The fight against Covid-19 has pushed health care to quickly adopt virtual tools at a rate much faster than normal. Some of the most promising pandemic-related health technologies go beyond just telehealth—for example, analytics and artificial intelligence (AI).
Advanced analytics is a broad category of technologies including AI, automation, predictive analytics, and prescriptive analytics that can detect meaningful patterns in data and help assess the future. AI refers to the capability of a machine to recreate "intelligent" decision-making using algorithms, pattern matching, and other techniques to produce outputs and improve over time.
This isn't the first time we've covered AI's potential against Covid-19. However, reflecting on the pandemic nearly two years after it started, there are even more lessons to be learned about AI and its potential against future pandemics.
There has been growing attention around using AI in health contexts in the areas such as administration, finance, research and development, and care maintenance, but we have not yet seen high levels of adoption in health care. Covid-19 was a turning point that has brought about new and additional uses of AI and analytics across all parts of pandemic response, from detection to treatment to prevention.
- Detect the pandemic at the earliest stages
Every day, there are enormous amounts of public population data being produced on both global and local scales through channels such as airline records, public health official statements, news reports, and social media. If filtered properly with AI, this data can spotlight abnormal patterns of disease spread much faster than standard surveillance methods.
Leverage big data from a variety of non-traditional sources
BlueDot used big data to correctly identify the pandemic on December 31, 2019—nine days before the CDC. Using a machine learning (ML) and natural language processing (NLP) model, they were able to process international data on a rapid cycle to quickly detect abnormality and the direction of spread. Metabiota similarly uses NLP to create visualizations of early disease spread patterns. In early 2020, they were able to analyze levels of fear around the virus to identify which countries around China were the next target of spread – before any of those countries reported cases.
Combine public data with ICD-10 codes for accurate forecasts for many infectious diseases
Optum's Infectious Disease Platform leverages AI to produce precision forecasts for various infectious diseases worldwide, from Influenza to Covid-19. Each disease forecast is generated using a unique set of forecasting models leveraging data from a wide variety of public and proprietary sources. The platform is developed to be more precise than standard public surveillance networks with the use of ICD-10 codes to distinguish the presence of one disease from other similar ones.
Track early cases on a local scale
Kaiser Permanente started using its telehealth call centers to track rates of calls about certain symptoms in local communities to see if volumes were higher than the expected norm, which can be an early sign of community transmission. Starting in February 2020, they noticed a jump from 200 to 3,500 calls a day for Covid-19 related symptoms and were able to predict and prepare for the surge in cases early on.
- Identify both symptomatic and asymptomatic cases at early stages of transmission
One of the most important parts of mitigating a pandemic is catching disease spread early for both symptomatic and asymptomatic cases. Several organizations have repurposed or developed AI programs for this exact purpose.
Zhongnan Hospital in Wuhan, China repurposed an Infervision Graphics Processing Unit-accelerated AI program (from its original task of detecting lung cancer) to detect signs of Covid-related pneumonia on CT scans. If pneumonia is detected in a patient, they can be diagnosed, isolated, and treated quickly without as much strain on staff and resources.
Detect unknown asymptomatic cases
MIT researchers adapted the Open Voice model for Covid-19 by analyzing the sounds of thousands of coughs from infected and non-infected people. This dataset trained an algorithm to detect the difference between Covid-positive and negative coughs based on key biomarkers of the virus' effects on the muscles and respiratory system. The model correctly detected 98.5% of coughs of those with Covid-19, including all asymptomatic cases. Ideally, this type of technology could be incorporated into users' personal device speakers to detect asymptomatic risk.
Improve contact tracing
Contact tracing aims to identify those who are exposed to infection, allowing them to self-isolate and curb transmission. Digital contact tracing apps do this relying primarily on Bluetooth capability. However, the baseline performance of Bluetooth for this purpose is often unreliable and can lead to significant errors in accuracy. AI for enhanced contact tracing is largely still in development, but with more advancements, could improve the accuracy of traditional tracing and mend additional concerns.
For example, Diveplane's Aware program uses its "Geminai" technology to apply a digital twin dataset that contains all real-world statistic properties for effective contact tracing without compromising personal details or the possibility of reverse engineering. Acquired Data Solutions and Kiana Analytics use location analytics for employers to assess each employee's location and dwell times in the workplace. If someone tests positive, the technology can then identify who was near them in the prior days and notify them of exposure.
- Monitor populations to proactively identify those most at risk:
Long-term, the key to preventing future pandemics will rely on population-level data insights, not just individual-level intervention. Many health organizations today are finding best ways to analyze data to proactively determine groups of highest risk and distribute preventive care resources.
Jvion is gathering data on 30 million patients to identify groups that are at highest risk for Covid-19, assessed across more than 5,000 variables including medical predispositions and social determinants of health. They then circulate to health systems lists of the highest-risk patients who should get proactive outreach so they can better understand risk and take precautionary care measures.
Parkland Center for Clinical Innovation used ML and big data to create the Proximity Risk Index for Parkland Health patients. The platform uses geo-mapping and hot spotting technology based on active Covid-19 cases to generate a dynamic risk score for each person. It also identifies which communities are at highest risk. Then, staff can send resources to areas of highest risk on staying safe, monitoring, testing, and other important information.
But what about privacy?
There may be an idea of a "trade-off" between privacy and public health needs during an emergency like a pandemic, as seen when the Department of Health and Human Services relaxed enforcement of certain HIPAA rules (as allowed under a Public Health Emergency demarcation). However, a 2021 Cisco survey shows that much of the public does not support the idea of this trade-off, with 42% of respondents saying they don't want any suspension of privacy rules in the name of a better pandemic response and around half saying they would have lower trust for an organization that uses AI to make major decisions.
Although much of privacy protection will depend on regulatory agencies, private organizations can still contribute to upholding privacy and building user trust by using informed consent, aggregating and de-identifying data, and building data oversight and privacy protection into their technologies.
AI's full potential for better managing and ending future pandemics will only be realized when organizations carefully and correctly collect, use, protect, and manage the data needed to power this technology.