Professional Documents
Culture Documents
MODELING
Dr. Dwi Endarti, M.Sc, Apt
Lab Manajemen Farmasi dan Farmasi Masyarakat
Departemen Farmasetika – Fakultas Farmasi UGM
1
Pendekatan studi Farmakoekonomi berdasarkan cara pengumpulan
data
Farmakoekonomi berbasis studi klinik
(“piggyback pharmacoeconomic study”/”pharmacoeconomic
study alongside clinical trial”)
2
Farmakoekonomi berbasis studi klinik
Economic Evaluation Alongside Clinical Trials (EEACTs) is direct observation of the impact
of a therapy on costs and effect
reimbursement
Farmakoekonomi berbasis studi observasional
“real world data” are observations of effects based on what happens after a prescriptive
(treatment) decision is made where the researcher does not or cannot control who gets
what treatment and does not or cannot control the medical management of the patient beyond
observing outcomes”
4
Farmakoekonomi berbasis modeling
5
Types of models commonly used
Decision tree
Markov model
Discrete event simulation
System dynamic model
Agent-based model/individual-based model
6
Decision trees
Earliest and most widely used
Structurally simplest
Simple directed graph without recursion
Pictorial representation of a decision situation
A chronological representation of the decision process
Present a sequence of decisions and chance events over time
Static model take a snapshot of a situation
Well suited to modeling outcomes that occur in the short-term and have a definite conclusion
7
Decision tree components
A network of three types of nodes:
Decision (choice) nodes
Chance (states of nature) nodes
Outcome (terminal or end) node
Probability
Each chance event is assigned a probability
Sum of alternatives =1 (100%)
Outcomes value
8
Example of a Decision Tree
Succes
Value_1
Medication A
Failure
Value_2
Disease X
Succes Value_3
Medication B
DECISION NODE
Failure
Value_4
9
CHANCE NODE TERMINAL NODE
Steps in decision tree analysis
1. Identify the decision
2. Structure the decision
3. Specify:
Consequences of each decision option over time
Probability that each consequence will occur
4. Determine the value of each outcome ($, LE, QALY)
5. Select the option with the highest expected outcome
10
Step 1: identify the decision
Who will be the decision maker?
What perspective will be considered?
What is the decision and what options will be considered?
Over what time span will the consequences be analyzed? (how long will study for the
decision?)
Example: consider a patient with signs and symptoms suggesting a disease X, there will be two
options:
Treatment with medication A or medication B
11
Step 2: structure the decision
Succes
Medication A
Failure
Disease X
Succes
Medication B
Failure
12
Step 3: specify probabilities of consequences for each options
Succes
p1=0.7
Medication A
Failure
Disease X p2=0.3
Succes
p3=0.8
Medication B
Failure
p4=0.2 13
Probability for each arm
1++ High-quality meta-analyses, systematic reviews of RCTs, or RCTs
More Reliability
with a very low risk of bias.
1+ Well-conducted meta-analyses, systematic reviews of RCTs, or
RCTs with a low risk of bias.
1- Meta-analyses, systematic reviews of RCTs, or RCTs with a high
risk of bias.
2++ High-quality systematic reviews of case control or cohort
studies. High-quality case control or cohort studies with a very
low risk of confounding, bias, or chance and a high probability
that the relationship is causal.
2+ Well-conducted case control or cohort studies with a low risk of
confounding, bias, or chance and a moderate probability that the
relationship is causal.
2- Case control or cohort studies with a high risk of confounding,
bias, or chance and a significant risk that the relationship is not
causal.
3 Non-analytic studies: for example, case reports, case series.
4 Expert opinion.
Less Reliability 14
Source: Based on Sackett and others (Canadian Task Force on the Periodic Health Examination)
Step 4: determine the value of each outcome (cost, LE, QALY)
Succes
u1
Medication A
p1
Failure
Disease X u2
p2
Succes
u3
Medication B
p3
Failure
u4
p4 15
Calculate the outcome values for each alternative
u1
Medication A
p1
p1u1 + p2u2
p2
Disease X u2
u3
p3
MedicationB
p3u3 + p4u4
p4
u4 16
Step 5: select the option with the highest expected outcome
17
Decision Tree Example
Illustrative example: Heparin for the prevention of deep vein thrombosis
(DVT) in hip replacement patients
Patients are at risk of DVT (and pulmonary embolism) post-surgery
Heparin can be injected pre-surgery and for 7-10 days post-surgery to try
and prevent clots
However, there are risks of bleeding
The research question:
‘Which is the more cost-effective treatment for hip replacement patients, heparin
or conventional treatment?’
