Patients with ataxia rarely arrive with their diagnosis. They arrive with a widened gait, falls they play down, and a complaint that three different systems can produce. This article deals with what is decided in the clinic: classifying the ataxia, rating it with the scale that the learned societies actually recommend, knowing what rehabilitation shifts and at what price in intensity, and recognising the rapid onset that is not a matter for physiotherapy.

• Updated August 2026
• Level clinical summary
• Sources 48 verified references (PMID / DOI)
• Reading time about 45 minutes

What does the cerebellum do, and what happens when it gives way?

The cerebellum does not command movement, it calibrates it. Understanding precisely what it calibrates explains why the ataxic patient trembles on arrival and not on departure, why the base of support widens, and why some rehabilitation strategies work while others fail by construction.

A predictive corrector, not an executor

The cerebellum receives a copy of the motor command and sensory information in return, and it constantly compares the predicted movement with the movement actually performed. That comparison serves two purposes: correcting the movement under way, and updating the internal model that will serve the next one. When it is damaged, what is missing is neither strength nor the programme, it is the adjustment. The patient knows what to do and does it badly.

The heaviest clinical consequence is not motor, it is educational. Amy Bastian's work has established that cerebellar damage consistently disrupts motor adaptation, the learning that works by error correction, whereas lesions of other brain regions often spare it17. In other words: the learning route on which a large part of our neurological protocols implicitly rests is precisely the one that is broken in this patient.

"Cerebellar damage consistently disrupts adaptation; lesions of other brain regions, often, do not." Bastian, Current Opinion in Neurology, 2008

This does not condemn rehabilitation: the trials show the opposite. But it steers its content. A patient who no longer adjusts finely through error makes progress through the repetition of whole tasks, through the building of substitution strategies, and through the upkeep of the capacities that do not depend on the cerebellum: strength, endurance, flexibility, a safe environment. That is exactly the profile of the programmes that worked in the trials, and it is also why very technical, very segmental protocols disappoint.

The cerebellar syndrome is not one, it is three

The reference clinical description, restated in 2026 by Manto, Mitoma, Burt and Schmahmann, divides the cerebellar syndrome into three categories7 :

• The cerebellar motor syndrome (CMS) : dysmetria, kinetic tremor, asynergia, dysdiadochokinesia, dyschronometria. This is the side that the SARA rates.
• The vestibulo-cerebellar syndrome (VCS) : saccadic dysmetria, saccadic smooth pursuit, inferior vertical nystagmus (downbeat), gaze-evoked nystagmus. This is what the ICARS rates in part and what the SODA scale rates specifically.
• The cerebellar cognitive affective syndrome, or Schmahmann syndrome (CCAS/SS) : impairment of executive function, spatial cognition, language processing and emotional regulation.

This third category deserves a word for the physiotherapist, because it is the one most often missed and the heaviest in practical consequences. A patient whose executive functions are impaired no longer organises their home programme, no longer plans their sequences, no longer transfers an instruction from one context to another. They are labelled "poorly adherent" when they are cerebellar right down to their capacity to adhere. A dedicated scale exists, the CCAS/Schmahmann Scale8 ; its use is the neuropsychologist's business, but simply knowing of it changes the way an exercise instruction is written.

How many patients, and which ones?

Degenerative ataxias are rare diseases, but taken together they are not so rare. The systematic review by Ruano and colleagues, which brought together 22 studies and 14,539 patients in 16 countries, puts the mean prevalence of autosomal dominant hereditary ataxias at 2.7 per 100,000 (range 1.5 to 4.0) and that of the recessive forms at 3.3 per 100,000 (1.8 to 4.9). SCA3, or Machado-Joseph disease, is the commonest dominant form, followed by SCA2 and SCA6; Friedreich's ataxia dominates the recessive forms2.

Two figures from that same review speak even more directly to practice. The first: bringing together hereditary ataxias and hereditary spastic paraplegias, the authors estimate that about one person in 10,000 is affected. The second, often left unsaid: in population series, 33 to 92 % of families with dominant transmission and 40 to 46 % of recessive families remain without an identified mutation after systematic genetic testing2. The patient who tells you "they didn't find anything" is not recounting a diagnostic odyssey: they are describing the statistical norm.

What has changed since 2023: SCA27B

A physiotherapy article has no business settling questions of genetics, but this one has to flag a recent shift, because it concerns exactly the patient seen in private practice: the 60-year-old adult who has been walking badly for a few years, and in whom "nothing was found".

In January 2023, a deep intronic GAA expansion of the FGF14 gene was identified as a cause of late-onset cerebellar ataxia. It was present in 61 % of French-Canadian index cases, 18 % of German, 15 % of Australian and 10 % of Indian cases of late-onset ataxia24. The clinical characterisation published the same year in 50 patients describes a late-onset pancerebellar syndrome, median age at onset 60.0 years, associated with an afferent sensory deficit in 55 % of patients and with dysautonomia in 28 %25.

Two findings from that series bear directly on rehabilitation. First the slowness: the measured progression is 0.29 SARA points a year, with a maximum score of 18 points even at advanced stages, and only half the patients use a unilateral walking aid after eight years of disease25. Then the treatment: 86 % of patients treated with 4-aminopyridine report a response that matters for their daily life, confirmed by three individual prospective experiments in on/off25. On this ground, "idiopathic" late-onset ataxia is no longer necessarily idiopathic, nor necessarily untreatable.

How do you tell cerebellar ataxia from proprioceptive or vestibular ataxia?

This is the question that decides everything else. The three ataxias produce an unsteady gait and an identical complaint, but they are not rehabilitated in the same way, and one of them responds far less well to coordinative training than the other two. The sorting is done with your hands, a chair, and five minutes.

The principle: which channel is missing, and what compensates

Balance rests on three inputs, visual, vestibular and somatosensory, and on one integrator, the cerebellum. Each ataxia corresponds to the failure of one link, and is signalled by the compensation the patient sets up spontaneously.

