18
© 2016 J Caro. All Rights Reserved. DICE Simulation for HTA: a new, unifying method J. Jaime Caro MDCM FRCPC FACP Adj. Prof Medicine & Epidemiology and Biostatistics, McGill University, Montreal Chief Scientist, Evidera (Discretely- Integrated Condition Event) Disclosure: I have no conflict of interest in relation to this topic or presentation.

Dice simulation for hta cadth

Embed Size (px)

Citation preview

Page 1: Dice simulation for hta cadth

© 2016 J Caro. All Rights Reserved.

DICE Simulation for HTA: a new, unifying method

J. Jaime Caro MDCM FRCPC FACPAdj. Prof Medicine & Epidemiology and Biostatistics, McGill University, MontrealChief Scientist, Evidera

(Discretely-Integrated Condition Event)

Disclosure:I have no conflict of interest in relation to this topic or presentation.

Page 2: Dice simulation for hta cadth

2

Modeling for HTA

HTA models

Dynamic transmission

SEIR

Agent-based

Disease course

Markov

Cohort

Microsim

DES

Constrained

Unconstrained

System dynamics

Stats

Infectious diseases

CEA/CUABIA

Disease epiPatient flowsPublic health

Page 3: Dice simulation for hta cadth

3

Generic Example

Country population

Diagnosis Disease X

Level A

Level B

surgery relapse

Tmt line 1

Death

Tmt line 2 Tmt line 3

incidence

Prop ATTrel

TT2 TT3

hDeath

hD1 hD2hD3

hD/Rel

Page 4: Dice simulation for hta cadth

4

What is DICE

A modeling technique that conceptualizes the decision-analytic problem in terms of two fundamental aspects:

Aspects that persist over timeHave levels, which can change & affect eventsMany conditions can be present at onceInterested in time spent at a given level (value)

Aspects that happen at a point in timeCan affect the level of a condition or other eventsMany can happen, at any timeInterested in number that happen (and when)

Conditions Events

discrete integration

Page 5: Dice simulation for hta cadth

5

Example

Country population

Diagnosis Disease X

Level A

Level B

surgery relapse

Tmt line 1

Death

Tmt line 2 Tmt line 3

incidence

Prop ATTrel

TT2 TT3

hDeath

hD1 hD2hD3

hD/Rel

Page 6: Dice simulation for hta cadth

6

The basic idea of DICE

Conditions

Name ValueProportion A

Disease type

Treatment

Age

Hazard death

Hazard relapse

Event: Start  Assigned Item ExpressionProportion A 0.8

Disease typeIf (rand()<.8, A (resectable), B)

TreatmentIf(Disease type = A,Surgery,Line1)

Age

Index(ageD,Match(rand(),freq)

SurgeryIf(Treatment=Surgery, Now, Never)

0.8

A (resectable)

Surgery

59.3

Event: Surgery  Assigned Item ExpressionSurgery NeverSurgery number Surgery number + 1Hazard relapse

Vlookup(Hazards,”Surgery”,2)

Relapse -ln(1-rand())/Hazard relapse… …

0.015

Surgery number 1

VBA

“Generic”Model

specifics are in the

tables

Page 7: Dice simulation for hta cadth

7

Example

Country population

Diagnosis Disease X

Level A

Level B

surgery relapse

Tmt line 1

Death

Tmt line 2 Tmt line 3

incidence

Prop ATTrel

TT2 TT3

hDeath

hD1 hD2hD3

hD/Rel

Page 8: Dice simulation for hta cadth

8

Cohort Markov approach

Disease-free Relapsed

Dead

Time (m)

Disease-free Relapsed Dead

0 1,000 0

1 1000-P 1000 x Prel 0xPdeath/Rel

2

… … … …

300 … … …

Prel

P death /Rel

Page 9: Dice simulation for hta cadth

9

DICE Cohort Markov version

Name ValueDisease-free 100Relapsed 0Death 0

Disease-free: 100%Transition: cycle

StartUpdate Disease-free, Relapsed, Death Select next eventTransition, End

Transition

Report all resultsEnd

TypeAssigned Item Expression

Condition Death Death+ Pdeath*Relapsed

Condition Relapsed

Relapsed*(1-Pdeath) +Disease-free*(Prel)

Condition Disease-free Disease-free*(1-Prel)Event Transition Now + cycle

Event End If ((Time=TimeHorizon), Now, Never)

Conditions Table Transition event Table

Name ValueTimeHorizon 10Never 99,999Now 0Prel .015Pdeath .001Cycle 1

Constants Table

%PF%P%D

Compact specification– Doesn’t change if longer time horizon– Or shorter cycles

o Can make these variable! Expressions written only once reducing error No programming! Transparent, easy to grasp

Page 10: Dice simulation for hta cadth

10

Cohort Markov approach

Disease-free Relapsed

Dead

Prel/BioM

P death /Rel

Age,sex,…

Microsimulation Markov approach

Pprog

P death /rel

Time (m)

