Artificial intelligence identifies people with COVID-19 who will become critically ill

Therapy Breakthroughs 5. okt 2020 6 min Clinical Associate Professor Eske Kvanner Aasvang, Groupleader, Associate Professor MSK, PhD Helge Bjarup Dissing Sørensen Written by Morten Busch

People hospitalized with COVID-19 can very rapidly change from stable improvement to extremely critical condition. This requires continuous monitoring of respiration and circulation since prompt treatment can be crucial to ensure survival. An intelligent wireless clinical support system for monitoring people with COVID-19 24/7 has now been developed. This monitoring can identify who will become critically ill.

Too few and too busy nurses and doctors combined with shortages of ventilators and personal protective equipment have cost many lives globally during the COVID-19 pandemic. The trajectory of the disease among individuals requires continuously monitoring vital parameters such as oxygen saturation , respiratory rate and circulation. Substantial deviation from the normal in one parameter does not necessarily mean danger. However, a combination of several moderate deviations may mean that the person is transitioning to a critical condition.

“Experienced doctors can identify these patterns, but we do not have enough doctors during a pandemic. We have now succeeded in developing and training an artificially intelligent monitoring system that can monitor people with COVID-19 24/7 and sound the alarm when the combination of vital signs is threatening. We have already observed the first 100 people with COVID-19 and are ready to implement the first version of the artificial intelligence–supported system now. In combination with new treatments such as remdesivir and corticosteroids, we hope to improve the survival rate and free up healthcare personnel to take care of those who are critically ill,” says Eske Kvanner Aasvang, Research Leader and Chief Physician , Department of Anaesthesiology, Rigshospitalet, Copenhagen and Clinical Associate Professor, Department of Clinical Medicine, University of Copenhagen.

Things can go wrong even when personnel do everything right

Eske Kvanner Aasvang is one of the three inventors of the multidisciplinary WARD project (Wireless Assessment of Respiratory and Circulatory Distress), which also includes Christian S. Meyhoff, Research Leader and Chief Physician, Department of Anaesthesia and Intensive Care, Bispebjerg Hospital and Clinical Associate Professor, Department of Clinical Medicine, University of Copenhagen, and Helge B.S. Sørensen, Biomedical Signal Processing & AI Research group leader, Associate Professor, Department of Health Technology, Technical University of Denmark, Kongens Lyngby. The project was actually being developed to enable monitoring of circulation and respiration after major surgery or hospitalization for severe medical disease, such as cancer and chronic obstructive pulmonary disease (COPD).

In recent years, the WARD project therefore developed a wireless monitoring system with various sensors that measure more than 10 important vital signs among inpatients 24/7 and connects this to AI to detect complications earlier than through standard intermittent manual observations .

“We are focusing intensely on building next-generation monitoring outside the post-anaesthesia and intensive care units. The idea is to replace how we monitor patients today: depending on a nurse having the time and equipment to do this, with many hours of no observation. This is why we developed the WARD system to enable early detection of illness and for early intervention, thereby reducing the frequency and severity of complications,” explains Eske Kvanner Aasvang.

Even if all the current standard procedures are followed during surgery, up to 30% of the patients may encounter complications with current post-surgery hospital routines such as checking the patients every 8 hours, and much can happen between these observations. The new WARD system can identify these complications and act immediately.

“Over time, artificial intelligence may even be able to predict complications, so that the person’s condition does not worsen, avoiding expensive treatment, prolonged hospitalization and permanent harm. We hope to spare the patient from discomfort and harm after surgery and to spare society from unnecessary expense,” says Eske Kvanner Aasvang.

Almost ready when COVID-19 struck

In the first phase of the WARD project, the researchers developed a system with wireless sensors that sends data to a cloud computer with artificial intelligence algorithms to eliminate false alarms and calculate whether to alert hospital personnel about critical conditions through mobile devices.

“WARD originally focused on patients undergoing surgery, but we also had a large parallel study monitoring patients hospitalized for severe deterioration in COPD. When COVID-19 struck Denmark back in March, we three initiators looked at each other and said that we think we have the ideal project to help COVID-19 patients and also help our colleagues in healthcare,” explains Eske Kvanner Aasvang.

When the COVID-19 pandemic began, the WARD system had already been tested on 700 inpatients. The researchers were therefore immediately able to use the equipment to monitor COVID-19 patients, and even if a nurse or doctor is not present the system can alert personnel that an adverse event is developing.

“This is especially critical in COVID-19, in which many people deteriorate acutely, rapidly and unpredictably. Since some of the vital signs that need to be monitored such as blood oxygen saturation and respiration and heart rates are the same, we adapted the system to COVID-19 in just a few weeks,” says Eske Kvanner Aasvang.

Doctors and nurses decide

The sensor system was developed based on current clinical practice, in which people with COVID-19 who are at high risk are measured manually a few times during a day. The WARD system monitors these patients through the sensors and then sends the raw data wirelessly to a cloud computer that can analyse them, inspired by the way doctors analyse the information on patients.

