Neuronal autoantibodies associated with poorer neuropsychological and motor outcomes 6 months after stroke: results from the PHYS-STROKE trial

Clinical Question (PICO)

In adults undergoing inpatient rehabilitation 5–45 days after a moderately severe ischemic or hemorrhagic stroke, does baseline serum neuronal autoantibody positivity, compared with seronegativity, identify worse motor, neuropsychological, or patient-reported outcomes at three or six months?

This is a prognostic exposure comparison, not a randomized test of antibody-directed treatment. The investigators examined any cell-based neuronal autoantibody and, separately, N-methyl-D-aspartate receptor GluN1 antibodies (NMDAR antibodies). A stroke-type subgroup analysis was exploratory; the paper does not establish a prespecified confirmatory subgroup hypothesis.

Bottom Line

The authors conclude that seropositivity was associated with poorer selected outcomes at six, but not three, months and requires confirmation in larger cohorts. The crucial qualification is that only baseline NMDAR positivity and poorer six-month EQ-5D-5L subjective mobility remained statistically significant after Benjamini–Hochberg correction: regression coefficient +1.04 mobility-category points, 95% CI 0.45–1.63, nominal P = 0.001, multiplicity-adjusted P = 0.046. Higher mobility-domain scores indicate more difficulty.

This is a signal for replication, not evidence to screen all stroke survivors or prescribe immunotherapy. The depression, walking-speed, and overall-health findings emphasized in the abstract did not survive the same correction. Full-text results, Section 3.5 and Table 3.

Design

Study type: Exploratory observational biomarker analysis nested within the multicenter, randomized, endpoint-blinded PHYS-STROKE rehabilitation trial. Antibody status itself was not randomized.
N: 200 enrolled in the parent trial.
- Aerobic fitness training: 105.
- Relaxation training: 95.
- Antibody testing at one or more visits: 199.
- Baseline antibody testing: 183.
- Antibody testing at all four visits: 109.
Randomization: The parent trial used 1:1 web-based allocation, stratified by center, age (≤65 versus >65 years), and functional ambulation category (≤3 versus >3). These details come from the primary PHYS-STROKE trial report, not from randomization of the biomarker exposure.
Setting: Seven inpatient rehabilitation centers in Berlin and Brandenburg, Germany.
Enrollment: September 2013–April 2017.
Follow-up: Scheduled visits at baseline, after the four-week intervention, and three and six months after stroke. The median final blood-sampling time was 180 days; this is not a reported mean follow-up duration.
Analysis: Generalized estimating equations for longitudinal serostatus; multivariable linear regression for three- and six-month outcomes. Models considered age, baseline NIHSS, the corresponding baseline outcome score, and education where applicable. Missing values were handled using 1,000 multiply imputed datasets with five iterations; imputed-model P values also underwent Benjamini–Hochberg adjustment.
Primary outcome: No single confirmatory primary endpoint was specified for this exploratory antibody analysis. The parent trial’s co-primary outcomes were change in Barthel Index and maximal walking speed from baseline to three months.
Blinding: Antibody-assay interpretation was blinded to patient data; the parent trial used blinded outcome assessors. Participants and treating personnel could not be blinded to treadmill versus relaxation sessions.

Population

Inclusion Criteria

Age ≥18 years.
Imaging-confirmed ischemic or hemorrhagic stroke, 5–45 days previously.
Barthel Index ≤65 at enrollment and ability to comply with study requirements.
Ability to sit unsupported for ≥30 minutes and medical suitability for aerobic exercise, as specified in the parent trial report.
Written informed consent.

Exclusion Criteria

The parent report identifies these key exclusions:

Intracranial hemorrhage caused by a ruptured aneurysm or arteriovenous malformation.
Inability to perform the required exercise.
Assisted walking before the index stroke.
Severe cardiac or psychiatric comorbidity.

These are the main-report criteria, not a claim to reproduce every item in Supplementary Table S1. The supplementary document was not available from the tested PMC asset endpoint. Parent trial eligibility.

Baseline Characteristics

Mean age: 69 ± 12 years.
Women: 81/200 (41%).
Ischemic stroke: 181/200 (91%); the cohort also included hemorrhagic stroke.
First-ever stroke: 162/200 (81%).
Median NIHSS: 8 (IQR 5–12).
Median modified Rankin Scale (mRS): 4 (IQR 4–4).
Median Barthel Index: 50 (IQR 35–60).
Baseline sampling occurred a median of 25 days after stroke (IQR 15–37); only 44 patients were tested within the first two weeks.
Baseline NMDAR-positive patients more often had prior cerebrovascular disease than seronegative patients (50% versus 26%) and had higher triglycerides. They were not simply a randomized or otherwise exchangeable comparison group.

