In the trial 401 patients aged 18-65 who reported an average of at least two headaches per month were recruited from general practices in England and Wales and randomly allocated to receive either up to 12 acupuncture treatments over three months from appropriately trained physiotherapists or usual care alone.^{6}

For the purposes of this evaluation we assume that the acupuncture intervention to be provided in the community by the NHS; hence we measure costs from both an NHS perspective and a societal perspective. We measured effectiveness in terms of the quality adjusted life years (QALYs) gained. For our base case, we have taken a conservative approach by excluding savings in productivity costs and by adopting a time horizon of 12 months, the length of the trial follow up. Given the time horizon, no need arose to discount costs or effects. We measured costs in UK prices (£) for 2002-3. We used the algorithm devised by Brazier et al,^{7} a single index measure of health related quality of life (HRQoL)—the SF-6D—to calculate for each patient at baseline, three months, and 12 months from patients' responses to the SF-36 at each of these time points.

The patients themselves reported unit costs associated with non-prescription drugs and private healthcare visits. We used the health component of the harmonised index of consumer prices to inflate these costs to 2003 levels.^{8}
details other unit costs. We used standard NHS costs for a specific service if these had been published.^{9} For NHS visits to practitioners of complementary or alternative medicine we used the mean cost of a private visit, as recorded in the trial. We recorded drug prescriptions for a subgroup of patients (n = 71) from the database of their general practitioner.

To estimate the cost of the study intervention we took the standard cost (including overheads, capital, and training) for an NHS community physiotherapist^{9} and multiplied it by the contact time for each individual patient with the physiotherapist trained in acupuncture. We did not include the cost of needles and other consumables as these are negligible compared with staff time.^{13} We assumed that acupuncture sessions on the NHS, but not by a study acupuncturist, had a duration equal to the mean duration of a study session, 31 minutes.

We used using linear regression (analysis of covariance, ANCOVA) with age, sex, diagnosis (migraine or non-migraine headache), severity of headache at baseline, number of years of headache disorder, site, and baseline SF-6D as covariates to estimate differences between groups for cost and effectiveness on the intention to treat principle. Exact methods for estimating confidence intervals for incremental cost effectiveness ratios are not possible, and we therefore used the net benefit approach to estimate parametric cost effectiveness acceptability curves.^{14}^{,15} Net benefit analysis usually requires any gain in outcome (for example, QALYs) from an intervention to be valued by using the ceiling ratio, λ, defined as the decision makers' willingness to pay for an additional unit of health outcome, and from this any additional costs are subtracted. A λ equal to £30 000 per QALY is a threshold of cost effectiveness consistent with decisions that have been taken by the National Institute for Clinical Excellence (NICE).^{16} The cost effectiveness acceptability curves show the probability that the incremental cost effectiveness is below λ, for a range of values of λ. We used SPSS for Windows, version 11.0.0, to perform statistical analysis and Microsoft Excel 2002 SP2 for the calculation of cost effectiveness acceptability curves.

For the base case we conducted no imputation for cases missing HRQoL data; therefore the cost effectiveness analysis sample was those patients who reported SF-36 completely in all three questionnaires and for whom QALYs could therefore be calculated. Data on use of resources and cost were available for a larger sample of cases, and for these variables we report statistics for all responding patients.

Economic evaluation is subject to uncertainty not just because of sample variation but also because of assumptions made and generalisability issues.^{17} We therefore conducted sensitivity analyses to test the robustness of the results to changes in the base case assumptions. We varied the staff time and grade associated with acupuncture treatment and used different strategies for missing data. We also added productivity costs by multiplying the number of days sick from work or other usual activity, as reported by the study patients, by the average earnings per day in England and Wales^{11} inflated to 2003 prices.^{12} The base case analysis does not project beyond the 12 months of observation. It is improbable that the difference in HRQoL observed at 12 months would disappear immediately. In the sensitivity analysis we assumed that, although the study acupuncture intervention was delivered as a one off package and not taken up again in subsequent years, the difference in costs (excluding acupuncture) and effectiveness would gradually subside at the same rate over varying periods of time. We discounted costs at 6% and QALYs at 1.5%, in keeping with the conventions of UK central government.^{18}