Evidence
Alert

New tool can predict the risks of surgery for people with COVID-19

A new tool, called CovidSurg, can predict the risks of surgery for people who develop COVID-19. It gives a score based on readily available information, and could help patients and professionals discuss and manage the likely risks. The tool, which was developed using machine learning, is free for surgical teams around the world to use. 

The COVID-19 pandemic has led to many surgical operations being cancelled. As a result, people are going untreated for longer, and may be more severely ill by the time of their surgery. But people who have COVID-19 around the time of surgery have higher rates of death than those without the virus. They also have more lung-related complications afterwards. A tool to predict each patient’s risk has been urgently needed. 

New research shows that the CovidSurg tool can estimate the risks of surgery for individuals who have COVID-19. Risks are based on readily available information such as the patient’s age, and whether they have received respiratory support (been on a ventilator).

The work can help identify an individual patient’s risk of death should they have surgery. This will help them make more informed decisions about whether to go ahead with the procedure.  It can also help staff to be more aware of the potential risks for each patient and to plan the care they are likely to need after surgery.  

What’s the issue?

Millions of surgical procedures have been cancelled since the start of the COVID-19 pandemic. This can lead to people’s conditions getting worse. To treat their condition, and to improve their quality of life, surgery needs to restart. This will also reduce the future impact on the NHS of needing to treat many people with worsened conditions.

People who develop COVID-19 around the time of their surgery are more likely than people without the infection to have complications and to die. Some surgery is not urgent. Previous studies estimated that between 1% – 9% patients having non-emergency surgery have COVID-19. More than half (51%) of these patients develop lung-related complications. Almost one in four (24%) die around the time of surgery.

To restart surgery, people need to understand the risks should they contract COVID-19. Now, researchers have developed sophisticated software that recognises patterns in past data and adapts (machine learning). They used the software tool to give people needing surgery, personalised risk scores.

The researchers hope this work will make healthcare professionals and patients feel more informed about the risks of surgery. They believe this work is another step towards more personalised discussions around surgical risk.

What’s new?

This study was conducted in 756 hospitals in 69 countries. It included 8492 patients undergoing any type of surgery. All patients had contracted COVID-19 around the time of surgery (in the 7 days before or the 30 days afterwards). Most (81%) surgeries were emergencies. More than half (57%) were for conditions other than cancer. The most common procedures were abdominal (41%), orthopaedic, for example of bones or joints (34%), and head and neck (10%).

The researchers listed 16 factors that might increase patients’ risk. They developed a model (using data from one group of patients) to explore which factors actually did increase their risk following surgery. They used data from another group of the patients in the study to check (validate) the findings. 

Five factors (of the original 16) predicted patients’ risk of poor outcomes:

    • patient’s age, with people over 70 most at risk 
    • needing respiratory support such as ventilation before surgery
    • score on the American Society of Anesthesiology (ASA) scale (a subjective assessment of a patient’s overall health, made by the anaesthetist, from completely fit to someone expected to die within 24 hours) 
    • type of surgery 
    • score on the Revised Cardiac Risk Index (an estimate of the risk of cardiac complications based on a patient’s history of, for example, heart disease).

The researchers looked at 26 combinations of the 5 factors. They tested each combination with 3 different types of algorithm to discover which model best predicted risk. The team found that only 4 of the factors were needed to predict a patient’s risk:

    • age
    • ASA score
    • revised cardiac risk index
    • preoperative respiratory support.

The CovidSurg model had similar accuracy to other predictive models used in healthcare. However, unlike other models, it uses current data to predict the risk future patients face. It is based on information that is easy to collect from patients. The research team hopes that the model will help patients and professionals understand each patient’s risk when discussing surgery.

Why is this important?

Understanding the risks relating to COVID-19 around the time of surgery will be important for the foreseeable future. Vaccination programmes could take years to deliver, especially in low- and middle-income countries. COVID-19-free surgical pathways are not likely to be available everywhere, especially for emergencies. 

