Base-Case Analysis (Fixed Stockpiles)
In our base-case model, we estimated that in a pandemic, ILI would develop in 28.1% of the population (16.8 million persons) (including 15 million pandemic influenza patients). Under the high CFR conditions of the 1918 scenario, ≈344,000 deaths would occur compared with ≈44,000 deaths under the 1957/69 scenario (). These scenarios would result in the loss of ≈2.2 or 0.4 million discounted QALYs, respectively, with a total discounted cost to the NHS of £113 million if no treatment program were initiated (no intervention). The treat-only program would reduce this loss by 700,000 or 90,000 QALYs at a cost of ≈£1,900 or £13,700 per QALY gained for the 1918 and 1957/69 scenarios, respectively, well below the £30,000 threshold. The test-treat program would further reduce this loss slightly by 30,000 or 4,000 QALYs but at a high cost of ≈£31,000 or £228,000 per QALY gained over the treat-only alternative. The test-treat option would be unlikely to be considered because cost-effectiveness is highly dependent on the fatality scenario.
Univariate Sensitivity Analysis of the Treat-Only Program (Fixed Stockpile)
Because the treat-only program was the most cost-effective program under both fatality scenarios, we carried out a univariate sensitivity analysis of the incremental cost-effectiveness of this program to variability in model parameters (Appendix Table
). AV drug efficacy for reducing complications and hospitalizations had minimal effect on the cost-effectiveness of the treat-only program, but this strategy was highly sensitive to AV drug efficacy for reducing death (). This was due to the relatively high QALY loss associated with pandemic influenza death (94% and 69% of the total QALY loss for the 1918 and 1957/69 scenarios, respectively). Because the value of this parameter is unclear, further studies of the potential protective effect of AV drugs against death are essential.
Figure 2 Univariate sensitivity analyses of the incremental cost-effectiveness of the treat only strategy over the no-intervention strategy to model parameters under the 1918 scenario (A) and 1957/69 scenario (B). OR, odds ratio; AV, antiviral; CAR, clinical attack (more ...)
The timing of the pandemic and the discount rate were influential parameters. However, variation in the timing of the epidemic is unlikely to change the recommendation that the treat-only strategy is cost-effective. If an epidemic occurs in 45 years, the costs per QALY gained would be approximately £3,800 and £28,000 for the 2 fatality scenarios (discount rate of 3.5%), still below the £30,000 threshold. At a discount rate of 6%, the treat-only option would be cost-effective for up to 30 years.
The efficiency of AV drug distribution is likely to be important. The program would however remain cost-effective if the probability of receiving AVs within 48 hours did not drop below 35% (£3,700 or £27,000 per QALY gained for the 1918 and 1957/69 scenarios), respectively.
The treat-only program was slightly more cost-effective in the summer than the winter or midwinter as the probability of pandemic influenza being the cause of ILI was higher (96%, 89%, and 84%, respectively). Wastage (wasted quantities, fraud, theft) of AV drug supplies as high as 25% had little effect on cost-effectiveness of the treat-only option ().
A lower CAR (15%) reduced the cost-effectiveness of AV drugs because some of the stockpile would not be used (surplus). Increasing the CAR above 25% had no effect on programs with fixed stockpiles because the same number of deaths and complications would be prevented at the same cost (the proportion of deaths prevented would be reduced, but the absolute number would remain the same).
Threshold Conditions for Test-Treat Option (Fixed Stockpiles)
For high CARs, where a fixed AV drug stockpile is less than the expected demand (as in the base-case), near-patient tests could be used to better target therapeutic courses. A univariate sensitivity analysis of the incremental cost-effectiveness of test-treat over treat only to variability in the near-patient test parameters, test sensitivity, specificity, unit cost, and shelf-life, was carried out. Under the 1918 scenario the test-treat strategy would require test sensitivity to exceed ≈90% (, panel A) and a test unit cost below £6 or a shelf-life above 3 years (, panel B) to be considered cost-effective. Test specificity would have little effect on the incremental cost-effectiveness because it has no effect on QALY loss. Under the 1957/69 scenario test-treat would never cross the cost-effectiveness threshold even with a 100%-sensitive or 100%-specific test, a test cost as low as £0, or a shelf-life as high as 4 years.
Figure 3 Univariate sensitivity analyses of the incremental cost-effectiveness of the test-treat strategy over the treat only strategy to A) near-test sensitivity and specificity and B) near-test unit cost and shelf-life. The test-treat program becomes cost-effective (more ...)
