markov model in health economic eva -...
TRANSCRIPT
Markov Model in Health Care
Septiara PutriCenter for Health Economics and Policy Studies
4/10/2015 1
Objectives
• Introduce and familiarize participants with basic Markov modeling for economic evaluation
• Familiarize participants with ‘step by step’ Markov model construction
• Prepare participants to conducting health economic models (Microsoft Excel application)
4/10/2015 2
Introduction “a health-care evaluation model as an analytic methodology that
accounts for events over time and across populations, that is based on data drawn from primary and/or secondary sources, and whose
purpose is to estimate the effects of an intervention on valued health consequences and costs” [1]
• Decision analysis: systematic-mathematical approach, informing clinical decisions, uncertainty in decision making
• Decision analytic model have been increasingly applied in health economic evaluation
• Markov modeling for health economic evaluation
4/10/2015 3
[1] Weinstein, Milton C., et al. "Principles of Good Practice for Decision Analytic Modeling in Health-Care Evaluation: Report of the ISPOR Task Force on Good Research Practices—Modeling Studies." Value in health 6.1 (2003): 9-17.
Decision Tree: An overview
4/10/2015 4
Which treatment should be chosen?
Disease A
Drug A
Drug B
Die
Survive
Die
Survive
Severe
Moderate
Mild
Severe
Nodes:decision point
between treatment options
possible events that patients experienced
terminal nodes
*Mutually exclusive for pathway*probability should be 1.0 in the end
Decision tree
4/10/2015 5
0.60
0.50
0.40
0.50
0.40
0.45
0.15
0.40
0.50
0.10
Hint: Expected Cost= probability x costExpected utility=probability x utility
Limitations of Decision Tree
4/10/2015 6
Limited for predicting continuous values Expected cost and effect/utility are less visible Limited to one output Difficulties for regression application and
representing long term prognoses Difficulties to apply when patients experiencing
recurrence events, particularly chronic disease Applying discounting and outcome over time
become issues. Inability to determine when events occur (start and
later?) Possibly excessive branches
Markov
Markov Process• Form of stochastic
process• Future event could
not be accurately predicted by past event
• Able to look at long sequence
• Memory less assumption
4/10/2015 7
Markov Chains
• Discrete time• Homogenous transition
probabilities • Statistical model
application real world
• Probability of particular transitionstransition matrix
Steps in ConstructingMarkov Model
Determine relevant states
• Mutually exclusive• Representative (clinically and
economically)• Available and consistent
evidence of clinical pathway-not contradicted, widely accepted
• Determine cycle length• Time horizon-long enough
4/10/2015 8
Asymptomatic
Progressive
Death
Steps in ConstructingMarkov Model
Determine transitions
• Determines relevant transition from each states
• Identify transition probabilities (most representative data)
• Cycle length should be reflective
4/10/2015 9
Asymptomatic
Progressive Death
Asymptomatic
Steps in ConstructingMarkov Model
Model parameters
• Outcomes? (Life years, QALYs)
• Determines transition probabilities, utilities, effectiveness and costs related to states
• To understand: transition matrix
4/10/2015 10
Progressive Death
A to P
P to D
A to D
Analyzing Markov Model • Model ParametersTransition Probabilities:A to P: 0.105P to D: 0.411A to D: 0.202
Costs (progressive state) : Rp 12,000,000 (drug A) Rp9,500,000 (drug B)
Utility (progressive state): 0.70 (drug A) 0.82 (drug B)Effectiveness: RR: 0.60Discounting: 3% (both cost and effect)
4/10/2015 11
Analyzing Markov Model • Transitions Matrix
4/10/2015 12
Asymptomatic Progressive Death
Asymptomatic A to A A to P A to D
Progressive 0 P to P P to D
Death 0 0 D to D
Asymptomatic Progressive Death
Asymptomatic 1-0.105-0.202 0.105 0.202
Progressive 0 1-0.411 0.411
Death 0 0 1
Cohort Simulation
4/10/2015 13
1000 00
105693 2022
480 135 385
Asymptomatic DeathProgressive Cycles
1
3
2
0.693=693
0.693=480
1.00=202
0.589=62
Analyzing Markov Model • Cohort Analysis
4/10/2015 14
Life Years
Asmptomatic Progressive Death Check Uncorrected Corrected
0 1 No Discounting
1 0,693 0,105 0,202 1,000 0,798 0,899
2 0,480 0,135 0,385 1,000 0,615 0,706
3 0,333 0,130 0,537 1,000 0,463 0,539
4 0,231 0,111 0,658 1,000 0,342 0,402
5 0,160 0,090 0,750 1,000 0,250 0,296
6 0,111 0,070 0,820 1,000 0,180 0,215
7 0,077 0,053 0,871 1,000 0,129 0,155
8 0,053 0,039 0,908 1,000 0,092 0,111
9 0,037 0,029 0,935 1,000 0,065 0,079
10 0,026 0,021 0,954 1,000 0,046 0,056
Incremental Cost Effectiveness Ratio
4/10/2015 15
Sensitivity Analysis [2]
4/10/2015 16
“There remains several types of uncertainty relatedmethods in analysis”
“Sensitivity analysis can help the reviewer to determinewhich parameters are the key drivers of a model’sresults”
“By reporting extensive outputs from sensitivityanalysis, modellers are able to consider a wide range ofscenarios and, as such, can increase the level ofconfidence that a reviewer will have in the model”[2] Taylor Matthew, 2009. Retrieved fromhttp://www.medicine.ox.ac.uk/bandolier/painres/download/whatis/What_is_sens_analy.pdf
One way sensitivity analysis• The simplest form of sensitivity
analysis• vary one value in the model by a
given amount, and examine theimpact that the change has on themodel’s results.
