University of São Paulo: Application that predicts the evolution of covid cases begins to be tested in 14 Brazilian hospitals
An application capable of predicting the diagnosis and clinical evolution of patients who contracted covid-19 will be tested in the coming weeks by 85 doctors from 14 hospitals across Brazil, in a survey coordinated by the Faculty of Public Health (FSP) at USP. The tool, called RandomIA , has an algorithm that uses data from covid tests to calculate the odds of infection, hospitalization and death in hospital units, providing estimates that help doctors make decisions about how to deal with the disease. At the moment, the researchers adjust the application to the needs of doctors in each hospital to carry out tests with the system based on clinical information from the patients themselves.
“The application, which can be accessed on cell phones, was developed to allow the inclusion of any Artificial Intelligence algorithm and the first tests will be carried out with data from covid-19”, says Professor Alexandre Chiavegatto Filho, from USP , to Jornal da USP. FSP, which leads the research. “The algorithm is first trained with real data about the disease, from the hospital where it is tested, and then incorporated into the application. We are currently in the validation phase of the application structure according to the physicians’ preferences.”
Alexandre Chiavegatto – Photo: Personal archive
Initially, a draw will be held at each hospital to define which doctors will receive the results of the application and which will not. “Next, the prognoses of patients made by the doctors who received the results will be compared with the prognoses of the doctors who did not have access to the application”, observes the professor. “The comparison is similar to studies with drugs and vaccines, where one group of patients receives the vaccine or the drug and another receives an ineffective substance, that is, a placebo.”
The app uses data from patients on whom RT-PCR exams were performed, regardless of the result, positive or negative. “It is important to mention that no patient data is shared, not even with the research coordinating center, being used only locally by the physician to test the application”, explains Roberta Wichmann, who carries out post-doctoral studies at FSP and oversees the operations of the project. “After filling out all the fields by the physician, the prediction can be made for both diagnosis and prognosis of the disease. The diagnostic and prognostic results are presented via graphs, with their respective probabilities of infection and need for mechanical ventilation, ICU admission and death.”
Ease of use
Mobile applications have been used in the health area (e-health) for various purposes, including decision support by health professionals. “On the other hand, it is noted that the concern with the usability of e-health applications is still little explored and, in relation to applications that use Artificial Intelligence to support clinical decisions, the lack of scientific information is even greater”, says Roberta. “Soon, RandomIA was created to fill the existing gap in the area and will only be made available after usability validation steps (ease of use) and a clinical study.”
According to the FSP researcher, the application seeks to improve decisions on priorities related to covid-19. “This includes response time and prediction of the risk of a positive diagnosis and/or prognosis of the disease, using only data routinely collected in the hospital, as well as optimizing the response time to the pandemic, as the diagnostic and prognostic prediction is provided instantly , enabling the physician to make a quick decision”, he emphasizes. “Other possible benefits of the work include identifying contributions to improving the use of e-health applications and generating machine learning solutions to assist physicians and the scientific community in decision making.”
“For testing purposes, RandomIA will undergo a clinical trial with doctors from all over Brazil, which will provide scientific results that can demonstrate the benefits of using the application in the health network in general”, points out Roberta. “Through this trial, it will be possible to identify whether there is benefit from using the application in making medical decisions for each patient.”
“We are now starting the first phase of the study, the usability test, which is to find out what information doctors would like to receive from an Artificial Intelligence algorithm, in order to facilitate its use”, concludes the FSP professor. The research is led by FSP’s Laboratory of Big Data and Predictive Analysis in Health (LABDAPS), directed by Chiavegatto Filho. The operational part of the study is led by Roberta, and the application’s development was funded by the National Council for Scientific and Technological Development (CNPq) and by Microsoft. The project has the collaboration of 85 doctors from 14 hospitals located in the five regions of Brazil.