Go back

Digital Health and AI in the Time of COVID-19

Digital Health and AI in the Time of COVID-19

This pandemic —and its associated strain on the healthcare system— is happening at a time of technological optimism and promise, in part through the digitalization of health data. Learn more!

By Dr. Mireia Crispin, Director of the “Future of Healthcare” program at the IE Center for the Governance of Change and Borysiewicz Fellow at the University of Cambridge.

 


 

At the time of writing, the COVID-19 pandemic has claimed close to 200,000 lives worldwide, infecting more than 2.5M people at an exponential rate. The high number of patients in critical condition has saturated intensive care units in the major outbreak areas, and doctors from all specialties —and, in some cases, even students and retired professionals— have been enlisted to assist. With a few counted exceptions, the response from most countries to the spread of the virus has relied on centuries-old measures including isolation, quarantines, and masks. While our powerlessness is humbling, and highlights how far we still need to go in our understanding of infectious diseases, the differences with respect to historical precedents are just as important. One of the reasons governments have taken such drastic measures is the confidence that, sooner rather than later, treatments will be improved, immunity tests developed, and a vaccine found.  The worldwide scientific community has united in an unprecedented manner, volunteering time, skill and effort to progress to this point as rapidly as possible.

This pandemic —and its associated strain on the healthcare system— is happening at a time of technological optimism and promise. The digitalization of health data, together with the advent of advanced data mining techniques, has brought forward the possibility of automating and even improving the tasks that healthcare professionals have traditionally conducted in a qualitative or semi-quantitative way. In particular, Artificial Intelligence (AI) techniques are increasingly being used in a wide variety of applications involving cognitive tasks, from image-based pattern recognition and data integration models for disease prognosis to triaging chatbots. The hope is that this new family of tools will alleviate the burden on an overstretched healthcare workforce and also enable new ways for patients to receive care that contribute to the long-term sustainability of the system.

Although the advances are promising, the associated challenges cannot be underestimated. AI algorithms must be robust enough to avoid biased learning, which can easily happen when training datasets are too small, too skewed, or poorly annotated.  This requires cross-disciplinary, international agreements for data sharing, standardization, curation, anonymization, validation, and continuous monitoring. Implementing the tools in the clinic also requires a digitally-trained workforce and widespread access to the latest technologies. At the same time, clinicians and patients have to be involved in the design and development process, as ultimately the tools will only be successful if they are comfortable using them. These are just a few examples of the hurdles faced by AI technologies, but they reveal one of the key common features: the need for a global effort.

The global dimension of the current health crisis has provided fertile ground for the rapid development of digital and AI technologies. When, early on in the pandemic, chest CT scans were found to reveal the extent of lung damage, efforts were established around the world to facilitate data sharing, model training and scan assessment. For example, the Tianhe-1 supercomputer in China was made accessible to anyone in the world in order to provide quick COVID-19 diagnoses based on chest scans. Similarly, in Europe,  a  collaboration of 30 international partners including the most affected areas in Italy and Spain created Imaging COVID-19 AI,  which aims to provide an automated diagnosis and quantitative analysis of COVID-19 based on imaging. The initiative is a collaboration between the European Society of Medical Informatics and the companies Robovision and Quibim.

In parallel, AI is being used to mine existing databases of medical information. As early as February 2020, Benevolent AI, a UK-based company, had proposed the use of existing drugs for COVID-19 treatment. In March, Kaggle, the world’s largest machine learning and data science community, launched a competition to analyze more than 47,000 scholarly articles about COVID-19 and related coronaviruses to learn about its origin, evolution, therapeutic, and social implications. Intrepida, a Swiss company, launched Ancora.ai, a web-based AI tool to match patients with relevant clinical trials.

Although not always AI based, telemedicine has boomed. NHS England recommended GPs to change face-to-face appointments to telephone or video in March. Some telemedicine platforms made their services available for free, including Kry in Sweden, Doctolib in France, and Adent Health in Denmark. Push Doctor, a UK company partnered with the NHS, claimed in March that usage of their product had increased by 70%. Kry claimed to have doubled their usual number of appointments in two weeks. The need for remote consultations has also laid the ground for virtual tools, some of them AI powered: symptom checkers based on user inputs have been launched by Babylon Health in the UK, Natural Cycles in Sweden, and Mediktor in Spain.

Regulation is also moving quickly. The French government and German health insurance companies have removed reimbursement restrictions on video consultations. In the US, Medicare has expanded its coverage to include telemedicine. The British Medicine and Healthcare Products Regulatory Agency authorized fast-track approval of medical devices during the outbreak, and the FDA stated that it did not intend to enforce requirements for “certain lower risk device software functions,” including symptom checkers.

In many cases, this rapid, global response is already providing critical and beneficial support. However, even at this relatively early stage there are some warning signs of the risks of this type of rapid development. In April, a review published in the British Journal of Medicine systematically evaluated 31 computational predictive models for COVID-19, finding that all of them were at high risk of bias, mostly due to non-representative sample selection. This means that, when tested in a different, more general population, the accuracy of the predictions could decrease significantly. Similar discussions are also happening in other scientific domains, for example for antibody tests, which have been proposed as the basis of “immunity passports”. Their performance can only be established confidently after trialling on large population samples; however, so far most tests assessments have only been performed on small groups of individuals, according to a recent news article of the journal Nature.

The progress that we have made in regards to AI and digital healthcare in a matter of weeks  could have, under normal circumstances, taken years.  This is indeed something to celebrate, even given the current circumstances.  However, the speed at which we have arrived to this point means that it is even more important than ever to monitor the advances carefully, to ensure that patients receive the best possible care, and to earn the trust of the clinical community and the general public.  Avoiding missteps at this time will be critical not just for the management of the pandemic, but also to ensure the credibility and the future of digital AI-powered healthcare.