COSMIC-19 [COVID-19]
Research type
Research Study
Full title
COntinuous Signs Monitoring In Covid-19 patients
IRAS ID
283425
Contact name
Fiona Thistlethwaite
Contact email
Sponsor organisation
The Christie NHS Foundation Trust
Clinicaltrials.gov Identifier
Duration of Study in the UK
0 years, 8 months, 1 days
Research summary
Summary of Research
This is a pilot study to assess whether artificial intelligence (AI) combined with continuous vital signs monitoring from CE marked wearable sensors can predict clinically relevant outcomes in patients with suspected or confirmed Covid-19 infection on general medical wards.\nPatients on general medical wards with COVID-19 infection considered to be at high risk of deterioration will be asked to wear vital signs sensors for the duration of their hospital stay. These sensors are an established method of recording patient vital signs and are CE marked. Patients enrolled in the study will continue to receive routine medical care as directed by their treating team.\n \nAll data recorded from the wearable sensors in this study will be analysed in conjunction with routine data collected during the patient’s treatment. Several models will be created using deep learning AI techniques with the aim of reliably predicting several important clinical outcomes.\nHigh quality predictive models rely on the data on which they are based and therefore as a secondary outcome we will report on the quality of the vital signs data recorded by the wearable sensors.\nThis study will identify whether continuous monitoring alone can improve identification of clinical deterioration compared to traditional vital signs and if the addition of AI technology / algorithms can provide an even earlier identification.
Summary of Results
The COVID-19 pandemic was an unprecedented challenge for the NHS. A significant number of people who caught COVID-19 became very unwell and needed admitting to hospital. It was important to monitor these inpatients carefully to assess whether they might need to be moved into the intensive care unit (ICU) where they could receive specialist help.
Wearable devices are a type of medical equipment that can be used to constantly (non-stop) monitor a patient’s vital signs. The COSMIC-19 study used wearable devices that can be stuck to a patient’s skin or worn like a wristwatch or a blood pressure cuff. Together they can measure pulse rate, breathing rate, temperature, oxygen levels and blood pressure. Below are pictures of the devices we used. The information from the devices is sent wirelessly to a digital tablet at the patient’s bedside. From the tablet it can then be sent to a central computer where it can be analysed.This pilot study was designed to test whether wearable devices are helpful to monitor patients with COVID-19 on the wards, whether patients find them comfortable enough to wear for long periods of time and whether the data that is gathered can be analysed using artificial intelligence (AI) techniques which may be able to predict which patients need transfer to the ICU.
Patients were recruited on admission to hospital, with suspected or confirmed COVID-19 infection. In total we enrolled 48 patients of whom 32 (67%) were men. On average patients were 51 years old and 28 out of 48 (58%) were of Black or Indian/Pakistani ethnicity. A quarter of the patients were diabetic and a quarter had high blood pressure.
All 48 patients agreed to wear the patch sensors and the wristwatch oxygen sensor but only 30 patients were willing to wear the blood pressure cuff. In total we were able to collect around 200 days’ worth of measurements from the patches and wristwatch device but only 72 days’ worth of data from the blood pressure cuff. On average patients wore the patch sensors and wristwatch oxygen sensor for 4 days but the blood pressure cuff was only worn for one day. From this we have been able to conclude that patients find an automatic blood pressure cuff much less acceptable than the other wearable sensors.During the study the wearable devices did not collect data 100% of the time. This is a common finding in wearable studies as the devices may fall off, patients may take them off, the batteries may need replacing and other technical problems can occur. On average we found that our devices recorded heart rate and breathing rate over 80% of the time, temperature over 90% of the time and oxygen levels over 60% of the time. There were more problems with blood pressure measurements which were captured around 30% of the time. Whilst the wearable sensors do not capture data perfectly, they do collect much more data than would be recorded by nurses on the ward who typically measure a patient’s vital signs only once every 4 to 8 hours.
Using pulse rate as an example, we found that only 1 in 5 patients would have a gap in their wearable measurements of over 4 hours during a 24-hour monitoring period. This is an important finding because it tells us that patch sensors can monitor patients reliably without requiring constant attention from staff working on a ward.
Comparing vital signs measured by wearable sensors and nurses
All patients in the study continued to have their vital signs measured by nurses on the ward. This allowed us to assess how the measurements made by the wearable sensors compared to those made by the nurses. We found that the average pulse rate difference between wearable measurements and nurse measurements was less than 1bpm (beats per minute) but sometimes the measurements disagreed by up to 20bpm. We found that the average breathing rate difference was 2 breaths/min but sometimes the measurements disagreed by up to 18 breaths/min. In the sickest patients we noticed that the wearable sensors often underestimated the breathing rate.These findings are important because they show us that wearable sensors measure vital signs differently to how nurses do it. Neither system is perfect because we already know that there are errors when nurses measure vital signs. The key point is that the two measurement methods are different so they can’t be used interchangeably. This means that more research is needed before we can roll out wearable sensors on all our hospital wards.
The wearable sensors captured a lot of information about our patients, much more than would normally be recorded on a hospital ward. Computers can analyse this information in much more detail than humans and can spot patterns that we might miss. We are using these so called “artificial intelligence algorithms” to see if we can predict the patients who needed intensive care by looking at their wearable sensor data. In total, 11 patients in our study needed to be admitted to the ICU and we noticed that they were much more likely to be of Indian/Pakistani ethnicity.
At the moment we are still refining our artificial intelligence algorithm but the preliminary results suggest that it can predict the patients who need intensive care with good accuracy. It is common to score such models on a scale from 0.5 to 1.0 (the “Area Under the Receiver Operating Curve”). A score of 0.5 would mean that our model is no better at predicting who needs intensive care than flipping a coin. A score of 1.0 would mean that our model predicts admission to intensive care perfectly. So far, our best models are scoring around 0.8. This result is better than some of tools that are in current use, many of which were created using data from lots more patients than we had in the study.
In summary our study has shown that some wearable sensors are well tolerated by hospital inpatients and they can collect large quantities of data without requiring lots of maintenance. There is a disagreement between measurements made by wearable sensors and nurses and this needs to be better understood. However, our early findings suggest that these sensors could be useful in helping to predict patients who need intensive care.
We appreciate the financial support from iMATCH (UKRI funded project) for this research. Finally, we would like to thank the participants for their contribution to this study.
REC name
Yorkshire & The Humber - Bradford Leeds Research Ethics Committee
REC reference
20/YH/0156
Date of REC Opinion
29 May 2020
REC opinion
Further Information Favourable Opinion