Selective Impairment of Motor Recovery from Typing Errors in Parkinson's Disease: A Survival Analysis

Navin Bondade

REVIEW 3 major objections 5 minor 13 references

Selective Impairment of Motor Recovery from Typing Errors in Parkinson's Disease: A Survival Analysis

T0 review · 3 major / 5 minor · reviewed 2026-08-02 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Parkinson's disease severity is linked specifically to slower motor recovery after a self-corrected typing error, while the keystroke instability that precedes the error remains unchanged—a dissociation detected with a continuous survival m

desk verdict A genuinely novel keystroke analysis with a solid core finding, but the 'genuine dissociation' claim outruns the evidence because the recovery measure's error-specificity is untested and the pre-error null is too weak to call intact. read the letter →

arxiv 2607.24796 v1 pith:BFUXME3V submitted 2026-06-30 q-bio.NC cs.AIcs.LG

Navin Bondade This is my paper

classification q-bio.NCcs.AIcs.LG

keywords Parkinson'sdiseasekeystrokedynamicserrormonitoringpost-errorrecoveryacceleratedfailuretimesurvivalanalysisdigitalbiomarkermotorcontrol

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper's central claim is that Parkinson's disease severity specifically harms the motor recovery that follows a self-corrected typing error, while leaving the detection of the error itself intact. Using the pause after a backspace as a proxy for motor recovery, the authors find that longer recovery pauses track higher disease severity, whereas the instability in keystroke timing immediately before the error does not. This dissociation is statistically robust: the two measures are uncorrelated, only the recovery measure survives a joint model, and the effect replicates across independent sub-cohorts and an external finger-tapping test. The paper also reports a methodological caution: modeling recovery as a discretized survival event destroys the signal, while a continuous accelerated failure time model preserves it, suggesting that the magnitude of the pause, not just whether recovery occurred, carries the information. If correct, this means ordinary typing can reveal a specific, circuit-relevant motor deficit in Parkinson's disease without any neurophysiological equipment.

What carries the argument

The central objects are backspace events as naturally occurring, self-corrected errors in free typing. Pre-error instability is computed as the mean absolute deviation of the three keystroke flight times before the backspace from the subject's median baseline, expressed as an excess over the same deviation in 500 random non-error windows per subject. Post-error recovery is the flight time of the first keystroke after the backspace, computed within a single session. The analytical engine is a log-normal accelerated failure time (AFT) survival model, which treats the continuous, right-skewed duration directly without a recovery threshold; a discretized time-to-event analog is shown to destroy

What would settle it

A direct test would compare the post-backspace pause against the flight time after randomly selected non-error keystrokes at matched positions in the same sessions; if the random-pause measure shows an equally strong association with disease severity in the same AFT model, then the 'post-error recovery' effect would be shown to reflect general motor slowing rather than error-specific restart.

Watch

Extended reading notes

Core claim

After a backspace-corrected typing error, the pause before the next keystroke lengthens with Parkinson's disease severity (UPDRS-III), while the keystroke instability that builds up before the error does not. The paper establishes this dissociation with a log-normal accelerated failure time (AFT) model on 1,570 error-correction events from 27 PD patients; the effect survives controlling for raw typing speed and replicates in both sub-cohorts, with an alternating finger-tapping test providing independent corroboration. A methodological lesson is that a discrete, Kaplan-Meier-style thresholding of recovery time wiped out the association at every tolerance tried, while the continuous AFT model

Load-bearing premise

The load-bearing premise is that the flight time of the keystroke immediately following a backspace measures post-error motor recovery specifically, and not just bradykinesia, cognitive pausing, or other task demands; the paper's control for raw typing speed partially addresses this but does not separate those alternatives.