• Proprioceptive (or sensory, or afferent) ataxia. The somatosensory input is missing. Vision compensates, effectively. The patient watches their feet, walks better in full light, and comes apart in the dark or in the shower. This is the mechanism of sensory neuropathies, posterior column lesions and certain deficiency states.
• Vestibular ataxia. The vestibular input is missing or delivers a false signal. The patient has spinning symptoms, nystagmus, a lateralised deviation, and compensates by fixing on the horizon.
• Cerebellar ataxia. It is the integrator that is damaged: all three inputs arrive, but they are no longer fused or converted into a calibrated command. No sensory compensation is possible, since nothing is missing upstream. That is why the patient is already unstable with the eyes open.

The proprioceptive patient deteriorates when you close their eyes. The cerebellar patient was already unstable before you closed them.

The Romberg, and what it is wrongly made to say

The Romberg test, standing with the feet together and the eyes closed, is taught as a test of proprioception, and so it is. But a review by Halmágyi and Curthoys published in 2021 recalls a decisive nuance: standing with the eyes closed on a soft surface rather than on a firm floor tests vestibular function more than it tests proprioception18. The foam, added to "toughen" the test, changes the question being asked. A pathological Romberg on foam in a patient who is normal on a firm floor points to the vestibule, not to the posterior columns.

Three practical points follow from this:

• Do the Romberg on a firm floor first, and record performance with the eyes open and with the eyes closed separately. It is the gap between the two that informs, not the absolute value.
• Do not conclude "positive Romberg" from mere unsteadiness. In the cerebellar patient there is sway with the eyes open and a slight worsening on closing them. In the proprioceptive patient the worsening is marked, often with a fall if you are not there.
• Add the foam afterwards, as a second question and not as a hardening of the first.

The sorting table

The last row of this table is not an opinion. In the founding trial by Ilg and colleagues, the patients whose afferent pathways were also affected made less progress than patients with pure cerebellar ataxia, in gait speed, lateral sway and intra-limb coordination; and balance regulation in static and dynamic tasks improved significantly only in the patients without afferent involvement11. The differential diagnosis is therefore not an academic exercise: it predicts part of the response to treatment.

A gait signature specific to the cerebellum

There is a kinematic marker that separates cerebellar ataxia from the other causes of instability, and it lights up clinical reasoning even without a gait laboratory. Ilg and colleagues compared the gait of 13 patients with cerebellar degeneration, 6 patients with peripheral vestibular loss and 8 patients with Parkinson's disease. All three groups shared balance-related gait abnormalities and increased general variability. But the increase in the temporal variability of intra-limb coordination was specific to cerebellar damage, markedly smaller in the two other groups16.

Translated into clinical observation: in a vestibular or parkinsonian patient, the instability produces an irregular step but a rhythm that is relatively preserved. In the cerebellar patient it is the tempo itself that breaks down: the stance and swing phases do not last the same from one step to the next. That is what you hear, literally, when you listen to an ataxic patient walking down a corridor.

Which scale should be used to rate ataxia: SARA, ICARS or BARS?

Three scales are in circulation, all valid, all published, and one of them is faster than the others. That does not make it the one the learned societies recommend. Here is what each of them measures, what their measurement properties are worth, and which one really serves physiotherapy follow-up.

What the Movement Disorder Society says

The Movement Disorder Society scales review committee sifted through 14 instruments, 9 rating scales and 5 functional tests, and classified them as "recommended", "suggested" or "listed" according to pre-set criteria. The result, published in Movement Disorders in 2021, is clear-cut6 :

• Spinocerebellar ataxias : ICARS and SARA "recommended".
• Friedreich's ataxia : FARS, ICARS and SARA "recommended".
• Ataxia-telangiectasia : ICARS and SARA.
• Brain tumour ataxia : SARA.
• Cerebellar signs in multiple sclerosis : ICARS.
• Multiple system atrophy, cerebellar type : UMSARS and ICARS.
• FXTAS syndrome : ICARS. CDG-PMM2 : ICARS.

The BARS appears in none of these recommendation lists. This is not a judgement on its intrinsic quality: it was validated by its authors, and its psychometric properties are good. It is an observation about status: in 2026, the reference scale for documenting an ataxia in a record, a prior-approval request or a publication remains the SARA, or possibly the ICARS.

The SARA in detail

The SARA (Scale for the Assessment and Rating of Ataxia) was developed and tested on two series of 167 and 119 patients with spinocerebellar ataxia. Its published characteristics10 :

• Administration time: 14.2 ± 7.5 minutes (range 5 to 40 minutes).
• Inter-rater reliability: ICC = 0.98. Test-retest reliability: ICC = 0.90.
• Internal consistency: Cronbach's alpha = 0.94, and factor analysis finds a single factor: the SARA measures one thing only.
• Linear relationship with global assessment on a visual analogue scale (r² = 0.98), strong correlation with the Barthel index (r = −0.80) and with part IV of the UHDRS (r = −0.89), but weak correlation with disease duration (r = 0.34).

This last point deserves a pause. A scale that correlates strongly with independence and weakly with how long the disease has lasted is exactly what follow-up needs: it describes a functional state, not a calendar. Two patients at the same chronological stage can have very different SARA scores, and it is the SARA that is right.

For the physiotherapist, one further argument, noted by Marquer, Barbieri and Pérennou in their systematic review: 18 of the SARA's 40 points bear on postural disorders (gait, standing and sitting)9. Nearly half the score is therefore played out on exactly the ground of rehabilitation. The remaining 22 points are split between speech and the four limb coordination tasks.

ICARS and BARS: what they are for

The ICARS (International Cooperative Ataxia Rating Scale) is the oldest of the three. Proposed in 1997 by an ad hoc committee of the World Federation of Neurology, it is semi-quantitative, scored out of one hundred points, and compartmentalised: posture and stance disorders, limb ataxia, dysarthria and oculomotor disorders are quantified separately, each compartment carrying a deliberately calibrated weight31. It therefore covers both the cerebellar motor syndrome and the vestibulo-cerebellar syndrome7, which explains why it remains the only one "recommended" in the indications where oculomotor function weighs heavily: multiple sclerosis, FXTAS, multiple system atrophy. Its cost in time, however, makes it barely realistic in private practice.

