Mobility patterns dominate explainable machine learning forecasts of
When a neurologist asks a patient to walk heel-to-toe down a hospital hallway, draw a spiral with a pen, or repeat a phrase so the doctor can judge whether their speech is slurred, the score is supposed to measure just one thing: how badly the cerebellum is working. The cerebellum is the part of the brain that helps control balance and movement. But a new article argues that this assumption may not hold true in many parts of the world. Writing in Frontiers in Neurology, a team of Indian specialists in movement disorders says that two widely used scales for doctors to rate ataxia may not measure the condition equally well across populations with different languages, education levels and cultures, even after careful translation. Ataxia is a condition that affects coordination and balance. The two scales are the Scale for the Assessment and Rating of Ataxia (SARA) and the International Cooperative Ataxia Rating Scale (ICARS). The authors work at the All India Institute of Medical Sciences in New Delhi, the National Institute of Mental Health and Neurosciences, and St. John’s Medical College in Bangalore. They wrote on behalf of the Movement Disorders Society of India National Ataxia Network. They introduce the idea of “contextual validity” to explain what translation alone cannot achieve: making sure a test measures the same thing in different settings.
This issue now matters greatly for medical research and treatment. Inherited forms of cerebellar ataxia are rare individually, but together they affect an estimated one in every 10,000 people worldwide. New treatments that may change how these diseases progress, along with clinical trials targeting specific genes, have made standardized ways of measuring outcomes essential for developing drugs. SARA was developed in 2006 at University Hospital Bonn in Germany. ICARS was created in 1997 by the Ataxia Neuropharmacology Committee of the World Federation of Neurology, with its main development center in Lyon, France. These are the most widely adopted and thoroughly tested tools used in ataxia research. Both measure problems with walking, standing, limb coordination, eye movements and speech by watching patients perform tasks. Because doctors observe what patients do instead of relying on what patients say about their own symptoms, these scales are often considered more objective and less affected by culture than questionnaires completed by patients. The new analysis challenges this belief at its core.
The authors carefully reviewed published translations and adaptations of SARA and ICARS from Europe, South America, North America and Asia. These included a Russian version published in 2026 and a Japanese version of SARA from 2009. They found a similar pattern in almost every case: teams translated the original text into another language and back again, asked experts to review it, and checked whether the wording, meaning and cultural references were equivalent. Some teams made instructions clearer, used wording that did not assume a person's gender, or suggested different cutoff scores for different ages. The Korean ICARS team also noticed that the size of the spiral used in a drawing task was not specified, but did not correct this issue. Most importantly, the review found little evidence that teams had made meaningful changes to the actual tasks or testing methods to account for differences in people's backgrounds beyond language. In other words, the scales have generally been taken around the world with their original tasks, testing rules and hidden cultural assumptions left in place.
Why is this a problem? The authors explain that a patient's score depends on the interaction between the examiner, the patient and the testing environment. It does not reflect brain function alone. Tasks such as walking, standing, following a moving finger, sliding the heel along the shin, quickly turning the hands back and forth, and speaking all depend on whether patients understand standard spoken instructions, know how to perform the activity, can follow a demonstration, feel motivated and communicate in ways that match the examiner's expectations. In clinics, doctors often need to translate instructions as they speak, explain them in different words or demonstrate the task. These changes can affect how a patient understands or performs the task, creating differences in scores that have nothing to do with the cerebellum. Even a perfectly translated scale may fail to measure the same thing in different settings if familiarity with the task or the way the examiner interacts with the patient regularly changes the results.
India, which the authors use as an example, shows how large this problem can be. Indian census data list 19,569 reported mother tongues, grouped into 1,369 language varieties, 121 major languages and 22 constitutionally recognized scheduled languages. At major specialist hospitals, patients and examiners often do not speak the same first language. As a result, an examiner may have to translate the original English instructions on the spot. The words and phrases used can differ between examiners and hospitals. Testing speech creates an additional challenge. SARA and ICARS rate how smoothly people speak, how clearly they form sounds and how they produce their voices. But accents, pronunciation, speech rhythm and articulation vary widely across India, sometimes even between nearby areas. An examiner who is unfamiliar with a patient's regional way of speaking might mistake normal dialect features for dysarthria, a speech problem caused by difficulty controlling the muscles used for speaking. This could confuse a person's lifelong language habits with a neurological problem.
Differences in education create another challenge. According to India's Periodic Labour Force Survey 2023-24, about one in five people aged seven and older, or 19.1 percent, cannot read and write. The rate is nearly twice as high among women, at 25.2 percent, as among men, at 12.8 percent. The PLFS 2025 reports that 32.2 percent of people aged 15 and older have not completed secondary education. Education levels also differ sharply between rural and urban areas and between men and women. India's National Family Health Survey 2023-24 finds that 53.6 percent of women and 45.4 percent of men have not completed ten or more years of schooling. These figures matter because some tasks on ataxia scales assume that people have learned certain skills through formal education. For example, the spiral-drawing task in ICARS assumes that people are familiar with drawing shapes on paper with a pen. People with little schooling, those who do manual work or those from rural backgrounds may have had few chances to practice this skill. As smartphone use increases, some groups may also have less experience with pen-and-paper tasks. A poor spiral drawing might therefore reflect a lack of familiarity with the task rather than a problem with limb coordination.
