COVID19-Osakidetza

By project TITTAN

Published on 14 July 2022

Spain

País Vasco

About this good practice

COVID19-Osakidetza is a prospective cohort study recruiting COVID-19 patients. This practice is about the process of how performing a collection of information to establish a baseline of sociodemographic characteristics, comorbidities and associated medications, vital signs, treatment received and lab test results of COVID-19 patients from baseline to discharge. This databse compiled, requirement of the intensive ventilatory support or invasive mechanical ventilation, and/or admission to a critical care unit and/or death during hospitalization. A curated Catboost model was able to predict the need of external ventilation of the patients. Based on 1568 patients included in the derivation cohort and 956 in the (external) validation cohort, the model was able to find the strongest predictors of prognosis; i) arterial blood oxygen pressure, ii) followed by age, iii) levels of several markers of inflammation (procalcitonin, LDH, CRP) and iv) alterations in blood count and coagulation, among other. The model predicted the progression to a score of 5 or higher on the WHO Clinical Progression Scale before patients required mechanical ventilation. This practice underlines not only the importance of systemic data collection but also the impact of a clinical validation of this type of computer models using the information included in the electronic health report. The process allowed the discovery of useful information for screening COVID19 patients to predict their evolution.

Resources needed

Electronic health report is essential to get the information which is used to train the machine-learning models. This requires of a close collaboration between several Departments in a healthcare system. In addition, expert research in machine learning and computer sciences should be involved.

Evidence of success

This practice represents a way to validate a machine-learning based prediction model with excellent performance properties to be implemented in electronic health report. Not only the goal is to predict progression considering the WHO Clinical Progression Scale prior to patients requiring mechanical ventilation, but also how to perform the process to find and validate these outcomes. Future steps are required, such as validation of the model in other settings and cohorts with clinical practice.

Potential for learning or transfer

The process demonstrated in this experience is useful to show the stepwise approach and to underline the importance of the need of a validation in different population cohorts before applying it in the clinic. However, this requires enough patients and independent validation process in different cohorts. Nevertheless, the procedure explained tackle the steps required to reach the success in the development of a machine learning model for the patient classification to predict their prognosis. Therefore, this information can be used for other regions to implement a stepwise approach.