The BARS (Brief Ataxia Rating Scale) was built in 2009 by Schmahmann and his team precisely to lift that obstacle. Their method is interesting: the authors first extended the ICARS by adding seven tasks (the MICARS), then searched computationally for the subset of five tasks that correlated best with the total, under the clinical constraint of keeping one per domain: gait, upper limb, lower limb, speech, oculomotor function. The result comes down to five tasks scored out of 30 points in all (gait 0 to 8, heel-shin and finger-to-nose rated on each side, dysarthria 0 to 4, oculomotor function 0 to 2), and to three figures13 : a correlation of 0.952 between the BARS and the rest of the MICARS; a Cronbach's alpha of 0.90 for the BARS against 0.92 for the SARA ; inter-rater reliability of 0.91 against 0.93. The authors conclude that the BARS is "valid, reliable, and sufficiently rapid and accurate for clinical use".

Two honest readings of this result sit side by side. The first: the BARS does almost as well as the SARA with fewer tasks, and it has the merit of including oculomotor function, which the SARA ignores. The second: "almost as well" on properties that are already excellent does not make up for the absence of a formal recommendation and for the thinner body of longitudinal data. In private practice the reasoning settles simply: the SARA remains the follow-up scale, the BARS can serve as a rapid screen when time is short, provided that measurements made with different scales are never compared with one another.

What the SARA does not rate, and what has to be rated elsewhere

Three blind spots, all of them clinically weighty:

• Oculomotor function. The vestibulo-cerebellar syndrome escapes the SARA entirely. A dedicated scale has existed since 2022, the SODA (Scale for Ocular motor Disorders in Ataxia)14.
• Cognition and affect. The CCAS/Schmahmann Scale explores them8 ; a review by the Task Force of the Society for Research on the Cerebellum has set out its conceptual framework15.
• The patient's lived experience. No clinician-rated scale says what the disease does to a life. Patient-reported measures, among them the PROM-Ataxia cited in the 2026 review of the scales7 fill this gap.

In everyday practice, though, the most profitable addition is none of these scales: it is a functional goal written with the patient. The Ilg 2009 trial and the Ilg 2010 trial both used a goal attainment score for each individual, and that is where the most telling gains were observed: the ability to achieve personally meaningful goals of everyday life11.

What goes into a physiotherapy assessment of ataxia that is of any use?

An assessment serves to decide and to compare. Anything that does neither costs session time. Here are the five measurements that carry the treatment decision, and the one history question that must never be asked open-endedly.

The question of falls, asked properly

"Do you fall?" gets "no, not really" from patients whose prospective diary shows the opposite. In the prospective study run within the EuroSCA cohort, 113 patients with SCA1, SCA2, SCA3 or SCA6 kept a falls diary between two annual visits: 84.1 % reported at least one fall over twelve months, injuries were frequent and their frequency rose along with the frequency of falls26. The authors had in fact run the same study retrospectively beforehand, precisely because recall bias made the reported figures unreliable.

One result from that same study steers the assessment: the presence of non-ataxic symptoms was associated with a higher frequency of falls26. In other words, the ataxia score is not enough to predict fall risk: it is the associated neurological comorbidities (spasticity, neuropathy, oculomotor disorders, cognitive impairment) that tip the balance. Earlier work by the same team had already established that falls are very frequent in degenerative cerebellar ataxias, often serious, and a source of fear of falling27.

In practice: ask how many falls there have been over the last three months, in what circumstances, with what consequences, and offer a diary. The diary is not a research gadget, it is the only instrument that makes the real frequency visible to the patient themselves.

The tandem, the earliest test

Tandem walking, one foot in front of the other, heel against toes, deserves a place of its own. The international consensus on ataxic gait establishes that it is impaired at the early stages of cerebellar disorders and can be particularly useful for assessing the pre-ataxic stages of progressive ataxias30. It is the test that gives way first, before free walking deteriorates. Its simplest measure: the number of consecutive steps taken without a deviation, out of the ten asked for.

That same consensus recalls a fact that structures the whole of management: ataxia of stance and gait is one of the very first deficits to appear in degenerative cerebellar disorders, it is disabling and it carries a high risk of falling30. Gait is not one goal among others, it is the goal.

The five measurements that are enough

A word about the functional independence measure (FIM). The Milne 2025 multicentre trial had made it its primary outcome, in its motor domain, the mFIM, and the result is worth knowing: rehabilitation improves it significantly at 7 weeks (mean difference 2.26 points, 95 % CI 0.26 to 4.26; p = 0.028) but no longer significantly at 30 weeks (1.74; 95 % CI −0.32 to 3.81; p = 0.098), even though the SARA score does remain improved at 30 weeks4. The two measures do not say the same thing and do not decay at the same rate: documenting both avoids concluding too quickly one way or the other.

Which red flags in an ataxia of rapid onset?

A degenerative ataxia sets in over years. Any other speed of onset changes the nature of the problem, and sometimes the time scale available for acting. This is the chapter to reread before taking on a new patient.

The speed of onset is the first sort

A normal MRI is no reassurance: the case of acute vestibular syndrome

The physiotherapist regularly meets the patient who was "seen in the emergency department, scan normal, they told me vestibular neuritis". Two publications call for caution.

The first is the original study of the HINTS examination. Kattah and colleagues prospectively examined 101 high-risk patients presenting with an acute vestibular syndrome (vertigo, nystagmus, nausea, intolerance of head movement, unsteadiness), of whom 25 turned out to have peripheral involvement and 76 central involvement, including 69 ischaemic strokes. The combination of three signs, a horizontal head impulse test that was normal, direction-changing nystagmus on eccentric gaze, or skew deviation was 100 % sensitive and 96 % specific for stroke. And above all: the initial diffusion MRI was falsely negative in 12 % of cases, all of them performed less than 48 hours after the onset of symptoms19.

A three-minute clinical examination proved more sensitive than early diffusion MRI for detecting stroke in acute vestibular syndrome.

The second publication tempers the first, and it would be dishonest to cite only one of them. A 2020 systematic review with meta-analysis, covering 5 studies and 617 participants, compared the performance of the HINTS examination according to who carried it out. Performed by neurologists: sensitivity 96.7 % (95 % CI 93.1 to 98.5) and specificity 94.8 % (91 to 97.1). Performed by a cohort that included emergency physicians: sensitivity 83 % (63 to 95) and specificity 44 % (36 to 51)20. The authors conclude that the HINTS examination, used on its own by emergency physicians, has not demonstrated sufficient accuracy to rule out a stroke.

