Assessment <strong>of</strong> cost-effectiveness evidenceTABLE 58 Cost-effectiveness <strong>of</strong> strategies for (n = 1000) troponin-negative patients, using data from Mills 155Strategy Total costs (£) Total QALYs ICER (£/QALY)No testing 374,040 11,891.14 –Exercise ECG 678,120 11,917.75 Extendedly dominatedH-FABP 544,340 11,911.26 8464CTCA 937,426 11,946.86 11,041ICA 1,705,790 11,950.36 219,532TABLE 59 Cost-effectiveness <strong>of</strong> strategies for troponin-negative patients, using data from the RATPAC trial 12Strategy Total costs (£) Total QALYs ICER (£/QALY)No testing 260,901 12,180.71 –Exercise ECG 590,601 12,181.78 DominatedH-FABP 449,520 12,182.57 101,408CTCA 876,680 12,184.20 262,061ICA 1,656,701 12,176.07 DominatedThe cost-effectiveness <strong>of</strong> CTCA therefore appears to depend on the assumed rate <strong>of</strong> subsequent death<strong>and</strong> non-fatal MI. Given the uncertainty in these risks, we performed ‘goal-seeking’ <strong>analysis</strong> to identify thelevel <strong>of</strong> risk at which the ICER for CTCA crosses the NICE threshold <strong>of</strong> £20,000–30,000/QALY comparedwith either H-FABP, ETT or no testing. We assumed a proportional relationship that risk <strong>of</strong> non-fatal MIis four times the risk <strong>of</strong> death. The results are shown in Table 60. Depending on the threshold used,CTCA is likely to be cost-effective if the combined risk <strong>of</strong> death <strong>and</strong> non-fatal MI within the time periodassumed to be influenced by initial diagnostic testing exceeds 2% (£30,000/QALY threshold) or 2.9%(£20,000/QALY threshold).Probabilistic results <strong>of</strong> the prognostic modelThe main probabilistic <strong>analysis</strong> for the prognostic model, using the 1-year event rates from Mills, 155is shown in Figure 41. CTCA had the highest probability <strong>of</strong> being cost-effective at thresholds above£10,000/QALY. Around £10,000 H-FABP had the highest probability, <strong>and</strong> below this level no testing hadthe highest probability <strong>of</strong> being cost-effective.The main probabilistic <strong>analysis</strong> for the prognostic model, using the 1-year event rates from RATPAC, 12 isshown in Figure 42. No testing was highly likely to be the most cost-effective strategy for all thresholds<strong>of</strong> < £100,000/QALY.Value <strong>of</strong> information analysesThere is always a chance that the wrong decision will be made as a result <strong>of</strong> the uncertainty in the existinginformation <strong>and</strong> the costs in terms <strong>of</strong> health benefit <strong>and</strong> resources forgone owing to this uncertainty canbe interpreted as expected value <strong>of</strong> perfect information (EVPI). Perfect information would eliminate thepossibility <strong>of</strong> making the wrong decision <strong>and</strong> therefore EVPI is determined jointly by the probability that adecision based on existing information will be wrong <strong>and</strong> the consequences <strong>of</strong> a wrong decision.The EVPI, although calculated for individual patients, can also be expressed for the total population <strong>of</strong>patients who st<strong>and</strong> to benefit, based on prevalence <strong>and</strong> the lifetime <strong>of</strong> the technology. This can also bethought as the maximum that the health-care system should be willing to pay for additional evidence toinform the decision in the future <strong>and</strong> thus is an upper bound on the value <strong>of</strong> conducting further research,102NIHR Journals Library
DOI: 10.3310/hta17010 Health Technology Assessment 2013 Vol. 17 No. 1TABLE 60 Threshold <strong>analysis</strong> to identify the cut-<strong>of</strong>f risksThreshold(£/QALY)Risk <strong>of</strong> nonfatalMIRisk <strong>of</strong> death20,000 0.023 0.005730,000 0.016 0.00411.0Probability <strong>of</strong> cost-effectiveness0.90.80.70.60.50.40.30.20.10.00510 15 20 25 30 35 40Willingness-to-pay threshold (£000)4550No TestingETTHFABPCTCAICAFIGURE 41 Probability <strong>of</strong> cost-effectiveness <strong>of</strong> strategies using data from Mills. 1551.21.0Probability <strong>of</strong> cost-effectiveness0.80.60.40.200 20 40 60 80100120 140 160No testingETTHFABPCTCAICA–0.2MAICER (£000)FIGURE 42 Probability <strong>of</strong> cost-effectiveness <strong>of</strong> strategies using RATPAC data. 12 MAICER, maximum acceptableincremental cost-effectiveness ratio.i.e. if the population EVPI exceeds the expected costs <strong>of</strong> additional research then it is potentially costeffectiveto conduct further research.Partial EVPI provides the value <strong>of</strong> reducing the uncertainty surrounding particular input parameters in thedecision model <strong>and</strong> this can be used to identify the parameters for which more precise estimates would bemost valuable to focus further research. However, this is computationally expensive for complex models.© Queen’s Printer <strong>and</strong> Controller <strong>of</strong> HMSO 2013. This work was produced by Goodacre et al. under the terms <strong>of</strong> a commissioning contract issued by the Secretary <strong>of</strong> Statefor Health. This issue may be freely reproduced for the purposes <strong>of</strong> private research <strong>and</strong> study <strong>and</strong> extracts (or indeed, the full report) may be included in pr<strong>of</strong>essional journalsprovided that suitable acknowledgement is made <strong>and</strong> the reproduction is not associated with any form <strong>of</strong> advertising. Applications for commercial reproduction should beaddressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials <strong>and</strong> Studies Coordinating Centre, Alpha House, University <strong>of</strong> Southampton SciencePark, Southampton SO16 7NS, UK.103