A trial published this week in JAMA Network Open asked whether cognitive rehabilitation can help people with Long COVID brain fog. The answer split in two.
Vanova and colleagues enrolled 78 adults with objectively confirmed cognitive impairment — at least one standard deviation below normal in two or more domains. Ten weeks of one-on-one video rehabilitation sessions. Three personally chosen functional goals. Six-month follow-up.
The results:
Vanova et al. JAMA Network Open 2026;9(7):e2620687 · Cohen's d at 3 months
Function improved enormously. Cognition barely moved. The authors' own explanation: gains reflect "acquisition of task-specific strategies... suggesting more efficient use of limited cognitive resources rather than increased capacity."
More efficient use. Not more capacity. This is not hedging — it is the most important sentence in the paper.
The blocked pathway
In "The Floor" I described the cognitive split: learning and memory recover, processing speed and executive function persist ~1.5 standard deviations below normal. Becker's Mount Sinai data showed the split clearly. The Vanova trial now shows the same split from the other direction — rehabilitation reaches function but not speed.
Why? Because the two depend on different brain circuits, and only one of them can still be repaired.
Processing speed and executive function rely on myelinated white matter tracts — long-range connections insulated by oligodendrocytes. Fernández-Castañeda and colleagues showed in 2022 that even mild COVID infection kills oligodendrocytes and strips myelin in mice. A 36-study DTI systematic review confirms widespread white matter tract damage in human Long COVID patients.
Normally, the brain repairs myelin damage through oligodendrocyte progenitor cells — OPCs that differentiate into new oligodendrocytes to replace the dead ones. But a 2026 paper from the Bergles lab in Science revealed something critical: OPC differentiation is constitutive, not damage-responsive. OPCs don't ramp up production when myelin is damaged. They differentiate at a constant rate, and only the oligodendrocytes that arrive where they're needed survive. Crucially, the paper found that inflammation negatively influences this constitutive rate.
Two hits. The factory runs at a fixed pace — it cannot speed up to meet demand. And inflammation slows even that fixed pace. COVID-driven neuroinflammation both creates the damage and suppresses the only repair mechanism.
The workaround
Goal attainment — the things Vanova measured at d = 1.57 — depends on cortical circuits: grey matter networks that can reorganize, reroute, and compensate. This is the brain working around the damage rather than fixing it.
The evidence for this workaround pattern comes from three directions:
A three-year post-COVID neuroimaging follow-up of 51 patients found increased basal ganglia neuronal activation despite continued volume loss — thalamus down 2.4%, putamen down 17.9%. The brain compensating harder as structure deteriorates. All inflammatory biomarkers (NfL, GFAP) had normalized. The damage was done; the rewiring was underway.
In children with MS and MOGAD, Vandermosten and colleagues found increased functional connectivity and neural synchronization as compensation for white matter compromise. The cortex recruits parallel pathways. It doesn't repair the damaged tracts.
And in chronic mild traumatic brain injury, where OPC capacity is intact, Karbasforoushan and colleagues showed that goal-oriented cognitive rehabilitation did produce white matter microstructural plasticity — actual tract repair. This is the control case. When the repair pathway works, rehabilitation produces structural change. When it doesn't — as in Long COVID — rehabilitation produces only cortical compensation.
What this means in practice
The Vanova trial is not a failure. It is a success — of a specific, limited kind.
Cognitive rehabilitation teaches patients to route around damage: external memory aids, task-chunking strategies, environmental modifications. The d = 1.57 effect on goal attainment is large by any standard. Patients reported reading longer, completing work tasks, maintaining social routines. These are real improvements in real lives.
But rehabilitation did not increase processing speed (d = 0.07) because processing speed depends on white matter integrity that cannot currently be restored. The floor I described in May — the persistent 1.5 SD deficit — is the biological limit that cortical strategies cannot breach.
This distinction matters clinically. It means rehabilitation should be offered and expected to help with daily functioning. It also means patients and clinicians should not interpret persistent cognitive test deficits as rehabilitation failure. The test scores measure the damage; the functional gains measure the adaptation. Both are real. They measure different things.
The decisive unrun experiment: pre- and post-rehabilitation neuroimaging in Long COVID patients with confirmed cognitive impairment. If the cortical compensation thesis is correct, you would see increased functional connectivity in grey matter networks with no change in white matter fractional anisotropy. No one has done this study.
What I don't know
The mechanistic bridge here crosses three conditions — MS, mTBI, and Long COVID — that share white matter pathology but differ in etiology and inflammatory profile. The inference that Long COVID rehabilitation works through cortical compensation rather than white matter repair is extrapolated, not directly demonstrated. It is the best explanation for the Vanova data, but it is an explanation, not a measurement.
Cognitive flexibility at d = 0.48 is ambiguous. It sits between the cortical and white matter domains, and I cannot assign it cleanly to either pathway. Whether the floor is permanent or merely long-lived remains unknown — the longest follow-up is 42 months. And the Vanova trial was unblinded with a 25% dropout rate, making the large goal attainment effect size less certain than the number suggests.
What I do know: rehabilitation works. It works by teaching the brain to use what it still has. That is not a small thing.