The lesson for the physiotherapist is twofold and unambiguous. This is not a test you make your own by reading an acronym : its performance collapses outside trained hands, and a "reassuring" result badly obtained is more dangerous than no test at all. But knowing its components remains useful: knowing that a head impulse test that is normal in an acutely vertiginous patient is a worrying sign and not a reassuring one is exactly the kind of counter-intuition that makes you ask for a further opinion.

The reversible causes, which on their own justify vigilance

The British guidelines on the progressive ataxias, drawn up by more than 30 health professionals and published in 2019, insist on a point that concerns everyone: more than a hundred different diseases can cause an ataxia, there is no disease-modifying treatment for most of them, but many aspects are treatable, and the early diagnosis of the few reversible causes is of paramount importance28.

The physiotherapist does not have to make these diagnoses. They do, however, hold a privileged observation post: they see the patient repeatedly, over time, and are often the first to notice a change of trajectory. A patient whose SARA gains three points in three months does not have a degenerative ataxia running its course: they have something else, and the only right answer is a letter to the general practitioner.

What does rehabilitation really change, and by how much?

The question is not whether rehabilitation "does good". It is by how many scale points, for how long, at what price in intensity, and how that compares with the speed at which the disease advances. The figures exist, they are modest, and they are more interesting than unmeasured enthusiasm.

The starting point: it is the only treatment

The 2014 international consensus on the management of degenerative cerebellar disorders, signed by fifteen teams, reaches a conclusion that has not been contradicted since: no drug has proved its efficacy, with the exception of the aminopyridines and acetazolamide in episodic ataxia type 2 and of the aminopyridines in certain downbeat nystagmus: "all authors agreed that the mainstays of the treatment of degenerative cerebellar ataxia are currently physiotherapy, occupational therapy and speech and language therapy"1.

The same text adds a sentence that this chapter sets out to honour: there is consensus that evidence-based guidelines for the physiotherapy of degenerative cerebellar ataxia remain to be developed1. Twelve years later, they still do not exist. What we have are trials and two meta-analyses.

The three founding trials

Ilg 2009: intensive coordinative training

Sixteen patients with progressive ataxia from cerebellar degeneration, four weeks of intensive coordinative training, an intra-individually controlled protocol in which each patient was their own control, with four assessments: eight weeks before, just before, just after, and eight weeks after the training. Result: significant improvement in motor performance and a reduction of ataxic symptoms on the clinical scales, maintained at the eight-week check11.

The authors conclude in terms worth quoting precisely: coordinative training improves motor performance and reduces symptoms, "enabling them to achieve personally meaningful goals of everyday life", and "for both groups, continuous training seems crucial to stabilise the improvements and should become a standard of care"11. Level of evidence stated by the review: class III.

Ilg 2010: durability at one year

The same team reassessed these patients one year after the intensive four-week block, which had been followed by a home exercise programme. The result is nuanced, and that is what makes it valuable: despite a progressive decline in motor performance and a progressive increase in ataxic symptoms due to disease progression, the improvements in performance and the gains in activities of daily living persisted12.

The disease went on advancing; the patient stayed above the trajectory they would have followed without training. That is the right mental model for talking to the patient: not "we are going to recover", but "we are going to shift the curve".

Miyai 2012: the randomised trial of intensive rehabilitation

Forty-two patients with pure cerebellar degeneration, randomised between an immediate group and a delayed-entry control group. The intervention: two hours of inpatient physiotherapy and occupational therapy a day on weekdays and one hour at the weekend, for four weeks, centred on coordination, balance and activities of daily living. Results21 :

• Significantly greater functional gains in the immediate group, on ataxia, gait speed and activities of daily living.
• The improvement in trunk ataxia was more marked than the improvement in limb ataxia.
• The gains on ataxia held at 12 weeks, those on gait at 24 weeks.
• At 24 weeks, 22 patients out of 42 still had at least one measure better than at inclusion; the authors note that functional status tended overall to return to baseline within 24 weeks.

It is this trial that the American Academy of Neurology retained, in its 2018 systematic review, to state that in patients with degenerative ataxias, four weeks of inpatient rehabilitation probably improves ataxia and function (a single class I study)29. That review now carries the label [RETIRED] in PubMed: the AAN has withdrawn it from its documents in force. It remains a valid historical source, it is no longer an active recommendation, and it would be misleading to cite it as one.

What the meta-analyses say

Three quantitative syntheses frame the question today, and they do not give the same figure, which is in itself a piece of information.

Milne 2017 reviewed 17 studies and 292 participants with genetic degenerative ataxias (148 dominant, 85 recessive), including 5 randomised trials, most of the work being of level III-3 or IV. Fifteen of the seventeen studies showed a statistically significant improvement on at least one measure of ataxia, function, gait or balance. The authors' conclusion: "there is consistent evidence that rehabilitation improves function, mobility, ataxia and balance". Major reservation, stated by them: fewer than half the studies assessed long-term outcomes23.

Winser 2023 meta-analysed 8 of the 26 studies included, with GRADE assessment. "Low" to "moderate" quality evidence in favour of therapeutic exercise for reducing disease severity as measured by the SARA: weighted mean difference −3.3 points (95 % CI −3.7 to −2.8; p < 0.01), and for improving balance as measured by the Berg scale: +2.6 points (95 % CI 1.1 to 4.2; p < 0.01). By contrast, the effect on functional independence is not significant : +1.6 points (95 % CI −1.5 to 4.6; p = 0.31)22.

Matsugi 2024 is the most recent and the most cautious. Eighteen randomised trials, 398 participants, of whom 315 were assessed with the SARA and included in the meta-analysis, with prior registration on the PROSPERO register. Overall effect of physiotherapy: −1.41 SARA points (95 % CI −2.16 to −0.66). The authors describe the risk of bias as "serious" and the certainty of the evidence as "low" to "very low", with high heterogeneity3.

The gap between Winser's −3.3 and Matsugi's −1.41 is not a contradiction: the populations, the inclusion criteria and the periods are not the same, and Matsugi restricted himself to randomised trials in cerebellar ataxia that is degenerative. The cautious reading is to keep the lowest and the most recent estimate (a gain of the order of 1.4 SARA points, with a low certainty of evidence).

