaningful analysis than aggregate meta‐analyses. This approach enabled the use of multivariable mixed‐effects modeling to account for patient‐ and study‐level heterogeneity, thereby providing more robust and generalizable estimates of factors associated with CMD detection.

Key limitations should be acknowledged. First, like all systematic reviews, our analysis is based on retrospective, heterogeneous data from published studies with varying methodologies, paradigms, and definitions of CMD, and without a universally accepted gold standard—introducing potential misclassification and between‐study variability [2]. Second, as we included only studies providing single subject data, several studies were excluded, in particular the largest cohort study of CMD [6]; however, single subjects' data on 170 of the 353 subjects presented in Bodien et al. [6] were extracted from other studies [12, 13, 16, 18, 21, 25]. Third, important confounders (e.g., sedation status, timing of assessment relative to injury, suspected severe language deficits, and clinical severity) were inconsistently reported and could not be fully adjusted for. Fourth, to detect as many studies as possible, our search included a range of alternative definitions of CMD. However, we only included studies explicitly investigating active command‐following paradigms. Finally, most data were cross‐sectional, limiting causal inference and insight into longitudinal recovery trajectories.

4.6. Clinical Implications

Our findings suggest that CMD detection is primarily determined by patient‐related factors rather than the specific diagnostic modality used. Clinically, this implies that negative results from a single modality should be interpreted with caution, particularly in patients with etiologies or clinical states associated with lower detection rates, e.g., patients with anoxic brain injury and UWS, respectively. Absence of evidence for CMD is not evidence of its absence [2, 3]. A multimodal assessment strategy, incorporating complementary techniques and paradigms together with longitudinal follow‐up, is likely necessary to maximize diagnostic sensitivity. As stated in the European guidelines, “a given patient should be diagnosed with the highest level of consciousness as revealed by any of the [available] approaches” [3].

5. Conclusions

In this large individual participant data meta‐analysis, CMD detection was independently associated with etiology and level of consciousness, but not with time since injury or diagnostic modality after adjustment. These findings emphasize the importance of patient selection and clinical context in interpreting CMD assessments and support the use of standardized, multimodal approaches to improve detection of CMD.

Author Contributions

Fawzi W. Abla: writing – original draft, methodology, formal analysis, writing – review and editing. Melika Hassani: methodology, writing – review and editing, formal analysis. Poul P. Laigaard: conceptualization, investigation, writing – original draft, methodology, writing – review and editing, formal analysis. Daniel Kondziella: conceptualization, methodology, formal analysis, supervision, writing – review and editing. Anna Kristina Eigenbrodt: methodology, writing – review and editing, formal analysis.

Funding

This work was supported by Rådet for Offerfonden (24‐610‐00245), Ehrenreichfonden, Lundbeck Foundation (R507‐2025‐286), and Independent Research Fund Denmark (5333‐00034B).

Ethics Statement

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1: Supporting Information.

Table S1: Multivariable mixed‐effects logistic regression for CMD detection.

Figure S1: Behavioral spectrum of consciousness and the potential region of cognitive motor dissociation (CMD). Overt behavioral evidence of consciousness increases from coma to UWS, MCS−, MCS+, and eMCS. CMD may occur in behaviorally non‐command‐following states. Clinical status may change over repeated examinations; therefore, the horizontal arrangement reflects typical behavioral presentation rather than a fixed patient trajectory.

Data Availability Statement

Raw data file “raw_data.xlsx” is available from the online Supporting Information files.