Markov Models

Question 1

Markov Models

Answer 1

• Handle simulations that decision trees cannot
• Diseases/conditions with more complex outcomes and longer follow up periods need to be modeled
• Can be a more realistic representation

Question 2

MM Definition

Answer 2

• Type of decision analysis used when patients transition from one health state to another
• Length of time in a health state depends on disease and interventions being evaluated
• Health states are “memoryless” - known assumption

Question 3

MM 5 Steps

Answer 3

1) Choose health states that represent possible outcomes of
intervention.
2) Determine possible transitions between health states.
3) Choose how long each cycle should be and how many cycles will be analyzed.
4) Estimate the probabilities associated with moving (transitioning) in and out of health states.
5) Estimate costs and outcomes associated with each option.

Question 4

Step 1

Answer 4

• Delineate the mutually exclusive health states that a patient might reasonably experience
• Patient cannot be in more than one health state during each cycle
• “Memoryless”

Question 5

Step 2

Answer 5

• Transitions between health states determined by clinical information
• Arrows are used to indicate which transitions are allowed
• Death is an absorbing state which indicates a patient cannot move to another state in a later cycle

Question 6

Step 3

Answer 6

• Outcomes past the initial health state can be followed further to examine future outcomes
• Each follow up interval = “cycle”
• Cycle is a time period considered clinically relevant to the disease or condition
• For chronic disease, a cycle length of 1 year is common
• Number of cycles depends on clinical relevance or situation

Question 7

Step 4

Answer 7

• Transition probabilities are used to eastimate percentage of patients moving from one health state during each cycle
• Probabilities come from previous research, clinical practice, or expert panels

Question 8

Step 5

Answer 8

• Outcomes for each state should be estimated and given a value
• Costs in each health state should be estimated as in simple decision analysis
• A cost can be assigned for each being in each health state
• Total costs and outcomes are then summed for all cycles

Question 9

Step 6

Answer 9

• Calculate cost-effectiveness
• ICERs
• One-way sensitivity analyses should be conducted: looks at uncertainty in assumptions in isolation (one variable at a time)

Question 10

Probabilistic Sensitivity Analyses

Answer 10

• Used to consider combined uncertainty in all model assumptions
• Run the model MANY times
• Each time randomly pull a different value for each of the assumptions based on the possible range of values for the assumptions
• Each iteration gets a different incremental cost, outcome, and ICER
• Multiple iteration results for incremental costs and outcomes are plotted on a “cost-effectiveness plane”

Question 11

Markov Disadvantages

Answer 11

• More complex than simple decision analysis/trees and therefore less transparent to decision makers
• Assumption: memoryless, not necessarily true in practice
• Data needed to estimate probabilities and costs, especially in the long term, often unavailable