Maxime Taquet : L'IA au service de la santé mentale: entre individus et collectifs скачать в хорошем качестве

Maxime Taquet : L'IA au service de la santé mentale: entre individus et collectifs

Maxime TAQUET (Oxford University) The application of artificial intelligence to large-scale electronic health records (EHR) covering millions of patients has accelerated research into brain health. In this talk, I will present three applications of EHRs for individual and collective brain health: drug repurposing, biomarker discovery, and risk prediction. First, EHRs can support the repurposing of drugs originally developed for other conditions. Leveraging a natural experiment created by the rapid transition from live to recombinant shingles vaccines in the USA, we found the recombinant vaccine was associated with a 17% increase in time lived dementia-free. A similar protective effect was observed for the recombinant RSV vaccine, which uses the same adjuvant, suggesting a potential neuroprotective role for the adjuvant itself. In an emulated target trial, we identified that prucalopride, a 5-HT4 receptor agonist used for constipation, was associated with reduced risk of depression. Additionally, we showed that the GLP-1 receptor agonist semaglutide (‘Ozempic’) was linked to lower risks of substance use disorders and cognitive impairment. These studies help support the prioritisation of candidate drugs for repurposing. Second, EHRs can help validate biomarkers. We identified elevated fibrinogen and D-dimer (relative to CRP) during acute COVID-19 as predictors of later cognitive deficits. Using EHR data, we confirmed this association in real-world settings and showed that elevated fibrinogen predicts cognitive issues even outside the context of COVID-19, suggesting shared pathways with other illnesses. Third, EHRs can enhance risk prediction. By operationalising “clinical instability” (i.e. fluctuations in the patient’s illness severity) using mental health records, we found that it predicts hospitalisation across psychiatric conditions independently from and as strongly as clinical severity. We also developed a score to predict treatment-resistant psychosis based on symptoms at first presentation, which generalised across 10 NHS mental health trusts. Together, these applications illustrate how EHRs can accelerate discovery and translational progress in brain health research.