Decision tree for heparin
Bleed
DVT
No bleed
LMW heparin
Bleed
No DVT
No bleed
Hip replacement patients
Bleed
DVT
No bleed
Conventional treatment
Bleed
No DVT
No bleed
Entering probabilities
Bleed
DVT 0.1
0.14 No bleed
LMW heparin 0.9
Bleed
No DVT 0.1
0.86 No bleed
Hip replacement patients 0.9
Bleed
DVT 0.01
0.25 No bleed
Conventional treatment 0.99
Bleed
No DVT 0.01
0.75 No bleed
0.99
Costs and Utilities
Costs assumed: Utilities assumed:
Cost of heparin - £300 DVT – 0.70
Cost of conventional treatment - £50 Bleed – 0.95
Cost of deep vein thrombosis event - DVT & bleed – 0.65
£2000 No event – 1.00
Cost of bleed - £500
Entering outcomes (QALYs) QALY
Bleed
0.65
DVT 0.1
0.14 No bleed
0.7
LMW heparin 0.9
Bleed
0.95
No DVT 0.1
0.86 No bleed
1.00
Hip replacement patients 0.9
Bleed
0.65
DVT 0.01
0.25 No bleed
0.7
Conventional treatment 0.99
Bleed
0.95
No DVT 0.01
0.75 No bleed
1.00
0.99
Bleed
2800
DVT 0.1
0.14 No bleed
2300
LMW heparin 0.9
Bleed
800
No DVT 0.1
0.86 No bleed
300
Hip replacement patients 0.9
Bleed
2550
DVT 0.01
0.25 No bleed
2050
Conventional treatment 0.99
Bleed
550
No DVT 0.01
0.75 No bleed
50
0.99
Full structure of cost-effectiveness analysis
Bleed
2800 / 0.65
DVT 0.1
0.14 No bleed
2300 / 0.7
LMW heparin 0.9
Bleed
800 / 0.95
No DVT 0.1
0.86 No bleed
300 / 1.00
Hip replacement patients 0.9
Bleed
2450 / 0.65
DVT 0.01
0.25 No bleed
2050 / 0.7
Conventional treatment 0.99
Bleed
550 / 0.95
No DVT 0.01
0.75 No bleed
50 / 1.00
0.99
Example: analysing the tree (output as QALYs)
Bleed
0.65
DVT 0.1 A
No bleed
A
0.14
0.7
LMW heparin 0.9
0.95 QALYs C
Bleed
0.95
No DVT 0.1 B
No bleed B
0.86
1.00
Hip replacement patients 0.9
LMW heparin : 0.95 QALYs
Bleed
0.65
DVT 0.01
No bleed
DD
0.25
0.7
Conventional treatment 0.99
0.92 QALYs FF Bleed
0.95
No DVT 0.01
EE
0.75 No bleed
1.00
0.99
Rollback Calculations
Work from terminal nodes towards decision nodes
31
Markov Model Transient or
Recurrent state: can
leave, stay, or return
Absorbing state:
individuals who enter
that state cannot
transition out of it
33
1. Identify Markov health states
Use the literature and expert opinions to identify health states
It may be better to define health state according to what makes sense from a
disease progression perspective
Definition of states must match cycle length
Different types of states:
Absorbing states: subjects remain there (“death”)
Non-absorbing states: subjects can exit state by getting worse or better
34
2. Identify cycle length
Cycle length must match the definition of the health states
To ensure that the length of the cycle reflects:
The nature of the disease and the timing of outcomes:
Acute disease require short cycles
When events can repeat frequently, a short cycle is also appropriate
Chronic disease may require longer cycles
Normally, the longest of cycle is 1 year
35
3. Identify allowable transitions
Focus is on: Example of defining allowable transitions:
Structure of the model Asymptomatic disease
Relationships between model states Asymptomatic disease
The best approach may be to: Asymptomatic disease Progressive
disease
Develop comprehensively the entire
clinical pathway from the initial to the Asymptomatic disease Death
final states Progressive disease Progressive
Create synthetic summary and disease
representation of the model Progressive disease Death
Death Death (absorbing state)
36
4. Identify transition probabilities
Transitional probabilities can be estimated from primary or secondary data sources:
Mortality: life tables and actuarial databases
Clinical trials, registries, retrospective databases
Literature
Subjective opinions from experts
Pure assumptions
Transitional probabilities:
Probability that of going from state A to state B in one cycle
Likelihood that an event will occur in a given time period (between 0 – 1)
Unitless
37
Example of transition probabilities
38
5. Identify outcomes value
Outcomes value = payoffs associated with the state:
Survival (Life expectancy)
QALY
Costs
39
6. Program the model and run the simulation
There is 2 approach can be used to program Markov model:
Markov cohort simulation:
determines the prognosis of an entire cohort of patient all at once, by splitting the initial cohort into all the
different states of the model
First order Monte-Carlo simulation:
Follows a large number of patients individually through the model; using random numbers and probability to
solve problems
40
7. Interpreting and Presenting the results
Average life expectancy for the cohort of patient simulated
= Sum of all person-years across all cycles and patients
Number of patients entering the simulation
Lifetime cost/QALY
= Total cost/QALY
Number of patients entering the simulation
41
Illustration: markov cohort simulation
42
State-Transition Diagram for Atrial Fibrillation
State of Next Cycle
State of Well Post- Dead
Current Cycle Stroke
p11=0.7
Well 0.7 0.2 0.1
p22=0.9
Post- Dead
Stroke p23=0.1
p33=1.0
The probabilities for all paths out of a state must sum to 1.0.
Evaluating Markov Models:
Cohort Simulation
State
Cycl Well Post- Dead Sum of Survival Check
e Stroke Years
Lived
0 10,000 0 0 10,000
1 7,000 2,000 1,000 9,000 0.9000 10,000
2 4,900 3,200 1,900 8,100 0.8100 10,000
3 3,430 3,860 2,710 7,290 0.7290 10,000
4 2,401 4,160 3,439 6,561 0.6561 10,000
5 1,681 4,224 4,095 5,905 0.5905 10,000
6 1,176 4,138 4,686 5,314 0.5314 10,000
7 824 3,959 5,217 4,783 0.4783 10,000
93 0 1 9,999 1 0.0001 10,000
94 0 0 10,000 0 0.0000 10,000
State
Cycle Well Post-Stroke Dead Sum of Years Survival Check
Lived
0 10,000 0 0 10,000
47