shows the primary results of the probabilistic analysis. The expected (average) total discounted costs and QALYs for all interventions are reported, which represent the estimated average costs and benefits (allowing for length and quality of life) per smoker, having accounted for uncertainty in the inputs to the economic model. This analysis includes disutilities and costs related to depression and self-harm. Interventions are ordered by increasing expected total cost, with NRT low having the lowest expected total cost and varenicline standard plus NRT standard having the highest expected total cost. E-cigarette low has the highest expected QALYs, followed by varenicline standard plus bupropion standard, and varenicline standard plus NRT standard. NRT low has the lowest expected QALYs.
Expected total costs, expected total utilities, ICERs and expected net benefit at a £20,000 willingness-to-pay threshold
We prefer interventions with lower costs and higher QALYs. Any intervention that has a higher expected cost and lower expected QALYs than another intervention is said to be dominated. As can be seen in , all treatments apart from NRT low are dominated by e-cigarette low, which is more effective, in terms of increased utility, and less expensive than the other interventions. If the funder is not willing to pay £56 per QALY, then NRT low is estimated to be most cost-effective. If the funder is willing to pay ≥ £56 per QALY, then e-cigarette low is estimated to be most cost-effective.
If the payer is willing to pay up to £20,000 per QALY, e-cigarette low has the highest expected net benefit (£7085), followed by varenicline standard plus bupropion standard (£6756), and varenicline standard plus NRT standard (£6591).
We present the uncertainty surrounding the cost-effectiveness of the various interventions using a CEAC (), which plots the probability that each intervention is the most cost-effective at a given willingness-to-pay threshold. Only interventions with a probability of being the optimal treatment of > 10% at any willingness-to-pay value are plotted.
Cost-effectiveness acceptability curve. Probability that treatment is optimal plotted against different willingness-to-pay per unit increase in utility (ceiling ratio). Based on 5000 Monte Carlo simulations. Std, standard.
shows that, at any willingness-to-pay value, e-cigarette low has the highest probability of being cost-effective, followed by varenicline low plus NRT standard. At any threshold above £20,000, the probability of e-cigarette low being the most cost-effective intervention is never > 30%, indicating a high degree of uncertainty about the optimal intervention.
The rank-o-grams presented in further demonstrate the uncertainty in the results. The lines are relatively flat for most interventions, showing that there is no strong probability that they will be the most or least cost-effective at a willingness-to-pay threshold of £20,000 per QALY. The exception is NRT low, which shows a clear probability that it is among the least cost-effective interventions if the payer is willing to pay £20,000 per QALY. There is a similar trend for bupropion low, bupropion standard and varenicline low, which have higher probabilities of being among the worst interventions than being among the best. The reverse trend is seen for e-cigarette low, e-cigarette high, varenicline low plus NRT standard, varenicline standard plus NRT standard and varenicline plus bupropion standard.
Rank-o-grams showing the probability that each intervention is ranked first, second, etc. based on net benefit at a willingness-to-pay threshold of £20,000 per QALY. Std, standard.
Value-of-information analysis
shows the results of the value-of-information analyses for the base-case model at a willingness to-pay per QALY threshold of £20,000. EVPI estimates the most the funder would be prepared to pay to eliminate uncertainty in the model input parameters. EVPI is helpful for understanding whether or not future research may potentially be of value. The per-quitter EVPI is £3645 and the population EVPI, representing all of the smokers attempting to quit in England, is £999M over a 1-year time horizon and £4994M over a 5-year time horizon. These values are substantial and suggest that future research studies to reduce parameter uncertainty in the model would be valuable, as the decision is clearly sensitive to uncertainty in the model inputs.
Expected value of perfect information and EVPPI for various subsets of model parameters, at a £20,000 willingness-to-pay value per QALY
Expected value of partial perfect information (EVPPI) estimates the most that the funder would be prepared to pay to eliminate uncertainty in a specific subset of model input parameters. Comparing EVPPI for different parameters allows us to identify the subsets of model inputs to which the decision is most sensitive. This can indicate where future research efforts may be invested most effectively. There is a high value per smoker in reducing uncertainty in all of the abstinence probabilities (£3053) but less of a value in reducing uncertainty in all of the AEs probabilities (£1654). EVPPI is marginally higher for cost parameters (£1216) than for utility parameters (£947).
We explored the potential value of a new trial comparing the two interventions with the highest expected net benefit, e-cigarette low and varenicline standard plus bupropion standard, which would provide information on the effectiveness of the interventions, costs and utilities. This gives a per-quitter EVPPI of £2342 and a population EVPPI of £642M over a 1-year time horizon and £3869M over a 5-year time horizon. Restricting to the collection of intervention effects only would reduce this value marginally to £1676, suggesting that a large trial, conducted well and adequately powered, may be a cost-effective area of future research, but that it may be most important to collect information on probabilities of abstinence and AEs. In particular, a trial comparing e-cigarettes with an active comparator such as varenicline standard plus NRT standard or NRT standard is likely to be a cost-effective investment.
