Population

The baseline clinical data were derived from the weighted mean of RCTs. Patients with active AS who had not been previously treated with targeted therapies (biologic-naïve patients) and who have inadequate response to NSAIDs therapy were selected for this study. Patients did not differ at baseline (p > 0.05). The mean age of the patients was 38.22 years. The Bath Ankylosing Spondylitis Disease Activity Index (BASDAI) and Bath Ankylosing Spondylitis Functional Index (BASFI) scores were 6.49 and 5.60, respectively. The male/female ratio of patients was taken from the epidemiological literature and was 73.68% for males and 26.32% for females [5]. This study did not involve animal or human population research. Clinical data were based on published RCTs, and thus the study did not require approval or ethical review.

Interventions and comparators

This study aimed to assess the cost-effectiveness of different treatment options for ankylosing spondylitis (AS). The analysis included a total of seven targeted therapies as interventions. These interventions included four TNFis (etanercept, adalimumab, infliximab, and golimumab), two IL-17 inhibitors (secukinumab and ixekizumab) and one JAK inhibitor (tofacitinib). In this study, the original drugs and generic drugs of seven targeted therapies were included at the same time. The detailed information is shown in (supplementary material Table S2.)

To evaluate the cost-effectiveness, we compared each targeted therapy intervention with conventional care (CC), which served as the comparator. In the target population, the intervention group would use targeted therapies such as biologics, and when targeted therapies fail, they would switch to conventional care, while the control group will continue to use conventional care until death. Furthermore, pairwise comparisons were conducted among the biotherapeutic interventions themselves to assess their cost-effectiveness. In the absence of head-to-head clinical trials of targeted therapies, this study the utilized the results of Network-Meta Analysis (NMA) in order to achieve indirect comparison between targeted therapies (supplementary material Table S1). BASDAI 50 model inputs were informed by a NMA of BASDAI 50 scores, with the timepoint of BASDAI 50 score taken as the primary endpoint of the relevant trial, provided this was between weeks 12 and 16. Since responder and non-responder baseline changes in BASDAI and BASFI scores during the initial treatment period were not reported separately in the clinical trials, the changes in these two indicators for TNFi responders and non-responders in this study were derived from previously published pharmacoeconomic studies. [6,7,8]

Model structure

This study used the York model established by the National Institute for Health and Care Excellence (NICE) in the UK [9], which is based on a systematic evaluation of the efficacy, safety and economics of TNFis. The model fully considers disease progression, incorporates the impact of adverse events on health outcomes, and has been used many times internationally to assess the economics of AS treatment. The model is cycled every year, thus simulating the patient’s life. Since AS is a chronic disease requiring long-term or even lifelong medication, the study was set to be lifelong, and 40 years was taken as the model cycle time based on the difference between the baseline mean age and the Chinese life expectancy [10]. The cost year was 2023, and costs and outcomes were discounted by 5% [11]. The model was constructed in Microsoft Excel® (Microsoft Corporation, Redmond, WA, USA).

As shown in Fig. 1, the York model is a 12-week decision tree model combined with the Markov model. Patients treated with targeted therapies were entered into separate Markov models based on their improvement in BASDAI score sat week 12. If a patient achieves BASDAI50 (50% reduction in BASDAI score from baseline), they will enter a three-state Markov model as a responder to the targeted therapy. In the Markov targeted therapy (Fig. 1a), the patient enters the intervention maintenance state, wherein the patient continues to be treated with the targeted therapy. These patients will then receive CC if the targeted therapies fail. They can progress directly to the death state in both the targeted therapies treatment state and the CC treatment state in Markov biology. If the patient does not reach BASDAI50, they will go directly to the CC state as nonresponders. In Markov CCs, patients may remain there or go to the death state. Patients in the control group went directly to the conventional treatment state in the Markov CC, where they will enter the death state.

