1. Benjamin Kearns, The University of Sheffield
What are the long-term costs and benefits of screening for ovarian cancer amongst postmenopausal women? 
Benjamin Kearns,
The University of Sheffield
13/04/2018
© The University of Sheffield

2. A quick introduction
A research fellow at The University of Sheffield, in the Health Economics and Decision Science section.
I work on a wide range of cost-effectiveness studies including cancer screening, vascular disease / services and mental health screening / treatment.
Interested in the application of statistics in HTA; including time-series and survival analysis.
Opportunity costs.
13/04/2018
© The University of Sheffield

3. Overview Introduction to health technology assessment (HTA)
Ovarian cancer screening: the decision problem
Health economic evaluation and extrapolation
Results and conclusions
13/04/2018
© The University of Sheffield

4. What is HTA?
A health technology is any technology that is used to improve people’s health.
HTA considers clinical effectiveness (does it work) and cost-effectiveness (is it value for money).
The NHS cannot afford to pay for all effective technologies; HTA helps with funding decisions.
Examples of health technologies include screening programmes, the use of drugs to treat a disease, and different methods for performing surgery.
HTA = framework to decide which health technologies to invest in.
13/04/2018
© The University of Sheffield

5. Outcomes in HTA
Key outcomes are costs to the healthcare system and benefits to patients.
In England, benefits are typically ‘Quality adjusted life years’ (QALYs).
QALYs weight years of life by their quality (utility). Utility values normally range between 1 (perfect health) and 0 (death).
Outcomes are in addition to clinical outcomes.
13/04/2018
© The University of Sheffield

6. Screening for ovarian cancer
Ovarian cancer is the fifth leading cause of cancer mortality amongst woman in the UK.
Early diagnosis is associated with improved survival → potential role for screening.
The UK Collaborative Trial of Ovarian Cancer Screening (UKCTOCS) investigated the effect of screening on ovarian cancer mortality.
Outcomes are in addition to clinical outcomes.
13/04/2018
© The University of Sheffield

7. UKCTOCS
Randomised 200,000 women to screening / no screening between 2001 and 2005.
Results presented December 2015 based on median 11 years follow-up.
No statistically significant effect of screening on mortality. However, a potential ‘delayed effect’.
Outcomes are in addition to clinical outcomes.
13/04/2018
© The University of Sheffield

8. Decision problem
Multimodal screening (MMS) associated with 15% relative reduction in ovarian cancer mortality (95% CI -3% to 30%).
Should screening for ovarian cancer be funded?
Need evidence on:
Costs to the healthcare system
Benefits to patients
Long-term effect on mortality
Outcomes are in addition to clinical outcomes.
13/04/2018
© The University of Sheffield

9. Evidence used Screening effect on mortality from UKCTOCS
Systematic reviews to identify:
Any other relevant trials
Costs of screening and treatment
Impacts on patients’ quality of life
Also input from clinical multidisciplinary teams, cancer registry and other experts.
By considering costs and health benefits gives a comprehensive review of potential benefits and harms.
13/04/2018
© The University of Sheffield

10. Economic evaluation Evidence synthesized in a mathematical model.
Markov model built in Excel.
Simulates a cohort of patients and follows their outcomes until death.
Outcomes are in addition to clinical outcomes.
13/04/2018
© The University of Sheffield

11. The need for extrapolation
We have evidence on the effectiveness of screening for 11 years of follow-up.
We need to know how effective it is for a woman’s lifetime (≈ 40 years follow-up)
Hence we need to extrapolate the observed data.
No ‘gold standard’ method.
Outcomes are in addition to clinical outcomes.
13/04/2018
© The University of Sheffield

12. Extrapolation methods
1&2) Standard parametric models: exponential, Weibull, Gompertz log-normal, loglogistic - choose best fitting.
Combine Royston-Parmar spline models with time-series methods.
As above, but with a model discrepancy term (only applied to MMS).
Guyot method
13/04/2018
© The University of Sheffield

13. Results: within-trial
Notes: no screening effect until near the end.
Different methods visually give a similar fit, although may not be flexible enough?
13/04/2018
© The University of Sheffield

14. Results: extrapolations
Control – 2 different param depending on if same or diff. MMS has same model for both.
13/04/2018
© The University of Sheffield

15. Clinical and cost-effectiveness
Absolute incremental (Δ) values: screening - no screening
Method
Costs
Life years
QALYs
Deaths*
RP-time series
£419
0.142
0.047
-1.77%
Model discrepancy
£435
0.141
0.035
-1.23%
Separate parametric models
£445
0.077
0.024
-0.95%
Same parametric models
£459
0.043
0.013
-0.46%
Life years range from 7.4 weeks to 2.2 weeks.
* Ovarian cancer deaths
13/04/2018
© The University of Sheffield

16. Should screening for ovarian cancer be funded?
Summarise outcomes by the incremental cost-effectiveness ratio (ICER) = Δcosts / ΔQALYs.
Technologies with ICERs < £13,000 per QALY may be cost-effective.
ICER helped to inform funding decisions
<13k = health gains offset the cost of the intervention.
13/04/2018
© The University of Sheffield

17. Should screening for ovarian cancer be funded?
Method
ICER
95% confidence interval
RP-time series
£ 8,864
(£2,600 to £51,576)
Model discrepancy
£12,643
(£3,734 to dominated)
Separate parametric models
£18,372
(£7,709 to £96,784)
Same parametric models
£36,769
(£13,888 to dominated)
ICER helped to inform funding decisions
<13k = health gains offset the cost of the intervention.
13/04/2018
© The University of Sheffield

18. Conclusions
Long-term cost-effectiveness estimates are sensitive to the methods chosen.
Results for ovarian cancer screening are promising, but substantial uncertainty in long-term effectiveness remains.
Further research is required into how to choose between alternative extrapolation methods.
13/04/2018
© The University of Sheffield