Editorial extensions

If this is right

Passive keystroke monitoring can separate error detection from error-recovery stages of motor control in Parkinson's disease, without EEG, intracranial recording, or structured clinical tests.
The continuous AFT modeling approach, rather than discrete survival thresholding, is necessary to preserve the signal in reaction-time-like behavioral data; the paper demonstrates this explicitly with a sweep of tolerances.
The post-error recovery pause is a severity-linked digital motor sign that is statistically independent of raw typing speed, suggesting it captures a distinct aspect of motor dysfunction.
The behavioral dissociation is consistent with (though not proof of) the idea that Parkinson's disease spares cortical error detection while impairing subthalamic-nucleus-linked post-error motor adjustment.
If replicated in external cohorts, the post-error pause could serve as an at-home, unobtrusive longitudinal marker of motor severity.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The same AFT-based dissociation could be tested in other movement disorders or in aging populations to see if post-error recovery impairment is specific to Parkinson's or reflects a general motor restart deficit.
The paper's null result on pre-error instability should be read cautiously: the proxy is crude (deviation from a median baseline), and a more sensitive measure using the exact error type, the surrounding context, or higher-resolution timing might reveal subtle error-monitoring changes that this coarse metric misses.
A natural next experiment is to record typing during on/off dopaminergic medication or during deep brain stimulation to see whether the post-error pause is modulated acutely by treatment, which would strengthen the link to subthalamic circuitry.
Because the AFT coefficient is distribution-dependent, comparing log-normal, Weibull, and semi-parametric continuous models on independent data would test how much of the effect relies on the chosen error distribution.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague, here's my take on arXiv:2607.24796.

The paper is worth a read. It's the first to treat backspace events as error-correction episodes in PD keystroke data, and it finds a robust association between the flight time after a backspace and UPDRS-III severity. The survival approach is sensible: they tried a discretized time-to-event version, saw it killed the signal, and switched to continuous AFT. That kind of transparency is rare and should be credited.

What's genuinely good: the subject-level permutation (p<0.0033) and bootstrap CI excluding zero give real support, and the finding replicates in two sub-cohorts. The confound check with raw typing speed is clean. The afTap cross-check is a nice independent validation. The joint model with pre-error instability is appropriate.

The soft spots are real but not fatal. The main issue is whether the post-backspace flight time specifically measures error recovery, or just general slowness/pausing around any disruption. Controlling for overall typing speed is a start, but it doesn't rule out event-level confounds. The paper doesn't check whether the same AFT association appears after non-error disruptions like long pauses or shift presses. Without that, the 'error-specific' interpretation is not secure. The pre-error null is also weak: n=27 subjects cannot establish 'intact' error monitoring; an equivalence test is needed. And the event-level p-values ignore within-subject correlation, though the subject-level tests are the ones that matter. No code is provided, which makes the exact pipeline hard to reproduce.

The dissociation claim—post-error recovery impaired, pre-error monitoring not—overstates the evidence. But the core positive finding is well-supported. With code, an equivalence test, and a control analysis for non-error disruptions, this could be a solid contribution.

I'd send it to peer review. It's the kind of paper that will generate useful discussion, and the methodological honesty is a point in its favor.

Referee Report

3 major / 5 minor

Summary. The paper analyzes backspace-triggered error-correction episodes in the MIT-CSXPD keystroke dataset (57 subjects, 27 PD with UPDRS-III) to test whether Parkinson's disease severity differentially affects pre-error keystroke instability (a proxy for error monitoring) and post-error recovery time (the flight time of the keystroke immediately after a backspace). The authors report a null correlation between pre-error instability excess and UPDRS (r=0.164, p=0.413), and a strong positive association between post-error recovery duration and UPDRS under a log-normal accelerated failure time (AFT) model (event-level p<1e-9 in each sub-cohort, subject-level permutation p<0.0033, bootstrap CI excluding zero). They argue that the two measures are statistically independent and that only the post-error measure survives a joint model, thereby claiming a genuine dissociation. The manuscript also documents an initial discretized survival formulation that destroyed the signal, motivating the continuous AFT approach.

Significance. If the dissociation holds, this would be a notable, low-cost digital behavioral marker that distinguishes error-monitoring stages from post-error motor recovery stages in PD, consistent with electrophysiological dissociations reported by Stemmer et al. and Siegert et al. The paper's strengths include transparent reporting of a failed analytic attempt, per-cohort replication, subject-level permutation and bootstrap checks, and explicit confound control for raw typing speed. The central inference is, however, not yet fully secure: the principal outcome's error-specificity is not established, the event-level p-values ignore within-subject correlation, and the pre-error null is not equivalence-tested. With appropriate additional analyses, the claims could be placed on much firmer footing.

major comments (3)

