Leveraging AI in the prediction of COVID-19 progressionAtellica® COVID-19 Severity Algorithm1

What if artificial intelligence could be used to predict the likely progression to severe disease in COVID-19 patients?

Imagine the benefits of being able to more accurately evaluate the likelihood of potential disease progressions and severe illness in COVID-19 patients. If we could better predict the progression of the illness in individual patients and help identify those at greater risk for severe disease in advance of significant clinical progression, healthcare workers could better plan the resources needed to support these patients and help target and implement earlier more effective treatment plans. 

14-500 Clinical cases

A new collaboration to combat a novel virus

Through a collaboration with a number of leading healthcare institutions across the globe, we initiated a year-long project to explore the validity of this concept and to develop a predictive model based upon machine learning to test the concept. Armed with deidentified COVID-19 patient data from more than 14,500 COVID-19 patients from our research partners at Houston Methodist Hospital (TX, US), Emory University Healthcare (GA, US), and La Paz University Hospital (Spain), we conducted a retrospective analysis which included various clinical, demographic and laboratory data. Drawing from our expertise in molecular, hemostasis, hematology, chemistry, and immunoassay testing, we began to combine and review data from selected lab parameters and explored their potential interdependent relationships in developing a model capable of predicting the likely progression to severe disease and life-threatening multiorgan dysfunction in COVID-19 patients. Nine clinically significant lab parameters were identified and selected for inclusion in the algorithm. In addition to patient age: D-dimer, Lactate dehydrogenase (LDH), Lymphocyte %, Eosinophil %, Creatinine, C-reactive protein (CRP), Ferritin, PT-INR and Cardiac Troponin-I.

65 AI powered apps

Transforming data into a predictive tool to potentially support physicians in their critical decision-making

Siemens Healthineers owns more than 700 patent families related to machine learning and is a leader in the field of healthcare artificial intelligence (AI) and deep machine learning with more than 65 applications powered by AI. By integrating talent from across the broader Siemens Healthineers in vitro diagnostics and digital innovations segments, we aggregated and analyzed the various patient data, including patient progression to acute respiratory failure, end-stage organ damage, and 30-day in-hospital mortality. Then, we began to “train” an AI system to analyze this data and how these lab values and patient age may interrelate algorithmically in determining the likely progression to these severe disease outcomes. With each additional set of patient data, the emerging AI system recognized patterns and applied analysis to new patient profiles, automatically perceiving, and analyzing values and outcomes to generate probabilities. This system “training” resulted in the Atellica® COVID-19 Severity Algorithm1 becoming more accurate over time.

The Atellica COVID-19 Severity Algorithm is intended for educational purposes only. It is not for clinical or patient care, diagnosis, treatment, or to cure or prevent any disease. Availability varies by country.

The Atellica® COVID-19 Severity Algorithm1 is now being evaluated by a number of laboratories around the world to help assess the potential clinical benefits that artificial intelligence may be able to offer in helping clinicians manage treatment plans for COVID-19 patients.

Web app

Get access to a fully functional Educational Use Only version of the algorithm in which you can enter a potential patient’s lab values and age to generate a COVID-19 clinical severity score, including projected probability of ventilator use, end-stage organ damage, and 30-day in-hospital mortality.

The algorithm has been designed for use with the GOOGLE CHROME web browser on computers, laptops, and tablets. The algorithm has not been optimized for use on smartphones and is therefore not recommended.  

Related information

1