Supplement 1: Value of Information Analysis
Journal name: Pharmacoeconomics
Article title: Breathing synchronized Hypoglossal Nerve Stimulation (HNS) with Inspire for Untreated Severe Obstructive Sleep Apnoea / Hypopnoea Syndrome (OSAHS): A simulated Cost-Utility Analysis from an NHS Perspective
Author List
Blissett, D B1*., Steier, J2., Karagama, Y3, Blissett, R. S1.
1MedTech Economics, 14 Marnhull Rise, Winchester, UK
2Guy’s & St Thomas’ NHS Foundation Trust, London, UK
3CHAPS, Faculty of Life Sciences and Medicine, King’s College London, Strand, London, UK Corresponding author details
Email: Deirdre.blissett@medtecheconomics.co.uk
Address: MedTech Economics, 14 Marnhull Rise, Winchester, SO22 5FH, UK Funding: This study was funded by Inspire Medical Systems
Conflict of interest
Deirdre Blissett and Rob Blissett are managing directors of MedTech Economics, a health economic consultancy that was commissioned by Inspire Medical Systems to develop the economic model described in this manuscript
Joerg Steier is a named inventor on a patent to treat sleep apnoea using transcutaneous stimulation:
Apparatus for treatment of snoring and sleep apnoea (WO2016124739A1) Karagama Yakubu has no conflicts of interest.
Methods
Oostenbrink et al (2008) [1] describe expected Value of perfect information (EVPI) as equal to the net monetary benefit (NMB) of the optimal strategy given perfect information, less the NMB of the strategy that would be adopted given the current information (base-case), averaged over all model iterations. EVPI was calculated at WTP thresholds ranging between £2,500 and £30,000, using 1,000 iterations.
Expected value of partial information (EVPPI) is used to quantify the value of removing uncertainty in a parameter or group of parameters in the decision to adopt a technology at a fixed WTP threshold.
In this analysis, EVPPI was run for two groups of variables, firstly grouping the HRs for cardiovascular disease (CVD) with severe and mild OSASH and secondly grouping all other inputs included in the PSA. The methods described by Brennan et al (2007) [2] were followed. A generalised Monte-Carlo sampling algorithm was applied in a nested simulation with an outer loop (n = 1,000) which sampled and subsequently fixed the parameters of interest. For each iteration of the outer loop the fixed parameters were carried through to an inner loop (n=1,000) which sampled from the remaining parameters in the model. The NMB of the intervention and comparator was averaged over the iterations of the inner loop. The highest NMB was used to determine the optimal adoption decision.
The average net benefit of the optimal adoption decision given perfect information on the parameters of interest was calculated by averaging the NMB of the optimal intervention over the iterations of the outer loop. Partial EVPI was determined by subtracting the average net mean benefit of the baseline adoption decision from the average NMB of the optimal adoption decision given perfect information on the parameters of interest.
Results
The results of the EVPI and two EVPPI per patient are reported in Figure 1. The EVPI and both EVPPIs are similar at very low thresholds, all reporting high values for perfect information. This is due to the low probability of a positive NMB for HNS with Inspire. The expected value of perfect or partial perfect information are very similar because reducing the uncertainty of any, or all the parameters is unlikely to change the baseline adoption decision.
Both the EVPI and both EVPPIs decrease at higher WTP threshold because there is greater certainty of HNS with Inspire being cost-effective.
The EVPPI_HR_CVD decreases substantially at a £30,000 WTP. This is consistent with the scenario that considered HNS having no CVD risk benefit for which we calculated an incremental cost- effectiveness ratio of £39,425. Therefore, it follows that the value of reducing the uncertainty at a WTP of £30,000 is low.
Figure 1 Expected value of Perfect Information (EVPI) and Expected Value of Partial Perfect Information (EVPPI).
References
1. Oostenbrink, J.B., et al., Expected value of perfect information: an empirical example of reducing decision uncertainty by conducting additional research. 2008. 11(7): p. 1070-1080.
2. Brennan, A., et al., Calculating partial expected value of perfect information via Monte Carlo sampling algorithms. J Medical Decision Making, 2007. 27(4): p. 448-470.
