© Health Economics Group 2016 Page 1 of 3
Incorporating Health Economics into Grant Proposals Health Economics Short Course
For more information and course dates, please visit our website: http://go.unimelb.edu.au/i8ba
Or email us: [email protected]
Modelling Extrapolation and Modelling Longer Term Outcomes and Costs
Centre for Health Policy Melbourne School of Population and Global Health
0.5
0.55
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
QA
LYs
Months
Standard practice treatment
Extrapolation - Outcomes
Normally the primary outcome is
compared at a particular point in
time for significance
But to find the benefit of an intervention versus standard practice we
need to know the difference at all points in
time
End of follow-
up
We need to work out the area between
these curves
Extrapolation - Outcomes
0.5
0.55
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
QA
LYs
Months
Standard practice treatment Standard practice (P) treatment (P)
But we don’t just want to know the benefit until the end of the follow-up but also the benefit of the intervention post follow-up
The benefit post follow-up can be as big if not bigger than the within follow-up benefit
End of follow-up
Extrapolation - Costs
Costs are no different to
outcomes – there may be large cost
savings in the difference between
treatment arms post follow-up
0
2000
4000
6000
8000
10000
12000
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Cos
t per
mon
th $
Months
Standard practice treatment Standard practice (P) treatment (P)
So it could be important to extrapolate
Our new treatment might be expensive in the short run but save resources in the long run
© Health Economics Group 2016 Page 2 of 3
Intervention to reduce youth binge drinking?
• QALYs in the short-term often do not change much • Most of the health gains are unlikely to be seen for many
years • So there is a need to estimate/predict them • We need to make sure we have a way to change short-
term changes in risk factors into long-term changes in health
• Often if we are improving long-term health this also means we are reducing long-term health expenditure (or at least delaying this expenditure) and so modelling the long-term cost implications of this is also important to determine the most approach treatment decision
Modelling
• Conceptually extrapolation into the future makes sense… • …similarly translating intermediate outcomes into final
outcomes • …but how do we do it?
• Ideally we want to predict the differences between the new treatment and standard practice for both costs and outcomes until death (for a lifetime)
Modelling Outcomes
• Extrapolation – Parametric modelling of survival analysis – Regression modelling – ‘Big data’ predictive approaches
• Translate impact on intermediate outcomes into final outcomes using: – Estimates from the literature e.g. relative risks on morbidity (QALYs)
and mortality – Information from other data sources e.g. registries or longitudinal
studies – Microsimulation modelling – combine evidence from a range of sources
to model life-course
Modelling Outcomes
• Estimate QALYs for people with Type 2 diabetes, based on profile of complications of each patient over their remaining lifetime
• Use UKPDS Outcomes Model which is based on an integrated set of parametric proportional hazard models to predict absolute risk of first occurrence of seven major diabetes-related complications, using: – patients’ characteristics – time varying risk factors (e.g. HbA1c)
10
Simulating lifetime outcomes
Average QALY for UKPDS patients by blood glucose policy
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44
Years (from diagnosis of diabetes)
QA
LY (U
tility
wei
ght)
Intensive Therapy
Conventional therapy
Med
ian
dura
tion
of th
e tr
ial
Minimumfollow-up
Maximumfollow-up
11
Costs are also attached to health events
Benefits are mostly post
follow-up
Small difference
in trial Here the red that we can see represents the lifetime benefit
Modelling Costs
• Often costs may be driven by health outcomes – once we have extrapolated outcomes we can link this to costs e.g. cost
of a CVD event such as a stroke
• Observational data sets are also very useful to estimate the resource implications of particular health outcomes – predict the costs (healthcare utilisation) associated with different stages
of COPD using routinely collected data sources
• Recruit an observational sample of those treated in the past (using standard practice) to better understand future cost (& health) implications of the disease
© Health Economics Group 2016 Page 3 of 3
Modelling - Uncertainty
• Modelling into the unknown often means that there is considerable uncertainty (so we need to make sure we capture and present this)
• But not predicting into the future means that we have
biased estimates and we are simply ignoring this uncertainty in terms of benefits & costs
• So when we are designing a study we need to think about data which are going to help us predict into the future