Fig. 1

During cycles in the York model, disease progression is reflected by changes in BASDAI and BASFI scores, where BASDAI scores show the degree of disease activity and BASFI scores show changes in the patient's somatic function. Additionally, disease-related costs and patient utility are related to the BASDAI and BASDAI scores. The main hypotheses included in this model are as follows:

(1)

Patients are directly switched back to conventional NSAID therapy when they do not respond to a targeted therapy during the intervention.

(2)

Targeted therapy responding and nonresponding patients had similar BASDAI and BASFI scores at pretreatment.

(3)

The BASDAI score varies only in the initial phase according to whether a person responds and remains unchanged during long-term progression.

(4)

The BASDAI and BASFI scores returned to baseline levels when patients moved from the intervention treatment state to the conventional treatment state.

In the scenario analysis, we considered different prices of medicines. The lowest market price (generic price) was used for all interventions in Scenario 1, and the highest market price (originator price) was selected for all interventions in Scenario 2 for simulation.

Clinical dataTreatment response

In the decision tree branch, each targeted therapy was entered into a different Markov model based on its BASDAI50 response at week 12 (Table 1). The BASDAI50 is a commonly used outcome metric in efficacy trials to determine whether a targeted therapy is clinically effective [16]. BASDAI50 responders will be admitted to the Markov Targeted therapies System.

Table 1 Parameters in the York modelShort-term health outcomes

Short-term health outcomes after the initial 12 weeks of treatment are captured by the BASDAI and BASFI scores, which reflect the effects of different interventions on disease activity as well as patient functioning. Changes in patients' BASDAI and BASFI scores after 12 weeks vary according to the patient's response to different interventions (Table 1).

Long-term health outcomes

In addition to reflecting short-term changes in the BASDAI and BASFI scores in the York model, the model captured the impact of treatment on long-term disease progression.

In patients with AS, patient function changes with age in response to the degree of disease activity and imaging progression, and this change in patient function is reflected by the long-term progression of the BASFI score. They are mainly related to the imaging process, and they are calculated by Eq. 1. In this case, the change in BASFI for a 1-unit change in modified Stoke AS Spine Score (mSASSS) is a fixed value of 0.057 [17], whereas the annual rate of change in mSASSS varies according to the treatment measure. According to the published literature, the annual rate of change in mSASSS is 0.42, which applies to all biologics [18]. Tofacitinib, while not a biologic, is different from NSAIDs and still follows this equation in published health economics studies [8]. In contrast, CC treatment is not considered to delay imaging progression, so the annual rate of change in the mSASSS score is 1.440 during natural disease progression [18].

$$Annual \, rate \, of \, BASFI \, change \, = \, Annual \, rate \, of \, mSASSS \, change*BASFI \, change \, with \, a \, 1 - unit \, change \, in \, mSASSS$$

(1)

Withdrawal of targeted therapy

Patients who respond to the initial intervention may be moved from targeted therapies maintenance treatment status to CC status in any subsequent cycle due to withdrawal. The withdrawal rate data for each intervention are shown in Table 1.

Adverse events

Adverse events in this study included tuberculosis and serious infections, which have an impact on both utility value and cost. The annual incidence data for tuberculosis and serious infections were obtained from a Cochrane systematic evaluation of adverse events for biologics and oral targeted therapy [19], and it was assumed that the incidence of tuberculosis and serious infections would be the same for all targeted therapies, at 0.22 and 3.5%, respectively.

Mortality

Patients can enter the death state at both the intervention maintenance state and the conventional treatment state in the Markov model. The baseline mortality data (0.737%) in the model were obtained from the China National Bureau of Statistics [20], and the mortality rate of AS patients was obtained by multiplying the baseline rate with the AS mortality odds ratio, which was obtained from the published literature; the standardized death ratio for men was 1.63 [9], and that for women was 1.38 [9].

