An economic assessment model was developed from a decision tree at 12 weeks combined with a Markov model of 12-week cycles, covering a 10-year time horizon. The model included hypothetical notional adult patients with an average age of 41 years, of whom 80.5% were women, who presented ≥ 4 migraine episodes per month, with one or more previous preventive treatment failures for EM, and three or more previous treatment failures for CM, one of which included onabotulinumtoxinA for CM treatment. The model was based on data from erenumab clinical trials (NCT02066415 [31], STRIVE, ARISE, LIBERTY [28,29,30] and HER-MES [26]) and adjusted to reimbursement conditions in Spain.

Populations of 1,000 notional patients each with EM or CM were assessed separately using topiramate and placebo as respective comparators, which were selected based on available clinical evidence from head-to-head comparison studies (NCT02066415 [31], STRIVE, ARISE, LIBERTY [28,29,30] and HER-MES [26]) and their reimbursement status within the Spanish Healthcare System. Topiramate is currently used as a preventive treatment for these patients in clinical practise in Spain and its comparison with erenumab was reported in the HER-MES study [26], whereas placebo was used as the control arm analysed in the main pivotal randomised control trials focusing on erenumab (NCT02066415 [31], STRIVE, ARISE and LIBERTY [28,29,30]). Erenumab and topiramate were compared among patients with EM with one or more previous treatment failures. For patients with CM, erenumab and placebo were compared after ≥ 3 failed treatments, with one of these treatments being onabotulinumtoxinA. For patients with EM and CM, prescription for symptomatic treatment was considered with triptans, aspirin, ibuprofen, ketoprofen, paracetamol, and paracetamol with codeine.

Outcomes for the cost-effectiveness analysis were incremental cost per QALY gained and incremental cost per migraine day (MD) avoided. The incremental cost-effectiveness ratio (ICER) was contrasted against the efficiency threshold per QALY in Spain more frequently described in the literature (€30,000/QALY) [32]. Although the primary analysis was performed from a healthcare system perspective, a sensitivity analysis performed from a societal perspective. The cost-effectiveness analysis followed international guidelines [33], and applied a discount rate of 3% for costs (year 2023) and effects [34].

The course of the disorder was represented by a decision tree with two alternative branches (responders and non-responders), followed by a Markov model comprising two health states (treatment and discontinuation) with cycles lasting 12 weeks (Fig. 1). Model health states were exclusive, i.e. patients could only be in one specific health state at a specific time. Responders were defined as patients with 50% reduction in the number of MMD at week 12, regardless of whether they suffered from EM or CM.

Model structure. All patients start preventive migraine treatment. Patients who at the end of 12 weeks have been classified as responders were assessed in the long term whether they remained on treatment or discontinued. Non-responders went directly to discontinuation status. Mortality was not considered as an individual health state, since this can occur to all patients, whatever their health status may be. Thus, the absorbing state for mortality is negative discontinuation

All patients entered the model at treatment initiation and, after the assessment period (12 weeks), their response (i.e. responders or non-responders) to treatment was determined. Patients who responded (responders) could either continue the prophylactic treatment initiated (state: treatment) or discontinue if they had experienced adverse events (state: discontinuation). Patients in the treatment state were assumed to remain in that state until discontinuation due to adverse events. Costs and utilities were assigned based on the number of MMDs at the end of each cycle.

The mortality of the different states was considered similar to that of the general population, adjusted by age and sex, because no additional mortality was associated with migraine [35].

Clinical data of the model were obtained from clinical trials of erenumab (Table 1). For the comparison between erenumab and topiramate in patients with EM, the HER-MES head-to-head study [26] was used. For the CM population, data for erenumab vs placebo were from Tepper et al. [31].

Data available in the HER-MES study [26] and the NCT02066415 clinical trial [31] were used to recalculate the probabilities of the adverse events applied in the model, with adjustments period required for this study (12 weeks). Mean values were weighted based on the sample size of the available studies.

Following discontinuation due to adverse events, data included in the model were obtained from the main clinical trials with erenumab [26, 28, 31]. The topiramate discontinuation values were based on those described in the HER-MES study [26]. For EM, discontinuation probability per cycle was 5.52% for erenumab [26, 28, 31] and 35.57% for topiramate [26]. For CM, discontinuation for placebo was 0.71% and for erenumab was 1.06% [31].

Since the model considered a 10-year time horizon and the selected studies had only short terms, the probability of moving from treatment to discontinuation state after 12 weeks was estimated from the available long-term data and assumed for the 10-year time horizon. This produced a probability of long-term discontinuation for each cycle (12 weeks) of 30.26% (standard error [SE] 0.0177) for placebo [36], 5.1% (SE 0.0112) for topiramate after estimating the increase in discontinuation between weeks 12 and 24 of the HER-MES study [26] and 0.5% for erenumab based on the data of the 5-year extension study [37].

