Overview: infectious disease modeling for decision-making

What is modeling?

A model is a simplified representation of a more complex system or process. While "modeling" can sound complicated, it's based on a way of thinking that most people do on a day-to-day basis. Models are used across many sectors: your cellphone's mapping app uses a model of road networks and traffic flows to estimate your journey time; meteorologists' models combine historical climate data and current observations to predict tomorrow's weather; and financial advisors use models of savings, stock market performance, and personal income to plan retirements. Regardless of the specific approach, modeling distills questions into simple parts so key elements can be identified and understood.

Take the example of financial planning: While not all of us are running complex simulations to compare retirement portfolios, many of us use a mental model to balance income against expected monthly bills and saving goals to make a budget. Like all models, that model will be a simplification of reality. Perhaps, in this case, you estimate this month's expenses based on your usual bills at this time of year. Unexpected expenses can still occur, and your simplified projection will likely never be perfect, but it can still be helpful when planning your grocery shopping.

What is infectious disease modeling?

Like all models, infectious disease models are simplifications of the real world. While the real world is too complex to perfectly replicate, simplified versions can help us explore a range of public health questions. Infectious disease models can be used to learn more about new outbreaks, understand how best to control them, and make predictions about the future. Modeling can also be used to fill gaps in current data and identify key sources of uncertainty, helping to guide data collection and research. Some examples of how infectious disease modeling has been used during public health responses are shown in the table below. Many different mathematical and statistical approaches can be used to understand the patterns of infectious disease transmission. Because of the breadth of areas where models can be useful, it is important to select an appropriate modeling method to answer a public health question.

It is important to recognize the strengths and limitations of infectious disease models to understand the role modeling can play in public health decision-making. Modeling is always limited in some capacity by the data available and the simplifications and assumptions used in the model. A model is not a magic box that can make decisions; it is critical to understand and evaluate the methods used in a model, and to consider modeling results alongside other data and insights from public health experts. Nevertheless, when used carefully, models help us make the most of what data we do have and understand clearly what we can and cannot say about the future.

How have infectious disease models been used to inform public health during outbreaks?

How can models guide public health decision-making?

Some models, specifically transmission models, can help decision-makers to explore the potential impact of different public health interventions on curtailing disease spread. Transmission models replicate the patterns of disease transmission in a population. This means that, given information about the effectiveness of different interventions, they can be used to help guide public health policy, such as which interventions should be used, how quickly, and which populations should receive them. Transmission models can, for example, be used to determine whether isolation and quarantine are sufficient to stop an outbreak, or if other interventions would be needed – as described further in Box 1. They can also be used to assess how effective interventions would need to be, or how many people would need to adhere to them, in order to control the spread of an outbreak.

Can models tell us what will happen in the future?

Nothing can truly see into the future. But as we see in weather forecasting, models can analyze trends and help make educated guesses about what will happen next. Short-term forecasting and scenario modeling are two approaches used to explore potential futures. Short-term forecasting makes these predictions on a shorter term, usually days to weeks, and can be used to help answer questions like "how many COVID-19 hospitalizations will there be in two weeks?". Short-term forecasts like these often attempt to predict healthcare outcomes (such as hospitalizations) based on chains of events already in motion (such as current infections).

Scenario modeling, meanwhile, explores potential futures over the longer-term, like months or years. This longer time horizon means scenario models are useful to answer questions like: "will there be more COVID-19 hospitalizations this winter compared to last?". Scenario models can also be used to explore different potential futures under different conditions—for example, the timing or size of respiratory disease seasons based on different assumptions about the emergence of viral strains.

What can models tell us when data are limited or incomplete?

Models are not a replacement for good data. If a model is based on flawed data, the results will also be flawed. However, models can still be used to help build intuition and guide policy when data are new or limited. For example, when a novel infectious disease or strain emerges, models can be applied to this data and used to learn new information about the new disease or strain (as highlighted earlier in the table of examples). Modeling can also enhance data that is incomplete or biased, based on information we know about reporting and disease trends. For example, nowcasting is a modeling approach that uses data on reporting patterns to adjust incomplete data and estimate current trends for disease metrics, even when reporting delays create the illusion of declining cases, as shown in Box 2. Modeling can also be used to show the range of potential outcomes under different assumptions or unknowns and can help understand the importance (or lack thereof) of key gaps in the data. For example, when modeling potential interventions to prevent outbreaks, the exact effectiveness of those interventions may not be known. In this case, models can be used to help frame upper and lower bounds of effectiveness based on that uncertainty.

Box 2: How can infectious disease modeling help improve seasonal disease monitoring?

Regularly collected data for common respiratory diseases like flu and COVID-19 often have reporting delays—which lead to the appearance of declining cases over time even when cases are actually increasing, as shown in this example from New Mexico1. This can be adjusted using nowcasting models, which can be incorporated into routine reporting structures to improve situational awareness for both public health decision makers and the public.

Summary

In the coming chapters, we will walk through steps to appraising a model, explain in more technical detail how different types of models work, and describe some basic elements of model validation and evaluation. We will also introduce key concepts in infectious disease epidemiology that come up frequently when using models to guide public health decision-making. Throughout, we will link to examples of modeling done by the Center for Forecasting and Outbreak Analytics (CFA) as well as other parts of CDC.