The trial that was missing: Milne 2025

Published in Annals of Neurology, this multicentre, randomised, single-blind superiority trial is the most robust work available. Five Australian sites, participants with dominant or recessive hereditary ataxia, aged 15 years and over, randomised 1:1 between 6 weeks of outpatient physiotherapy followed by 24 weeks of a home exercise programme (n = 39) and continuation of usual activity (n = 37). Intention-to-treat analysis on 71 participants, with blinded assessors4.

• At 7 weeks : mFIM improved by 2.26 points (95 % CI 0.26 to 4.26; p = 0.028) and SARA by −1.21 points (95 % CI −2.32 to −0.11; p = 0.032).
• At 30 weeks : SARA held at −1.51 points (95 % CI −2.76 to −0.27; p = 0.017), but mFIM not significant (1.74; 95 % CI −0.32 to 3.81; p = 0.098).
• Frequent adverse effects in both groups : fatigue, pain and falls.

This last point is to be read carefully. Falls and fatigue are among the frequent adverse events on both sides (in the patients who were rehabilitated as much as in those who carried on with their usual activity). Rehabilitating an ataxic patient is not a risk-free activity, and saying so to the patient is part of the contract.

The result one would rather keep quiet

Seven years earlier, the same team had run a delayed-start randomised trial in Friedreich's ataxia: six weeks of outpatient rehabilitation followed by six weeks of a home programme. It included only 19 participants: of 159 people assessed for eligibility, 92 were excluded and 48 declined to take part. Result: no significant difference on the functional independence measure between the groups at six weeks (2.00 ± 3.16 against 0.56 ± 4.06). One quality-of-life subscale, the "body movement" domain of the Friedreich Ataxia Impact Scale, did by contrast improve significantly (p = 0.003). And above all: the within-group improvements obtained after the supervised phase were not maintained after the home programme phase39.

Two lessons follow, and neither is comfortable. The first concerns recruitment: when 140 people out of 159 do not take part, the trial population is not the practice population. The second concerns the home programme: prescribing it is not enough for it to preserve the gains. In the Ilg 2010 trial it preserved them over a year; in this one it did not preserve them over six weeks. The difference probably lies in the support given to the programme rather than in its content, and that is exactly where the physiotherapist's work is played out.

The comparison that gives it meaning

A gain of 1.4 SARA points, on its own, says nothing to anyone. Set against the speed at which the disease advances, it takes on immediate meaning. The European EUROSCA cohort followed 462 patients with at least one follow-up visit, over a median observation period of 49 months, and established an annual progression of the SARA score5 of 2.11 points for SCA1, 1.56 for SCA3, 1.49 for SCA2 and 0.80 for SCA6.

The right thing to say to the patient then becomes: "what we get through rehabilitation is roughly the equivalent of one year of progression of your disease. It does not stop it, and you will have to keep going to hold on to that gain." It is true, it is quantified, and it is more motivating than a vague promise.

Which modalities for which level of evidence?

Session time is a scarce and non-expandable resource. This chapter allocates it according to what has been measured, distinguishing what has a demonstrated effect, what has none, and what is still entirely unknown: three categories that are readily confused.

The effects by modality, as measured

The subgroup analysis of the Matsugi 2024 meta-analysis is the most directly usable document, because it compares the modalities with one another on the same outcome measure, the SARA3.

Three readings force themselves on you in front of this graph.

First, the three effective modalities are effective within a hair's breadth of one another : −1.65 for aerobic training, −1.59 for multi-component training, −1.58 for balance. None dominates the others. There is no superior "method" to be discovered, there is training time to be spent.

Second, the confidence interval for multi-component training is very wide (−5.15 to −0.03): the upper bound skims zero. The estimate is significant by a whisker, on five heterogeneous trials. It has to be treated as such.

Third, dual tasking is the only modality whose point estimate goes the wrong way (+0.24), on a single trial and with an interval that covers practically the whole imaginable spectrum (−6.4 to +6.88). That is not proof of ineffectiveness, it is a total absence of information. Faced with a patient who falls, devoting session time to an unevaluated modality when three evaluated ones exist is a choice you must be able to justify.

The modality by level of evidence table

The grading, in plain terms

The particular case of aerobic training

This is the surprise of the recent literature, and it deserves some space, because it is counter-intuitive: why would pedalling improve an ataxia?

A randomised trial, single-blind for the assessor, compared six months of home aerobic training (19 patients) with balance training (17 patients) in ambulant patients with cerebellar ataxia. Thirty-one participants completed the trial, adherence exceeded 70 %, and there were no serious adverse events related to the training. Mean improvement: 1.9 SARA points (standard deviation 1.62) in the aerobic group against 0.6 points (standard deviation 1.34) in the balance group. The authors conclude that a phase III trial is warranted32.

One must be careful not to over-interpret a feasibility trial of 36 randomised patients, and its authors are explicitly careful themselves. But two practical lessons stand up. The first: home aerobic training is feasible and safe in the ambulant ataxic patient, with high adherence. The second: it is the modality that costs the least in supervision and is the easiest to keep up over time, and keeping it up over time is what decides the result.

The modality that produces the best result is not the most sophisticated one: it is the one the patient will still be doing in six months.

And in children, and at advanced stages?

Two populations are regularly left aside, and two studies document them.

In children, a trial showed that coordinative training based on body-controlled video games improves ataxia in children with degenerative ataxia34. The 2014 consensus already relied on this result to state that coordinative training improves motor function in the adult as well as in the juvenile patient1.

At advanced stages, where people give up soonest, a rater-blinded, intra-individually controlled trial studied ten young subjects with advanced degenerative ataxia, unable to stand or barely able to do so, with twelve weeks of home training using body-controlled video games, structured in two six-week phases so as to fit the training to individual progression. Postural control improved33. Ten patients do not make a proof, but they are enough to forbid the sentence "at this stage there is nothing more to be done".

How do you build a programme, session by session?

The trials that worked share one architecture: a dense block that creates the gain, a home relay that preserves it. This chapter turns that architecture into real sessions, with the published doses and not invented orders of magnitude.

