Meta-modelling Markov Model Simulations for cost effectiveness - - PowerPoint PPT Presentation
Meta-modelling Markov Model Simulations for cost effectiveness analyses ICTR-PHE 2012 Daniel Abler 1 Steve Harris 2 Jim Davies 2 1CERN 2Department of Computer Science, University of Oxford daniel.abler@cern.ch , { steve.harris, jim.davies }
ICTR-PHE 2012 Daniel Abler 1 Steve Harris2 Jim Davies 2
1CERN 2Department of Computer Science, University of Oxford daniel.abler@cern.ch, {steve.harris, jim.davies}@cs.ox.ac.uk
01.03.2012
Daniel Abler (CERN) Meta-modelling Markov Model Simulations 01.03.2012 1 / 6
. Cost-Effectiveness studies... . . . . . ..., it is recommended not to adopt particle therapy as standard treatment in NSCLC yet. More evidence is needed ...
[Grutters et al., The cost-effectiveness of particle therapy in non-small cell lung cancer: Exploring decision uncertainty and areas for future research, Cancer Treatment Reviews, 36, 6, 2010]
. ... and reporting guidelines . . . . . information about: Study Design Data Collection Analysis and interpretation of results such as: effectiveness, quality, costing data details on modelling ...
[Drummond et al. Guidelines for authors and peer reviewers of economic submissions to the BMJ, BMJ 1996;313:275]
. . . . . . . . .
Daniel Abler (CERN) Meta-modelling Markov Model Simulations 01.03.2012 2 / 6
. Cost-Effectiveness studies... . . . . . ..., it is recommended not to adopt particle therapy as standard treatment in NSCLC yet. More evidence is needed ...
[Grutters et al., The cost-effectiveness of particle therapy in non-small cell lung cancer: Exploring decision uncertainty and areas for future research, Cancer Treatment Reviews, 36, 6, 2010]
. ... and reporting guidelines . . . . . information about: Study Design Data Collection Analysis and interpretation of results such as: effectiveness, quality, costing data details on modelling ...
[Drummond et al. Guidelines for authors and peer reviewers of economic submissions to the BMJ, BMJ 1996;313:275]
. Objective . . . . . To facilitate exchange, interpretation and re-use of Markov Model Simulations (MMS) by creating a candidate model sufficiently expressive to describe modelling assumptions, data input and computational specifications.
Daniel Abler (CERN) Meta-modelling Markov Model Simulations 01.03.2012 2 / 6
. Markov Models . . . . . model stochastic processes here: disease progress of a patient costs and utilities assigned to states and transitions probabilities assigned to transitions between states . . s1 . start . assymptomatic . s2 . progressive disease . s3 . death . p12 . p23 . p13 . p11 . p22 . p33 = 1 . . State . Transition . PayOff .
2
.
1..*
. * .
0..*
. * .
0..*
Daniel Abler (CERN) Meta-modelling Markov Model Simulations 01.03.2012 3 / 6
. features . . . . . Instance of Markov Model Simulation specifies
simulation settings Markov Model instances (and thus Values) used in simulation results obtained by simulation
→ can serve as processing instruction for simulation programme and documentation of computed MMS . . Metamodel of Markov Model Simulations . specific Markov Model Simulation . specific Markov Model Simulation . specific Markov Model Simulation . instance . .
Markov Model Simulation
.
simulation settings
e.g. halfCycle correction, initialAge, modalities,...
results
.
Markov Model
.
modalities payOff classes cycle duration
.
Values
.
distribution of data source unit ...
.
*
.
1..*
.
*
.
1..*
.
Results results for individual markov model comparative results
.
State
.
Transition
.
PayOff
.
2
.
1..*
.
*
.
0..*
.
*
.
0..* Daniel Abler (CERN) Meta-modelling Markov Model Simulations 01.03.2012 4 / 6
. .
markovModelSimulation AnExampleSimulation { s i m u l a t i o n S e t t i n g s{ simulationType : d e t e r m i n i s t i c numberOfCycles {markovModel G ru tte rs 20 10 1 : 1 , markovModel G rut te rs 20 10 2 : 5} h a l f C y c l e C o r r e c t i o n : 1 u s e M o d a l i t i e s {modalityType p r o t o n , modalityType carbon} useMarkovModels {markovModel G r u t t e r s 2 0 1 0 1 , markovModel Gr ut te r s2 01 0 2} t r a n s f e r {markovModel G ru tt e r s20 10 1 : s t a t e State treatmentDeath − > markovModel G ru tt e r s20 10 2 : s t a t e S t a t e d e a t h , . . . } markovModel G ru tt e r s2 010 1 { . . . } markovModel G ru tt e r s2 010 2 { . . . } }
.
