Impact on total population health and societal cost, and the implication on the actual cost-effectiveness of including tumour necrosis factor-α antagonists in management of ankylosing spondylitis: a dynamic population modelling study

Abstract

Background: Sequential treatment of ankylosing spondylitis (AS) that includes tumour necrosis factor-α antagonists (anti-TNF agents) has been applied in most of the Western countries. Existing cost-effectiveness (CE) models almost exclusively presented the incremental CE of anti-TNF agents using a closed cohort while budget impact studies are mainly lacking. Notwithstanding, information on impact on total population health and societal budget as well as on actual incremental CE for a given decision time span are important for decision makers. This study aimed at quantifying, for different decision time spans starting from January 1, 2014 in the Dutch society, (1) impact of sequential drug treatment strategies without and with inclusion of anti-TNF agents (Strategies 1 and 2, respectively) on total population health and societal cost, and (2) the actual incremental CE of Strategy 2 compared to Strategy 1.

Methods: Dynamic population modelling was used to capture total population health and cost, and the actual incremental CE. Distinguishing the prevalent AS population on January 1, 2014 and the incident AS cohorts in the subsequent 20 years, the model tracked individually an actual number of AS patients until death or end of the simulation time. During the simulation, data on patient characteristics, history of drug use, costs and health at discrete time points were generated. In Strategy 1, five nonsteroidal anti-inflammatory drugs (NSAIDs) were available but anti-TNF agents withdrawn. In Strategy 2, five NSAIDs and two anti-TNF agents continued to be available.

Results: The predicted size of the prevalent AS population in the Dutch society varied within the range of 67,145-69,957 with 44-46 % of the patients receiving anti-TNF agents over the period 2014-2034. The use of anti-TNF agents resulted in an increase in the annual drug costs (168.54-205.28 million Euros), but at the same time caused a decrease in the annual productivity costs (12.58-31.21 million Euros) and in annual costs of healthcare categories other than drugs (7.23-11.90 million Euros). Incremental cost (Euros) per QALY gained in Strategy 2 compared to Strategy 1 corresponding to decision time spans of 5, 10, 15 and 20 years improved slightly from 75,379 to 67,268, 63,938 and 61,129, respectively. At willingness-to-pay thresholds of 118,656, 112,067, 110,188 and 110,512 Euros, it was 99 % certain that Strategy 2 was cost-effective for decision time spans of 5, 10, 15 and 20, respectively.

Conclusions: Using the dynamic population approach, the present model can project real-time data to inform a healthcare system decision that affects all actual number of AS patients eligible for anti-TNF agents within different decision time spans. The predicted total population costs of different categories in the present study can help plan the organization of the healthcare resources based on the national budget for the disease.

Keywords: Ankylosing spondylitis; Budget impact; Cost-effectiveness; Cost-utility; Discrete event simulation; Health impact; Microsimulation; Modelling; Population dynamics; Tumor necrosis factor.

Fig. 1

Main components and simulation process of the patient-level model used for tracking a patient with ankylosing spondylitis. For more information, see Tran-Duy et al. [20]

Fig. 2

Conceptual model of the dynamic population simulation. Each solid circular dot on the left margin represents an AS patient at the time point when he or she enters the simulation. The patients are differentiated into groups, including the prevalent population on 1 January 2014 and incident cohorts in years 2014–2034, whose sizes are calculated based on the sizes of the general population and prevalence and incident rate. Each solid horizontal line represents the temporal movement of each patient, which ends when the patient dies (marked with a solid diamond) or reaches January 1, 2034 (marked with an open circle on the vertical bar on the right margin). The open circles represent the patients in the prevalent population at a time point of interest who are included in the summary statistics of disease measures, costs and health utilities

Fig. 3

Changes over time in the number of patients with BASDAI within a specific interval in two treatment strategies. Strategy 1 consists of five available non-steroidal anti-inflammatory drugs (NSAIDs) and Strategy 2 consists of the same available NSAIDs as in Strategy 1 and two tumour necrosis factor-α antagonists

Fig. 4

Scatter plot of incremental total population costs against incremental total population quality-adjusted life year (QALYs) in Strategy 2 (alternative) compared to Strategy 1 (reference). Strategy 1 consists of five available non-steroidal anti-inflammatory drugs (NSAIDs) and Strategy 2 consists of the same available NSAIDs as in Strategy1 and two tumour necrosis factor-α antagonists. The clouds corresponding to January 1 of 2019, 2024, 2029 and 2034 were obtained from simulations with four different time spans of decision, 5, 10, 15 and 20 years, respectively. In each cloud, each data point was obtained from one run of simulation for all the AS patients appearing in the period from January 1, 2014 to the end of the corresponding time span with a set of model parameter values sampled from appropriate probability distributions; 10,000 runs were executed which resulted in 10,000 data points. All the data points in the four clouds lie in the north-east quadrant of the plane

Fig. 5

Cost-effectiveness acceptability curves for two treatment strategies based on the uncertainty in cost and quality-adjusted life year (QALY) on January 1 of 2019, 2024, 2029 and 2034. Strategy 1 consists of five available non-steroidal anti-inflammatory drugs (NSAIDs) and Strategy 2 consists of the same available NSAIDs as in Strategy 1 and two tumour necrosis factor-α antagonists (anti-TNF agents). The four curves were obtained from simulations with four different time spans of decision, 5, 10, 15 and 20 years