Cultural practices can also affect what happens during an examination. India has a documented diversity of 4,635 distinct communities, and many people continue to wear clothing such as sarees, mekhela chadors, dhotis, lungis, mundus, ghagras and burqas. During tests of walking, standing and moving the heel toward the knee, these garments may affect stride length, leg movement or balance even when the cerebellum is working normally. Some patients may also prefer not to change their traditional clothing for personal, cultural or religious reasons. Neither SARA nor ICARS gives clear instructions about clothing or footwear. Yet walking barefoot, wearing slippers or wearing shoes can change the way someone walks and maintains balance. Practical conditions at hospitals matter too. Walking heel-to-toe on a polished floor that a patient does not know, performing an examination in a crowded outpatient area without enough space for a ten-meter walk, or testing eye movements when the scales do not specify the target's position, speed, distance or lighting can all make results less consistent between hospitals in multi-center studies.
These problems could have serious effects on clinical trials. Researchers increasingly use SARA as a main or secondary measure in studies that track how inherited ataxias develop over time and in trials of new treatments. Such studies assume that the same score means the same level of impairment for every participant, whatever their background. If a person's background or testing conditions regularly affect how they perform a task, differences between hospitals or countries might reflect measurement errors rather than real biological differences. This could distort estimates of how quickly a disease progresses, how well a treatment works, how many participants a trial needs and whether a change in score is important for patients. The authors stress that there is no need to go back and change established outcome measures automatically. Instead, researchers can study performance on individual test items, examine differences between centers, use standard instruction scripts in several languages, train and certify examiners, and make testing procedures consistent. These steps can reduce many sources of variation in ongoing studies without changing the scales themselves.
The authors also set out a broad plan for improving how these scales are adapted for different cultures. They argue that reliable adaptation must go beyond translation to include interviews that check how people understand the questions or instructions, small pilot studies and thorough testing of how well the adapted scale works. This process should follow established guidance such as COSMIN, a framework for evaluating measurement tools used in health research. An adapted scale should show that different examiners, and the same examiner on different occasions, can produce reliable results. It should also measure what it is meant to measure, agree with relevant established measures, detect changes in a patient's condition and measure the same underlying problem as the original scale. The authors recommend bringing together several hospitals and research centers, with neurologists, rehabilitation specialists, psychologists, language experts and patient representatives working together. They also suggest that multilingual AI tools could provide standardized instructions in different languages while doctors continue to score patients. In the future, AI-assisted speech analysis might measure features of the voice and timing of speech to help distinguish neurological dysarthria from differences between dialects. However, the authors emphasize that these tools would need careful testing across languages, dialects, education levels and recording conditions. Wearable sensors and digital measures collected at home and work could add to clinic-based assessments and reduce the effects of conditions specific to a clinic. Researchers should also test whether measurements are equivalent both for the full scale and for each individual task, because tasks that depend on spoken instructions are likely to be more affected by context than simple movement tasks.
The authors' main message is that neurological outcome measures should measure problems with the cerebellum, not differences in language, education, culture or life experience. Although their argument focuses on India, the country is a useful example because it has a large population affected by these conditions, growing access to genetic and molecular diagnosis, and increasing participation in international research. The same principles apply wherever neurological scales administered by clinicians are used across different languages and cultures. They also apply to tests based on physical performance in other neurological conditions. The authors conclude that adapting a scale for different cultures should mean more than making its language familiar. The goal is to ensure that the same level of neurological impairment produces the same measurement, no matter where or under what conditions the patient is assessed. This is essential for fair patient care, reliable research across multiple centers and a fair assessment of the treatments now being developed for ataxia.
Subject of Research: Cross-cultural validity of ataxia rating scales administered by clinicians
Article Title: Are clinician-administered ataxia scales truly culture-free? Rethinking contextual validity in diverse populations
Article References: Narzari, H., Yadav, R., Srivastava, A. K., Pal, P. K., & Garg, D. (2026). Are clinician-administered ataxia scales truly culture-free? Rethinking contextual validity in diverse populations. Frontiers in Neurology, 17, Article 1991845. https://doi.org/10.3389/fneur.2026.1991845
Image Credits: AI-generated
DOI: 10.3389/fneur.2026.1991845
Keywords: ataxia, SARA, ICARS, cerebellar ataxia, contextual validity, cross-cultural adaptation, clinical outcome measures, psychometric validation, measurement invariance, India, clinical trials, movement disorders