HTA - lecture 5 - introduction to modelling

Question 1

probabilities

Answer 1

Conditional probabilities:  condition on something else that could happen
Probabilities in tree are conditional probabilities, e.g.
Given treatment A, P(cured)=0.2 = 20%
Given cured, P(cancer back)=0.1 = 10%

Path probabilities:  entire population
Path probabilities are unconditional probabilities, e.g.
P(Treatment A & cured & cancer back& cured) = 0.2% = 0.002

Question 2

model structures

Answer 2

• Decision analytic models (decision trees):
  o Short periods/ single events
   Outcome of genetic test for example
   Chronic diseases are not good for a decision tree
• Markov models:
  o Reflects continuous risk of event over longer period
   Osteoporosis: patient continuously ‘at risk’ for fracture
   High blood pressure: patient continuously ‘at risk’ for cardiovascular disease
   COPD: patients are ‘at risk’ to deteriorate over time (natural disease progression) of improve when new treatment is started
   Cancer patients are at risk of progression (cancer comes back)

Question 3

markov modelling

Answer 3

• Reflects continuous risk of event over longer period
• Organized around health states rather than pathways
• Probabilities relate to transitions between health states
• Cycle length defines period of transition

Question 4

the markov assumption

Answer 4

• The Markov assumption states that the probability of moving from one state to the other only depends on the current state
  o All people have the same chance of staying healthy, even though you were ill before
• So:
  o Previous health states are disregarded
  o Time spent in current state is disregarded
• Helpful for continuous risks

Critique: people sometimes only get a disease once in their life. When modelling a group, the group can be heterogeneous

Question 5

modelling pros

Answer 5

Modelling pro’s
* Makes explicit definition of relevant patient group, clinical events, patient outcome, costs etc. necessary
* Shows what data and information are lacking
* Makes it possible to examine the impact of input uncertainty on outcome
* Relatively fast and simple (compared to empirical research), and relatively cheap

Question 6

modelling cons

Answer 6

Modelling con’s
* (Over)simplifies the complicated, real world
* Model structure subject to bias
* Model input subject to bias
* Misinterpretation of the results is easy