The two-stage architecture, as it was tested

This graph says one thing and one thing only, but it is decisive: the only protocol whose benefit can still be read at one year is the one that included a home programme12. The most intensive inpatient protocol, with no relay, saw functional status return towards its starting level within 24 weeks21. Time spent building the home programme is not time stolen from the exercises: it is what determines whether the exercises will have been of any use.

The published doses

The content of a typical session

No publication prescribes a session plan. What follows is a transposition of the content of the positive protocols: coordination, balance, gait training, strengthening, activities of daily living3 and it is flagged as such: this is a proposed way of organising a session, not a trial result.

1. Balance, in a real progression (15 to 20 minutes)

The progression follows the SARA's own scoring, which has the advantage of making progress measurable with the follow-up tool: feet naturally apart, then feet together, then tandem, then single leg. At each level, vary four variables and only one at a time: surface (firm then foam), vision (open, reduced, closed), task (static then dynamic), hand support (two hands, one hand, light touch, none).

One safety point that is not negotiable: in a patient 84 % of whose peers fall within the year26, balance exercise is done in a corner of the room, between parallel bars, or with a gait belt. Falls are among the frequent adverse events of the Milne 2025 trial4 ; they happen in the session too.

2. Gait and transfers (15 minutes)

The tandem is both the earliest test30 and an exercise that transfers directly. Add to it: walking with changes of direction, counted turns, stepping over low obstacles, stops on command, going up and down stairs. Since trunk ataxia is the component that responds best to intensive training21, sitting work without a backrest, on an unstable surface and then with upper limb movements, deserves a fixed place.

3. Limb coordination (10 minutes)

Tasks of the finger-to-nose type, finger chase, fast alternating movements, heel-shin slide, that is to say the very items of the SARA, worked as exercises. Progressive amplitude, progressive speed, then a precision constraint (targets, templates). Recall here the fundamental constraint: in the cerebellar patient, learning by fine error correction is the route that is damaged17. The instruction is better aimed at the whole task and its result than at segmental correction of the movement.

4. Aerobic endurance (15 to 20 minutes)

This is the part most often sacrificed, even though it carries one of the three significant effects of the meta-analysis3 and a direct trial gives it the advantage over balance training alone32. The equipment matters less than safety: stationary bike, arm ergometer, treadmill with support, supervised walking. In the patient who cannot stand, a seated ergometer remains within reach.

5. Building the home programme (5 to 10 minutes, every session)

Three to five exercises, written down, illustrated, doable without equipment and without any risk of an unsupervised fall. Since fatigue is a frequent adverse event in both arms of the Milne 2025 trial4, the home dose is negotiated with the patient rather than prescribed.

What belongs to other professionals, and must be referred on

The international consensus puts physiotherapy, occupational therapy and speech and language therapy on the same level1. Two referrals deserve to be offered systematically.

Speech and language therapy, for the dysarthria. The state of the evidence has to be presented honestly, though: the Cochrane review devoted to the treatment of speech disorders in Friedreich's ataxia and the other hereditary ataxias identified 14 trials and 721 participants, of which thirteen evaluated drug treatments ; no meta-analysis was possible, for want of two studies sharing the same assessment procedure44. Ataxic dysarthria remains a field without established evidence, which does not excuse you from referring, but does forbid you from promising.

Occupational therapy, for home adaptation and assistive equipment. It is the component that the Miyai 2012 trial never separates itself from21 : the protocol combined physiotherapy and occupational therapy, and the gains also bore on activities of daily living.

What do real clinical cases teach us?

Four published cases, all verifiable, chosen as much for what they contradict as for what they show. One ends better than predicted, another less well than hoped, and the third does not work at all, and that is the most useful one.

How do you apply this in the clinic?

This chapter is the operational translation of everything above: what you do at the first session, what you write in the record, what you say to the patient, and when you hand over.

The first session, in 45 minutes

1. Focused history (10 min). Date and mode of onset. Number of falls over three months, circumstances, consequences. Current medication, in particular antiepileptics, lithium, benzodiazepines. Alcohol intake. Family history of gait disorders. Whether genetic testing was done, and when: the answer "before 2023" has a value of its own.
2. Differential sorting (10 min). Cerebellar motor signs in the limbs and in speech; Romberg on a firm floor, eyes open then closed; distal deep sensation; a search for nystagmus. The sorting tree of chapter 2 unfolds in this order.
3. SARA (15 min). Rated in full, recorded item by item and not as an overall score: it is the split between posture and limbs that steers the programme.
4. Two functional measures (5 min). Tandem over ten steps, and one gait measure.
5. Written goals (5 min). Two or three, framed by the patient, in their own words, with an observable success criterion.

What to write in the record

The overall SARA score on its own cannot usefully be reread. What can be reread in six months: the date, the postural subtotal (gait + standing + sitting, out of 18) and the limb subtotal, the number of tandem steps, the number of falls in the quarter, the walking aid used, and progress on the goals. Six lines.

Three true things to say to the patient

• On the expected effect. "Rehabilitation does not cure the disease and does not stop it. What it does, as measured in the trials, is take back roughly the equivalent of one year of progression. That is little if you hope for a cure, and a great deal if you count in years of independence."
• On duration. "The benefit fades within a few months if you stop. The only published protocol whose effect can still be seen a year later included a programme to do at home."
• On risk. "We are going to work on balance, so at the limit of your stability. In the trials, falls and fatigue are among the frequent adverse effects. That is why we work in a space that has been made safe, and why I will ask you not to repeat the hardest exercises on your own."

When to hand over, and to whom

The five mistakes that keep coming back

• Concluding "positive Romberg" without having compared eyes open and eyes closed on a firm floor. It is the gap that informs, and foam asks a different question18.
• Treating a mixed ataxia as a pure cerebellar ataxia. The afferent component predicts a poorer response to coordinative training11 and calls for a substitution strategy on top.
• Devoting the session to limb coordination. It is the trunk that responds best to intensive training21, and it is gait that decides independence.
• Neglecting aerobic work. It is one of the three significant modalities in the meta-analysis3, and the easiest to carry on with at home.
• Stopping at the end of the prescribed sessions without having set up the home programme. Without a relay, the gain erodes within a few months21 ; with a relay, it holds at one year12.