[Language Workbench Spoofax: http://strategoxt.org/Spoofax]
.
[example from: Grutters et al., The cost-effectiveness of particle therapy in non-small cell lung cancer: Exploring decision uncertainty and areas for future research, Cancer Treatment Reviews, 36, 2010] Daniel Abler (CERN) Meta-modelling Markov Model Simulations 01.03.2012 5 / 6
. .
markovModelSimulation AnExampleSimulation { s i m u l a t i o n S e t t i n g s{ simulationType : d e t e r m i n i s t i c numberOfCycles {markovModel G ru tte rs 20 10 1 : 1 , markovModel G rut te rs 20 10 2 : 5} h a l f C y c l e C o r r e c t i o n : 1 u s e M o d a l i t i e s {modalityType p r o t o n , modalityType carbon} useMarkovModels {markovModel G r u t t e r s 2 0 1 0 1 , markovModel Gr ut te r s2 01 0 2} t r a n s f e r {markovModel G ru tt e r s20 10 1 : s t a t e State treatmentDeath − > markovModel G ru tt e r s20 10 2 : s t a t e S t a t e d e a t h , . . . } markovModel G ru tt e r s2 010 1 { . . . } markovModel G ru tt e r s2 010 2 { . . . } }
.
State_init, PO_treatTime_CI, PO_treatCost_CI S_treatDeath , PO_DeathTreat State_treatNoAcuteAes, PO_UtilityNoAeDurTreat 1 - TPacPneumGt3 + TableTPacOesophGt3, POCostsDyspnPerYear State_death transfer State_woDysp, PO_FollowUp, PO_UtilityNoAeAfterTreat transfer PO_DeathOther TPdiseaseMort, PO_DeathCancer State_withDysp, PO_CostsDyspnPerYear, PO_FollowUp, PO_UtilityDyspAfterTreat TPotherMort TPotherMort, PO_DeathOther TPdiseaseMort, PO_DeathCancer
. Visualisation of states and transi- tions used in this study, generated from the MMS description. .
[Language Workbench Spoofax: http://strategoxt.org/Spoofax]
.
[example from: Grutters et al., The cost-effectiveness of particle therapy in non-small cell lung cancer: Exploring decision uncertainty and areas for future research, Cancer Treatment Reviews, 36, 2010] Daniel Abler (CERN) Meta-modelling Markov Model Simulations 01.03.2012 5 / 6
. .
.
Meta-modelling Markov Model Simulations for cost-effectiveness analyses
Daniel Abler1, Steve Harris2, Jim Davies2
1European Centre for Nuclear Research, Geneva CH-1211, Switzerland 2Department of Computer Science, University of Oxford, Oxford OX1 3QD, UKdaniel.abler@cern.ch, {steve.harris, jim.davies}@cs.ox.ac.uk . . Objective . . . .
Markov Model Simulations (MMS) By creating candidate model sufficently ex- pressive to describe modelling assumptions, data input and computational specifications. Motivation .
.
Economic assessment of novel health technologies plays an increasing role in the planning of future research programmes and the establishment of reimbursement policies, e.g. to investigate which of several alternative treatment strategies makes best use of available
resource use and outcomes where no directly comparable data is available. Markov Model Simulations (MMS) are frequently used for such health economical assessment. Typically, proprietory specialised software or general purpose spreadsheet applications are used for the evaluation of MMS. The lack of interchange standards and well-defined semantics, respectively, makes communication of the methodology, modeling assumptions, parameters used in a particular simulation and the verification and re-use of existing simulations difficult. Markov Models .
.
Markov models (MM) are used to model random processes that evolve over time. They can be used to model the disease progression of a patient, represented by a finite number of mutually exclusive health states corresponding to clinically and economically important events. Health states are connected by transition probabilities. . s1 . start . . assymptomatic . s2 . . progressive disease . s3 . . death . p12 . p23 . p13 . p11 . p22 . p33 = 1 . ’PayOffs’ such as costs and utilities can be attributed to states and transitions. . Probabilities are assigned to transitions between states. Markov Model Simulations .
