Despite considerable efforts and substantial investment, there are still no effective treatments for many neurological and psychiatric diseases, including autism and other developmental syndromes. There are 2 main reasons for this
I) These syndromes are born in utero, which modifies the pathogenesis of the syndrome. As I have already explained, the triggering element will modify the maturative sequences and result in alterations to the construction of the brain well before birth. The concept of Neuroarchaeology that I developed 17 years ago translates into sets of displaced and poorly connected neurons that remain ‘immature’, generating activities that disrupt brain function. This probably also explains the heterogeneity of autism and of syndromes with a significant genetic component, such as Fragile X, Rett and even Trisomy 21.
II) This heterogeneity is also at the root of clinical trial failures. As a reminder, a drug is marketed after a sequence consisting of phase 1 – testing the safety of the treatment, phase 2 – a limited trial to test its efficacy and phase 3 – a very large trial in several countries to validate its efficacy. However, for autism, 9 phase 2 trials based on our work have been successfully carried out, showing that Bumetanide reduces the severity of autism (more than 1,000 children worldwide have been successfully treated). Despite this, phase 3 failed, preventing further development.
Based on the idea that the heterogeneity of the syndrome could be at the root of this failure, we used an AI-type approach (or rather Machine Learning – ML) to re-analyse the phase 3 data. It should be noted that in this phase, a single diagnostic scale had to be improved compared with the placebo. This scale is the average of many parameters (more than 60 for the SRS), which weakens the conclusions of this trial because an average is never just an average! We therefore re-analysed all the phase 3 data – by which we mean all the subscales of the clinical criteria – in order to test the hypothesis that some of these parameters could be attenuated by the treatment. And indeed, we can identify 20 to 40% of children who respond to the treatment. This is very promising, as it is likely to reduce the incidence of autism in many children (incidence close to 2%). This conclusion is also valid for many other syndromes/diseases, including those with a major genetic component. Large clinical trials fail, even though phase 2 trials were successful, because these syndromes are heterogeneous, contrary to popular belief. It is likely that these same limitations will be observed for syndromes such as Fragile X, Rett, and even Down syndrome, as has already been the case for Fragile X syndrome.
However, it should be remembered that any ML-type study must be validated to avoid the results being correlative, due to ‘chance’ as it were. We therefore need to carry out a new phase 2/3 with inclusion criteria based on the parameters we have identified. The problem is financial, given that this study will cost around €3-4 million, which we need to raise. If our trial validates our ML data, the next step is almost guaranteed: a phase 3 trial will be successful, enabling us to have the 1st treatment for autism. That’s what I’m working towards.
The lesson here is that we need to learn from our failures, as we do everywhere else in science. It is going to become essential to identify classes of syndromes among patients in order to focus clinical trials on them, and this is where AI is going to prove essential, making it possible to rapidly analyse large cohorts of data and break down a syndrome into a series of syndromes sharing the same sensitivity to a particular treatment. Claiming that a single drug will be effective in treating all autistic children (or others) is a pipe dream and/or a promise that will not be kept. But managing to treat a proportion of patients is not a bad goal in itself, and for those involved it is a great source of support. Here, too, a certain amount of humility is called for, as is the case with everything to do with neurological and psychiatric illnesses.
