Researchers from the Hospital Clínic-IDIBAPS coordinated a study that made it possible to determine with an accuracy of 96% whether a kidney transplant patient admitted to hospital with COVID-19 would need an ICU bed and/or more intensive treatment. This was achieved thanks to an algorithm with variables corresponding to these patients, which were collected in accident and emergency departments. This aspect was critical during the first wave of the pandemic, which is when the prediction method was developed.

The study was led by Dr. Ignacio Revuelta, nephrologist, head of the living donor kidney transplant programme at the Hospital Clínic de Barcelona and leader of the IDIBAPS
Translational research in post-transplant neoplasia research group. The study was published in a journal considered a point of reference in the field of artificial intelligence (Artificial Intelligence Review), and was carried out in collaboration with professionals from the Hospital Clínic-IDIBAPS responsible for dealing with infections in immunosuppressed patients (Dr. Moreno and Dr. Bodro), and from the Free University of Bozen-Bolzano (Prof. Dr. Santos-Arteaga) and the University of Trento (Prof. Dr. Di Caprio).

By applying a prior algorithm optimization phase (Data Envelopment Analysis [DEA]) for the subsequent development of an Artificial Neural Network (ANN) to the prediction model, it was possible to increase the efficiency of the data collected in the accident and emergency departments corresponding to kidney transplants patients admitted with COVID-19. This increased the sensitivity of the model and thus made it possible to work with smaller cohorts, with various outputs, allowing for the application of more dynamic and efficient Artificial Intelligence. This aspect is of key importance when, in new situations that have a significant impact on health, information is required to make decisions without having to wait until a large amount of data are obtained.

The predictions, which are made using the aforementioned hybrid prediction model, were validated using a battery of multiple Machine Learning techniques. Although, before the prior optimization, these techniques could provide predictions with an accuracy of around 80%, they were far lower than with our model, which achieved an accuracy of up to 96%.

The study developed a prediction model applied to hospital admissions in a cohort of kidney transplant patients who were admitted with COVID-19. This model allowed us to predict the clinical course of the disease, which enabled us to identify those patients in danger of progressing to a severe stage of the disease. During the first wave of COVID-19, the Hospital Clínic Nephrology and Kidney Transplant Service used a telematics system to monitor the patients, most of them, over 1,006 patients, 38 of whom had to be admitted to the Hospital Clínic due to COVID-19 between 3 March and 24 April 2020.

The development allows the patients' progress to be categorized using the values from the analyses carried out when they are admitted to hospital. The prediction model can help guide the management of COVID-19 in this group of patients with specific characteristics, through the identification of key predictors that permit a management of resources in a patient-centred model. In short, the study enables the optimization and design of algorithms for making predictions on the evolution of this type of patient with an accuracy of 96% and for predicting whether these patients will have to be admitted to intensive care units or require an intensification of treatment during the hospitalization, a critical aspect during the first wave of the pandemic.

Moreover, this study allowed for the identification of new indicators, collected when kidney transplant patients infected with COVID-19 were admitted to hospital, in order to predict their evolution, which had not been identified with the application of classical statistics.


According to Dr. Ignacio Revuelta, “in an overwhelming demand scenario, as in the case of the SARS-CoV-2 pandemic, the pressure on healthcare systems may exceed the latter’s predicted capacity for coping with such extreme situations. Healthcare systems need scientific evidence and validated predictive models to improve the management of patients and optimize resources. Moreover, it is essential to be able to develop agile models that can give us precise information quickly, because any delay in obtaining this information could have a negative impact." “In short, we are moving towards personalized predictive medicine," concludes Dr. Revuelta.

Ultimately, the study allows for the extrapolation of the potential evolution of patients in order to improve the efficiency of patient management processes and the allocation of resources in hospitals.

Link to the study

Subscribe to Directory
Write an Article

Recent News

El diagnóstico genético neonatal mejor...

Un estudio con datos de los últimos 35 años, ind...

Más de 1.500 cambios epigenéticos en e...

Un equipo de investigadores de la Universidad Juli...

Tuneable reverse photochromes in the sol...

A new technique allows the design of solid materia...

Highlight

Eosinófilos. ¿Qué significa tener val...

by Labo'Life

​En nuestro post hablamos sobre este interesante tipo de célula del...

Un estudio de INCLIVA muestra el efecto ...

by INCLIVA

Han desarrollado un estudio para evaluar la correlación entre el teji...

Photos Stream