.
MMS simulate the disease progression of a fictuous pa- tient population based on a markov model. With esti- mates for resource use (e.g. cost of being in a particular health state) and health outcome (e.g. quality of life in a particular health state) assigned to states and transitions, estimations of long term cost and outcomes associated with a particular health care intervention can be obtained. The results of multiple simulations (based on markov mod- els corresponding to the disease progress following differ- ent interventions) can be used to obtain measures for the relative cost effectiveness of different treatment strategies. Meta-Modelling .
.
Different diseases and treatment strategies require differ- ent markov models, input parameters and simulation set-
tion required for computing and interpreting the results of a markov model simulation, a language is needed that al- lows to express the fundamental features of MMS so that any specific MMS can be described in this language. . Metamodel of Markov Model Simulations . ’template’ for describing MMS . specific Markov Model Simulation . specific Markov Model Simulation . specific Markov Model Simulation . instance . specific MMS, e.g. cost-effectiveness of particle therapy in lung cancer . . Meta Model for Markov Model Simulations .
. .
. . Features . .
exist independently from Markov Model Simulations, this allows reuse of existing Value and Markov Model instances.
full record of the MMS by specifying simulation settings, the Markov Model instances used in the simulation and the Results obtained by the simulation.
as processing instruction for a simulation programme and as documentation of computed MMSs. Meta Model Structure .
.
. Markov Model Simulation .
. Markov Model .
. Values .
. * . 1..* . * . 1..* . Results
. State . Transition . PayOff . 2 . 1..* . * . 0..* . * . 0..* Language for MMS . . Based on the meta-model, a language for MMS has been developed using the Spoofax Language Workbench [1]. The listing on the right shows an extract of the descrip- tion of a published [4] cost-effectiveness study comparing radiotherapy modalities for non-small cell lung cancer in this language. Data Model . . An xml-schema data model has been created to facilitate the exchange of MMS descriptions and to allow for the documentation of the simulation results. Extract from example MMS description, based on [4]. . .
markovModelSimulation AnExampleSimulation { s i m u l a t i o n S e t t i n g s { simulationType : d e t e r m i n i s t i c numberOfCycles {markovModel Grutters2010_1 : 1 , markovModel Grutters2010_2 : 5} h a l f C y c l e C o r r e c t i o n : 1 useModalities {modalityType proton, modalityType carbon} useMarkovModels {markovModel Grutters2010_1, markovModel Grutters2010_2} t r a n s f e r {markovModel Grutters2010_1 : s t a t e State_treatmentDeath −> markovModel Grutters2010_2 : s t a t e State_death, . . . } markovModel Grutters2010_1 { . . . } markovModel Grutters2010_2 { . . . } }. Visualisation of states and transi- tions used in this study, generated from the MMS description.
(partial view for space reasons) ..
State_init, PO_treatTime_CI, PO_treatCost_CI S_treatDeath , PO_DeathTreat State_treatNoAcuteAes, PO_UtilityNoAeDurTreat 1 - TPacPneumGt3 + TableTPacOesophGt3, POCostsDyspnPerYear State_death transfer State_woDysp, PO_FollowUp, PO_UtilityNoAeAfterTreat transfer PO_DeathOther TPdiseaseMort, PO_DeathCancer State_withDysp, PO_CostsDyspnPerYear, PO_FollowUp, PO_UtilityDyspAfterTreat TPotherMort TPotherMort, PO_DeathOther TPdiseaseMort, PO_DeathCancerBibliography .
.