Probabilistic Sensitivity Analysis (Fixed Stockpiles)
Model parameters were varied (Appendix Table
) in a probabilistic sensitivity analysis, which suggests that for fixed AV drug and test stockpiles, the probability is high that the treat-only option would be cost-effective, irrespective of the fatality scenario (). The test-treat option would result in small QALY gains (and often losses) but at substantial additional costs. The probability of this strategy being cost-effective is low compared with the treat-only option, particularly for the 1957/69 fatality scenario.
Figure 4 Probabilistic sensitivity analysis of the incremental cost-effectiveness of the treat-only over the no-intervention strategy and the test-treat strategy over the treat-only strategy for the A) 1918 and B) 1957/69 death scenarios (1,000 iterations). Cost-effective (more ...)
Incremental Cost-effectiveness during a Pandemic Wave (Fixed Stockpiles)
The probability that an ILI case will be due to pandemic influenza will vary over the time course of a pandemic (assumed to peak between wk 6 and 7 for a wave lasting 15 wk) (2
). Therefore, near-patient testing may be useful during early stages of a pandemic when clinical judgment is low and inappropriate AV administration is high. The cost-effectiveness of test-treat over a pandemic wave was analyzed.
The AV drug stockpile was assumed to remain fixed at 14.6 million courses (1
), and the test stockpile was varied with the cumulative number of ILI cases expected per week of the pandemic wave. shows the total incremental cost-effectiveness of the test-treat strategy over treat only for each test stockpile for a CAR of 25%. Test-treat would be cost-effective (<£30,000 per QALY gained) for test stockpiles up to 12.1 million (the expected no. of cumulative ILI cases at wk 8 of a pandemic) under the 1918 scenario. Test-treat may even be considered for test stockpiles up to 13.7 million (wk 9 of a pandemic) as the cost-effectiveness was ≈£32,700 per QALY gained. However, under the 1957/69 scenario test-treat would not be cost-effective at any stage of the pandemic, although it may be considered for test stockpiles up to ≈35,000 (wk 2 of a pandemic) as the cost was ≈£34,000 per QALY gained.
Figure 5 Incremental cost-effectiveness of the test-treat strategy over the treat-only strategy during a pandemic wave (antiviral [AV] stockpile = 14.6 million courses, test stockpile = number of cumulative influenza-like [ILI] cases, clinical attack rate = 25%). (more ...)
For a CAR of 15% (data not shown), test-treat would not be cost-effective throughout a pandemic as the AV drug stockpile would exceed demand (cumulative ILI cases). For a CAR of 35% (data not shown), although test-treat would be cost-effective for test stockpiles up to 16.4 million (wk 8) under the 1918 scenario, it would not be cost-effective at any stage of the pandemic under the 1957/69 scenario. Therefore in the short-term, stockpiling enough tests for the first 2 weeks of a potential influenza pandemic (≈35,000 tests) could help conserve limited AV drug stockpiles. However, this is highly dependent on the CAR and CFR. In the long-term, a more cost-effective plan would increase AV drug stockpiles to cover expected demand (CAR + background ILI + wastage, see optimal Stockpiling).
In the long term, it may be more cost-effective to increase stockpiles to cover expected demand (CAR + background ILI + wastage). and the Appendix Figure
show the expected costs and QALY losses under a range of different AV drug and test stockpiles and CARs (base-case test characteristics and AV drug efficacy assumed). Each point represents 1 scenario (test and AV drug stockpile size). Points on the efficiency line were potentially cost-effective strategies. Strategies that increase cost but reduce QALY loss (moving from left to right on the efficiency line) should be considered until the slope of the line exceeds the threshold of £30,000 per QALY (efficiency line ends). For each CAR, this process suggested that the optimal strategy was treat only, stockpiling enough AV drugs to meet demand (CAR plus background ILI plus AV drug wastage). Therefore, for a CAR of 25%, the optimum stockpile was ≈20 million AV drugs only () because the expected number of ILI cases would be 16.8 million (15 million of which would be pandemic influenza) and the expected AV drug wastage would be 2.2 million. The test-treat strategies were never on the efficiency line, even for a perfect test (100% sensitivity and specificity), because they resulted in similar QALY loss as treat only but at increased costs. Indeed, when the size of the AV drug stockpile exceeded the demand, test-treat resulted in increased QALY loss (if the test is not 100% sensitive), because some true pandemic influenza case-patients would be denied treatment even though there was a surplus of AV drug courses.
Figure 6 Optimal cost-effectiveness of antiviral (AV) and test stockpiling (0–30 million units) for a clinical attack rate (CAR) of 25% under the A) 1918 and B) 1957/69 scenarios. The composite test (Test A) and a perfect test of 100% sensitivity and 100% (more ...)