Contribution
A retrospective pooled analysis (“meta-analysis”) of every patient the Tübingen group had trained with a brain–computer interface since 1995: 35 patients, 29 with ALS, trained with slow-cortical-potential, sensorimotor rhythm or P300 BCIs. It shows that BCI performance does not decline continuously with paralysis: it is flat from minor impairment to the locked-in state and collapses in the completely locked-in state (CLIS). It then states a hypothesis to explain the collapse, extinction of goal-directed thinking through lost response–outcome contingency, and the observation that would refute it.
Key insight
The failure in CLIS is cast as operant extinction, not as loss of perception or of cognition. Evoked potentials show the CLIS brain still discriminates and processes heard speech; what goes is the voluntary, reward-driven production of a response. The mechanism the authors name is the patient losing “the perception of the contingency between the required physiological behaviour … and its consequences”, compounded by a reward whose value is blunted because artificial feeding and ventilation keep the body from ever being out of balance. On this view, the inbound channel is intact and the outbound loop is what has died.
Assumptions
- Defined by the absence of a channel. CLIS is “a state in which no communication and thus, no social interaction is possible”; LIS patients keep “residual voluntary control over few muscles, such as eye movement, eye blinks, or twitches with the lip”.
- Categorical outcome. Performance is coded into four ordinal levels (chance, above chance, criterion ≥ 70%, independence), not correct response rates, and the best level across BCI types is taken for each patient.
- Heterogeneous training. Session counts, BCI versions and interruptions differ across patients; number of sessions is not modelled. Eleven patients used two or three BCI types.
- Cognition read from passive ERPs. Cognitive function in CLIS is inferred from oddball P300 and semantic N400 responses to auditory stimuli, since the patients cannot respond.
- The curarized-rat analogy. The hypothesis rests on Dworkin and Miller’s failure (1986) to replicate operant control of autonomic responses in chronically curarized rats, and on the claim that such rats and CLIS patients “share many similarities”.
Key results
- Sample. 35 patients: 29 ALS, 1 chronic Guillain–Barré, 1 muscular dystrophy, 1 cerebral paresis, 1 hypoxic brain damage after cardiac arrest, 2 stroke. Impairment: 2 minor, 8 moderate, 8 major, 10 LIS, 7 CLIS (Table 2; Table 4 has N = 34 because one patient’s data were lost). SCP-BCI: 25 patients; SMR-BCI: 5; P300-BCI: 11.
- Mean correct responses. SCP 66% (N = 23), SMR 71% (N = 5), P300 66% (N = 11) (Table 3).
- Impairment and performance. Kruskal–Wallis χ² = 13.4, df = 4, p < .01; η² = .40 (F4/29 = 4.78, p < .01). Without the CLIS group: χ² = 3.5, df = 3, p = .32; η² = .12 (F3/23 = 1.0, p = .41). Adding other groups’ patients (Table 5), still without CLIS: χ² = 3.9, df = 3, p = .27; η² = .07.
- The CLIS group. “in the CLIS group, 5 of 7 patients performed at chance level and only two achieved above chance level control, which was with 59 and 62%, far below criterion level control.” They were in CLIS from the start: “we were unable to restore basic communication (yes/no) in patients who were in the complete locked-in state at the beginning of training”.
- Ventilation is not the cause. Among non-CLIS patients, ventilated and non-ventilated did not differ (U-test Z = −.87, p = .43).
- Cognition in CLIS. “all patients had the N1-P2 complex indicating functionality of primary auditory cortex”; in the oddball paradigm “all but one patients’ response to the target stimuli was significantly different from that to the standard stimuli”; “Responses to semantic mismatch were absent in two CLIS patients”. In the Discussion: “With one exception, all CLIS patients of the sample reported here had ERP-responses to one or more of the complex cognitive tasks”, which “do not prove or disprove normal information processing in CLIS but suggest some intact ‘processing modules’ in most ALS patients with CLIS despite a reduced general arousal”.
