In most countries, the Covid-19 outbreak is being managed with a similar strategy. At the beginning a cautious response was enough to tackle the small number of infections, then a ramp up of healthcare capacity was needed to address the critical phase, while lockdown measures were taken to flatten the epidemic curve. Now, as the curve has peaked almost everywhere, governments are discussing how to manage de-confinement and the following recovery steps.
Can advanced technologies such as Artificial Intelligence (AI) support decision-making by predicting Covid-19 trends and possible relapses? In a population-level observational study published on The Lancet Digital Health, a team of researchers from Fogarty International Center, US National Institutes of Health, gathered data from the Chinese healthcare-oriented social network DXY.cn, other national and international sources to analyze the outbreak progression across China and make predictions. They proved AI and machine-learning techniques are pretty good at providing useful insights about patients and treatments (for instance profiling infected people, tracking movements and relations, assessing delays between symptom onset, comparing hospital and clinic care, etc.), but have limited forecast capability due to some data quality issues.
Other studies suggested the correlation of multiple data sources through machine-learning models could be used to measure individual clinical risks. Think of the possibility to correlate parameters such as age, basic medical histories, the presence of co-morbidities such as diabetes or hypertension, and other data such as household composition to conjecture about possible Covid-19 severe outcomes and the probability an individual would need intensive care if infected. Different social distancing and protection policies could be defined for “high risk” and “low risk” people, of course considering available resources, medical liability risks, and other trade-offs. The same classification could inspire the differentiation of de-confinement strategies at later stages of the pandemic.
However, as reported by Harvard Business Review, individual medical records are not always easily accessible, and any prediction model needs to be “trained” before returning relevant, actionable information. A better approach might be to share data model across cities or countries whose populations resemble each other, presuming the virus impact to be similar. If the dataset is large enough to enable some level of personalized prediction, the quality of outcomes improves as more data is added.
Privacy and cybersecurity regulations also come into play, and pandemic management approaches based on personal data might be inapplicable in some countries. Can we agree to suspend normal privacy laws to allow the sharing of anonymized data during a pandemic? Some government are moving this way, and many people might be willing to exceptionally and temporarily provide their data, through appropriate and secure channels, to feed emergency-response models.
Although the question is not simple to be unraveled, the availability of large, accurate datasets is the starting point for AI to mitigate the fallout from Covid-19 pandemic and prepare for the next one.