Mechanism-based pharmacoeconomic modelling

Professor Dyfrig Hughes

Centre for Health Economics & Medicines Evaluation

Empirical modelling

• Data collection

• Specify correlation structure between variables

• Use a numerical technique to find parameters for the structure such that the correlation between the data is maximised

• Validate the model with external data

Mechanism-based modelling

• Use fundamental knowledge of the interactions between variables to define the model structure

• Perform experiments to determine the parameters of the model

• Collect data from the process to validate the model

Principles of pharmacology

• Pharmacokinetics

– the study of drug disposition in the human body

• Pharmacodynamics

– the study of the pharmacological effects of drugs and the mechanisms of their actions

Pharmacokinetics (PK)

– Absorption

– Distribution

– Metabolism

– Excretion

Population PK

Pla

sma

con

cen

trat

ion

(m

g/l

)

Predicted plasma concentration (mg/l)

Pharmacodynamics (PD)

nn

n

DEC

DEE

50

max

Sigmoidal Emax model that describes the ‘dose response curve’ derived from the receptor occupancy theory

PK-PD model

Time (months)

Clin Pharmacokinet 2008; 47 (7): 417-448

Applications

• Dose selection • Dosing schedules • Dosing in special populations

– Renal impairment – Genetic polymorphism in metabolising enzymes – Children

• Study of drug interactions • Model-based drug development

– Clinical trial simulations – Informing decisions along the drug development

pathway

Aims

• To develop the methods for linking PKPD models with economic evaluations

• To conduct proof of concept studies

Lewis Sheiner Prize, PAGE 2011, Athens

Case study 1: Rituximab

Economic models

• Population PK and PD models from Ternant – Br J Clin Pharmacol 2009; 68: 561-73

PK equations

PD equation

Integrate PKPD with PE models

Maintenance therapy

First line therapy Trial-based results

Simulation results

Conclusions

Case study 2: NOACs

• Newer oral anticoagulants (dabigatran, rivaroxaban, apixaban) set to revolutionise management of patients with atrial fibrillation

• Meanwhile, pharmacogenetic-guided warfarin dosing is an improvement on clinically dosed warfarin

• Aim to assess comparative cost-effectiveness

Key results from pivotal trials

• Primary outcome of stroke or systemic embolism

– superiority of dabigatran (1.11% v 1.71% /year; p<0.001) and

– apixaban (1.27% v 1.60% /year; p=0.01) and

– non-inferiority of rivaroxaban (2.1% v 2.4% /year; p=0.12)

• Rates of major bleeding not significantly different between dabigatran 150mg and warfarin or between rivaroxaban and warfarin, but apixaban associated with a lower risk of major bleeding (2.13% v 3.09% per year; p<0.001)

Economic model - DES

Generate patient characteristics: Age, risk

factors, prior event history

Duplicate patient across the arms of the evaluation

Simulate next patient event (e.g. stroke, MI, discontinuation)

Calculate costs, utilities for that event and the time

between events

Update patient characteristics, health state,

current medication

• Where possible, clinical data were taken from the RE-LY study and the FDA submission based on it: – Initial patient characteristics

– Event rates for strokes, bleeds, PE, TIA, MI, CHF, vascular death, adverse events whilst on warfarin or dabigatran

– Discontinuation rates

• Mortality from non-vascular causes, and incidences of hypertension and diabetes taken from UK population data

• Event rates for aspirin come from a published meta-analysis comparing warfarin and aspirin

Clinical data

Ann Intern Med 2007;146:857-67

• Utility decrements related to AF and other clinical events taken from a published index of utility scores: – Strokes, bleeds, TIA, PE and MI all lead to a temporary loss of utility

– Strokes, MI and ICH also lead to a permanent loss of utility

• Utility losses from taking warfarin and aspirin come from a published cost-utility study comparing the two treatments, and an assumption was made that the utility loss associated with dabigatran was the same as that for aspirin

Utility data

Med Decis Making 2006;26:410-20

• Costs of outcome events are all taken from NHS reference cost data, following the methodologies of previously published HTA submissions

• All events come with a one-off cost; strokes and MIs also have costs that continue over time

• Costs for warfarin therapy were derived from a warfarin pharmacogenetics study

• List price of dabigatran

Cost data

Pharmacogenet Genomics 2009;19:800-12

Event Dabigatran 110mg Dabigatran 150mg Warfarin

Non-disabling stroke 0.52 (0.50) 0.36 (0.37) 0.58 (0.58)

Disabling stroke 0.94 (0.94) 0.66 (0.66) 1.00 (1.00)

Myocardial Infarction

0.72 (0.72) 0.74 (0.74) 0.52 (0.53)

Pulmonary embolism

0.11 (0.12) 0.14 (0.15) 0.09 (0.09)

Major bleed 2.72 (2.71) 3.10 (3.11) 3.38 (3.36)

Minor bleed 13.14 (13.16) 14.85 (14.84) 16.39 (16.37)

Death – vascular cause

2.43 (2.43) 2.27 (2.28) 2.70 (2.69)

Results: simulated vs. (trial)

Strategy Warfarin Dabigatran

110mg Dabigatran

150mg

Stroke or systemic embolism 0.241 0.208 0.174

MI 0.061 0.072 0.073

Major bleed (including ICH)

0.331 0.318 0.361

Life years 10.851 10.940 11.051

QALYs 6.416 6.484 6.536

Total cost £6,480 £10,529 £9,850

Main results

Dabigatran 110mg bid dominated by dabigatran 150mg bid Dabigatran 150mg versus warfarin Incremental cost-effectiveness ratio: £23,082

