The study, published in The British Journal of Psychiatry, proposes a clear, easy-to-interpret, and useful predictive tool for clinical practice, which in terms of performance surpasses a diverse set of complex artificial intelligence algorithms.
This research was motivated by a clinical need: despite advances made in the treatment of psychosis, between 20% and 30% of patients do not respond well to medication and fewer than 40% achieve remission of their symptoms. Currently, when a person arrives at a medical facility with their first episode, clinical teams do not have at their disposal reliable tools for knowing which patients will require a more intense intervention and which will evolve favourably in a short time.
To address this challenge, Sergi Mas is leading FarmaPRED-PEP, a multicentre personalised and precision medicine project coordinated by IDIBAPS. The group analysed data from two large Spanish cohorts covering a total of over one thousand patients, and incorporated 47 clinical and cognitive variables and 87 genetic markers. Based on this information, they compared traditional statistical models with artificial intelligence techniques. The result was clear: the best model was also the simplest. Just six variables, which can be routinely collected in clinical practice, are sufficient for predicting early recovery: duration of untreated psychosis, days of prior treatment, level of functioning, insight, executive function, and cognitive reserve.
As Sergi Mas explains, "sometimes what we need in the clinic are not opaque algorithms, but transparent tools that are easy to interpret. Our model shows that the information that we already collect routinely can be very powerful in anticipating the patient's evolution and bringing us closer to the application of precision psychiatry.”
The model performed better than more sophisticated automatic options. Not even genetic markers, a highly promising field, improved predictive capacity. In fact, these biomarkers still present problems of generalisation between cohorts and need more research before they can be clinically useful.
This study’s principal contribution is allowing us to anticipate what type of intervention each person will need. Patients at high risk of not recovering quickly could benefit from a more intensive approach from the outset: more psychotherapy sessions, closer pharmacological follow-up, or specific cognitive rehabilitation programmes. On the other hand, those with a good prognosis could actively participate in such significant decisions as the rate at which treatment is reduced or the intensity of follow-up.
For Laura Julià, first-named author of the study and researcher at IDIBAPS, "this model provides a practical tool that can help personalise care from day one. Identifying those patients who may have a more complicated evolution allows us to optimise resources and offer them the support that they really need.”
With the aim of facilitating the validation and use of this model everywhere, the team has made the complete code of the analysis public and has developed an online tool that allows the six variables to be entered in order to obtain an individual prediction of recovery.
