How has AI helped in the COVID pandemic?
Where it's used
AI researchers around the globe responded quickly to find way to use Machine Learning and AI algorithms to help defeat the COVID 19 pandemic. Applications were developed in a five key areas, Forecasting, Diagnosis, Containment and Monitoring, Drug Development and Treatments, Medical and Social Management. Some, such as in Forecasting, were aimed at predicting new outbreaks to assist policy making, others were aimed at reducing the impact on health services by remote triaging.
Forecasting
On December 31, BlueDot’s infectious disease risk assessment algorithm, that uses both human and artificial intelligence, alerted it’s customers to the outbreak in China, nine days before WHO.
BlueDot uses foreign language news reports, animal and plant disease networks as well as global ticketing data to help predict where and when infected residents are going next. The algorithm doesn’t make use of social media and was able to predict that the virus would jump form Wuhan to Bankok, Seoul, Taipei and Tokyo within the next few days (Bogoch 2020).
Forecasting is difficult using AI based approaches because they rely on very large data sets to be effective and there is no guarantee that one virus will behave like another.
Some attempts have been made to use social media data to model and predict the progress of COVID 19 but the challenge with such data is that it accumulates noise. Google’s Flu predictor platform, based on analysing peoples search trend, estimated more than double the number of doctor visits in 2014. One of the conclusions from this work was that “big data” doesn’t substitute for traditional data collection and analysis. It was dubbed the “big data hubris”.
Diagnosis
The ability to diagnose and screen for the Corona virus disease is a critical part of managing the pandemic. Many AI based systems were developed to provide rapid diagnosis, even remotely for crude cough tests, using smart phones or an ordinary telephone. One of the challenges in developing remote diagnosis is how to deliver it to the whole population without discriminating against those that don’t have access to the required technology, such as a smart phone. These applications have not been scaled and have only been tested in small scale trials.
.... almost all published prediction models are poorly reported, and at high risk of bias such that their reported predictive performance is probably optimistic.
One method of diagnosis relies on the use of image recognition to detect pneumonia like symptoms from X ray images and computerised tomography (CT) scans. Several literature surveys have been conducted to evaluate the reported results of these applications and they reach the same disappointing conclusions.
Our review finds that none of the models identified are of potential clinical use due to methodological flaws and/or underlying biases.
Containment and Monitoring
Several approaches have been used to contain populations and to monitor their movements or proximity to people known to be infected. Various proposals were made for ways to ensure social distancing, such as the use of drones with facial recognition.
Predictions of the movement and spread of the disease using Machine Learning were used to restrict movement in an attempt to contain the spread of the virus. These approaches varied from using Machine Learning on travel data and even monitoring social media.
In Asia, checking the temperature of people on a massive scale was carried out, mostly in public places like train stations or entrances to buildings. People with an elevated temperature were refused access. Not all approaches to tracking and containment used AI algorithms but China used a traffic light system using self reported data combined with the persons health records to control their movements. A red light prohibited access, orange required quarantining whilst green allowed free movement.
Various applications have been proposed for monitoring and controlling the movement of people from drones, facial recognition and GPS tracking. The most widely used approach, but not requiring AI, has been one that relies on smart phones to detect proximity to infected people, registered on the track and trace system, either using GPS or less intrusive bluetooth technology.
.jpg)
A schematic of app-based COVID-19 contact tracing (Fig. 4 from Ferretti et al. 2020). CC BY 4.0 via Wikimedia Commons.
The live threatening nature of the global pandemic as well as the threat to health care systems and the economy has been the main justification by governments for what, in normal times, would be regarded as an intrusion on privacy and freedom of movement and association.
This is quite different from the traditional debate about whether confronting security threats to our way of life merits sacrificing the values of freedom and privacy that define us. Covid-19 is not an ideology, and rebalancing the contract between citizens and the state to take advantage of new technologies is not capitulation.
Whether GPS or bluetooth based, concern has been expressed by civil liberties groups over the privacy of the data and the potential for wide scale tracking of people, perhaps even after a pandemic, once the apps have become established. A group of nearly 300 academics from around the world signed a “Joint Statement on Contact Tracing” to express their concerns about the adoption of the technology and to propose privacy principles for its use.
It is vital that, in coming out of the current crisis, we do not create a tool that enables large scale data collection on the population, either now or at a later time.
It is interesting to note that there has not been a good take up of tracing apps around the globe, except in more authoritarian regimes where citizens can be forced to comply.
Although these applications did not require AI algorithms, it is to be expected that any centralised database will be used for Machine Learning to detect and predict patterns of behaviour and disease spread.
In European the average adoption rate of contact tracing apps by the population was 22% with a large variation between countries such as Finland (51%) and Croatia (2%).
Between a 60% and 80% take up is regarded as being necessary to make such track and trace systems effective. The study in Europe did however report lower cumulative infection rates where there was a higher, but not ideal adaption rate of the governments track and trace application.
Higher adoption rates were associated with lower cumulative infection rates.
Drug Development and Treatments
Drug development is a long and extremely expensive process, taking on average 10 years and requiring a series of medical trials to evaluate the effectiveness and side effects of any proposed new treatment. Only around 1 in 1000 possible drug treatments progress from preclinical testing to full clinical trials and of those only 1 in ten make it to the approval stage.
AI algorithms are used in the early stages of drug development to help to reduce the initial number of compounds considered, including predicting likely adverse reactions.
There are deep worries that a drug could prove ineffective or worse, dangerous, when used on populations that have a different response to the biomarker that was used to develop or validate that drug.