A model to predict risk will enable patients and professionals around the world to discuss surgery. If an individual’s risks are too high, surgery could be delayed. If surgery is needed urgently, the tool allows an informed discussion with the patient about the risks. Critical care following surgery can be planned.

The CovidSurg Mortality Score tool can also identify people at low risk, whose surgeries can proceed as planned. It uses information that is available in all resource settings. 

Decisions to undertake surgery always balance risks and benefits. This tool gives more information to guide full discussions with patients and inform the management of their care. The researchers say their work is a step towards more individualised care for patients.

What’s next?

The research team has made the predictive tool freely available for any hospital to download and use.

Good internet access is needed to use the tool which may be a barrier to some hospitals accessing it. However, collaborators in low- and middle-income countries have started using the tool . The researchers say it is simple and quick to use. They have raised awareness of the model in the global community and are expecting it to be used globally.

While the tool is valid and usable now, the researchers would welcome other researchers to test it further. 

You may be interested to read

This NIHR Alert is based on: CovidSurg Collaborative. Machine learning risk prediction of mortality for patients undergoing surgery with perioperative SARS-CoV-2: the COVIDSurg mortality score. British Journal of Surgery 2021;108:11

Another paper in the same series (CovidSurg) detailing more about the methods used: COVIDSurg Collaborative. Mortality and pulmonary complications in patients undergoing surgery with perioperative SARS-CoV-2 infection: an international cohort study. The Lancet 2020;396:27-38 

CovidSurg: information from GlobalSurg & The NIHR Global Health Research Unit on Global Surgery about the impact of COVID-19 in surgical patients and services.

CovidSurg Risk Stratifier: information on the project from Health Data Research UK.

 

Funding: This research was funded by a NIHR Global Health Research Unit Grant. 

Conflicts of Interest: The study authors declare no conflicts of interest.

Disclaimer: NIHR Alerts are not a substitute for professional medical advice. They provide information about research which is funded or supported by the NIHR. Please note that views expressed in NIHR Alerts are those of the author(s) and reviewer(s) and not necessarily those of the NHS, the NIHR or the Department of Health and Social Care.

Commentaries

Study author

We were not surprised that the tool is accurate. But we hadn’t expected to find that so few factors could predict someone’s risk. In fact, the four that emerged from this research are very easy to find out from patients or their records, so this outcome is better than we thought. 

We need to be able to personalise people’s treatment. Our main finding is that surgeons can use this tool to predict risk of surgery for their patients. This means we can use resources as effectively as possible.

Elizabeth Li, NIHR Doctoral Research Fellow, Surgical registrar, University of Birmingham 

Member of the public 

At my age, you never know when you might need an operation! I think this paper is highly relevant to surgeons and patients about to undergo surgery, including emergency interventions.

The long list of collaborators suggests that the findings will be widely accepted. Many countries were involved and I think the findings will be implemented where the data were collected. Differences in national settings and practices could have a big influence and I wonder whether the model will be used in the UK.

As well as the intrinsic merits of this research, this paper will help establish that there is a rich field for machine learning to exploit in health, yielding future patient benefit.

John Walsh, Public Contributor, Swindon 

Researcher in the same field 

The paper presents a novel analysis and is a unique dataset allowing assessment of a considerable number of COVID-19 surgical patients. Currently, no other such dataset exists.

Further work is needed to build on this work. Risk needs to be communicated to patients and decision-making needs to be joint. Individual level factors need to be considered such as the type of surgery and how urgent it is. Clinicians discussing with patients need to make clear that the model gives an indication of risk. It does not give a direct prediction.

There was missing data so the algorithm is partly based on estimates. Using different estimates, the same patient factors would give a different prediction of risk. The authors acknowledge that there are likely to be differences across study sites due to patient factors, the available resources and systems and so on.

More work is needed to independently validate this tool. Plus, over time, there will be changes in variants of the virus, and changes in clinical options.

Yize Wan, NIHR Clinical Lecturer in Intensive Care Medicine and Anaesthesia, Queen Mary University of London and Barts Health NHS Trust, London