Utility refers to the value that an individual places on a health state: A score of 1 represents the best possible quality of life while a score of 0 represents a state equivalent to death. In this analysis, QALYs were calculated using estimated utilities obtained from the individual data collected from the trials [28, 31], by applying a linear regression based on MMD reduction, and considering the mortality of the general population. Consequently, utilities corresponding to each MMD state of the spectrum analysed (0–28 MMD) were calculated to give a mean reduction in utility per MMD of 0.0176 (SE 0.0035). Patients without migraine showed a utility of 0.85, patients with migraine and < 3 MMD showed a utility of 0.79–0.83, EM patients with < 7 MMD showed a utility of 0.72–0.78, EM patients with ≥ 8 MMD showed a utility of 0.6–0.71, CM patients with < 21 MMD showed a utility of 0.48–0.58, and CM patients with ≥ 21 MMD showed a utility of 0.36–0.46.

Based on the utility by MMD, a utility loss was applied for adverse events [28, 31, 38]. Utility losses applied to each treatment were based on frequency of onset as follows: brain fog -0.097 (SE 0.13), fatigue -0.061 (SE 0.097), exercise intolerance -0.048 (SE 0.092), insomnia -0.048 (SE 0.088), neck stiffness and pain -0.045 (SE 0.077), muscle weakness -0.034 (SE 0.058), sleepiness -0.03 (SE 0.059), constipation -0.029 (SE 0.06), drooping eyelids -0.024 (SE 0.067), respiratory tract infection -0.012 (SE 0.033), paraesthesia -0.012 (SE 0.045), dizziness -0.01 (SE 0.041), dry mouth -0.01 (SE 0.044), injection site pain -0.008 (SE 0.025) and itchiness -0.006 (SE 0.023) [38].

Therapy costs were estimated using approved dosages (dose and frequency of use) and unit costs [39]. Ex-factory prices were used for hospital-dispensed treatments, and retail prices were applied for the remaining treatments, always using the reimbursed price. The corresponding discount according to Spanish Royal Decree 08/2010 [40] was applied.

For acute treatment with triptans, the pharmacological cost was estimated using the weighted average of almotriptan, eletriptan, frovatriptan, naratriptan, rizatriptan, sumatriptan and zolmitriptan, as well as the cost of the consumption of aspirin, ibuprofen, ketoprofen, paracetamol, and paracetamol with codeine.

The cost per health state, including hospitalisations, emergency room visits, appointments with specialists or primary care doctors, and concomitant medication (Table 2), was based on the average unit costs of each Spanish autonomous region [41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57] and the use of resources provided, which were validated by an expert panel comprising specialists in neurology, and hospital pharmacists with knowledge of the disorder. Hospitalisation costs were calculated considering the average cost of a complete stay due to migraine and other headaches (DRG 54) at the neurology unit of the 2020 Minimum Basic Database Set [58], which is an administrative database on hospitalised patients. All costs were inflated to 2023 costs using the Consumer Price Index [59].

For the societal perspective analysis, which corresponds to an alternative scenario, indirect costs, such as absenteeism and presenteeism, were included in the model. Out-of-pocket expenses were not considered as evidence is lacking for this in the field of migraine. The relationship between absenteeism measured by the MIDAS scale [60] and MMD was determined through a regression model, and the same was performed for presenteeism. After determining the volume, the cost of one day of lost work for patients with absenteeism was applied, along with half a day of work for patients with presenteeism [61].

To estimate the daily labour cost, the total monthly wage cost (€2,847.10) [62] was adjusted to 22 working days, considering an unemployment rate of 12.7% [63]. In the sensitivity analysis, the valuation of indirect costs was incorporated using WPAI [64] methodology.

These models are inevitably subject to uncertainty, thus we conducted both deterministic and probabilistic sensitivity analyses, following the recommended guidelines [34]. To evaluate the potential impact of each assigned variable, a univariate analysis was performed, varying the parameters by 20%. The results are presented in a tornado diagram [65]. A univariate sensitivity analysis was also performed using different scenarios (number of treatment failures, absence of prior failure to onabotulinumtoxinA, societal perspective, measurement of absenteeism/presentism). The model was complemented with a probabilistic sensitivity analysis by applying lognormal distributions for odds ratio, beta distributions for probabilities and utility loss, multivariate normal distributions for utilities and for the regressions assessed, and gamma distributions for resource use and discontinuation rates [66]. The probabilistic sensitivity analysis was supported by a 1000-iteration Monte Carlo simulation, which draws random values of the parameters per iteration to provide a theoretical probability distribution. The probabilistic sensitivity analysis results were plotted into a cost-effectiveness acceptability curve [67].