Key Epidemiological Models

Why This Matters

Epidemiological models are the backbone of how public health officials predict, respond to, and control disease outbreaks. You're being tested on your ability to understand why different models exist, when each is most appropriate, and how they inform real-world interventions like vaccination campaigns or social distancing policies. These models connect directly to core epidemiology concepts: disease transmission dynamics, population immunity, outbreak prediction, and intervention effectiveness.

Don't just memorize model names and their compartments. Know what assumptions each model makes, what types of diseases each best represents, and how changing parameters like $R_0$ affects predictions. Exam questions often ask you to select the appropriate model for a given scenario or explain why one model captures certain disease dynamics better than another.

---

Compartmental Models: The Foundation

Compartmental models divide a population into distinct groups based on disease status and track how individuals move between those groups over time. The core idea: disease dynamics depend on the relative sizes of each compartment and the rates at which people transition between them.

SIR Model (Susceptible-Infectious-Recovered)

The SIR model is the simplest and most foundational compartmental model. It splits the population into three groups:

• S (Susceptible) — people who can catch the disease
• I (Infectious) — people who currently have the disease and can spread it
• R (Recovered) — people who have recovered and gained permanent immunity

The critical assumption here is lifelong immunity after recovery. Once someone moves to R, they never return to S. This makes SIR a good fit for diseases like measles, mumps, and rubella, where one infection generally protects you for life. The basic reproduction number $R_0$ emerges naturally from SIR dynamics, since transmission depends on how many susceptible people remain.

SEIR Model (Susceptible-Exposed-Infectious-Recovered)

The SEIR model adds a fourth compartment, E (Exposed), between S and I. People in the E compartment are infected but not yet infectious — they're in the incubation period.

This matters because many diseases have a meaningful delay between when someone gets infected and when they can spread it to others. COVID-19, influenza, and Ebola all have incubation periods ranging from days to weeks. Without the E compartment, a model would overestimate how quickly new infections appear. SEIR gives public health officials more realistic timing predictions for when exposed individuals will become contagious and when an epidemic will peak.

Like SIR, SEIR still assumes permanent immunity after recovery.

SIS Model (Susceptible-Infectious-Susceptible)

The SIS model drops the recovered compartment entirely. After infection, individuals return directly to the susceptible pool, meaning they can be reinfected immediately.

This captures diseases where lasting immunity doesn't develop: the common cold, gonorrhea, chlamydia, and many bacterial infections. Because there's no permanent immunity building up in the population, the disease can persist indefinitely without needing to be reintroduced from outside. This is what makes SIS useful for modeling endemic diseases that circulate continuously.

Extending the Framework

Compartmental models are flexible. You can add compartments to match specific disease characteristics:

• V (Vaccinated) for modeling immunization campaigns
• M (Maternally immune) for newborns with passive immunity
• Age-stratified compartments when transmission rates differ by age group

The transition rates between compartments represent biological and behavioral parameters. The infection rate $\beta$ captures how quickly susceptible people become infected, while the recovery rate $\gamma$ captures how quickly infectious people recover. Simulating interventions means changing these rates: a vaccine moves people from S to R (or V), while treatment increases $\gamma$.

---

Modeling Approaches: Deterministic vs. Stochastic

The choice between deterministic and stochastic approaches depends on population size and how much real-world variability matters for the question you're trying to answer. Deterministic models give you the average expected outcome; stochastic models show you the range of possible outcomes.

Deterministic Models

Deterministic models use fixed equations that produce the same output every time you run them. There's no randomness involved. The standard SIR model, for example, uses differential equations like:

$\frac{dS}{dt} = -\beta S I$

This equation says the susceptible population shrinks at a rate proportional to both the number of susceptible and infectious people. Because these equations are smooth and continuous, deterministic models work best for large populations where random variation averages out. If you're modeling influenza across a city of 500,000 people, a deterministic approach gives reliable predictions.

Stochastic Models

Stochastic models incorporate randomness to reflect real-world uncertainty. Instead of fixed transmission rates, they use probability distributions, so each simulation run produces a slightly different epidemic trajectory.

