• Tarsila Machado

AI in Public Health, Part One: COVID-19 Case Estimation

Updated: Jul 4, 2021

Throughout the COVID-19 pandemic, artificial intelligence (AI) has been utilized to model patterns of infection cases, both in the U.S. and globally. Epidemiological systems based in AI have proven to be useful in identifying patterns of disease spread across populations in real-time and forecasts future trends. Although AI is showing promises of being effective in healthcare data modeling during COVID-19, it still presents certain barriers and challenges within its functions. These machine learning systems are not yet perfect, but they can greatly improve upon current methods of disease pattern recognition and infection case estimates across the globe. Having case estimations that reflect the true incidence of disease as closely as possible is key in helping public health professionals craft emergency response and preparedness plans that best serve their communities and most effectively combat the COVID-19 virus.

A recently published article discusses how a machine-learning framework designed to assist in counting daily cases and deaths from COVID-19 detected significant underestimations in both U.S. states and abroad. The authors express that having so many undocumented cases of infections and deaths obscures the true value and known severity of affected populations, which hinders efforts to control the virus. The AI algorithm created for this framework was used to adjust for underreporting of cases and deaths to reveal the actual impact of COVID-19 and identify potential hotspots to target with necessary interventions and decision-making.

This machine-learning system found that in 25 out of the 50 countries analyzed, the actual number of cases was estimated to be around five to 20 times greater than previously reported figures. For countries such as Belgium, the U.S., and Brazil, it also identified that around 10 percent of the population had already been infected at least once. This framework was able to provide updated estimates of cases daily, along with the actual fractions of currently infected individuals per region. Having more accurate current case estimations and predictions is essential in the planning of tailored interventions like contact tracing to prevent future outbreaks. The estimations calculated by this AI framework are critical for identifying the true severity (i.e. actual number of cases) of COVID-19 across geographical regions across the world, which might otherwise be miscalculated through traditional epidemiological methods alone.

Additionally, AI and machine learning tools can efficiently complement traditional epidemiological tools and hold several applications in the healthcare world. The public and clinical health applications of AI can include assisting in disease diagnosis, drug design, vaccine design, identifying epidemiological trends, tracking and forecasting disease spread, and predicting criticality in disease outcomes. While it is not yet widely used in the global healthcare industry, these tools can help enhance public health professional’s abilities to better understand and respond to crises such as COVID-19 by reducing the individual burden on each of those working towards disease case modeling. Experts point out that potential flaws exist due to AI relying on input data and assumptions by the humans using it, which can vary as real-life public health events unfold. Additional measures must be taken to ensure maximum accuracy of AI-based predictions; it should not be seen as a replacement to human intelligence, but rather as a tool to assist health professionals in their studies and surveillance.

AI frameworks and machine learning cannot and should not completely replace human intelligence in observation or decision-making by health professionals. However, its potential benefits as a complementary tool to traditional methods of case estimation and disease control are numerous. These tools are capable of assisting with epidemiological modeling of disease outbreaks, forecasting needs for healthcare infrastructure (I.e. clinics, testing sites), medical supplies (I.e. medications, testing kits), and human resources (I.e. contact tracers, varied health specialists). Combined with human expertise, AI-based public health approaches for case estimations can play a substantial part in the fight to control the spread of COVID-19 and more accurately understand the global impacts of this pandemic.


Speaking Plainly:

  • AI-based data modeling systems present the potential for recognizing, tracking, and predicting patterns in regional disease spread worldwide.

  • Additional AI and machine learning applications include assistance in disease diagnosis, vaccine and drug development, and disease outcome forecasting.

  • These tools can be used to more accurately understand the nature of disease outbreaks, demographics, and factors behind disease spread patterns amid the COVID-19 pandemic.

  • While these frameworks still present some accuracy flaws in their predictions due to the data put into them, they prove useful when used in conjunction with traditional epidemiological methods and human expertise.

  • Having greater accuracy in COVID-19 case estimations with the use of AI-based tools can help health professionals better respond to regional trends with appropriate resources and policy decisions based on need.