Interventions

Aerobic fitness arm: Bodyweight-supported treadmill exercise for 25 minutes, five days per week, for four weeks, in addition to standard inpatient rehabilitation.
Relaxation arm: Muscle-relaxation sessions of the same duration and frequency, also in addition to standard rehabilitation.
Biomarker exposure: Serum screened against 40 neuronal antigens using commercial cell-based assays and against cerebellar tissue using tissue-based assays. NMDAR testing included IgM, IgA, and IgG; other cell-based antigens were assessed with IgG. Seropositivity was defined as a titer ≥1:10; high titers as ≥1:100.
No antibody-targeted therapy: There was no randomized immunotherapy, antibody-removal intervention, or antibody-guided rehabilitation strategy.

Outcomes

Principal Prognostic Finding

There was no prespecified primary antibody endpoint. In the imputed, covariate-adjusted six-month models, the following associations were reported; the final column is the Benjamini–Hochberg-adjusted P value, not the nominal P value.

Exposure and six-month outcome	Regression coefficient β	95% CI	Nominal P	Multiplicity-adjusted P
NMDAR positivity: EQ-5D-5L mobility	+1.04	0.45 to 1.63	0.001	0.046
Any neuronal antibody: EQ-5D-5L mobility	+0.51	0.04 to 0.99	0.033	0.407
Any neuronal antibody: CES-D depression score	+3.93	0.03 to 7.82	0.048	0.407
NMDAR positivity: overall-health visual analogue rating	−9.06	−17.96 to −0.16	0.046	0.407
NMDAR positivity: maximum walking speed	−0.30 m/s	−0.57 to −0.03	0.027	0.407

These are regression coefficients, not odds ratios, risk ratios, or randomized treatment effects. The EQ-5D-5L mobility coefficient concerns an ordinal response modeled linearly; it is not a one-point change in the aggregate EQ-5D utility index. The authors state that the depression-score difference did not reach their threshold for a minimally clinically important difference. Table 3 and Discussion.

Other Clinical Outcomes and Subgroups

No statistically significant antibody–outcome associations were reported at three months.
At six months, NMDAR positivity was not significantly associated with mRS (β +0.09, 95% CI −0.37 to 0.54; nominal P = 0.714), Barthel Index (β −0.48, 95% CI −9.28 to 8.33; P = 0.915), or Montreal Cognitive Assessment (β +1.60, 95% CI −0.33 to 3.54; P = 0.104).
The exploratory ischemic-versus-hemorrhagic analysis was described as showing that the observed associations were driven by ischemic stroke. The main paper does not provide the subgroup coefficients, confidence intervals, or a formal interaction estimate; these reside in Supplementary Table S11. With only 19 hemorrhagic-stroke participants in the parent cohort, this is not evidence that the relationship differs reliably by stroke type.

Antibody Frequency and Longitudinal Models

Any cell-based seroreactivity: 39/183 (21.3%) at baseline versus 18/137 (13.1%) at six months. The denominators differ.
Tissue-based reactivity: 4/183 (2.2%) versus 9/137 (6.6%).
NMDAR positivity at any visit: 30/199 (15.1%); baseline NMDAR positivity: 24/183 (13.1%).
Across follow-up, NMDAR IgM was detected in 21 patients, IgA in 14, and IgG in only two; categories overlap.
Any cell-based positivity at six months versus baseline: imputed GEE OR 0.59, 95% CI 0.36–0.98, nominal P = 0.041, adjusted P = 0.246. The nominal temporal trend did not survive correction in Table 2.
Female sex and longitudinal NMDAR positivity: OR 3.00, 95% CI 1.34–6.71, nominal P = 0.008, adjusted P = 0.144. This is an exploratory exposure association, not a demonstrated sex-specific treatment effect.

Adverse Events / Safety

There was no statistically significant difference in serious adverse events between baseline NMDAR-positive and seronegative patients, including hospital referral, serious cerebrovascular or cardiovascular events, and death. Event counts and comparative estimates were not reported in the main text; lack of significance does not establish equivalent risk.
The article notes more serious adverse events in the parent aerobic-fitness arm at three months. That concerns the rehabilitation intervention and should not be attributed to antibody status.
No immunotherapy safety or efficacy data were generated by this analysis.

Figures

Longitudinal frequency of cell-based and tissue-based neuronal autoantibody reactivity — *Figure 1.* Antibody frequency across the six-month follow-up. Red denotes cell-based reactivity pooling all immunoglobulin classes, yellow denotes cell-based IgG reactivity, and blue denotes tissue-assay reactivity. Measurements precede rehabilitation at baseline, follow its completion at V1, and occur at three months (V2) and six months (V3). The populations tested at successive visits differ. Caption paraphrased from the publication.