“We asked doctors to scientifically describe how they analyse the information they receive and identify complications. Then we developed mathematical models that automatically estimate these complications and their early biomarkers,” explains Helge B.D. Sørensen.

The close collaboration with doctors and health professionals has been key to developing the system. In fact, according to Helge B.D. Sørensen, the whole basis for the research to succeed is taking the existing challenges of the health sector as the starting-point. Then the researchers can try to develop new, innovative solutions.

“Doctors and nurses are assessing and informing us about their requirements for the user interface on their smartphone, and this is a good example of the current collaboration. They have been fully involved in the user interface, and we have given them free rein to tell us how they want it designed, because we need to ensure that they want to use it, and we tailor it as much as possible to what they want,” says Helge B.D. Sørensen.

Very confused situation

Eske Kvanner Aasvang compares the current system of periodic monitoring with a blindfolded airplane flight in which the passenger only looks down rarely and randomly.

“The new system keeps our eyes open all the time and alerts a nurse or doctor that they need to treat the very serious complications among people with COVID-19. This system is reliable enough that we could theoretically send many of the less severely ill patients home, because we monitor and hospitalize them if they show signs of becoming critically ill,” explains Eske Kvanner Aasvang.

“The WARD system also has the great benefit of eliminating the need for physical contact between patients and healthcare personnel. We can monitor remotely, reducing the risk of infecting the personnel. Further, personal protective equipment, which was in short supply in the early days of the pandemic, does not need to be used to the same extent,” he adds.

Although the WARD system was developed for a different purpose, it is ready to be rolled out for the current and future viral pandemics. So far, the WARD system has monitored 100 COVID-19 patients through August, and the data together with that from future patients will be used to train the AI specifically to target COVID-19 patients.

“This will also free up doctors and nurses to carry on with their real core task – treating the patients who are severely ill. With the development of new types of drugs such as remdesivir, which must be administered just before a patient enters the most critical condition to be effective, the WARD system can identify the patients who can benefit from the medicine and tell us to give it to them at just the right time,” says Eske Kvanner Aasvang. “An experienced doctor is required to discover the same patterns the WARD system can identify, and once we have trained the system, we may be able to detect patterns we have not identified previously that outperform the manual assessment of even the best doctors, so attention can be focused on the right patients.”

Making medication more effective

The researchers tested the system on the people who are most severely ill with COVID-19 and found that about 70% have had very prolonged and frequent episodes with reduced blood oxygen saturation during a day.

“These can last for up to 45 minutes with oxygen saturation so low that this may affect other organ systems, especially the heart but also the brain. These patients are very confused when they are admitted because their brains lack oxygen. Now we can detect these episodes immediately and can intervene more rapidly,” explains Eske Kvanner Aasvang.

The researchers are evaluating the trajectory of COVID-19 among people monitored by the WARD system versus standard care. The data show that the system potentially can shorten the duration of severe desaturation among very severely ill patients.

“Some patients can manage with a little extra oxygen delivered through their nose. Some patients improve with lung physiotherapy, but some will also need a ventilator. The personnel must decide what to do, but a system like ours ultimately incorporates real-time dedicated machine learning,” explains Eske Kvanner Aasvang.

Based on the system’s great potential for detecting complications among people with COVID-19, Denmark’s administrative regions have decided to provide DKK 5.5 million for the project to purchase initial hardware for all the regions, and the current capacity enables monitoring of 250 patients simultaneously, but the system is fully scalable and could potentially monitor a limitless number of patients.

“We believe that there is a huge unfulfilled need to take this to the next generation. The fact that we still monitor and measure like we did in the early 1980s is quite old-fashioned. Phones became wireless many years ago, so it is strange that we are still using old technology in hospitals. Wireless sensors are the future, and integrating wireless sensors, artificial intelligence and this very clinical focus will certainly save many people’s lives,” concludes Eske Kvanner Aasvang.

Eske Kvanner Aasvang, Christian S. Meyhoff and Helge B.D. Sørensen received DKK 3.9 million for the project COVID-19 WARD from the Novo Nordisk Foundation’s emergency coronavirus initiative to mitigate the adverse health effects of the COVID-19 pandemic. In 2018, Innovation Fund Denmark invested DKK 18 million in the WARD project, which is also supported by Denmark’s administrative regions, The Danish Cancer Society, Steno Diabetes Centers, Copenhagen Center for Health Technology (CACHET), Radiometer Medical ApS, A.P. Møller Foundation and internal funding.

Department of Clinical Medicine's research is carried out at research centres and hospitals where our staff have their principal place of employment....

Helge B.D. Sørensen (HBDS) conducts research on biomedical signal processing for better solutions for individuals concerning monitoring, diagnostics,...

© All rights reserved, Sciencenews 2020