1. Value and uncertainties in identifying the cost-effective sequence of tests
Rita Faria, MSc
Centre for Health Economics
University of York, UK
@RitaINdeFaria

2. Acknowledgements
This presentation refers to 2 projects, both funded by the National Institute for Health Research (NIHR) Health Technology Assessment Programme. The views expressed here are mine and not necessarily those of the NIHR or the Department of Health and Social Care. The project on the cost-effectiveness of cascade testing protocols to diagnose familial hypercholesterolaemia (FH) is a collaboration between the University of York, University of Nottingham, among others. For more details, see The project on the cost-effectiveness of magnetic resonance imaging for the diagnosis of prostate cancer (the PROMIS study) is a collaboration between the University of York, University College London, among others. For more details, see

3. Steps in a (model-based)-CEA
Problem structuring
Model building
Analysis

4. Problem structuring for the CEA of diagnostic tests
How does diagnosis inform management?
What are the consequences of
Different management options
Given the patients’ true disease status
And the tests’ results
How can the tests be used to diagnose patients?

5. Problem structuring Hypothetical example
Options
No test
Treat all
Treat none
Test
Treat acc. to test results A
Test
Healthy
Don’t treat
Diseased
Treat
No test

6. Problem structuring Our first real example
Familial Hypercholesterolaemia (FH)
FH mutation
High LDLC
High risk of CVD
1st degree relatives have 50% chance of having the FH mutation.
Cascade testing = screening of relatives
What is the cost-effective way of doing cascade testing?

7. Problem structuring Our first real example
Is the decision problem…
Cascade testing vs no cascade testing?

8. 4 Ways to get relatives into the cascade = 4 options
Stage 1: Select indexes
Yes No
2 “Tests”
= 7 options
Stage 2: Get relatives tested
Yes No
4 Ways to get relatives into the cascade
= 4 options
Stage 3: Diagnosis of relatives FH
History of CVD
Low CVD risk
High CVD risk
2 “Tests” test relatives in the cascade
= 5 options
As FH patient
Primary prevention CVD
Secondary prevention CVD
No management
4 management options

9. X Test cut-offs to classify test positive vs test negative
292 options
X Test cut-offs to classify test positive vs test negative
= 1,047 strategies
Project ongoing…

10. Problem structuring Our second real example
MRI and biopsy for the diagnosis of clinically significant prostate cancer
Clinically significant cancer should be treated
Non-clinically significant cancer can be monitored and treated only if it progresses
The goal is to diagnosis cancer and distinguish between the 2 types
We can use MRI, TRUS biopsy or TPM biopsy (perfect but costly)
But treatment requires confirmatory biopsy

11. Problem structuring Our second real example
Is the decision problem…
MRI vs TRUS-biopsy?
MRI + TRUS-biopsy vs biopsy alone?

12. Non-significant cancer= monitoring
Believe the test
Test again
What is the 1st test?
TRUS
MRI
TPM
What is the 2nd test?
TRUS
MRI
TPM
Believe the test
Test again
What is the 3rd test?
TRUS
MRI
TPM
No cancer = No treatment
CS cancer= treatment
Non-significant cancer= monitoring

13. X Test cut-offs to classify lesions
32 test sequences
X Test cut-offs to classify lesions
= 383 strategies

14. Model building Parameterisation
Accuracy of the tests
MRI
TRUS-biopsy (standard biopsy)
TPM biopsy (perfect biopsy; resource intensive)
Direct impact of tests on QoL and costs
Long-term health consequences and costs

15. Direct impact of the tests
Healthcare utilisation due to AEs
Literature
Prices
NHS reference costs
HRQoL
PROMIS study

16. Accuracy in a test naïve population
TRUS-biopsy +
Perfect biopsy
PROMIS study
MRI
Accuracy of TRUS-biopsy and MRI in a diagnostic test naïve population
Example:
Probability (MRI=CS cancer | true status=CS cancer)
Probability (TRUS biopsy = CS cancer | true status=CS cancer)

17. What about test accuracy in men who had had a prior test?
Literature
Relative sensitivity where possible
Absolute accuracy otherwise
1st Biopsy
P(B1=CS| truth=CS)
P(B1=NCS| truth=CS)
1-sum
P(B2=CS| truth=CS & B1=NCS)
2nd
Biopsy
P(B2=NCS| truth=CS & B1=NCS)
1-sum

18. Parameterisation Pay-offs of diagnosis
No cancer
Non-CS cancer
CS cancer
True status
No cancer
Non-CS cancer
CS cancer
Diagnosis
Discharged
Monitoring
Treatment