Individual NMDAR IgM, IgA and IgG antibody titers across follow-up visits — *Figure 2.* Individual trajectories of NMDAR antibody titers, separated by immunoglobulin class. Colors encode each patient's starting titer, allowing the same trajectory to be followed across subsequent visits. Caption paraphrased from the publication.

Criticisms

The parent RCT does not randomize the biomarker. This secondary analysis remains observational and cannot show that antibodies cause impaired recovery.
Multiplicity substantially changes the story. Only one six-month outcome association survives correction, with adjusted P = 0.046. A broad claim of established neuropsychological and motor harm would overstate the results.
Limited exposed sample: Just 24 baseline patients were NMDAR-positive; only two patients had NMDAR IgG detected across visits. Pooling predominantly IgM/IgA positivity does not establish an autoimmune-encephalitis mechanism.
Variable biological starting point: Sampling 5–45 days after stroke cannot distinguish pre-existing antibodies from a response to the stroke. Blood–brain barrier integrity was not measured in this analysis.
Missingness: Blood testing fell from 183 patients at baseline to 137 at six months. Only 109 had testing at every visit. Multiple imputation helps under its assumptions but cannot guarantee removal of bias from informative attrition.
Modeling choices: Subjective mobility is ordinal but was modeled using linear regression. Rehabilitation assignment was excluded from outcome models to reduce overfitting; residual confounding and model dependence remain possible.
Prediction is not yet validated: No externally validated prognostic rule, demonstrated incremental discrimination, decision-curve benefit, or antibody-guided intervention was reported.
External validity: Findings come from selected, exercise-capable inpatient rehabilitation patients, predominantly with ischemic stroke. They do not establish applicability to all stroke survivors, severely medically unstable patients, or hemorrhagic stroke specifically.

Funding

The article’s funding statement declares no financial support for this work and/or its publication. It does not separately describe a funder’s role. This statement should not be interpreted as proof that the original PHYS-STROKE trial had no funding.

Disclosures include employment of Kristin Rentzsch by the laboratory that performed antibody testing and author grants or honoraria declared outside the submitted work. The parent trial was registered as NCT01953549; the accompanying BAPTISe study was registered as NCT01954797. Funding and conflict-of-interest statements.

The paper

Authors. Charlotte Pietrock et al.
Title. Neuronal autoantibodies associated with poorer neuropsychological and motor outcomes 6 months after stroke: results from the PHYS-STROKE trial.
Journal. Frontiers in Immunology, volume 16, article 1678840.
Year. Assigned to the 2025 collection; electronically published March 18, 2026, according to the PMC JATS record. The collection year and online publication date are not interchangeable.
DOI. 10.3389/fimmu.2025.1678840
PMCID. PMC13038563

Deep Dive — click to expand

What this is

This look-back examines a biomarker paper published online in March 2026 using a rehabilitation cohort enrolled in 2013–2017. Its important contribution is not a treatment breakthrough but a longer view of recovery: serum neuronal autoantibodies were associated with selected six-month outcomes even when three-month associations were absent. The strongest finding was approximately one additional level of self-reported mobility difficulty among baseline NMDAR-positive patients. That merits investigation, but the study provides neither causal proof nor a validated clinical screening pathway. Primary full text.

1. Shadow Audit

The abstract’s plural conclusion is broader than the multiplicity-corrected result. Depression was nominally worse by 3.93 CES-D points; maximum walking speed was nominally lower by 0.30 m/s; overall health was nominally lower by 9.06 visual-analogue points. All three had adjusted P = 0.407. Only NMDAR positivity versus subjective mobility retained adjusted P <0.05, at 0.046. Nor was there a significant mRS, Barthel Index, or MoCA association. The paper’s central signal therefore concerns a patient-reported domain, not a demonstrated generalized deterioration in functional or cognitive recovery.

The other concealed boundary is antibody identity. Of 199 patients ever tested, 30 were NMDAR-positive, but only two had IgG detected; most positivity involved IgM or IgA. A reader who mentally substitutes “pathogenic anti-NMDAR encephalitis” for this pooled serum exposure has changed the biological question without evidence. Tables 1 and 3.

2. Inversion Engine

For the corrected mobility result to support the opposite directional conclusion, the estimated coefficient would have to move from +1.04 toward or below 0: from greater mobility difficulty to no difference or less difficulty. The reported unadjusted 95% CI has a lower bound of +0.45, so this model does not contain a beneficial-direction effect. But the multiplicity-adjusted P of 0.046 is close to the conventional 0.05 threshold; it does not guarantee robustness to different missing-data assumptions, ordinal modeling, or residual confounding.