This matters most for small populations. When case numbers are low, random chance can determine whether an outbreak takes off or fizzles out on its own. A deterministic model might predict 15 cases, but a stochastic model might show that in 40% of simulations the outbreak dies after 3 cases, while in others it reaches 30. That range of possibilities is critical information for outbreak response.

Reed-Frost Model

The Reed-Frost model is a classic stochastic, discrete-time model that simulates infection spread in small, closed populations. It works in generations: during each time step, every susceptible individual has some probability of being infected based on their contact with infectious individuals.

It's particularly useful for modeling outbreaks in bounded settings like schools, nursing homes, or households, where you can reasonably define who contacts whom. Because it's stochastic, it captures the real possibility that an introduction into a small group might not cause an outbreak at all.

---

Individual-Level Models: Capturing Complexity

When population-level averages aren't enough, individual-level models simulate how differences in behavior, social connections, and decision-making affect disease spread. These models trade simplicity for realism.

Agent-Based Models

Agent-based models (ABMs) simulate individual agents, each with their own characteristics: age, occupation, daily routine, risk behaviors, health status. Each agent follows a set of rules governing their movement, contacts, and responses to infection or interventions.

The strength of ABMs is capturing heterogeneity. Not everyone in a population behaves the same way or faces the same risk. ABMs let you test "what if" scenarios with precision: What happens if you close schools but keep workplaces open? What if only 60% of people comply with a mask mandate? This makes them powerful tools for evaluating targeted interventions, though they require significant data and computing power.

Network Models

Network models represent a population as a set of nodes (individuals) connected by edges (contacts or relationships). Instead of assuming everyone mixes randomly with everyone else, network models map the actual structure of who interacts with whom.

Network topology has a direct impact on disease spread. Highly connected individuals (hubs) can become superspreaders, transmitting to many others. Clustered groups can experience intense local outbreaks while other parts of the network remain unaffected. Network models are especially useful for identifying which connections to disrupt for maximum intervention impact, such as targeting vaccination at hub individuals.

---

Key Metrics: Reproduction Numbers

Reproduction numbers quantify transmission potential and guide intervention thresholds. Understanding the difference between $R_0$ and $R_t$ is essential for interpreting outbreak data and evaluating control measures.

Basic Reproduction Number ($R_0$)

$R_0$ is the average number of secondary infections caused by one infectious person in a completely susceptible population. It represents the theoretical maximum transmission potential of a disease in a given population.

The threshold value is 1:

• $R_0 > 1$: each case generates more than one new case, so the outbreak grows
• $R_0 < 1$: each case generates less than one new case, so the disease dies out
• $R_0 = 1$: the disease is stable (endemic equilibrium)

$R_0$ also determines the herd immunity threshold, which is the proportion of the population that needs to be immune to stop sustained transmission:

$p = 1 - \frac{1}{R_0}$

For measles ($R_0 \approx 12-18$), this means roughly 92-95% of the population needs immunity. For a disease with $R_0 = 3$, the threshold is about 67%.

Effective Reproduction Number ($R_t$)

$R_t$ is the real-time version of $R_0$. It reflects actual transmission at a specific point during an outbreak, accounting for current immunity levels, active interventions, and behavioral changes in the population.

Unlike $R_0$, which is a fixed property of the pathogen-population combination in a naive population, $R_t$ is dynamic. It changes as more people recover or get vaccinated, as public health measures take effect, and as people alter their behavior. If $R_t$ drops below 1 after implementing controls, the outbreak is declining.

---

Quick Reference Table

---

Self-Check Questions

1. A disease allows reinfection after recovery and persists in the population indefinitely. Which model best captures this dynamic, and why would SIR be inappropriate?

2. Compare $R_0$ and $R_t$: If a disease has $R_0 = 3$ but current $R_t = 0.8$, what does this tell you about the state of the outbreak and population immunity?

3. You're modeling a potential outbreak in a nursing home with 50 residents. Would you choose a deterministic or stochastic approach? Explain your reasoning.

4. Both SEIR and SIR models assume permanent immunity. What distinguishes them, and for which type of disease would choosing SIR over SEIR lead to inaccurate predictions?

5. An FRQ asks you to evaluate an intervention targeting "superspreaders." Which modeling approach would best capture this strategy's effectiveness: compartmental, agent-based, or network models? Justify your choice.