recent research activities at uef mixed treatment ...drugis.org/files/kuopio2011/pres3.pdf ·...

Recent research activities at UEF

Mixed treatment comparison of triptans

and cost-effectiveness evaluation as an

example

Christian Asseburg

Lecture at 10:00 on 9.6.2011

Study group members

Piia Peura, Finnish Medicines Agency (Fimea)

Janne Martikainen, Juha Turunen, Timo Purmonen, Emma Pänkäläinen, Tuija Oksanen, Christian Asseburg (UEF)

Funding

Finnish Medicines Agency (Fimea)

9.6.2011Triptans MTC / Christian Asseburg 3

Overview

Background

•Fimea, Triptans

•Prior work on triptans cost-effectiveness

Mixed treatment comparison

•Systematic review

•Clinical outcomes

Model development

Model fitting

•WinBUGS

Results

•Strengths and weaknesses


BackgroundFinnish Medicines Agency (Fimea)

”Fimea is the national competent authority for regulating pharmaceuticals. *…+ Fimea’s aim is to improve the pharmaceutical service for the population and the safety, appropriateness and economy of pharmacotherapy.” (www.fimea.fi)

However, in Finland pharmaceutical prices are set by:

•Kela, the social insurance institution, for pharmacy prescription products: Pharmaceutical companies apply for reimbursement

• Individual hospital districts: Pharmaceutical companies can negotiate bulk pricing individually

When treatment innovations move costs from the hospital sector to the pharmacy sector, it is unclear how the current system of reimbursement decisions can allocate societal spending optimally.


Background

Pharmaco-economics

?

Diagram:www.fimea.fi


Background: Triptans

Triptans are selective serotonin 5-HT1B/1D receptor agonists that were introduced in the early 1990s and are recommended as first-line treatment of severe migraine.

• In Finland, the following triptans are available (2009): almotriptan, eletriptan, frovatriptan, naratriptan, rizatriptan, and sumatriptan

•Sumatriptan went generic in 2008.

•Cost-effectiveness of these triptans hasn’t been assessed in Finland.

Prior work:

Ramsberg, J and Henriksson, M. The cost-effectiveness of oral triptan therapy in Sweden. Cephalalgia. 2007; 27: 54-62

Ferrari, M D, et al. Triptans (serotonin, 5-HT1B/1D agonists) in migraine: detailed results and methods of a meta-analysis of 53 trials. Cephalalgia. 2002; 22: 633-658

BackgroundPrior work: Ramsberg and Henriksson 2007

Cost-effectiveness model

•24 hour horizon

•Direct medication cost

•Indirect costs due to lostproductivity

•Nodes:

– Adverse events (AE)

– Pain-free 2 hours (PF2h)

– Recurrence (Rec)

– Sustained pain-free no AE (SNAE)


Begin treatment

No AE

AE

PF 2h

No PF 2h

No Rec

Rec

No Rec

Rec

PF 2h

No PF 2h

1

2

3

4

5

6

BackgroundPrior work: Ferrari et al. 2002

Meta-analysis on 53 trials

•Pain free 2h, response 2h, recurrence of headache, AEs, and sustained pain free assessed individually.

•No network meta-analysis

•Placebo-controlled studies on the samedrug are pooled (random effects)

•Results presented as placebo-subtractedas well as absolute probabilities

•Qualitative discussion of head-to-head active comparator trials



Mixed treatment comparisonSystematic review

We updated the Ferrari et al. systematic review.

•Several studies that were included in Ferrari et al. failed ourinclusion criteria (usually because the studies had not beenpublished). We identified additional studies, including several thatwere more recent than the previous review.

•56 studies qualified for inclusion.

•54 studies reported on ”Response 2h”.

•49 studies reported on ”Pain free 2h”.

•35 studies reported on ”Recurrence”.

•45 studies reported on ”Adverse events”.

•Mixed treatment comparison is required.


Mixed treatment comparisonNetwork of evidence (shown for “pain free 2h”)


Mixed treatment comparisonClinical outcomes

Unclear relationship between different outcomes, e.g.

•“Pain free 2h” is conditional on “Response 2h”.

•Are treatments with higher probability of response (i.e. fast onset)

– more likely to result in sustained pain-free (i.e. generally effective),

– or less likely (i.e. effect wears off fast),

– or no relationship between these? or treatment-dependent?

• Is there a link between placebo response and verum response, i.e. in trials with high placebo response rate, is a high response to active treatment

– more likely?

– less likely?

– no link?


Model development

The model presented here builds on ideas from two papers:

•Lu, G and Ades, AE. Combination of direct and indirect evidence in mixed treatment comparisons. Statistics in Medicine. 2004; 23: 3105-24.

– This will be covered in Geert van Valkenhoef’s presentations.

•Arends, LR, Vokó, Z and Stijnen, T. Combining multiple outcome measures in a meta-analysis: an application. Statistics in Medicine. 2003; 22: 1335-1353

– Meta-analysis of surgery or traditional treatment for stroke prevention

– Outcomes:(1) Number of events in the traditional group,(2) Number of events in first month after operation,(3) Number of events post-1-month after operation


Model development (2)

Thoughts about fixed/random effects, correlations, conditionality...

• In a Bayesian random-effects model, both the treatment effect and the between-trials variation needs to be estimated.

– Requires at least a few trials for each comparator (or assumption thatbetween-trials variation is known/identical for different comparators)

•Estimating correlation between outcomes:

– Correlation between the baselines (e.g., placebo arms in trials)

– Correlation between study-specific baseline and treatment effects on the same outcome

– Correlation between baselines and treatment effects on differentoutcomes

• Ideally model conditionally independent outcomes

Model fitting

Bayesian model coded in BUGS language

•“Standard” code from Lu and Ades for MTC.