[1] The spoofax language workbench. http://strategoxt.org/Spoofax. [2] Alan Brennan, Stephen E Chick, and Ruth Davies. Online companion for: A taxonomy of model structures for economic evaluation of health technologies. http://faculty.insead.edu/chick/papers/HTAModel-AppTable.html. [3] A. Briggs and M. Sculpher. An introduction to markov modelling for economic evaluation. PharmacoEconomics, 13(4):397–409, April 1998. [4] Janneke P.C. Grutters, Madelon Pijls-Johannesma, Dirk De Ruysscher, Andrea Peeters, Stefan Reimoser, Johan L. Severens, Philippe Lambin, and Manuela A. Joore. The cost-effectiveness of particle therapy in non-small cell lung cancer: Exploring decision uncertainty and areas for future research. Cancer Treatment Reviews, 36(6):468–476, October 2010.. . Conclusion . . . . A meta-model for the description of Markov Model Simulation (MMS) studies for health economic evaluations has been developed. A first description language and data model has been implemented and tested on published MMS studies comparing radiotherapy modalities. Language and data model could be used for describing MMS to automated computing services as well as for reporting purposes where documentation of the modelling assumptions, simulation parameters and results is required. Testing on further usecases is needed to improve the meta-model and to ensure its applicability to MMS studies in other medical areas. This research project has been supported by a Marie Curie Early Initial Training Network Fellowship
(PITN-GA-2008-215840-PARTNER). Webpage: http://partner.web.cern.ch
.
. Conclusion . . . . . developed first candidate model for description of Markov Model Simulation studies implemented textual language and data model based on the candidate model, tested
could be used for describing MMS to automated computing services as well as for reporting purposes next... validation on further use cases, also beyond radiotherapy
Daniel Abler (CERN) Meta-modelling Markov Model Simulations 01.03.2012 6 / 6
. .
.
Meta-modelling Markov Model Simulations for cost-effectiveness analyses
Daniel Abler1, Steve Harris2, Jim Davies2
1European Centre for Nuclear Research, Geneva CH-1211, Switzerland 2Department of Computer Science, University of Oxford, Oxford OX1 3QD, UKdaniel.abler@cern.ch, {steve.harris, jim.davies}@cs.ox.ac.uk . . Objective . . . .
Markov Model Simulations (MMS) By creating candidate model sufficently ex- pressive to describe modelling assumptions, data input and computational specifications. Motivation .
.
Economic assessment of novel health technologies plays an increasing role in the planning of future research programmes and the establishment of reimbursement policies, e.g. to investigate which of several alternative treatment strategies makes best use of available
resource use and outcomes where no directly comparable data is available. Markov Model Simulations (MMS) are frequently used for such health economical assessment. Typically, proprietory specialised software or general purpose spreadsheet applications are used for the evaluation of MMS. The lack of interchange standards and well-defined semantics, respectively, makes communication of the methodology, modeling assumptions, parameters used in a particular simulation and the verification and re-use of existing simulations difficult. Markov Models .
.
Markov models (MM) are used to model random processes that evolve over time. They can be used to model the disease progression of a patient, represented by a finite number of mutually exclusive health states corresponding to clinically and economically important events. Health states are connected by transition probabilities. . s1 . start . . assymptomatic . s2 . . progressive disease . s3 . . death . p12 . p23 . p13 . p11 . p22 . p33 = 1 . ’PayOffs’ such as costs and utilities can be attributed to states and transitions. . Probabilities are assigned to transitions between states. Markov Model Simulations .
.
MMS simulate the disease progression of a fictuous pa- tient population based on a markov model. With esti- mates for resource use (e.g. cost of being in a particular health state) and health outcome (e.g. quality of life in a particular health state) assigned to states and transitions, estimations of long term cost and outcomes associated with a particular health care intervention can be obtained. The results of multiple simulations (based on markov mod- els corresponding to the disease progress following differ- ent interventions) can be used to obtain measures for the relative cost effectiveness of different treatment strategies. Meta-Modelling .
.
Different diseases and treatment strategies require differ- ent markov models, input parameters and simulation set-
tion required for computing and interpreting the results of a markov model simulation, a language is needed that al- lows to express the fundamental features of MMS so that any specific MMS can be described in this language. . Metamodel of Markov Model Simulations . ’template’ for describing MMS . specific Markov Model Simulation . specific Markov Model Simulation . specific Markov Model Simulation . instance . specific MMS, e.g. cost-effectiveness of particle therapy in lung cancer . . Meta Model for Markov Model Simulations .
. .
. . Features . .
exist independently from Markov Model Simulations, this allows reuse of existing Value and Markov Model instances.
full record of the MMS by specifying simulation settings, the Markov Model instances used in the simulation and the Results obtained by the simulation.
as processing instruction for a simulation programme and as documentation of computed MMSs. Meta Model Structure .
.
. Markov Model Simulation .