[Section 4.4, Figure 1, Table 2] The outcome, flight time of the keystroke immediately following a backspace, is interpreted as specifically measuring post-error motor recovery. However, no event-level control demonstrates error-specificity. An effect of UPDRS on flight time following any non-error disruption (e.g., a long pause, punctuation, or a Shift press) would produce the same results even if recovery from errors per se is unaffected. The only confound reported is subject-level raw typing speed, which cannot adjust for event-level confounds such as key identity, hand/finger, or surrounding context. Please add control events matched for position and context, or include event-level covariates, or at minimum show that the association disappears for non-error flight times.
[Section 5.2, Table 1; Section 5.3, Table 2] The headline p-values (p<1e-9) are computed at the event level with n=1570 events from 27 subjects, but the AFT models do not cluster by subject. Within-subject correlation will deflate standard errors and inflate significance. The subject-level permutation (p<0.0033) and bootstrap partially address this, but the 300-permutation resolution is coarse. Report the permutation p-value with more permutations or with an exact formulation, and quote the subject-level/bootstrap CI as the primary inferential result rather than the event-level p-values. The abstract and conclusion should be reworded accordingly.
[Sections 4.2, 5.1, 7] The conclusion that error monitoring is 'intact' or that there is a 'genuine dissociation' rests largely on the null pre-error correlation (r=0.164, p=0.413 with n=27). A null result with n=27 cannot establish absence of an association without an equivalence bound or a power analysis. Report the confidence interval for r and specify the smallest effect the study could detect; otherwise the dissociation claim overstates the evidence. A pre-registered or clearly bounded equivalence test would make the claim defensible.

minor comments (5)

[Section 4.2] The pre-error instability sanity check (pooled Wilcoxon p<0.0001) is also event-level and ignores subject clustering. As a sanity check this is acceptable, but it should be labeled as a pooled event-level comparison.
[Table 2] The confound row reports p=5.2e-11 for UPDRS after adding typing speed, but does not give the UPDRS coefficient or CI. Adding these would help readers assess effect size.
[Figure 1] The caption says 'at the subject level' yet the main test in panel B is an event-level AFT model. Clarify what is plotted and what the statistical contrasts actually use.
[References] Goldberger et al. (2000) appears in the reference list but is not cited in the text; either cite it where the dataset provenance is described or remove it.
[Section 4.3] The choice of a ±50% tolerance for the discretized recovery definition is presented without justification; consider a brief rationale or a sensitivity statement.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: recovery time is a measured keystroke feature and UPDRS severity is an external clinical score; the AFT coefficient is estimated after outcome definition, not used to construct the outcome.

full rationale

The paper's derivation chain is empirical rather than definitional. The post-error recovery outcome is the flight time of the keystroke immediately following a backspace (Section 4.4), a directly measured quantity that is not defined in terms of UPDRS-III or of any fitted parameter. UPDRS-III is an independently administered clinical severity score, and the AFT coefficient is estimated from the joint distribution of these two variables; the outcome is not a function of the regression fit, so no fitted input is renamed as a prediction. The pre-error instability measure is likewise constructed from keystroke deviations against the subject's own baseline, without using disease severity, and its null result is then compared against the post-error finding. The 'genuine dissociation' claim is based on the empirical uncorrelatedness of the two measures and a joint model, which is a statistical comparison, not a circular identification. The only self-citations (Bondade, 2026) are used for dataset/literature context and a companion analysis of missingness; they do not supply the post-error recovery result or any uniqueness/forbidden-alternative argument, so they are not load-bearing. External citations to Stemmer et al. and Siegert et al. provide mechanistic interpretation after the fact, not premises that define the measures. The reported discretization failure (Section 4.3) is transparently a methodological negative result, not evidence that the final signal was manufactured by construction. The reviewer-identified weaknesses—whether the post-backspace flight time is specifically motor recovery rather than general bradykinesia, and the absence of cluster-robust event-level inference—are validity and evidence-strength concerns, not circularity: they question what the measured quantity means and how confident the inference is, but they do not show that the outcome equals an input by definition. Accordingly, no circular step can be quoted and exhibited, and the appropriate score is 0.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

The analysis depends on several interpretive assumptions about what keystroke timings measure, plus a standard statistical assumption about UPDRS-III. No new entities are introduced, and no ad hoc numeric parameters were tuned to achieve significance—the AFT coefficients are estimated as the inferential quantity, and the tolerance sweep was reported as null.