19. PF Literature review Radical treatment vs watchful waiting
Localised prostate cancer
Digitise survival curves
Obtain progression risk to metastatic cancer
Obtain mortality risk from metastatic cancer D PF M
Markov model
Obtain costs and QALYs
Calibration model
Obtain transition probabilities
Survival curve from Wilt TJ, Brawer MK, Jones KM, et al. Radical prostatectomy versus observation for localized prostate cancer. N Engl J Med 2012;367: DOI: /NEJMoa

20. Analysis

21. Cost-effective strategy ICER~7k/QALY
Strategies detect < 80% CS cancers
Cost-effective strategy
ICER~7k/QALY
MRI 1st then up to 2 TRUS-biopsies

22. Key drivers of cost-effectiveness
Sensitivity of TRUS-biopsies as the 2nd or 3rd test.
Probability that patients misclassified as having non-CS cancer are diagnosed in the future.
Health losses of misclassifying patients with CS cancer as having no cancer.

23. Limitations CEA includes only 3 tests Unit costs of tests
Data on accuracy of TRUS-biopsy as 2nd or 3rd test
Data on long-term outcomes

24. Learnings for future research
Sources of data on long-term outcomes
No data on patients incorrectly classified as not having the disease.
Long-term outcomes may be subject to classification error.
Tests are often used in combination
Difficult to map out all ways to use tests in combination
Difficult to find data to inform the performance of tests in combination in the model
Challenge in communicating methods and results
Mismatch between options in the model and clinical practice.
Structural uncertainty from assumptions.

25. For more details on this work see:
Rita Faria
Centre for Health Economics
University of York
@RitaINdeFaria

26. Additional slides

27. Additional cost per CS cancer detected

28. Cost-effectiveness results M7 223, T7 223, M7 222, P4 2
Strategy
Biopsy def
mpMRI def
mpMRI cut-off
QALYs
Costs
ICER
M7: mpMRI for all men; TRUSB in men with suspicion of CS cancer. Re-biopsy with TRUSGB those in whom CS cancer was not detected 2 3
8.66
£5021
£5,501
T7: TRUSB for all men; Men classified as NC or non-CS receive a mpMRI. Men with suspicion of CS cancer receive a 2nd TRUSB
8.69
£5194
£5,778
M7: mpMRI for all men; TRUSB in men with suspicion of CS cancer. Re-biopsy with TRUSB those in whom CS cancer was not detected
8.72
£5367
£7,076
P4: TRUSB in all men and TPMB in men in whom CS cancer was not detected
Not applicable
8.74
£5968
£30,084

29. Sensitivity analysis Key results
Strategies starting with TRUS-biopsy (T7 222, T9 222) are cost-effective if
1st MRI-targeted TRUS-biopsy has lower sensitivity
2nd MRI-targeted TRUS-biopsy has higher sensitivity
Higher ICER of radical treatment vs monitoring (radical treatment 15% less effective or monitoring detects at least 45% of missed CS cancers)
Strategies where all or most men have TPM-biopsy (P1, P4 2) are cost-effective if
Risk of death from TRUS-biopsy > 0.5%.
Men incorrectly classified as no cancer have at least 0.1 lower QALYs.
Also tested
Lower prevalence of cancer – no changes
Lower HQRoL due to TRUS-biopsy – no changes
Bivariate sensitivity analysis on costs of the tests – mostly no changes

31. Limitations and key uncertainties (1) Accuracy and short-term costs
Limited data on the sensitivity of TRUS-biopsies post-mpMRI:
Generic MRI-targeted TRUS-biopsy although there are variations
Assumed that TRUS-biopsy post-MRI has the same cost as blind TRUS- biopsy, but has better sensitivity.
Tests costed with NHS reference costs, which may not reflect true costs to the NHS and lack of capacity to offer mpMRI to all men in a timely basis
Only included mpMRI, TRUS-biopsy and TPM-biopsy, whilst there are other tests and biomarkers that can be used in diagnosis

32. Limitations and key uncertainties (2) Long-term outcomes and costs
Summary data on time to progression and death  calculation of long-term costs and QALYs assumed constant risks over time.
No data on progression of men with missed cancers  assumed equivalent to PIVOT’s arm on watchful waiting but tested in SA
No data on NICE active surveillance protocol  assumed equivalent to PIVOT’s arm on watchful waiting.
Long-term outcomes relate to men diagnosed with imperfect test (TRUS-biopsy)