Utility

Health-related quality of life in AS patients has been proven to be dependent on BASDAI scores, BASFI scores, age and sex [7], whereas long-term BASDAI and BASFI scores were captured in the York model. Therefore, this evaluation (Eq. 2) was similarly able to model health state utility using a regression model approach. The negative utility of severe infection or tuberculosis is shown in Table 2.

$$Utility = 0:9610 - 0:0442 \, * \, BASDAI \, \, 0.0330 \, * \, BASFI \, \, 0.0111 \, * \, Sex \, [1 = male,0 = female] \, + 0:0005*Age$$

(2)

Table 2 Base-case deterministic cost-effectiveness results (targeted therapies vs. CCs)Cost

The financial burden that patients need to bear includes direct costs and indirect costs. Direct costs include direct medical costs related to disease treatment, such as drug costs, injection fees, examination fees, and hospitalization fees. Indirect costs involve the patient's labor loss due to the AS disease, such as the loss of wages of patients and caregivers. This study calculates direct health care costs based on the Chinese health system perspective, including drug and injection costs, outpatient fees, examination fees, disease-related costs and adverse event costs. Drug and injection costs, outpatient fees and examination fees are defined as initial treatment costs for the first 12 weeks and continued care costs. The median prices of targeted therapies were selected for the base analysis. The prices of drug originators and generics were considered in the scenario analysis. The drug prices were obtained from the Chinese market [21], while the injection costs, outpatient fees and adverse event costs were obtained from the literature (Table 1). The relevant fees for outpatient visits and examinations are from the price list of medical service items for each province in China. In addition, disease-related costs (Eq. 3) and adverse effect treatment costs were included in this study. Disease-related costs (Eq. 3) are dependent upon the extent of disease progression as measured by the BASFI score [9]. The cost parameters are shown in (supplementary material Table S3).

$$Disease - related \, cost \, = \, 143.53 \, * \, exp \, \left( \right)$$

(3)

Model outcomes

The final results of the model in this study included total costs incurred cumulatively after the study timeframe and quality-adjusted life years (QALYs), both calculated at a discount rate of 5%. Incremental cost-effectiveness ratios (ICERs) and incremental net monetary benefits (INMBs) were calculated to compare the economics of different treatment options. According to the World Health Organization (WHO) and China Guidelines for Pharmacoeconomic Evaluations [11] recommendations, the willingness to pay (WTP) will be 1–3 times the per capita gross domestic product (GDP) in China in 2023, namely, ¥89,358/QALY− ¥268,074/QALY. If the ICER < WTP, then the intervention was cost-effective. INMB > 0 indicates an economical intervention, and the calculation methods for the ICER and INMB are shown in Eqs. 4 and 5.

$$ICER = (C_ - C_ )/ \, \left( - E_ } \right)$$

(4)

$$INMB = WTP*\left( - E_ } \right) \, - \, \left( - C_ } \right)$$

(5)

Sensitivity analysis

We performed a one-way sensitivity analysis (OWSA) to explore the cost-effectiveness of each regimen when parameters changed between the upper and lower limits, and a tornado diagram was generated to depict the analysis results. We conducted probabilistic sensitivity analysis (PSA) by 1000 iterations of Monte Carlo simulation. We used scatter plots and cost-effectiveness acceptability curves (CEACs) to analyse the economics of targeted therapies at different WTP levels. The range of parameters included in the sensitivity analyses is shown in Table 2. Additionally, since some parameters did not have a reported range, this study selected a 10% fluctuation as the parameter variation range based on reference guidelines and expert opinions.

Scenario analysis

To gain a comprehensive understanding of the pricing dynamics in the Chinese pharmaceutical market, this study conducted scenario analyses based on two specific scenarios. The first scenario focused on the analysis of generic drugs available in the Chinese market for the treatment of AS. By examining the pricing of these generic drugs, we aimed to investigate the pricing strategies employed by pharmaceutical companies for AS generic drugs in China, considering factors such as competition, regulatory requirements, and cost-saving potential. The second scenario involved the analysis of imported innovative drugs, focusing on seven targeted therapies used to treat ankylosing spondylitis (AS) that are available in the Chinese market. Through these two distinct scenarios, our study aimed to provide a comprehensive understanding of the pricing landscape for AS treatments in the Chinese pharmaceutical market, encompassing both imported innovative drugs and generic drugs alternatives.