assumptions (6)

domain assumption Backspace events are self-detected errors and can be treated as natural error-correction episodes.
Section 3/4.1 assumes every backspace marks a self-detected error; in practice backspace may be used for editing or other reasons, but the entire analysis rests on this interpretation.
domain assumption The flight time of the keystroke immediately following a backspace measures post-error motor recovery.
Outcome definition in Section 4.4; not directly validated as a recovery measure beyond its correlation with severity.
domain assumption The mean absolute deviation from a subject's baseline across the 3 pre-error keystrokes captures error-monitoring-related destabilization.
Pre-error instability measure in Section 4.2; they sanity-check that pre-error windows differ from random windows, but not that this specifically reflects error monitoring.
domain assumption UPDRS-III can be treated as a numeric severity covariate with linear effect on log-time.
Standard AFT assumption in Section 4.4; not tested for linearity; small n.
domain assumption Unmeasured confounders (age, gender, medication status, session effects) do not drive the association.
The paper notes no age/gender fields (Section 6); medication status not mentioned; this assumption is load-bearing because the outcome may relate to overall motor slowing.
domain assumption Event-level observations are independent for the AFT likelihood (or the subject-level permutation test mitigates the violation).
The headline p-values are event-level; within-subject clustering is handled only by the subject-level permutation/bootstrap, not in the AFT itself.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Selective Impairment of Motor Recovery from Typing Errors in Parkinson's Disease: A Survival Analysis." pith.science (2026). https://pith.science/paper/BFUXME3V

@misc{pith2026260724796,
  author       = {Pith},
  title        = {Pith review of: Selective Impairment of Motor Recovery from Typing Errors in Parkinson's Disease: A Survival Analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BFUXME3V}},
  note         = {Machine review of arXiv:2607.24796}
}

abstract

Parkinson's disease (PD) affects multiple, dissociable stages of motor and cognitive control. We ask whether passively-collected keystroke dynamics can distinguish two of these stages: noticing a self-generated error (error monitoring) versus recovering normal motor rhythm afterward (motor restart). Using backspace events as naturally-occurring error-correction episodes in the public neuroQWERTY MIT-CSXPD dataset (57 subjects with sufficient backspace data, 27 with PD and UPDRS-III scores), we find no evidence that pre-error keystroke instability differs by disease severity ($r=0.164$, $p=0.413$), but strong evidence that post-error recovery time does, modeled as a continuous accelerated failure time (AFT) survival outcome ($p<10^{-9}$ in each of two independent sub-cohorts; permutation $p<0.0033$; bootstrap 95% CI excluding zero). The two measures are uncorrelated with each other ($r=-0.065$), and in a joint model only post-error recovery remains significant, confirming a genuine dissociation rather than two redundant signals. The effect survives controlling for raw typing speed and replicates against an independent clinical motor test (alternating finger-tapping, $p=0.011$). An initial attempt to model recovery as a discretized time-to-event outcome (analogous to Kaplan-Meier survival curves) destroyed the signal regardless of threshold choice; switching to a continuous AFT formulation, which also fits the data's right-skewed distribution far better than ordinary regression, recovered it. We relate this behavioral dissociation to existing electrophysiological evidence that error detection and post-error motor adjustment in PD are mediated by distinct neural circuits, the former largely spared, the latter linked to subthalamic nucleus activity, and argue that this dissociation is detectable through everyday typing alone.

Figures

Figures reproduced from arXiv: 2607.24796 by the authors.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

13 extracted references · 1 linked inside Pith

[1]

Adams, W.R. (2017). High-accuracy detection of early Parkinson's Disease using multiple characteristics of finger movement while typing. PLOS ONE, 12(11), e0188226

2017
[2]

Alfalahi, H., Khandoker, A.H., Chowdhury, N., Iakovakis, D., Dias, S.B., Chaudhuri, K.R., & Hadjileontiadis, L.J. (2022). Diagnostic accuracy of keystroke dynamics as digital biomarkers for fine motor decline in neuropsychiatric disorders: A systematic review and meta-analysis. Scientific Reports, 12, 7690

2022
[3]

Bondade, N. (2026). Inverse Reinforcement Learning for Interpretable Keystroke Biomarkers in Parkinson's Disease. arXiv:2606.25270 [cs.LG]

arXiv 2026
[4]

Davidson-Pilon, C. (2019). lifelines: survival analysis in Python. Journal of Open Source Software, 4(40), 1317

2019
[5]

Desmurget, M., Gaveau, V., Vindras, P., Turner, R.S., Broussolle, E., & Thobois, S. (2004). On-line motor control in patients with Parkinson's disease. Brain, 127(8), 1755--1773

2004
[6]