. Markov Model .
. Values .
. * . 1..* . * . 1..* . Results
. State . Transition . PayOff . 2 . 1..* . * . 0..* . * . 0..* Language for MMS . . Based on the meta-model, a language for MMS has been developed using the Spoofax Language Workbench [1]. The listing on the right shows an extract of the descrip- tion of a published [4] cost-effectiveness study comparing radiotherapy modalities for non-small cell lung cancer in this language. Data Model . . An xml-schema data model has been created to facilitate the exchange of MMS descriptions and to allow for the documentation of the simulation results. Extract from example MMS description, based on [4]. . .
markovModelSimulation AnExampleSimulation { s i m u l a t i o n S e t t i n g s { simulationType : d e t e r m i n i s t i c numberOfCycles {markovModel Grutters2010_1 : 1 , markovModel Grutters2010_2 : 5} h a l f C y c l e C o r r e c t i o n : 1 useModalities {modalityType proton, modalityType carbon} useMarkovModels {markovModel Grutters2010_1, markovModel Grutters2010_2} t r a n s f e r {markovModel Grutters2010_1 : s t a t e State_treatmentDeath −> markovModel Grutters2010_2 : s t a t e State_death, . . . } markovModel Grutters2010_1 { . . . } markovModel Grutters2010_2 { . . . } }. Visualisation of states and transi- tions used in this study, generated from the MMS description.
(partial view for space reasons) ..
State_init, PO_treatTime_CI, PO_treatCost_CI S_treatDeath , PO_DeathTreat State_treatNoAcuteAes, PO_UtilityNoAeDurTreat 1 - TPacPneumGt3 + TableTPacOesophGt3, POCostsDyspnPerYear State_death transfer State_woDysp, PO_FollowUp, PO_UtilityNoAeAfterTreat transfer PO_DeathOther TPdiseaseMort, PO_DeathCancer State_withDysp, PO_CostsDyspnPerYear, PO_FollowUp, PO_UtilityDyspAfterTreat TPotherMort TPotherMort, PO_DeathOther TPdiseaseMort, PO_DeathCancerBibliography .
.
[1] The spoofax language workbench. http://strategoxt.org/Spoofax. [2] Alan Brennan, Stephen E Chick, and Ruth Davies. Online companion for: A taxonomy of model structures for economic evaluation of health technologies. http://faculty.insead.edu/chick/papers/HTAModel-AppTable.html. [3] A. Briggs and M. Sculpher. An introduction to markov modelling for economic evaluation. PharmacoEconomics, 13(4):397–409, April 1998. [4] Janneke P.C. Grutters, Madelon Pijls-Johannesma, Dirk De Ruysscher, Andrea Peeters, Stefan Reimoser, Johan L. Severens, Philippe Lambin, and Manuela A. Joore. The cost-effectiveness of particle therapy in non-small cell lung cancer: Exploring decision uncertainty and areas for future research. Cancer Treatment Reviews, 36(6):468–476, October 2010.. . Conclusion . . . . A meta-model for the description of Markov Model Simulation (MMS) studies for health economic evaluations has been developed. A first description language and data model has been implemented and tested on published MMS studies comparing radiotherapy modalities. Language and data model could be used for describing MMS to automated computing services as well as for reporting purposes where documentation of the modelling assumptions, simulation parameters and results is required. Testing on further usecases is needed to improve the meta-model and to ensure its applicability to MMS studies in other medical areas. This research project has been supported by a Marie Curie Early Initial Training Network Fellowship
(PITN-GA-2008-215840-PARTNER). Webpage: http://partner.web.cern.ch
.
.
. Conclusion . . . . . developed first candidate model for description of Markov Model Simulation studies implemented textual language and data model based on the candidate model, tested
could be used for describing MMS to automated computing services as well as for reporting purposes next... validation on further use cases, also beyond radiotherapy Thanks to Jim Davies, Steve Harris, Ken Peach (PTCRi), Manjit Dosanjh (CERN)
. . .
This research project has been supported by a Marie Curie Early Initial Training Network Fellowship of the European Community’s Seventh Framework Programme under con- tract number (PITN-GA-2008-215840-PARTNER). Webpage: http://partner.web.cern.ch Daniel Abler (CERN) Meta-modelling Markov Model Simulations 01.03.2012 6 / 6