Frequently asked questions

Elsewhere in neurology

This article treats cerebellar ataxia as a clinical entity and deals with its differential diagnosis. The neurological conditions within which an ataxia may sit, and the forms of vertigo it is confused with, are dealt with separately:

• Rehabilitation after a stroke (for ataxias of vascular origin, including cerebellar infarction).
• Multiple sclerosis: the physiotherapist's role (cerebellar involvement is one of the components of the picture there).
• Parkinson's and physiotherapy (instability of another mechanism, often confused with this one from a distance).
• Vestibular neuritis and benign paroxysmal positional vertigo (the vestibular side of the differential diagnosis).
• Peripheral neuropathy, particularly diabetic (the leading cause of proprioceptive ataxia in private practice).
• The Boubée scale (for rating sitting balance in other neurological contexts).

Bibliography: 48 verified references

1. Ilg W, Bastian AJ, Boesch S, et al. Consensus paper: management of degenerative cerebellar disorders. Cerebellum. 2014;13(2):248-268. PMID 24222635
2. Ruano L, Melo C, Silva MC, Coutinho P. The global epidemiology of hereditary ataxia and spastic paraplegia: a systematic review of prevalence studies. Neuroepidemiology. 2014;42(3):174-183. PMID 24603320
3. Matsugi A, Bando K, Kondo Y, et al. Effects of physiotherapy on degenerative cerebellar ataxia: a systematic review and meta-analysis. Front Neurol. 2024;15:1491142. PMID 39866519
4. Milne SC, Roberts M, Williams S, et al. Goal-Directed Rehabilitation Versus Standard Care for Individuals with Hereditary Cerebellar Ataxia: A Multicenter, Single-Blind, Randomized Controlled Superiority Trial. Ann Neurol. 2025;97(3):409-424. PMID 39520242
5. Jacobi H, du Montcel ST, Bauer P, et al. Long-term disease progression in spinocerebellar ataxia types 1, 2, 3, and 6: a longitudinal cohort study. Lancet Neurol. 2015;14(11):1101-1108. PMID 26377379
6. Perez-Lloret S, van de Warrenburg B, Rossi M, et al. Assessment of Ataxia Rating Scales and Cerebellar Functional Tests: Critique and Recommendations. Mov Disord. 2021;36(2):283-297. PMID 33022077
7. Manto M, Mitoma H, Burt AL, Schmahmann JD. Cerebellar clinical syndromes: the triad and rating scales. J Neurol. 2026;273(2):129. PMID 41663552
8. Hoche F, Guell X, Vangel MG, Sherman JC, Schmahmann JD. The cerebellar cognitive affective/Schmahmann syndrome scale. Brain. 2018;141(1):248-270. PMID 29206893
9. Marquer A, Barbieri G, Pérennou D. The assessment and treatment of postural disorders in cerebellar ataxia: a systematic review. Ann Phys Rehabil Med. 2014;57(2):67-78. PMID 24582474
10. Schmitz-Hübsch T, du Montcel ST, Baliko L, et al. Scale for the assessment and rating of ataxia: development of a new clinical scale. Neurology. 2006;66(11):1717-1720. PMID 16769946
11. Ilg W, Synofzik M, Brötz D, Burkard S, Giese MA, Schöls L. Intensive coordinative training improves motor performance in degenerative cerebellar disease. Neurology. 2009;73(22):1823-1830. PMID 19864636
12. Ilg W, Brötz D, Burkard S, Giese MA, Schöls L, Synofzik M. Long-term effects of coordinative training in degenerative cerebellar disease. Mov Disord. 2010;25(13):2239-2246. PMID 20737551
13. Schmahmann JD, Gardner R, MacMore J, Vangel MG. Development of a brief ataxia rating scale (BARS) based on a modified form of the ICARS. Mov Disord. 2009;24(12):1820-1828. PMID 19562773
14. Shaikh AG, Kim JS, Froment C, et al. Scale for Ocular motor Disorders in Ataxia (SODA). J Neurol Sci. 2022;443:120472. PMID 36403298
15. Argyropoulos GPD, van Dun K, Adamaszek M, et al. The Cerebellar Cognitive Affective/Schmahmann Syndrome: a Task Force Paper. Cerebellum. 2020;19(1):102-125. PMID 31522332
16. Ilg W, Golla H, Thier P, Giese MA. Specific influences of cerebellar dysfunctions on gait. Brain. 2007;130(Pt 3):786-798. PMID 17287287
17. Bastian AJ. Understanding sensorimotor adaptation and learning for rehabilitation. Curr Opin Neurol. 2008;21(6):628-633. PMID 18989103
18. Halmágyi GM, Curthoys IS. Vestibular contributions to the Romberg test: Testing semicircular canal and otolith function. Eur J Neurol. 2021;28(9):3211-3219. PMID 34160115
19. Kattah JC, Talkad AV, Wang DZ, Hsieh YH, Newman-Toker DE. HINTS to diagnose stroke in the acute vestibular syndrome: three-step bedside oculomotor examination more sensitive than early MRI diffusion-weighted imaging. Stroke. 2009;40(11):3504-3510. PMID 19762709
20. Ohle R, Montpellier RA, Marchadier V, et al. Can Emergency Physicians Accurately Rule Out a Central Cause of Vertigo Using the HINTS Examination? A Systematic Review and Meta-analysis. Acad Emerg Med. 2020;27(9):887-896. PMID 32167642
21. Miyai I, Ito M, Hattori N, et al. Cerebellar ataxia rehabilitation trial in degenerative cerebellar diseases. Neurorehabil Neural Repair. 2012;26(5):515-522. PMID 22140200
22. Winser S, Chan HK, Chen WK, et al. Effects of therapeutic exercise on disease severity, balance, and functional independence among individuals with cerebellar ataxia: A systematic review with meta-analysis. Physiother Theory Pract. 2023;39(7):1355-1375. PMID 35212247
23. Milne SC, Corben LA, Georgiou-Karistianis N, Delatycki MB, Yiu EM. Rehabilitation for Individuals With Genetic Degenerative Ataxia: A Systematic Review. Neurorehabil Neural Repair. 2017;31(7):609-622. PMID 28595509