Falkenstein, M., Hohnsbein, J., Hoormann, J., & Blanke, L. (1990). Effects of crossmodal divided attention on late ERP components. II. Error processing in choice reaction tasks. Electroencephalography and Clinical Neurophysiology, 78(6), 447--455

1990
[7]

Gehring, W.J., Goss, B., Coles, M.G.H., Meyer, D.E., & Donchin, E. (1993). A neural system for error detection and compensation. Psychological Science, 4(6), 385--390

1993
[8]

Giancardo, L., S\'anchez-Ferro, A., Arroyo-Gallego, T., Butterworth, I., Mendoza, C.S., Montero, P., Matarazzo, M., Obeso, J.A., Gray, M.L., & San Jos\'e Est\'epar, R. (2016). Computer keyboard interaction as an indicator of early Parkinson's disease. Scientific Reports, 6, 34468

2016

Show all 13 references

[9]

Goldberger, A., Amaral, L., Glass, L., Hausdorff, J., Ivanov, P.C., Mark, R., Mietus, J.E., Moody, G.B., Peng, C.K., & Stanley, H.E. (2000). PhysioBank, PhysioToolkit, and PhysioNet: Components of a new research resource for complex physiologic signals. Circulation, 101(23), e...

2000
[10]

Rabbitt, P.M.A. (1966). Errors and error correction in choice-response tasks. Journal of Experimental Psychology, 71(2), 264--272

1966
[11]

ucke, C., Huebl, J., Schneider, G.-H., Ullsperger, M., & K\

Siegert, S., Herrojo Ruiz, M., Br\"ucke, C., Huebl, J., Schneider, G.-H., Ullsperger, M., & K\"uhn, A.A. (2014). Error signals in the subthalamic nucleus are related to post-error slowing in patients with Parkinson's disease. Cortex, 60, 103--120

2014
[12]

Stemmer, B., Segalowitz, S.J., Dywan, J., Panisset, M., & Melmed, C. (2007). The error negativity in nonmedicated and medicated patients with Parkinson's disease. Clinical Neurophysiology, 118(6), 1223--1229

2007
[13]

Tripathi, S., Arroyo-Gallego, T., & Giancardo, L. (2022). Keystroke-Dynamics for Parkinson's Disease Signs Detection in an At-Home Uncontrolled Population: A New Benchmark and Method. IEEE Transactions on Biomedical Engineering

2022

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.

[1] [1]

Adams, W.R. (2017). High-accuracy detection of early Parkinson's Disease using multiple characteristics of finger movement while typing. PLOS ONE, 12(11), e0188226

2017

[2] [2]

Alfalahi, H., Khandoker, A.H., Chowdhury, N., Iakovakis, D., Dias, S.B., Chaudhuri, K.R., & Hadjileontiadis, L.J. (2022). Diagnostic accuracy of keystroke dynamics as digital biomarkers for fine motor decline in neuropsychiatric disorders: A systematic review and meta-analysis. Scientific Reports, 12, 7690

2022

[3] [3]

Bondade, N. (2026). Inverse Reinforcement Learning for Interpretable Keystroke Biomarkers in Parkinson's Disease. arXiv:2606.25270 [cs.LG]

arXiv 2026

[4] [4]

Davidson-Pilon, C. (2019). lifelines: survival analysis in Python. Journal of Open Source Software, 4(40), 1317

2019

[5] [5]

Desmurget, M., Gaveau, V., Vindras, P., Turner, R.S., Broussolle, E., & Thobois, S. (2004). On-line motor control in patients with Parkinson's disease. Brain, 127(8), 1755--1773

2004

[6] [6]

Falkenstein, M., Hohnsbein, J., Hoormann, J., & Blanke, L. (1990). Effects of crossmodal divided attention on late ERP components. II. Error processing in choice reaction tasks. Electroencephalography and Clinical Neurophysiology, 78(6), 447--455

1990

[7] [7]

Gehring, W.J., Goss, B., Coles, M.G.H., Meyer, D.E., & Donchin, E. (1993). A neural system for error detection and compensation. Psychological Science, 4(6), 385--390

1993

[8] [8]

Giancardo, L., S\'anchez-Ferro, A., Arroyo-Gallego, T., Butterworth, I., Mendoza, C.S., Montero, P., Matarazzo, M., Obeso, J.A., Gray, M.L., & San Jos\'e Est\'epar, R. (2016). Computer keyboard interaction as an indicator of early Parkinson's disease. Scientific Reports, 6, 34468

2016