Artificial Intelligence
Even before the COVID-19 pandemic hit health systems worldwide, hopes were high that the widespread development and deployment of AI within healthcare could help overstretched care providers through the development of new drugs, the optimisation of data and information flows and the personalised and timely delivery of care.11 With the pandemic in full swing, it is timely to reflect on how AI can help (or has helped) health systems to manage the crisis and to consider the role of AI as countries prepare for a potential second wave of infections linked to coronavirus.
At the outset of the current crisis, innovative AI-based analysis of social media data and news reports helped to predict the spread of the outbreak. Canadian company Blue Dot is credited with being first to recognise an unusual cluster of pneumonia cases in Wuhan before official sources confirmed this as COVID-19. Large amounts of data can be gathered and aggregated quickly from a range of sources, such as Twitter, Facebook, local news outlets and public health statistics to reconstruct and then potentially predict the spread and the behaviour of the COVID-19 outbreak. These early successes at modelling and predicting disease behaviour are encouraging, but questions need to be asked about the reliability and quality of the data that go into the AI.
Social media analysis could potentially be triangulated further with mobile phone data that capture people’s movements to give a real-time prediction of risk and disease spread. Such tracing of movement could support the public with complying more easily with social distancing by being routed away from crowded areas. Apple and Google have formed a partnership to develop an app to support contact tracing. This app takes a decentralised approach, where data are stored locally on each person’s phone. In the UK, NHSX has rejected this partnership’s design and opted for the development of a proprietary app where data will be held centrally on NHS servers.12 This raises ethical and privacy concerns, in particular, around the potential for data sharing beyond the immediate COVID-19 pandemic. There are also uncertainties about the actual utility of contact tracing due to the lack of adequate, validated risk models and due to the need to ensure widespread use of the app within the population.
Babylon Health, already a controversial player in the AI healthcare market prior to COVID-19, extended its symptom-checking app with a specific COVID-19 decision algorithm that might help with supporting patients in getting better and more targeted advice. This could potentially reduce unnecessary attendances at emergency departments and community walk-in centres. However, there is as yet no rigorous evidence available.
A key strength and application area of AI has been imaging and diagnostics, and this is something that could be put to good use during the pandemic. For example, a Chinese team trained a deep learning neural network to identify COVID-19 from chest X-rays and to distinguish this from other forms of pneumonia.13 If applied successfully in clinical practice, such an AI-supported approach could help protect healthcare staff and speed up the process of isolating and potentially tracking patients. However, care needs to be taken with results reported at this early stage. A review of 31 diagnostic and prediction models found that all of the models were at high risk of introducing bias and that the accuracy and performance estimates were likely to be overly optimistic.14 In order to speed up training of algorithms and to enhance their performance, shared data repositories should be built up globally, and the transparency of reporting needs to be enhanced.
Lastly, AI has the potential to support the treatment of COVID-19 through the development of new drugs and the redeployment of existing drugs. For example, large numbers of research papers accessible through the COVID-19 Open Research Database can be analysed quickly using machine learning to extract relevant knowledge about drugs that might be beneficial for the treatment of COVID-19. AI has also been used in the race for the development of vaccines and treatments. Hong Kong-based company Insilico Medicine reported that it had developed six new molecules that could potentially halt viral replication.
AI has potential to help health systems to fight COVID-19 through these initiatives around predicting and reducing spread, and by supporting diagnosis and treatment. There are open questions about data quality, transferability of results across settings and health systems, the performance of algorithms when actually used in clinical systems, and about access to data and protection of privacy. The crisis provides us with an opportunity to gain a glimpse of the future and to ponder these questions.