24. Pellerin D, Danzi MC, Wilke C, et al. Deep Intronic FGF14 GAA Repeat Expansion in Late-Onset Cerebellar Ataxia. N Engl J Med. 2023;388(2):128-141. PMID 36516086
25. Wilke C, Pellerin D, Mengel D, et al. GAA-FGF14 ataxia (SCA27B): phenotypic profile, natural history progression and 4-aminopyridine treatment response. Brain. 2023;146(10):4144-4157. PMID 37165652
26. Fonteyn EM, Schmitz-Hübsch T, Verstappen CC, et al. Prospective analysis of falls in dominant ataxias. Eur Neurol. 2013;69(1):53-57. PMID 23146840
27. van de Warrenburg BPC, Steijns JAG, Munneke M, Kremer BPH, Bloem BR. Falls in degenerative cerebellar ataxias. Mov Disord. 2005;20(4):497-500. PMID 15645525
28. de Silva R, Greenfield J, Cook A, Bonney H, Vallortigara J, Hunt B, Giunti P. Guidelines on the diagnosis and management of the progressive ataxias. Orphanet J Rare Dis. 2019;14(1):51. PMID 30786918
29. Zesiewicz TA, Wilmot G, Kuo SH, et al. Comprehensive systematic review summary: Treatment of cerebellar motor dysfunction and ataxia [RETIRED]. Report of the Guideline Development, Dissemination, and Implementation Subcommittee of the American Academy of Neurology. Neurology. 2018;90(10):464-471. PMID 29440566 (document withdrawn by the AAN, cited here as a historical source).
30. Cabaraux P, Agrawal SK, Cai H, et al. Consensus Paper: Ataxic Gait. Cerebellum. 2023;22(3):394-430. PMID 35414041
31. Trouillas P, Takayanagi T, Hallett M, et al. International Cooperative Ataxia Rating Scale for pharmacological assessment of the cerebellar syndrome. J Neurol Sci. 1997;145(2):205-211. PMID 9094050
32. Barbuto S, Kuo SH, Winterbottom L, et al. Home Aerobic Training for Cerebellar Degenerative Diseases: a Randomized Controlled Trial. Cerebellum. 2023;22(2):272-281. PMID 35303255
33. Schatton C, Synofzik M, Fleszar Z, Giese MA, Schöls L, Ilg W. Individualized exergame training improves postural control in advanced degenerative spinocerebellar ataxia: A rater-blinded, intra-individually controlled trial. Parkinsonism Relat Disord. 2017;39:80-84. PMID 28365204
34. Ilg W, Schatton C, Schicks J, Giese MA, Schöls L, Synofzik M. Video game-based coordinative training improves ataxia in children with degenerative ataxia. Neurology. 2012;79(20):2056-2060. PMID 23115212
35. Benussi A, Dell'Era V, Cantoni V, et al. Cerebello-spinal tDCS in ataxia: A randomized, double-blind, sham-controlled, crossover trial. Neurology. 2018;91(12):e1090-e1101. PMID 30135258
36. Benussi A, Cantoni V, Manes M, et al. Motor and cognitive outcomes of cerebello-spinal stimulation in neurodegenerative ataxia. Brain. 2021;144(8):2310-2321. PMID 33950222
37. Serrao M, Casali C, Ranavolo A, et al. Use of dynamic movement orthoses to improve gait stability and trunk control in ataxic patients. Eur J Phys Rehabil Med. 2017;53(5):735-743. PMID 28627859
38. Bultmann U, Pierscianek D, Gizewski ER, et al. Functional recovery and rehabilitation of postural impairment and gait ataxia in patients with acute cerebellar stroke. Gait Posture. 2014;39(1):563-569. PMID 24119775
39. Milne SC, Corben LA, Roberts M, et al. Can rehabilitation improve the health and well-being in Friedreich's ataxia: a randomized controlled trial? Clin Rehabil. 2018;32(5):630-643. PMID 29072092
40. Cha JM, Kim HS. Functional Improvement after Taking Rehabilitation Program in Cerebellar Ataxia in a Patient with Systemic Lupus Erythematosus: a Case Report. Brain Neurorehabil. 2020;13(2):e11. PMID 36744189
41. Wilson CM, Mitchell CL, Hebert KM. Cerebellar Stroke Occupational Therapy and Physical Therapy Management from Intensive Care Unit to Outpatient: A Case Report. Cureus. 2017;9(12):e1949. PMID 29468104
42. Monteiro LHF, Rêgo INS, da Conceição ABS, et al. Physiotherapy in Spinocerebellar Ataxia Following COVID-19: A Biomechanical and Biopsychosocial Case Report. Physiother Res Int. 2026;31(1):e70122. PMID 41267344
43. Freund JE, Stetts DM. Use of trunk stabilization and locomotor training in an adult with cerebellar ataxia: a single system design. Physiother Theory Pract. 2010;26(7):447-458. PMID 20649489
44. Vogel AP, Folker J, Poole ML. Treatment for speech disorder in Friedreich ataxia and other hereditary ataxia syndromes. Cochrane Database Syst Rev. 2014;2014(10):CD008953. PMID 25348587
45. Sinha S, Kataria A, Kolla BP, Thusius N, Loukianova LL. Wernicke Encephalopathy — Clinical Pearls. Mayo Clin Proc. 2019;94(6):1065-1072. PMID 31171116
46. Edlow JA, Newman-Toker DE, Savitz SI. Diagnosis and initial management of cerebellar infarction. Lancet Neurol. 2008;7(10):951-964. PMID 18848314
47. McKeon A, Tracy J. Paraneoplastic movement disorders. Handb Clin Neurol. 2024;200:211-227. PMID 38494279
48. Mitoma H, Adhikari K, Aeschlimann D, et al. Consensus Paper: Neuroimmune Mechanisms of Cerebellar Ataxias. Cerebellum. 2016;15(2):213-232. PMID 25823827

Every reference in this article was checked one by one against the PubMed database on 14 August 2026: existence of the identifier, accuracy of the journal, the year and the pagination, and agreement between the claim cited and the actual content of the publication. The quantified estimates reproduce the values published by the authors, with their confidence intervals and the reservations they themselves state about the risk of bias and the certainty of the evidence.