Mitigation Strategies for Communities with Covid-19 Transmission in Lesotho Using Artificial Intelligence on Chest X-Rays and Novel Rapid Diagnostic Tests (MistraL)
Project partners developed and tested disease detection and control strategies specifically tailored to Lesotho. Following an initial disease screening and triage phase, individuals received a suite of integrated tuberculosis (TB) COVID-19 and human immunodeficiency virus (HIV) diagnostic services and were referred for treatment if needed.
The Need for Context-Specific Preparedness and Mitigation Strategies in Low- and Middle-income Countries
During the COVID-19 pandemic, a lack of healthcare resources including intensive care beds, ventilators, personal protective equipment and testing facilities in low- and middle-income countries created very challenging conditions. A further problem was the threat of disruption to essential disease control and prevention programmes such as those dedicated to human immunodeficiency virus (HIV) and tuberculosis (TB). Collateral damage to healthcare services such as these had been observed in previous epidemics, such as the Ebola virus outbreak of 2014 – 2015 in West Africa. Preparedness and mitigation strategies such as those involving testing, contact tracing, quarantine and triage must be adapted according to the context in which they are implemented. In low- and middle-income countries, limited resources and fragile healthcare systems often dictate what is feasible.
The MistraL research consortium, comprised partners from the Swiss Tropical and Public Health Institute, the Swiss non-profit organisation SolidarMed, Radboud University and the Foundation for Innovative New Diagnostics (FIND) – a global non-profit organisation. The aims of the project were first to develop disease detection and control strategies that were specifically tailored to benefit Lesotho, a lower middle-income country with a high prevalence of HIV and TB and a relatively weak health system capacity, and second to evaluate at that time novel diagnostic tools for COVID-19. The diagnostic tools were (i) an artificial intelligence (AI)-based software called CAD4COVID designed to provide automated evaluation of chest X-rays to signal where COVID-19 is a likely diagnosis, (ii) new rapid antigen COVID-19 tests and (iii) a novel bio-aerosol sample collection device.
Evaluating Diagnostic Tools
One of the diagnostic tools assessed as part of the integrated testing strategy was an artificial intelligence (AI) software tool called CAD4COVID. This tool is trained with patient data such that it can provide automated evaluation of chest X-rays to signal where COVID-19 or TB pneumonia is a likely diagnosis. CAD4COVID itself was a weak predictor of the early stages of COVID-19. However, the research team also tested the effects of training AI systems with additional data that can easily be obtained using inexpensive point-of-care blood tests. The diagnostic performance of AI tools was compared using combinations of X-ray data, white blood cell (immune cell) count and C-Reactive Protein (the concentration of which increases in response to inflammation) levels. It was found that an AI system which used only white blood cell counts was most effective but not dissimilar to systems which used a combination of data sources.
A further component of the project involved an evaluation of promising rapid antigen-based SARS-CoV-2 diagnostic tests. These tests offer significant advantages over polymerase chain reaction (PCR) tests, particularly in resource-constrained settings as they do not require specialised equipment and provide results quickly. However, they generally have lower accuracy rates. It was found that there was good agreement between the results obtained using rapid antigen tests which employed nasal and nasopharyngeal sampling methods. However, the tests’ sensitivity (the rate at which they accurately identified positive cases) did not reach the World Health Organisation’s minimum requirement. Promising results were obtained in a small-scale sub-study which tested a novel bio-aerosol sample collection device.
Promising Results and Further Research
This project provided compelling evidence that an integrated approach to testing for COVID-19, TB and HIV is feasible in routine programmatic settings at the district level in Lesotho; a finding which has potential relevance for future epidemic or pandemic situations and for other low- and middle-income countries. Indeed, the MistraL project led to a follow-up project, also funded by the BRCCH entitled ‘Improving Access to SARS-CoV-2 Screening and Testing through Community-based COVID-19 Case-Finding and the Use of Digital Solutions in Lesotho and Zambia’.
Banner image above: Digital X-ray results of the lung region of a patient. Software CAD4COVID uses artificial intelligence algorithms to produce a heatmap (on right) indicating regions of abnormality as well as a score (0-100) for suspicion of COVID-19.
