COVID-19, artificial intelligence and the benefits of multi-method modeling

The idea is to integrate different methods of modeling to overcome the limitations of individual research approaches – and to gain the most public health insights from each.
By Bill Siwicki
12:46 PM

Dr. Lauren Neal, leader of Booz Allen's health AI practice

Photo: Booz Allen

Dr. Lauren Neal, leader of research and consulting firm Booz Allen's health AI practice, is a proponent of taking a multi-method approach to modeling COVID-19 disease dynamics in artificial intelligence.

She believes a multi-method approach provides a better understanding of COVID-19 and other infectious diseases – how they spread and impact communities, with the goal of being better prepared for future public health threats.

She also believes a "virtual laboratory" can be used to investigate a wide range of what-if scenarios, and easily adapted to future high-consequence public health threats.

Healthcare IT News sat down with Neal to talk about these approaches and how AI can help with the COVID-19 pandemic.

Q. When it comes to artificial intelligence and COVID-19, how is a multi-method approach to modeling COVID-19 disease dynamics better than other approaches?

A. We have long employed simulation modeling to further increase our understanding of complex infectious diseases as well as their development, spread dynamics and potential treatments. Examples include models for zoonotic diseases such as Zika, Ebola, West Nile Virus, SARS, MERS and the recent COVID-19.

Two modeling techniques, system dynamics (SD) and agent-based modeling (ABM), have been frequently used in recent years to investigate the complex nature of infectious diseases despite their limitations. For example, SD operates at a high level of abstraction by compartmentalizing the human population into different disease stages such as susceptible (S), infected (I) and recovered (R), among others while assuming everyone behaves the same way within each compartment.

ABMs tend to address this limitation by tracking each individual member of the population and simulating granular profiles of individual interactions and movements within the population. However, this high level of model fidelity comes with a handful of trade-offs, including intensive cost of computation for large populations as well as increased model uncertainty due to a myriad of model assumptions.

We believe that effectively choosing between modelling methods is a question of minimizing trade-offs in the model creation, verification and validation process. The idea of multi-method modeling is to integrate different methods of modeling to overcome the limitations of individual methods and get the most from each one.

Booz Allen's multi-method model for COVID-19 combines the advantages of SD and ABM, allowing the simulation of spatially explicit scenarios representing future states of disease transmission within different local communities and testing risk management policies across a wide range of scenarios using "what-if" analysis.

Q. What is a virtual laboratory and how can it be used to investigate public health threats?

A. Historically, randomized control trials, cohort studies and case-control studies were commonly used methods to investigate the epidemiology of public health threats as well as potential intervention options to mitigate the risks. However, performing large trials and studies to achieve generalizability and sufficient statistical power is quite difficult, time-consuming and costly.

Therefore, a comparable, reliable and easy-to-use planning tool is needed to assess interventions and their impacts. A virtual laboratory is a special type of simulation model that can be used to represent the dynamics of COVID-19 spread within a community and facilitate "what-if" simulations that explicitly represent the uncertainty in supporting data and assumptions about risk factors associated with onset of the disease within the community.

A virtual laboratory is a risk-free environment, in which ideas on intervention strategies for a particular public health threat (for example, social distancing, partial lockdown and vaccination, among others) can be tested in a systematic manner without the time, costs and risks associated with experiments conducted in a real-world setting.

Virtual laboratories can have many uses, and present many possibilities for innovation, but it is their capability to provide real-time insight, enable forecasting and provide decision support for live operations that is most immediately accessible. With these abilities, community, state and federal public health decision-makers can be more effective, improve efficiency and deliver cost savings while protecting lives.

Q. What is a multi-criteria decision analysis (MCDA) framework, and how is it used with artificial intelligence and COVID-19?

A. Decision-making regarding implementation of public health interventions can sometimes be heuristic, and it can be argued that decisions based on a single criterion disregard important information about other relevant related outcomes. In managing the COVID-19 pandemic, several compelling narratives seem to have played a significant role in the decision-making processes regarding which risk intervention and management measures to implement.

During the pandemic, public authorities have had to make decisions based on uncertain quantitative evidence and expert scientific evidence (for example, possible future scenarios), on assessments of the health system capacity (for example, ICU beds) and on expected public adoption of more or less restrictive measures such as social distancing and lockdown measures as well as reopening of local communities and businesses.

When empowered by real-time data harnessed using artificial intelligence and machine learning techniques, as well as forecasted disease dynamics based on simulation modeling, multi-criteria decision analysis (MCDA) can help decision-makers make data-driven decisions based on multiple, sometimes conflicting criteria in a transparent and systematic manner.

For example, Booz Allen has used an MCDA framework considering local decision criteria such as new daily infections, decline in new daily deaths, new hospitalizations and ICU bed utilizations to systematically analyze simulated forecasts obtained from our multi-method model and generate risk maps for individual states.

These risk maps could potentially be used by public health decision-makers to target available surveillance and infection control measures based on the perceived levels of COVID-19 risks in local communities.

Q. How does all of this apply to the work of healthcare provider organizations' C-suite executives and caregivers on the frontlines of the pandemic?

A. The COVID-19 pandemic has brought us unprecedented and evolving challenges since its onset. We have made considerable efforts to address these challenges using a suite of data-driven tools, including artificial intelligence and simulation modeling.

While early efforts have been focused on epidemiological modeling of COVID-19 spread at global, national and state levels, the pandemic has raised many more localized challenges that our data-driven approaches can also address.

For example, the rapid onset of the COVID-19 crisis has shown increased demand and risks for healthcare provider organizations due to continuously changing and unpredictable circumstances. Simulation modeling and virtual laboratories can be utilized to proactively manage risk to healthcare organizations during the current pandemic and future large-scale public health threats.

We can investigate a wide range of scenarios to enhance our preparedness by optimizing hospital workflow structures, developing new processes, managing staffing levels, procuring equipment, bed management, and enforcing consistency of medical management of patients, among others.

In these ways, a virtual laboratory can be used as both a learning tool (for example, better understanding how a hospital as well as frontline healthcare providers function under a local community COVID-19 outbreak) and an evaluation tool (for example, testing complex scenarios like optimal patient throughput for an emergency department).

Virtual laboratories can effectively support executive-level decisions made at the healthcare provider organizational level to create capacity and manage scarce resources for the effective care of critically ill patients, while testing scenarios to evaluate the ability of the health system capacity to cope with expected and unexpected demands during the pandemic.

Twitter: @SiwickiHealthIT
Email the writer: bsiwicki@himss.org
Healthcare IT News is a HIMSS Media publication.

Want to get more stories like this one? Get daily news updates from Healthcare IT News.
Your subscription has been saved.
Something went wrong. Please try again.