- The hypothesis, in the authors’ words (Discussion): “Chronically curarized rats and people with longer time periods in CLIS may loose the perception of the contingency between the required physiological behaviour (SMR decrease, changes of the SCP amplitude or heart rate increase) and its consequences (brain stimulation reward in the curarized rat and cursor control or letter selection in the patient).” Extinction occurs “because experimental trials with a clear contingency between the physiological response (e.g., heart rate) and the rewards are too rare in a 24 hour day … the remaining 22 hours without any contingency provide enough time for extinction. Together with the proposed blunting of the reward value due to the lack of homeostatic imbalances, acquisition of voluntary control of the desired physiological response becomes impossible”.
- The prevention it predicts. “We propose that learning BCI-control or any other contingency before onset of CLIS, which can be transferred from LIS to CLIS should prevent extinction in CLIS.”
- The conclusion and its refutation condition. “we may conclude that passive sensory information processing is intact in CLIS even at the most complex semantic processing levels. It is the complete lack of motor control and feedback which might be responsible for the cessation of voluntary cognitive activity, goal directed thinking and imagery supporting a ‘motor theory of thinking’ already discussed by William James … A single CLIS patient who learns to communicate with a BCI or any other communication method will disprove our hypothesis.”
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | BCI performance falls with physical impairment only because of the CLIS group | moderate | Kruskal–Wallis and η² with and without CLIS; small groups, ordinal coding, unequal training |
| C2 | None of seven patients who entered BCI training already in CLIS reached yes/no communication | strong (as a report) | Table 1 and 4; two reached 59% and 62% |
| C3 | Passive auditory and some semantic processing survive in most CLIS patients | moderate | ERP battery; one patient without, two without N400; “do not prove or disprove normal information processing” |
| C4 | Loss of perceived response–outcome contingency, with blunted reward value, extinguishes voluntary goal-directed thinking in CLIS | weak | analogy with curarized rats (whose result failed to replicate); stated as a hypothesis (“might be responsible”) |
| C5 | Contingency learned before CLIS and transferred into it should prevent extinction | weak | prediction from C4; no data |
| C6 | Ventilation does not explain impaired BCI learning | weak | non-significant U-test excluding the CLIS group |
Concepts
- LIS / CLIS. LIS: almost completely paralysed “with residual voluntary control over few muscles”; CLIS: “all motor control is lost”, equivalently “no communication and thus, no social interaction is possible”. CLIS can also arise from disorders of consciousness; one of the seven (PR, hypoxia after heart attack) “was probably in such an altered state of consciousness but has never been diagnosed as such”.
- Criterion level control. ≥ 70% correct, the accuracy the authors take as needed for binary communication.
- Extinction. In the learning-theory sense: a learned response fades when it is no longer followed by its consequence.
- Motor theory of thinking. Attributed to James (1890): voluntary thought depends on motor control and its feedback.
Connections
It pools the group’s earlier patient reports (Kübler et al. 1998, 1999, 2004, 2005; Birbaumer et al. 1999; Neumann and Birbaumer 2003; Nijboer et al.; Hill et al. 2006) and compares them with other groups’ (Wolpaw, Pfurtscheller, Neuper, Sellers and Donchin, Piccione, Hoffmann). The extinction hypothesis is Birbaumer’s (2006, “Breaking the silence”), resting on Miller (1969) and Dworkin and Miller (1986). The ERP battery is Kotchoubey et al. (2002–2005). Chaudhary et al. (LIT-1200, read in NOTE-951), with Birbaumer as senior author, cite it as the hypothesis their patient tests.
Bearing on the record
- QUESTION-121: which of the two things does the hypothesis concern? Contingency between intention and outcome, as the question says, and the paper is explicit that the other one survives. What is lost is “the perception of the contingency between the required physiological behaviour … and its consequences”; what is kept is “passive sensory information processing … even at the most complex semantic processing levels”. The extinction mechanism is operant: too few trials in which a response is followed by its consequence, and a reward that no longer restores a homeostatic imbalance. So the question’s separation is the authors’ own, and its inference holds: if the hypothesis is true, what extinguishes goal-directed thinking is not a shortage of evidence the loop did not select. One qualification: the conclusion names “the complete lack of motor control and feedback”, and “feedback” there plausibly includes the somatic and proprioceptive feedback of movement (the paper does not say). That is still feedback contingent on the patient’s own output, not unselected input.
