Despite a growing body of research on the subject, long-term Covid is still difficult to diagnose. It can cause a large number of very diverse symptoms (more than fifty), the intensity of which varies over time. And while a large number of patients hospitalized with covid develop these symptoms (up to 50%, according to a recent Chinese study), patients who have never had covid also appear to develop them (admittedly to a lesser degree). ). as shown by this American preprint. So symptoms alone are not enough to identify a case of prolonged Covid. This is problematic, especially for identifying study participants for this disease, which is necessary to find effective treatments. In an article published in The Lancet Digital Health, researchers from the University of Colorado (USA) propose a new tool to facilitate this long-term detection of Covid patients – an artificial intelligence that analyzes all the medical data of patients before and after coronavirus infection and who manages to detect long-term cases of Covid with a high probability of success.
Big data chasing long Covid
The researchers trained their AI model on demographic and clinical data from 597 patients seen at three long-term Covid clinics in the US. Their model learned to identify these long Covid patients by comparing medical records before and after they had Covid-19. In this way, he determined whether these data had changed after infection: new drugs, more frequent visits to the doctor, new symptoms, etc. As a control, the model used data from almost 100,000 patients with covid from the cities where these long covid were located. -clinics, of which 19,000 were hospitalized with COVID-19 and 78,000 were not hospitalized.
Despite a growing body of research on the subject, long-term Covid is still difficult to diagnose. It can cause a large number of very diverse symptoms (more than fifty), the intensity of which varies over time. And while a large number of patients hospitalized with covid develop these symptoms (up to 50%, according to a recent Chinese study), patients who have never had covid also appear to develop them (admittedly to a lesser degree). ). as shown by this American preprint. So symptoms alone are not enough to identify a case of prolonged Covid. This is problematic, especially for identifying study participants for this disease, which is necessary to find effective treatments. In an article published in The Lancet Digital Health, researchers from the University of Colorado (USA) propose a new tool to facilitate this long-term detection of Covid patients – an artificial intelligence that analyzes all the medical data of patients before and after coronavirus infection and who manages to detect long-term cases of Covid with a high probability of success.
Big data chasing long Covid
The researchers trained their AI model on demographic and clinical data from 597 patients seen at three long-term Covid clinics in the US. Their model learned to identify these long Covid patients by comparing medical records before and after they had Covid-19. In this way, he determined whether these data had changed after infection: new drugs, more frequent visits to the doctor, new symptoms, etc. As a control, the model used data from almost 100,000 patients with covid from the cities where these long covid were located. -clinics, of which 19,000 were hospitalized with COVID-19 and 78,000 were not hospitalized.
In summary, the model identified the most predictive factors for long-term Covid, including respiratory symptoms or treatments for those symptoms; certain non-respiratory symptoms characteristic of long-term Covid, such as cardiovascular problems or sleep disturbances; and risk factors such as diabetes or kidney disease. Finally, the model was tested on a fourth city cohort with a long-term covid clinic (more than 30,000 patients, including 125 with long-term covid). The model correctly identified long-term Covid patients with an 82% success rate.
A tool to advance research on this chronic disease
After the model was validated, it was tested on a cohort of nearly 2 million people over the age of 18 who had COVID at least 90 days prior to the analysis date. The medical data of these patients were collected in the N3C cohort database, created in collaboration with several health institutes in the United States, which, under the auspices of the American Institute of Health (NIH), pooled clinical data from their COVID-19 patients. such studies. According to an NIH press release, this approach would identify more than 100,000 people in this cohort with long-term Covid.
With this new tool, researchers hope they can quickly identify people who could be involved in research on long-term Covid and its possible treatments. In particular, to be able to classify them into long-term Covid subphenotypes, with their own symptoms and requiring special treatment. At the same time, it is clarified that their approach still has limitations: for example, people who do not have access to the healthcare system are not represented, as well as patients who were infected at the beginning of the pandemic and have no evidence of their infection. But this artificial intelligence, which will improve with time and the number of people analyzed, can certainly speed up research on the long Covid, which is still a mystery without a cure.