• This is known as one-way sensitivityanalysis, since only one parameter ischanged at one time.
[2] Taylor Matthew, 2009. Retrieved fromhttp://www.medicine.ox.ac.uk/bandolier/painres/download/whatis/What_is_sens_analy.pdf[3] Tornado diagram showing results of the one-way sensitivity analysesfor the difference (vaccination versus no vaccination) in total cost pervaccinated child. The vertical line indicates where the incremental costsfor the vaccine strategy were more than €0.Knoll et al. Health Economics Review 2013 3:27 doi:10.1186/2191-1991-3-27
4/10/2015 17
Tornado diagram
Two way sensitivity analysis
4/10/2015 18
Smith, T., Jordaens, L., Theuns, D. A., van Dessel, P. F., Wilde, A. A., & Hunink, M. M.(2013). The cost-effectiveness of primary prophylactic implantable defibrillatortherapy in patients with ischaemic or non-ischaemic heart disease: a Europeananalysis. European heart journal, 34(3), 211-219.
• Assesing the impact of pairs of variables in a CEA by varying them across a plausible range of values and combinations with all other variables constant at their baseline value.
• If varying more than two, it is called multi-way sensitivity analysis'.
Probabilistic sensitivity analysis (PSA)
4/10/2015 19
“A form of sensitivity analysis inwhich probability distributionsare applied to the ranges for amodel’s input parameters, andsamples from these distributionsare drawn at random togenerate an empiricaldistribution of the relevantmeasure of cost-effectiveness.”
Andronis, L., Barton, P., & Bryan, S. (2009). Sensitivity analysis in economicevaluation: an audit of NICE current practice and a review of its use andvalue in decision-making.
Pharoah, P. D., Sewell, B., Fitzsimmons, D., Bennett, H. S., & Pashayan, N. (2013). Cost effectiveness of the NHS breast screening programme: life table model. BMJ: British Medical Journal, 346.
Probabilistic sensitivity analysis (PSA)
4/10/2015 20
Useful Reading Lists• Briggs, Andrew, Mark Sculpher, and Karl Claxton. Decision modelling
for health economic evaluation. Oxford university press, 2006.• Philips, Zoe, et al. "Good practice guidelines for decision-analytic
modelling in health technology assessment”• Weinstein, Milton C., et al. "Principles of Good Practice for Decision
Analytic Modeling in Health-Care Evaluation: Report of the ISPORTask Force on Good Research Practices—Modeling Studies." Valuein health 6.1 (2003): 9-17.
• Weinstein, Milton C. "Recent developments in decision-analyticmodelling for economic evaluation." Pharmacoeconomics 24.11(2006): 1043-1053.
• Briggs, Andrew, and Mark Sculpher. "An introduction to Markovmodelling for economic evaluation." Pharmacoeconomics 13.4(1998): 397-409.sment." Pharmacoeconomics 24.4 (2006): 355-371.
4/10/2015 21
Practical
• All the parameters information has already filled
• Move to sheet 3• Follow the available directions• Remember that all parameters are named,
just put the formula directly and running your model
• If your sum probability is not 1.0 yet…just relax!, try once more time
4/10/2015 22