While there is still no approved drug treatment to reduce the main symptoms of autism, a study published in the journal Translational Psychiatry shows that artificial intelligence could be a game changer. By reanalyzing data from large clinical trials previously considered negative using machine learning algorithms, researchers show that, when certain clinical profiles are taken into account, treatment with bumetanide is significantly more effective than placebo in 30 to 40% of children. This innovative approach paves the way for more targeted clinical trials based on precision medicine and opens up a new therapeutic opportunity in a field that has so far lacked a drug solution.
Autism spectrum disorders (ASD) now affect more than 1.5% of children, a proportion that is trending toward 2%, according to the most recent epidemiological data. Despite this growing prevalence, no drug treatment has yet been validated to reduce the core symptoms of autism in children.
A new study, the result of a collaboration between Neurochlore, a research company dedicated to autism spectrum disorders, and B&A Biomedical, which specializes in the analysis of clinical data using artificial intelligence, and published in the journal Translational Psychiatry, now sheds new light on the evaluation of autism treatments. Using artificial intelligence, the researchers show that 30 to 40% of children included in large clinical trials that were considered negative actually showed significant improvement in their disorders.
Why do clinical trials fail in autism?
Autism covers a range of disorders, with highly heterogeneous profiles. Each child has a different combination of symptoms, which can affect, to varying degrees, their social interactions, communication, behavior, and adaptation, as well as their sensitivity and response to sensory stimuli.
To evaluate a potential treatment for autism, research relies on standardized clinical scales, such as CARS or SRS, which are based on the assessment of numerous cognitive and behavioral parameters. Efficacy is generally estimated by comparing the average scores obtained on these scales between a treated group and a control group (placebo). However, the therapeutic benefits observed in some patients may be diluted or even go unnoticed when analyzed across all patients studied, due to the high clinical heterogeneity of autism, which results in highly variable responses from one individual to another.
“The failure of a clinical trial does not necessarily mean that the treatment is ineffective. It may simply reflect the extreme diversity of clinical profiles in autism,” explains Prof. Yehezkel Ben-Ari, neurobiologist and corresponding author of the study.
The case of bumetanide: positive results… then failure
Prof. Ben-Ari and Dr. Éric Lemonnier showed that bumetanide (a drug already used for other indications) could reduce the severity of certain symptoms of autism. These results were confirmed in several Phase 2 clinical trials involving a total of more than 1,030 children treated worldwide. However, two large phase 3 clinical trials, conducted in children aged 2 to 17 over a period of six months in nearly fifty centers across thirteen countries in Europe, Australia, and Brazil, were unable to confirm these results.
Artificial intelligence changes the interpretation of data
Faced with this contradiction between the results of phases 2 and 3, the researchers revisited all the raw data from the phase 3 trials, but with a radically different approach.
Instead of considering all participants as a homogeneous group, they used an artificial intelligence algorithm (Q-Finder) capable of classifying participants into subgroups and examining the clinical response to treatment in each of the subgroups. The result: the analysis reveals that, in 30 to 40% of children included in these phase 3 trials, treatment with bumetanide is significantly more effective than the control group, effects that did not appear in the overall analysis.
“Our results confirm a simple idea: the same treatment cannot work for all children with autism. We need to identify, from the outset, the clinical profiles that can benefit from it,” emphasizes Prof. Ben-Ari.
Which children could benefit from the treatment?
The study shows that children with autism who are likely to respond better to treatment share certain identifiable clinical characteristics, already observed during consultations. In particular, they present:
- Moderate difficulty adapting to change, for example when an unexpected event disrupts the organization of their day;
- Repetitive behaviors, which may include a need for routines, repetition, or stable reference points in order to feel secure;
- Marked social difficulties, such as limited interaction with other children, difficult social communication, or a tendency to withdraw in certain situations.
The key finding of the study is that the combination of certain specific clinical traits can predict response to treatment. These criteria are already commonly used in assessments by healthcare professionals and do not require any additional examinations or genetic testing, but rather a more detailed reading of individual profiles.
Towards a more personalized approach to autism treatment
These results support a new approach: tailoring clinical trials, and in the future treatments, to children’s profiles, rather than seeking a one-size-fits-all solution. To confirm these data and take a decisive step forward, the researchers now want to launch a new targeted clinical study, including only children with profiles identified by artificial intelligence.
“This validation study is essential if we hope to eventually offer the first drug treatment for a clearly identified population of autistic children,” says Prof. Ben-Ari.
Next step: a validation study
The team is currently looking for a public or private industrial or financial partner to support this validation phase and the launch of a new clinical study.
Key points
- Autism affects nearly 2% of children
- No drug treatment is currently approved for its core symptoms
- Artificial intelligence reveals that, taking certain clinical profiles into account, treatment with bumetanide is significantly more effective than placebo in 30 to 40% of children.
- A personalized approach could pave the way for targeted treatment for an identified population of autistic children
To read publication: https://eproofing.springer.com/ePj/journals/cTjEa3xG8ZxNzPqzY8ySWkMvoLlF39bwPgZhontQAnkBxwgr7LiRCo8_mW5FXdpYxi0-_jv4LhwH7bmHpkeejvv6LsTHBPgBKs44o6S2wE0m2HqU1O17cTH9ktNSIx31
More informations: www.babiomedical.com/en
