Long COVID Brain Fog: Rehab Beat Usual Care on Goal Scores

It is the end of an ordinary Tuesday, and someone who used to run a department is standing in a kitchen holding a page they have now read four times. The infection was brief, and it was two years ago. What stayed is harder to describe to a doctor than a cough: the word that will not arrive, the meeting whose thread slips halfway through, the familiar recipe that has become genuinely difficult. The fear underneath it is not really about a diagnosis. It is whether this is the permanent shape of things now.

So a headline reporting that cognitive rehabilitation may improve the cognitive outcomes of long COVID lands with unusual weight. Behind that headline is a trial called CICERO, run by University College London across three sites in England, and its result is both better than this field has had before and narrower than the coverage suggests.

CICERO was a multicentre, single-blind, two-arm, parallel-group randomized clinical trial conducted between February 2023 and March 2024. It enrolled adults aged 30 to 60 who had had COVID-19 and who had objective cognitive impairment, defined as scoring at least one standard deviation below the age norm in two or more cognitive domains. That entry criterion matters more than it looks: this was not a study of self-reported fog. Everyone admitted to it had measurable deficits on testing.

Seventy-eight participants were randomized one to one, 38 to cognitive rehabilitation and 40 to treatment as usual. Their mean age was 47.3 years (SD, 7.2); 54 were female (69.2%) and 24 male (30.8%). The trial had calculated that it needed 88 people, split 44 and 44, to detect a conservative effect of 0.7 on the goal-attainment score at three months, so it closed ten participants short of its own recruitment target.

The intervention was ten individual one-hour sessions, once a week, with a trained researcher, applying evidence-based strategies to three functional goals that each participant had selected as personally meaningful. Treatment as usual was, in the authors' own word, variable; most of those participants had access to specialist memory clinics. The primary outcome was participant-reported goal attainment at three months, measured with the Bangor Goal-Setting Interview. Analysis was by intention to treat, using multilevel mixed-effects models with two-sided 95% confidence intervals.

At three months, goal attainment was significantly greater in the rehabilitation group than in the usual-care group: an adjusted mean difference of 2.88 points (95% CI, 2.03-3.73; P < .001), which corresponds to a Cohen d of 1.57. That is a large effect by any conventional reading, and the confidence interval stays a long way clear of zero.

The difference was still there at six months, and it was smaller: an adjusted mean difference of 1.72 points (95% CI, 0.86-2.57; P < .001; Cohen d = 0.91). The gap between the two groups half a year out is a little over half what it was at three months. The authors' stated conclusion is that individualized, goal-oriented rehabilitation led to significant and sustained improvement in goal attainment among people with long COVID-related cognitive impairment.

One absence is worth naming, because it governs how far the result travels. The trial set out to evaluate cognitive and clinical outcomes alongside goal attainment, but the published abstract reports the goal-attainment result only. Whether the objective cognitive test scores moved, and by how much, is not something that abstract settles.

The Bangor Goal-Setting Interview does something unusual for a primary endpoint: it scores a person against goals that person chose. One participant's three goals might be following a recipe from start to finish, getting through a work meeting without losing the thread, and finishing a chapter of a novel. Another participant's three would be entirely different. The score asks whether those specific things became easier.

That is a defensible choice — it measures what people actually lose — and it carries a structural vulnerability. This was a single-blind trial, which means the assessors were blinded and the participants were not. The people scoring the primary outcome knew whether they had just spent ten hours working with a researcher on precisely these three goals. Some portion of a 2.88-point difference is the rehabilitation, and some portion is knowing one received it.

The comparison arm compounds that problem. Usual care was variable rather than attention-matched, so the trial sets ten hours of structured individual attention against whatever each participant's local service happened to provide. It cannot separate the specific evidence-based strategies from goal-setting itself, or from ten hours of a professional's undivided focus.

The authors open by stating that cognitive impairment in long COVID has no proven treatments to date. That is the baseline this result lands against. A field with nothing now has a randomized trial, conducted in people with measured impairment rather than self-described symptoms, showing a large effect on everyday function that remained detectable three months after the sessions stopped.

The distance between that and a treatment someone can actually obtain is mostly logistical rather than scientific. The sessions came from a trained researcher inside a trial, not from a standing clinic service, and the trial ran in England, where most of the comparison group could already reach a specialist memory clinic. What clinicians weigh next is whether a ten-session, goal-led protocol survives contact with ordinary staffing, and whether the improvement represents capacity regained or the same capacity spent more efficiently — a distinction a goal-attainment score is not built to draw.

The desk's position is that this is a real result being asked to carry more than it can. CICERO is the first randomized evidence that anything helps long COVID-related cognitive impairment, and the effect is large, sustained to six months, and found in people with documented deficits rather than self-reported fog — no small thing in a field whose honest summary until this week was that nothing was proven. But the primary outcome is self-scored by unblinded participants against goals they chose themselves, measured against a control arm that received no matched attention, and that is precisely the configuration in which a Cohen d of 1.57 should be read as a direction rather than a magnitude. The number worth carrying forward is therefore not 2.88 points. It is that goal-led rehabilitation has become the first intervention in this field worth the cost of a larger, attention-controlled trial.