Sensitivity analysis with results based on abstinence alone
shows the primary results of the sensitivity analysis when the impact of depression and self-harm is removed from the model. In this case, bupropion low has the lowest expected total cost. Varenicline standard plus NRT standard, again, has the highest expected total cost. Varenicline standard plus NRT standard has the highest expected QALYs, followed by varenicline low plus NRT standard. NRT low has the lowest expected QALYs.
Expected total costs, expected total utilities, ICERs and expected net benefit at a £20,000 willingness-to-pay threshold, based on abstinence alone
An intervention is said to be ‘extendedly dominated’ if a mix of two interventions can provide the same QALYs at a lower cost. As can be seen in , all treatments apart from NRT high, e-cigarette high, e-cigarette low, varenicline low plus NRT standard and varenicline standard plus NRT standard are dominated by bupropion low, which is more effective, in terms of increased utility, and less expensive than the other interventions.
The interventions on the efficiency frontier (i.e. those that are not dominated or extendedly dominated) are NRT low, e-cigarette low and varenicline standard plus NRT standard. If the payer is not willing to pay £159 per QALY, then bupropion low is estimated to be most cost-effective. If the payer is willing to pay between £159 and £1302 per QALY, then e-cigarette low is estimated to be most cost-effective, and if the willingness to pay per QALY is above £1302, then varenicline standard plus NRT standard is estimated to be most cost-effective.
At a willingness-to-pay threshold of £20,000, varenicline standard plus NRT standard has the highest expected net benefit (£9895), followed by varenicline low plus NRT standard (£9759).
We present the uncertainty surrounding the cost-effectiveness of the various interventions, using a CEAC (). Only those interventions with a probability of being the optimal treatment of more than 10% at any willingness-to-pay value are plotted. shows that, at any willingness-to-pay value, varenicline low plus NRT standard has the highest probability of being cost-effective, followed by varenicline standard plus NRT standard. At any threshold above £20,000, the probability of any intervention being the most cost-effective intervention is never > 40%, again, indicating a degree of uncertainty in the optimal intervention.
Cost-effectiveness acceptability curve. Probability that treatment is optimal plotted against different willingness-to-pay per unit increase in utility (ceiling ratio). Based on 5000 Monte Carlo simulations. Sensitivity analysis based on abstinence alone. (more...)
The rank-o-grams are presented in .
Rank-o-grams showing the probability that each intervention is ranked first, second, etc. based on net benefit at a willingness-to-pay threshold of £20,000 per QALY. Sensitivity analysis based on abstinence alone. Std, standard.
Sensitivity analysis with only UK-licensed interventions included
shows the primary results of the sensitivity analysis including only interventions that are licensed in the UK (NRT low, standard and high, bupropion low and standard, and varenicline low and standard). In this case, NRT low has the lowest expected total cost and varenicline low has the highest expected total cost. Varenicline standard has the highest expected QALYs, followed by NRT standard. NRT low has the lowest expected QALYs.
Expected total costs, expected total utilities, ICERs and expected net benefit at a £20,000 willingness-to-pay threshold, based on licensed interventions only
As can be seen in , all treatments apart from NRT low, bupropion low and NRT standard are dominated by varenicline standard, which is more effective, in terms of increased utility, and less expensive than the other interventions. Bupropion low is extendedly dominated by NRT standard.
The interventions on the efficiency frontier are NRT low, NRT standard and varenicline standard. If the payer is not willing to pay £32 per QALY, then NRT low is estimated to be most cost-effective. At a willingness to pay per QALY above £32, but below £15,665, NRT standard is estimated to be most cost-effective. At a willingness to pay per QALY above £15,665, varenicline standard is estimated to be most cost-effective.
At a willingness-to-pay threshold of £20,000, varenicline standard has the highest expected net benefit (£3697), followed by NRT standard (£3663).
We present the uncertainty surrounding the cost-effectiveness of the various interventions using a CEAC (). Only those interventions with a probability of being the optimal treatment of more than 10% at any willingness-to-pay value are plotted. shows that, at any willingness-to-pay value above £5000, NRT standard has the highest probability of being cost-effective, followed by varenicline standard.
Probability treatment is optimal plotted against different willingness-to-pay per unit increase in utility (ceiling ratio). Based on 5000 Monte Carlo simulations. Sensitivity analysis based on licensed interventions. Std, standard.
The rank-o-grams are presented in . These show that, at a willingness-to-pay value of £20,000, NRT standard and varenicline standard have the highest probabilities of being the most cost-effective treatment. NRT low and varenicline low have very low probabilities of being the most cost-effective.
Rank-o-grams showing the probability that each intervention is ranked first, second, etc. based on net benefit at a willingness-to-pay threshold of £20,000 per QALY. Sensitivity analysis based on licensed interventions only. Std, standard.