Cost effectiveness acceptability curve

Sub-group ICER (£/QALY)

RE-LY population 23 082

CHADS2 score 2 20 207

CHADS2 score ≥3 15 895

Centres’ time in therapeutic range ≥65.5% 42 386

Centres’ time in therapeutic range <65.5% 20 396

Creatinine clearance <30-50 mL/min 18 647

Previous stroke or TIA 17 286

Vitamin K antagonist naive 22 517

Age ≥75 years 17 857

Results: subgroups

• Dabigatran 110mg bid appears to be inferior to 150mg bid in almost all circumstances

• In our base case analysis, dabigatran 150mg bid is approximately on the boundary of being cost-effective

• Cost-effectiveness varies considerably based on the patients’ stroke risk and the quality of INR control

• Uncertainty in stroke rates is the largest single source of uncertainty

Summary

Other NOACs

• Each pivotal trial compared NOAC with clinically-dosed warfarin

• Adjusted, indirect comparison

– to assess relative benefits and harms, and help guide treatment selection

Methods

• Stroke risk profile of the US atrial fibrillation population, in terms of CHADS2 scores

• Bucher method of adjustment for indirect comparisons among trials

• Probabilities of treatment discontinuation

• Utility scores from the US Medical Expenditure Panel Survey of several thousand patients

Probabilities of treatment with highest net benefit by patient subgroup

Main findings

• Apixaban achieved highest net benefit (expressed as QALYs), followed by dabigatran, rivaroxaban then warfarin

• No subgroup in which the probability of apixaban being the most effective is below 50%, and none where the probability of warfarin being the most effective is above 5%

What about PGx warfarin?

• Variability in response to warfarin can be partly explained by genetic polymorphisms in – CYP2C9 , VKORC1

• People with variant alleles are at an increased risk of over-anticoagulation and bleeding

• Dosing algorithms based on PGx may result in better INR control, and hence better clinical outcomes

• No RCTs comparing PGx-warfarin with NOACs

• Multiple dosing algorithms possible

• Population PKPD model of warfarin used to predict time below, in and above INR range based on a range of algorithms (NONMEM)

• Data from a systematic review used to link time in range to clinical endpoints

• Health outcome model used to extrapolate to a lifetime horizon and compare different treatments in terms of QALYs accrued

– Based on discrete event simulation described earlier

Simulation structure

• From Hamberg et al, CPT 2010;87:727-34 which predicts INR measurements based on dose, age and genetic information

• Patient characteristics based on those of the UK atrial fibrillation population

• Model allows for explicit incorporation of non-adherence

Population PKPD model

• Loading phase – To achieve correct INR range as quickly as possible

without over-anticoagulating

• Predicted maintenance dose – To predict the most likely dose to maintain a patient in

range in the long term

• Maintenance phase – Further dose adjustments are made based on INR at

clinic visits

• Genetic information can be used in each stage

Dosing algorithms

• Loading dose: 10, 10, 5mg (days 1,2,3)

• Predicted maintenance dose: IWPC algorithms

– A clinical algorithm based on age, height, weight, ethnicity, use of amiodarone and enzyme inducers

– A pharmacogenetic algorithm which uses all these variables and genetic information

• Doses adjusted with the Fennerty algorithm

Algorithm selection - Example

Population PKPD results – INR

time in range (INR 2-3)

time above range (INR >3)

time below range (INR <2)

Open symbols (dashed lines) clinical algorithms Filled symbols (solid lines) pharmacogenetic algorithm

Results

Life extension (months)

∆QALYs (95% CR)

PGx warfarin 0.003 0.0031 (0.1649, 0.1327)

Rivaroxaban 1.11 0.0957 (-0.0510, 0.2431)

Apixaban 2.06 0.1298 (-0.0290, 0.2638)

Dabigatran 1.47 0.1065 (-0.0493, 0.2489)

All compared with warfarin dosed according to clinical algorithm

Sub-group analysis

Probability of each treatment accruing the largest number of QALYs

Economic results

• Rivaroxaban was dominated as a treatment option by both dabigatran and apixaban

• Dabigatran was extendedly dominated by apixaban

• ICER for genotype guided warfarin versus clinical algorithm dosed warfarin was £13,226 per QALY gained, and the ICER for apixaban versus genotype guided warfarin therapy was £19,858 per QALY gained

Simulation - Time in INR range

Genotype-guided group Control group

• Early estimation of cost-effectiveness

• Cost effectiveness estimates in specific populations

• Inform pricing decisions

• Inform pricing decisions for different doses

• Assessing impact of inter-patient variability and protocol deviations on cost-effectiveness

• Inform protocol design

– Value of information analysis to quantify the value of future research in reducing parameter uncertainty

Potential applications

Disease model

•Biology

•Biomarker / outcome relationship

•Natural progression

Drug model

•Pharmacokinetics

•Pharmacodynamics

•Co-variate effects

Population model

•Patient demographics

•Drop-outs

•Adherence

Health outcomes modelling

•Health state utilities

•Benefit-risk assessment

•Economic appraisal

A natural extension to MBDD

Milligan et al, CPT 2013 doi:10.1038/clpt.2013.54

Acknowledgements

• MRC funding (NW HTMR)

• Munir Pirmohamed, Steven Lane (University of Liverpool)

• Joshua Pink (formerly Bangor University)