- CLAIM-339. Its quotation is accurate, but it gives the hypothesis as “the judgement loop, on this view, runs down when action is gone”. The paper says less and something more specific. Less: it is “might be responsible”, from seven patients, by analogy with a rat result that itself failed to replicate. More specific: what runs down is voluntary, reward-driven response production (“voluntary cognitive activity, goal directed thinking and imagery”), not perception or semantic processing. The paper therefore supports the claim’s own concession (“What it does not say”, second bullet) more than its empirical paragraph: it is evidence that action carries the contingency that sustains goal-directed activity, not evidence that action is the route by which the loop receives unselected evidence.
- CLAIM-341, defeater (b). The paper offers a candidate for (b): lucid goal-directed judgement failing when every contingency between intention and outcome is removed. But it does not establish it. The patients’ consciousness was not shown to be otherwise preserved (one was probably in a disorder of consciousness, two lacked the N400, arousal was “reduced”, and “It is often difficult to decide whether the patient is awake or in sleep stage 1 or 2”), and the failure was a failure to acquire a new operant skill, not a demonstrated loss of judgement. The thesis’s distinction also bites: what the paper says is extinguished is production of a response that acts on something (a cursor, a letter); whether judgement about what is heard survives is exactly what the paradigm cannot see. So the paper leaves (b) untested, which is the CASE’s asymmetry.
- CLAIM-342. No direct bearing. The paper’s scoring of patient performance is fixed (chance thresholds computed in advance), but that is the observers’ anchor, not the patient’s.
- CASE-107.
- Hearing intact; vision not tested. All CLIS patients had the N1-P2 auditory response, so primary auditory cortex was working. All stimuli were auditory; vision is not assessed. The SCP and SMR BCIs gave visual cursor feedback, which matters for CLIS patients whose eyes may be closed or dry; the paper does not discuss this.
- “None of seven CLIS patients regained yes/no communication”. Correct in substance, but “regained” should read that they entered training already in CLIS. None had learned a BCI in LIS and carried it over, which is precisely the route the authors predict would prevent extinction. Chaudhary et al.’s patient took that route (he used an EOG-based BCI with the same speller before CLIS), so his success fits the hypothesis’s own proviso rather than simply refuting it.
- “Not every contingency goes.” The paper sharpens this against the CASE. Its second mechanism is that artificial feeding and ventilation keep bodily functions “artificially constant”, so “imbalances of the equilibrium do not occur” and reward loses its homeostatic value. The visceral regulation the CASE counts as surviving paralysis is, in the ventilated and tube-fed CLIS patient, partly clamped from outside. That bears on QUESTION-121’s candidate 2 (the body) as well.
- For the anthology. No instruction for machine-learning practice.
Limitations
- Seven CLIS patients, heterogeneous in cause (one hypoxic, one GBS), BCI type and training length; several were trained only briefly (“days”, “one session”).
- The hypothesis is an analogy; the authors give no direct measure of perceived contingency or of goal-directed thinking in any patient.
- Ordinal coding of performance and multiple entries per patient; the authors “refrain from inferential statistics” across BCI types.
- The count of other groups’ patients is given as 27 in the Results and Table 5 and as 23 in the Discussion.
- Visual feedback BCIs were used with patients whose vision was not assessed.
Open questions
- Does a contingency learned in LIS survive the transition into CLIS? The paper predicts yes; Chaudhary et al. is the closest test.
- Is the CLIS failure extinction, arousal (sleep-like EEG slowing), cognitive decline or a failure of the feedback channel? The paper cannot separate them.
- For the record: is judgement about heard material, which needs no response contingency, preserved in CLIS? Only a paradigm whose readout does not depend on operant learning could say.