One of the success areas in the use of Machine Learning algorithms has been in the repurposing of drugs already approved for use in a particular condition. An example of this is the repurposing of a drug used in treating rheumatoid arthritis to alleviate the severe symptoms that can occur in some patients with COVID 19. The usefulness of AI in pattern matching as well as in discovering patterns, or connections, that are difficult for humans to spot easily, make it a powerful tool in learning from and searching millions of published texts and trials data.
In vaccine development, AI is being used along with systems biology to screen candidates and predict immune system response, through simulation, significantly cutting down the time normally taken in the development of a new vaccine. Very high levels of safety are required in vaccine development as they are being given to a very large population of healthy patients. It normally takes many years of rigorous development and testing of different vaccine candidates on animals, and then humans, before a safe and effective vaccine is licensed.
Machine Learning has also been used for taxonomic and hierarchical classification of COVID strains. Google owned DeepMind have used their deep learning algorithms in a program called Alphfold, to identify protein structure linked to COVID 19 that might be valuable for vaccine formulation.
Medical and Social Management
The Allen Institute for AI, in partnership with the CORD 19 Open Research Dataset, used its Natural Language Processing tools to allow academics and the general public to search publications related to COVID 19. Microsoft, also a CORD 19 partner, has created a COVID 19 data lake, providing researchers access to various data sets related to the pandemic.
Aimed at the general public, Natural Language based Chatbots have been deployed by many public health authorities around the world, including the World Health Organisation and the Centre for Disease Control, to provide health advice and information about the COVID 19.pandemic.
Limitations of AI
Whilst AI has been and continues to be useful in some areas of managing the COVID 19 pandemic, such as drug repurposing and forecasting infection spread, there are clear limitations to its use in a clinical setting.
While AI has been practically applied for the identification of candidate drugs for drug repurposing and contact tracing , its application and utility for COVID-19 in clinical settings have been insignificant to date.
AI has not yet been impactful against COVID-19. Its use is hampered by a lack of data, and by too much data (noisy and outlier). Overcoming these constraints will require a careful balance between data privacy and public health, and rigorous human-AI interaction. It is unlikely that these will be addressed in time to be of much help during the present pandemic.
Problems around data availability, access and standardisation spanned the entire spectrum of data science activity during the pandemic. .... better data would enable a better response. ... issues of inequality and exclusion related to data science and AI arose during the pandemic. These included concerns about inadequate representation of minority groups in data, and low engagement with these groups, which could bias research and policy decisions.
References
Bogoch II, Watts A, Thomas-Bachli A, Huber C, Kraemer MUG, Khan K. Pneumonia of unknown aetiology in Wuhan, China: potential for international spread via commercial air travel. J Travel Med. 2020 Mar 13;27(2):taaa008. doi: 10.1093/jtm/taaa008. PMID: 31943059; PMCID: PMC7107534.
Chee ML, Ong MEH, Siddiqui FJ, Zhang Z, Lim SL, Ho AFW, Liu N. Artificial Intelligence Applications for COVID-19 in Intensive Care and Emergency Settings: A Systematic Review. Int J Environ Res Public Health. 2021 Apr 29;18(9):4749. doi: 10.3390/ijerph18094749. PMID: 33947006; PMCID: PMC8125462.
Data science and AI in the age of COVID-19: Reflections on the response of the UK’s data science and AI community to the COVID-19 pandemic, Alan Turing Institute, 2021.
Kahnbach L, Lehr D, Brandenburger J, Mallwitz T, Jent S, Hannibal S, Funk B, Janneck M
Quality and Adoption of COVID-19 Tracing Apps and Recommendations for Development: Systematic Interdisciplinary Review of European Apps
J Med Internet Res 2021;23(6):e27989
Track & Trace Diagram from Ferretti, Luca; Wymant, Chris; Kendall, Michelle; Zhao, Lele; Nurtay, Anel; Abeler-Dörner, Lucie; Parker, Michael; Bonsall, David; Fraser, Christophe (2020-03-31). “Quantifying SARS-CoV-2 transmission suggests epidemic control with digital contact tracing” CC BY 4.0
Naudé, W. Artificial intelligence vs COVID-19: limitations, constraints and pitfalls. AI & Soc 35, 761–765 (2020). https://doi.org/10.1007/s00146-020-00978-0
Patrick McGee and Hannah Murphy, Financial Times, April 28 2020, Coronavirus apps: the risk of slipping into a surveillance state, https://www.ft.com/content/d2609e26-8875-11ea-a01c-a28a3e3fbd33
Prediction models for diagnosis and prognosis of covid-19: systematic review and critical appraisal, BMJ 2020; 369 doi: https://doi.org/10.1136/bmj.m1328 (Published 07 April 2020) BMJ 2020;369:m1328 (CC BY 4.0)
Ramamoorthy, A., Pacanowski, M. A., Bull, J., & Zhang, L. (2015). Racial/ethnic differences in drug disposition and response: review of recently approved drugs. Clinical Pharmacology & Therapeutics, 97(3), 263-273.
Roberts, M., Driggs, D., Thorpe, M. et al. Common pitfalls and recommendations for using machine learning to detect and prognosticate for COVID-19 using chest radiographs and CT scans. Nat Mach Intell 3, 199–217 (2021). https://doi.org/10.1038/s42256-021-00307-0
Šćepanović, S., Aiello, L. M., Zhou, K., Joglekar, S., & Quercia, D. (2021). The Healthy States of America: Creating a Health Taxonomy with Social Media. Proceedings of the International AAAI Conference on Web and Social Media, 15(1), 621-632. Retrieved from https://ojs.aaai.org/index.php/ICWSM/article/view/18089