A group of nephrologists from Fundació Puigvert, in collaboration with the technology company IOMED, has carried out research in which it has created an artificial intelligence (AI) system validated in 1.2 million patients, capable of tracking the clinical course to identify hereditary kidney diseases whose diagnosis may be delayed for years.
The work, presented in Murcia during the 56th National Congress of the Sociedad Española de Nefrología (S.E.N.), focuses on complex conditions that are difficult to recognize, since their first signs do not always have a direct connection with the kidney and may present themselves in isolation over time. According to the data presented, the average delay until reaching diagnosis is between five and seven years.
To address this, rather than limiting itself to reviewing specific symptoms or analytical parameters at a given point, the proposed model examines the patient’s entire clinical journey: what problems have emerged, in what sequence, and at what time. To do so, it integrates information from the electronic medical record, laboratory results, and pharmacological treatments. The algorithm has been applied to four hereditary kidney diseases: Alport syndrome, Fabry disease, primary hyperoxaluria, and tuberous sclerosis complex.
In order to evaluate the tool’s performance, the specialists carried out a retrospective validation in patients treated in different Spanish hospitals. In cases where diagnosis had already been established, subsequent clinical events after confirmation of the disease were removed, so that the system would have to identify the pattern solely from information recorded beforehand.
In a cohort of 72,611 people with chronic kidney disease (CKD), the AI has achieved a sensitivity of more than 84 percent across the four conditions analyzed—that is, it has made it possible to detect more than 84 out of every 100 affected patients for each of them. In addition, the authors estimate that it would be necessary to review 30 patients identified by the system to find a potential case, which must then undergo specific study to verify the diagnosis.
Complement to traditional clinical diagnosis
After highlighting that the main innovation of the study is to incorporate the temporal dimension into the analysis of clinical data, the researchers have detailed that the model has already begun its prospective evaluation in clinical practice, with genetic confirmation of suspected cases. In this way, although it is not intended to replace clinical or genetic diagnosis, the goal is "to function as a safety net that makes it possible to select, within very large populations, the people in whom it is worth investigating the possible existence of an inherited kidney disease," they have stated.
"This approach can help reduce diagnostic delay and enable earlier interventions," they have said, adding that "the system can start looking for patterns of a particular disease starting from initial groups of only 10 to 22 already diagnosed patients". "This capability is especially relevant in rare diseases, in which the small number of available cases makes it difficult to develop tools capable of identifying them in large populations," they have concluded.