39.1 Running Patients through the model

You can run individual patients through a model via Microsimulation allowing you to incorporate individual data elements into the model. This fundamentally expands what you can do with your model by allowing patient characteristics and patient history events to affect future values -such as probabilities, costs and utilities.

Microsimulation runs different individuals through the model via a random walk using Monte Carlo Simulation. Some other names this analysis is also referred to as:

  • Microsimulation - a common name in this text

  • 1st-order trials

  • Individual-level simulation

  • Random walk

  • Monte Carlo Simulation (we try to avoid using this term in this context because Monte Carlo Simulation is a broader term referring to several types of simulation)

With Markov Cohort Analysis/Expected Value Analysis, there is no way to support different patient characteristics or to use prior events to drive future events, which can limit your ability to model complex diseases. Microsimulation expands the modeller's toolkit to be able to consider how a patient’s characteristics and past events can impact the future events, traits and outcomes of the model.

This chapter focuses on many important aspects of building microsimulation models, including:

  • Introducing heterogeneity via patient characteristics

  • Tracking patient events via trackers

  • Analysing the model via Microsimulation and interpreting the results

  • Running Sensitivity Analysis on microsimulation models

Random Walk

When you run Microsimulation on a model, you send a batch of individual patients into the model. Each patient will run through the model until death or until reaching the termination condition for the Markov model (typically a time horizon limit).

As the patient proceeds through the model, they will take a random walk through a certain pathway. Every subsequent patient will likely take a different pathway based on the “random walk” points in a model – specifically Chance Nodes and Markov Nodes.

The branches of both Chance Nodes and Markov Nodes require probabilities to control patient flow. In Markov Cohort Analysis/Expected Value Analysis, the cohort would be split into fractions based on the probabilities. With Microsimulation, you cannot break a person into fractions. Instead, a random number is drawn to determine to which branch the patient should proceed.

For example, a patient could reach a chance node with three branches.

  • Branch 1 – 20% (random range 0-0.2)

  • Branch 2 – 30% (random range 0.2-0.5)

  • Branch 3 – 40% (random range 0.5-1.0)

When the patient reaches that chance node, a random number is drawn between 0 and 1. Based on the random number, the patient will move to the appropriate branch.

  • If 0 <= random number <= 0.2, proceed to Branch 1

  • If 0.2 < random number <= 0.5, proceed to Branch 2

  • If 0.5 < random number <= 1, proceed to Branch 3

This process is repeated at every Markov and Chance node in the model as the patient proceeds on the random walk.

Patient Outcomes

When the patient finally finishes his/her pathway through the model, he/she will have accumulated value along the way in payoffs and trackers (described later). The primary outputs for the patient will typically be the total cost and total effectiveness accumulated over the entire patient run.

Each patient’s individual cost and effectiveness is meaningless as isolated values; however, when you aggregate those values into means for a large cohort of patients, you get a good estimate of the expected value (average cost, effectiveness) for each strategy. These average values from the Microsimulation then feed into Decision Analysis to identify the optimal strategy.

Later sections in this topic will examine individual and aggregate cost and effectiveness values when interpreting the Microsimulation results.

Complex models may require a large number of patients to guarantee that the mean values accurately represent the expected values for each strategy.