A useful inversion is not “prove antibodies help.” It is “could positivity be a correlate rather than a cause?” Prior cerebrovascular disease was more common in the NMDAR-positive group, baseline sampling varied substantially, and blood–brain barrier status was unavailable. To strengthen a causal explanation, a future study must show that the association persists under tighter sampling, isotype-specific analysis, and richer adjustment. No sensitivity threshold or confounding-strength bound was calculated here, so none should be invented. Methods and limitations.

3. Second-Order Catalyst

Interpretive implication: Academic rehabilitation services and prospective cohort investigators should move first, by measuring six- and twelve-month patient-reported mobility alongside objective walking and mood outcomes. The most defensible initial population resembles the study: moderately impaired, predominantly ischemic-stroke patients entering inpatient rehabilitation—not an unselected hyperacute stroke population.

The immediate change is to a research follow-up protocol, not a routine antibody order set. A confirmatory multicenter cohort can standardize sampling time, distinguish IgM/IgA from IgG, and prespecify one patient-centered endpoint. Routine clinical adoption should wait for external validation and evidence that knowing the assay result changes a decision beneficially. The paper does not supply an adoption timetable or implementation trial.

4. Asymmetric Leverage

The opportunity is a targeted prognostic research question, not a large-denominator treatment effect. Baseline NMDAR positivity occurred in 24/183 tested patients (13.1%). That identifies a manageable research subgroup, but there is no absolute risk reduction, number needed to treat, or cost-effectiveness result because no antibody-guided treatment was tested.

The lower-risk payoff lies in better outcome measurement. A six-month patient-reported mobility association was detectable when mRS and Barthel Index associations were not. Adding a structured mobility question to research follow-up can test whether global scales miss an important experience of disability without first expanding antibody testing across the health system. Any claim that antibody testing itself improves care remains unproven. Table 3.

5. Paradigm Destroyer

This paper challenges two shortcuts: “a trial-derived biomarker association is randomized evidence,” and “a negative three-month association settles every later recovery outcome.” Neither is justified. Conversely, it does not overturn standard secondary prevention, rehabilitation practice, or criteria for diagnosing autoimmune neurological disease.

An appropriately modest two-sentence protocol amendment would be: “In a prospective rehabilitation outcomes protocol, retain a six-month assessment of patient-reported mobility and mood rather than relying exclusively on three-month global disability. Do not add routine neuronal-autoantibody screening or immunotherapy on the basis of this exploratory analysis; biomarker testing for this hypothesis belongs in a confirmatory research pathway.” This is an editorial proposal for evidence generation, not a guideline recommendation.

MVP — Minimum Viable Proof

The numerical finding worth trying to reproduce is baseline NMDAR seropositivity associated with greater six-month mobility difficulty, here β +1.04 (95% CI 0.45–1.63), with adjusted P = 0.046. The minimum next proof is an independent, prospectively specified study showing a clinically meaningful association after multiplicity control, appropriate ordinal analysis, and transparent handling of missingness.

For bedside screening, association is still insufficient: the assay must improve a validated clinical prediction strategy and identify an action with patient-important benefit. For immunotherapy, the minimum proof is different and higher—a randomized intervention study demonstrating benefit and acceptable harm in a clearly defined biomarker-positive population. This paper meets neither clinical-utility requirement.

Best Combination

Pair this result with domain-specific longer-term evidence rather than treating all post-stroke outcomes as interchangeable. In the independently retrieved DEMDAS cohort report, 58/569 participants were NMDAR-antibody-positive. Positivity was associated with twelve-month memory impairment (adjusted OR 3.8, 95% CI 1.33–10.82), but not global cognitive impairment. That study and PHYS-STROKE concern different endpoints and do not justify a pooled mobility or cognition effect estimate.

Together they motivate a research program with explicit outcome domains, standardized assay methods, isotype-specific results, and follow-up beyond three months. They do not establish that circulating antibodies are pathogenic, that every positive assay signals autoimmune disease, or that suppressing the immune response improves recovery. The parent PHYS-STROKE randomized intervention should remain conceptually separate from both observational biomarker analyses.

Overvalue Warning

Do not turn “associated with poorer outcomes” into a proven multi-domain syndrome. Only one outcome association survived the article’s multiplicity correction; the depression difference was also judged below the authors’ clinically important threshold. Absence of other statistically significant associations is not proof of absence of effects.
Do not turn serum NMDAR positivity into a treatment indication. Only two participants had IgG detected across visits, antibody status was not randomized, and no antibody-targeted therapy was studied. Neither routine screening nor immunotherapy follows from these data.

Educational use notice. The Time Is Brain Review is dedicated to medical education. Figures from published articles are embedded for teaching purposes, with attribution to the original publication. We do not claim copyright over any embedded figure; all rights remain with the original publisher and authors under the article's published license. If you are a rights holder and would like a figure removed, please contact the editorial team. See our editorial notice for the full policy.