•Adapted to random baselines, fixed treatment effects.

•Own code for multiple outcomesand correlated baselines

•Data entry in Excel, transfer to OpenBUGS using R and statconn.

model {# ND observations. XS, XO, XT index the study, outcome and treatment# into the observation vector.# Modelled probability p is indexed by study, outcome and treatment.for (i in 1:ND) {

r[i]~dbin(pred[i],n[i])pred[i]<-p[XS[i],XO[i],XT[i]]

}

# Calculation of trial-specific outcomes, NS studiesfor (i in 1:NS) {

# Correlated random baseline for the four outcomeslos[i,1:NO]~dmnorm(mu[],Omega[,])for (j in 1:NO) {

for (k in 1:NT) {# Log-odds scale. See Lu and Ades (2004) for matrix A.logit(p[i,j,k]) <- los[i,j]+

inprod(A[,k],tx[((j-1)*(NT-1)+1):(j*(NT-1))])}

}}

# Priors: Vague priors on all the baselines and tx effectsfor (i in 1:NO) {

mu[i]~dnorm(0,0.001) # Random effects -> los}for (i in 1:(NO*(NT-1))) {

tx[i]~dnorm(0,0.0001) # Fixed treatment effects}

# Construct the variance-covariance matrix for the baselines,# a priori uncorrelated.for (i in 1:NO) { for (j in 1:NO) {

varcovmu[i,j]<-equals(i,j)}}# Vague Wishart priorOmega[1:NO,1:NO]~dwish(varcovmu[,],NO)

# Parametrisation of treatment effects, see Lu and Ades (2004).# From their matrix A, I moved the placebo treatment to the end.for (i in 1:(NT-1)) { for (k in 1:NT) {

A[i,k] <- equals(i,k) - 1/NT}}

}


Model fittingWinBUGS / OpenBUGS

Numerical sampling, requires expertise and skill – not an automatic procedure. From the OpenBUGS manual:

• “Potential users are reminded to be extremely careful if using this program for serious statistical analysis. We have tested the program on quite a wide set of examples, but be particularly careful with types of model that are currently not featured. If there is a problem, OpenBUGS might just crash, which is not very good, but it might well carry on and produce answers that are wrong, which is even worse. Please let us know of any successes or failures.”


Model fittingGoodness-of-fit

Graph of empirical against theoretical quantiles:

• If the model specifies the empirical distribution correctly, the data points should follow the unit line.

• Here, an S-shape is apparent –especially regarding data points where the observed probabilities were lower than predicted.

• Closer examination revealed that almost all these “outliers” were on placebo arms.

– Assumption of fixed effect may not be appropriate for the placebo arms.


0,00

0,10

0,20

0,30

0,40

0,50

0,60

0,70

0,80

0,90

1,00

0,00 0,20 0,40 0,60 0,80 1,00


ResultsMixed treatment comparison

Illustrating just one outcome:Sustained pain-free, no adverse event (SNAE)

0 0,05 0,1 0,15 0,2 0,25 0,3

placebo

zolmitriptan 5 mg

zolmitriptan 2.5 mg

sumatriptan 100 mg

sumatriptan 50 mg

rizatriptan 10 mg

rizatriptan 5 mg

naratriptan 2.5 mg

frovatriptan 2.5 mg

eletriptan 40 mg

almotriptan 12.5 mg

SNAE

ResultsCost-effectiveness: Cost per QALY gained


0,0%

20,0%

40,0%

60,0%

80,0%

100,0%

0 10000 20000 30000 40000 50000 60000 70000 80000 90000 100000

Pro

bab

ilit

y o

f co

st-

effe

ctiv

enes

s

Willingness-to-pay per QALY

eletriptan 40 mg

rizatriptan 10 mg

sumatriptan 50 mg

sumatriptan 100 mg

Frontier

Treatment Total costs (€) Total utility (QALYs*10-5) Health-economic summary

Naratriptan 2.5 mg 28.91 (23.10, 36.15) 12.7 (4.4, 22.4) Dominated

Frovatriptan 2.5 mg 27.46 (21.72, 34.72) 17.8 (6.8, 31.5) Dominated

Almotriptan 12.5 mg 27.66 (22.22, 34.49) 25.2 (18.4, 32.7) Dominated

Sumatriptan 50 mg 21.31 (16.06, 27.90) 31.1 (24.7, 38.2) Dominated

Zolmitriptan 2.5 mg 26.81 (21.56, 33.36) 32.6 (25.2, 40.9) Dominated

Sumatriptan 100 mg 20.86 (15.75, 27.25) 37.0 (30.0, 44.7) Base-case

Zolmitriptan 5 mg 28.46 (23.41, 34.78) 38.3 (29.3, 48.3) Dominated

Rizatriptan 10 mg 26.37 (21.48, 32.46) 47.3 (39.0, 56.5) Dominated

Eletriptan 40 mg 23.64 (18.94, 29.52) 51.1 (42.2, 61.0)

ICER to Sumatriptan 100 mg:

€19,659


Strengths and WeaknessesStrengths

•Sound methodology of mixed treatment comparison

•Simultaneous model for all relevant clinical endpoints


Strengths and weaknessesWeaknesses

•Model code had to be written specifically for this project

– Prone to errors

– MTC model design decisions may not be universally valid for allprojects

•No automatic model-fitting

– Manual supervision necessary

– Model-fitting requires some expertise and results may vary slightlybetween runs

•Poor integration between Excel and fitting software (WinBUGS)

•No standardised goodness-of-fit tests etc.

Thank you!Questions, comments?

[email protected]

www.uef.fi

recent research activities at uef mixed treatment ...drugis.org/files/kuopio2011/pres3.pdf ·...

Documents