Disease-free Relapsed Dead

0 1,000 0

1 1000-P 1000 x Prel 0xPdeath/Rel

2

… … … …

300 … … …

Age, sex, determinants

Biomarker

Page 11: Dice simulation for hta cadth

11

NameInitial Value

Disease-free NRelapsed 0Death 0

Name Initial ValueDisease Disease-freeAge Pick ProfileSex Pick ProfileBiomarker Pick Profile

Disease-free: 100%Transition: cycle

StartUpdate Progression-free, Progression, Death Select next eventTransition, End

Transition

Report all resultsEnd

DICE Microsimulation version

Conditions TableAge Sex BioM45 Male 12445 Female 216… …

Profiles Table

Assignment Type

Assigned Item Expression

Condition Disease

If (Disease=“Disease-free”,if(rand()<Prel,“Relapsed”, “Disease-free”),If(rand()<Pdeath,“Dead”, “Relapsed”)

Event Transition Now + cycleEvent End If ((Time=TimeHorizon), Now, Never)

Transition event Table

Disease: disease-free Disease

Page 12: Dice simulation for hta cadth

12

Example

Country population

Diagnosis Disease X

Level A

Level B

surgery relapse

Tmt line 1

Death

Tmt line 2 Tmt line 3

incidence

Prop ATTrel

TT2 TT3

hDeath

hD1 hD2hD3

hD/Rel

Page 13: Dice simulation for hta cadth

13

Tmt line 1Tmt line 1Tmt line 1Tmt line 1Tmt line 1Tmt line 1Tmt line 1

Microsimulation Markov approach

Disease-free Relapsed

Dead

Prel/BioM

P death /Rel

Age,sex,…

Time-to-event approach (“DES”)

RelapsedRelapsedRelapsedRelapsedRelapsedRelapsedRelapsed

Tmt line 1

Relapsed

Page 14: Dice simulation for hta cadth

14

Assignment Type

Assigned Item Expression

Condition Disease

If (Disease=“Disease-free”,if(rand()<Prel,“Relapsed”, “Disease-free”),If(rand()<Pdeath,“Dead”, “Relapsed”)

Event Transition Now + cycleEvent End If ((Time=TimeHorizon), Now, Never)

Transition event Table

Update DiseaseSelect next eventTransition, End

Update DiseaseSelect next eventSwitch, Death

Start Transition

Report all resultsEnd

DICE Time-to-Event Model

DeathDisease: Disease-freeTransition: cycle

Update DiseaseSelect next eventEnd

Relapse

TTE: Relapse, Death

Start event TableAssignment Type

Assigned Item Expression

Event Relapse -ln(1-rand())/hazardrel

Event Death -ln(1-rand())/hazardotherDeathRelapse event TableAssignment Type

Assigned Item Expression

Condition Disease RelapsedEvent Death Min(-ln(1-rand())/hazarddeath/rel,Death)

EventTreatment switch Now

Select new treatmentSwitch

Treatment switch event Table

Page 15: Dice simulation for hta cadth

15

Assignment Type

Assigned Item Expression

Condition Disease RelapseEvent Death -ln(1-rand())/hazarddeath/relapse

EventTreatment switch If (Treatment=“SoC”, Now, Never)

Output QALY +Time x UtilityOutput LY +TimeOutput Cost +(Time x CostTmt) + CostRel

Condition Utility UtilRel

DICE outputs

Conditions Table

Relapse event Table

NameInitial Value

Disease

Disease-free

Utility UDF

QALY 0LY 0Cost 0

Type Name Discount

Accumulator

QALY 3%

Accumulator

LY 3%

Accumulator

Cost 3%

Counter

Tmtswitch

0%

… …

Outputs Table

Page 16: Dice simulation for hta cadth

16

• Very flexible & natural• Can combine cohort,

individual & time-to-event approaches

• Transparent, simple to communicate

• Standard framework (easy to learn)

• Less error-prone• Enables structural sensitivity

analysis• Straightforward to review• Fast to create, easy to modify

• Excel is slow• No individuals,

interactions• No resources, queues• (lacking experience,

validation, publications)

Advantages & limitations

Page 17: Dice simulation for hta cadth

17Economic analyses

What can DICE be used for?

Epidemiologic modeling

Patient Flow

RCT Simulation

Portfolio optimization

MCDA

Page 18: Dice simulation for hta cadth

18

DICE transforms the way we develop models

Old wayPick a technique

(e.g., Markov)

Adapt decision problem To selected technique

Spend wks/mprogramming

Final Model• Complex• Huge• Tricky to verify• Hard to explain• Forget changing

structure!

New wayFocus on decision

problem

Design model to suit decision problem

ImplementDICE in days

Final Model• Straightforward• Compact• Easy to verify• Simple to explain• Change structure

any time