A machine learning algorithm developed by researchers can assist in identifying individuals who are most vulnerable to contracting COVID-19 while being treated in a hospital.
In a research, the artificial intelligence (AI) tool had an accuracy rate of 87% in identifying patients who were at a high risk of acquiring COVID-19. Researchers at Imperial College London and the Infection Prevention and Control (IPC) division of Imperial College Healthcare NHS Trust created the technology.
The programme was created by researchers utilising typical patient and hospital data. They instructed it to recognise COVID-19 risk indicators such age, gender, interactions with other infectious patients, bed locations, and patient movement patterns within the hospital.
Predicting which hospital patients are likely to contract an infection can assist stop the spread of the infection to other patients and personnel.
Data from patients admitted to hospitals run by Imperial College Healthcare during the first two waves of COVID-19 infection were used to test the tool, and data from Geneva University Hospitals was used to validate it. The study, which was released in the Lancet Digital Health, is the first to successfully estimate the probability of COVID-19 emerging in a hospital setting using patient contact information.
The Medical Research Foundation and the Imperial Biomedical Research Centre of the National Institute for Health and Care Research (NIHR Imperial BRC) provided funding for the study. It made use of information from the Clinical Analytics, Research and Evaluation (iCARE) environment at Imperial College Healthcare, which makes routinely gathered, anonymized healthcare data available for research with immediate patient benefits.
tool for surveillance
To identify patients who are most likely to contract COVID-19 and to take action to prevent negative patient outcomes, predictive models must be developed.
Mathematics Department of Ashleigh Myall at Imperial College London
The research team thinks this technique could be used to treat additional illnesses that some hospitalised patients may be at risk of getting, like Clostridium difficile (C difficile), a kind of bacteria that can cause diarrhoea.
Ashleigh Myall, the study’s lead author from Imperial College London’s Department of Mathematics, said:
“During the COVID-19 epidemic, several individuals contracted the infection while they were being treated in hospitals. To identify patients who are most likely to contract COVID-19 and to take action to prevent negative patient outcomes, predictive models must be developed. This can be done in addition to the usual precautions to reduce outbreaks and new transmissions.
“We have developed a machine learning system that, with up to 87 percent accuracy, can identify patients who are most at risk of contracting COVID-19. This paradigm could be incorporated into a variety of surveillance systems to improve infection control, prevention, and management efforts, particularly in the winter when COVID-19 infections are more likely to spread.
The training and testing phase took place before the UK’s vaccination rollout, so the study has limitations in that it does not consider a person’s vaccination status when estimating risk. Our approach, nevertheless, is quite predictive and applicable to other contagious viruses.
Operational head for antimicrobial stewardship, surveillance, and epidemiology at Imperial College Healthcare NHS Trust and research co-author Sid Mookerjee said:
The COVID-19 pandemic has motivated academics and physicians to devote more time, resources, and labour to addressing existing and upcoming pandemics.
“A much sought-after clinical solution is accurately forecasting patients’ risk of contracting infections like COVID-19, C difficile, and other infectious disorders.
“With this work, we provide a novel and highly predictive method for identifying patients who are at risk of contracting infections while receiving hospital care. This can support the development of safe and efficient patient care pathways and bed management, enabling the provision of top-notch healthcare.
specifics of the research
It is generally known that COVID-19 infections can spread within healthcare facilities. According to reports, 12–15% of all COVID-19 cases in healthcare settings and up to 16.2% during the pandemic’s heights are COVID-19 infections that arise after hospital admission.
In the past, it has been common practise to predict infections that may arise in healthcare settings by identifying risk factors like age, gender identity, comorbidities, and the length of the patient’s stay; this practise has not taken into account patient contacts, locations, or patient flow through hospitals.
The spread of HCAIs depends heavily on the patient’s contacts, which can vary, despite the fact that these approaches alone can identify predicted risk factors of HCAIs quite effectively.
The group aimed to determine whether patient risk for HOCI could be predicted using patient contact information using a machine learning method. In order to estimate patient risk of HOCIs, the team integrated patient contacts data based on bed allocation with clinical and hospital data from the iCARE system, which is supported by the NIHR Imperial BRC.
HOCIs are characterised as infections in patients who tested positive for SARS-CoV-2, a coronavirus type that produces COVID-19, three days or more after admission. Patient contacts were defined as patients who shared a room, ward, or building on the same day, regardless of the use of PPE or environmental ventilation as COVID-19 preventive measures.
Testing
The tool was unable to take into consideration how common national IPC practises including the use of PPE, ventilation, hand hygiene, and cleaning reduce patients’ risk of contracting COVID-19. The study was carried out prior to the widespread national rollout of COVID-19 immunisation, which significantly lowers the chance of contracting COVID-19.
The system analyses patient hospital data and risk factors related to COVID-19 infections to produce a predicted risk score between 0 and 1.
During the first two UK COVID-19 surges, the model was tested using information from 51,157 patients at Imperial College Healthcare NHS Trust hospitals (March to May 2020 and September 2020 to April 2021). During this time, 3749 patients tested positive for COVID-19 three or more days after being admitted to the hospital, and 87% of these cases were predicted correctly by the machine learning system.
The team contrasted this with a control group that had negative COVID-19 tests. After that, they used data from an external dataset from Geneva University Hospitals from 2021 to validate the system.
