Modelling the case for cost-effectiveness of interventions to improve medication adherence in patients with difficult to control asthma

Patient and Public Involvement (PPI)

While the analysis reported in this paper is based on a hypothetical cohort of patients PPI formed a fundamental part of the inspiration of the research idea. Data collected from interviews and focus groups, conducted as part of the Straight Talking Asthma Information Project, suggested that there was a variable level of knowledge about asthma care, with many myths and misconceptions circulating around the condition.

Interviews from the South Asian community in Kirklees uncovered that language played an enormous role in community members not knowing where to access help for asthma, including what to do in an emergency situation. In some cases there were low levels of understanding when attending doctors’ appointments, how and when to take medications. Community members also stated that they often felt excluded and under- represented in the health literature that is available to them. Much health literature did not visibly represent them, nor did they cover the culturally specific questions that they wished to ask.

These insights from patients and communities directly led to the collaboration between the authors of this paper and the underlying aim of this study to better understand the value case for investment in interventions which improve the adherence to asthma medication.

The model overview

This research employs a refined Markov model from Zafari’s study (Zafari et al., 2016), which categorizes patients into three symptom control states—controlled, partially controlled, and uncontrolled—alongside states for exacerbation and death. We extend its application to examine asthma management dynamics over a 20-year horizon, using weekly cycles to capture short-term changes and long-term outcomes. Asthma control criteria are aligned with the Global Initiative for Asthma (GINA) guidelines (Asthma, 2023). Notably, the model accounts for asthma-related mortality exclusively in the context of exacerbations, distinguishing other deaths as non-asthma related. Figure 1 illustrates the operational structure of the model, delineating the transitions between different states of asthma control, exacerbation, and mortality. The model's computations are executed within Excel.

Figure 1. Schematic representation of the modified Asthma Markov Model.

Asthma control, as per GINA guidelines, is assessed weekly through a combination of patient diary entries and exacerbation data, focusing on: (1) absence of night-time symptoms, (2) two or fewer days of daytime symptoms per week, (3) limited reliever usage, (4) maintaining Peak Expiratory Flow at or above 80% of the predicted normal, and (5) no limitations on daily activities. A week is classified as 'Controlled' if all criteria are met without severe exacerbation. 'Partially Controlled' weeks feature one or two unmet criteria without exacerbation, while weeks with three or more unmet criteria or any exacerbation are deemed 'Uncontrolled.'

Cohort patients

The study specifically addresses adults with difficult-to-control asthma from the NHS and personal social services (PSS) perspective, recognizing the heightened risk and crucial need for adherence in this subgroup (Murphy et al., 2012). Utilizing data from the 2020 Annual Asthma Survey (Asthma UK, 2020), the initial state distribution reflects the current UK asthma population: 40.1% uncontrolled, 39.4% partly controlled, and 20.5% fully controlled. This baseline facilitates an accurate assessment of intervention impacts under prevailing conditions. The study further categorizes exacerbations by severity and hospital admission status—39% admitted versus 61% non-admitted—highlighting the diverse management needs within the difficult-to-control asthma demographic (Losappio et al., 2019).

Model assumptions

Three key assumptions underlie our model's design:

1) Exacerbation Duration and Hospital Stays: The length of acute exacerbations and associated hospitalizations does not affect transition probabilities outside of control status and medication adherence. This assumption simplifies the model without compromising its ability to capture critical dynamics.

2) Constant Adherence Levels: Adherence is considered static over time for each patient, despite potential real-world variations influenced by symptom severity and exacerbation frequency. This facilitates a straightforward comparison of long-term outcomes across different adherence scenarios.

3) Exclusivity of Exacerbation Severity: While patients may experience multiple exacerbation types within a single cycle, the model treats exacerbation severity as mutually exclusive. This approach allows for a direct comparison between cohorts with varying adherence levels, elucidating the distinct long-term outcomes tied to adherence.

Transition probabilities and treatment effectiveness

The transition probabilities within the Markov model were derived from a meta-analysis conducted by Chongmelaxme (Chongmelaxme et al., 2020), focusing on the odds ratio of care discontinuity. The non-asthma related mortality rates were sourced from the National Life Tables by the Office for National Statistics of the United Kingdom (Statistics, 2021b). Given that asthma severity fluctuates but does not directly correlate with mortality in the context of this study, no differential mortality was assumed between the health states other than via correlation with exacerbations.

Under scenarios of full adherence, the transition probabilities among the three asthma control states and exacerbation rates were informed by Bateman (Bateman et al., 2010), while mortality rates specific to asthma exacerbations were adopted (Engelkes et al., 2020). Transitions from the exacerbation state back to the control states were based on findings from Bateman (Bateman et al., 2008). Under scenarios of zero adherence (i.e. no treatment), the probabilities of transition were calculated using odds ratios from Chongmelaxme's meta-analysis (Chongmelaxme et al., 2020), providing a comprehensive understanding of care discontinuity impacts on asthma management.

This study assumes a non-linear relationship between treatment adherence and effectiveness, where effectiveness is exponentially related to adherence rates. This allows for a nuanced comparison of intervention cost-effectiveness across varying adherence levels, modelled after the systematic review and meta-analysis by (Chongmelaxme & Dilokthornsakul, 2019). Figure 2 illustrates the relationship between treatment adherence and effectiveness, serving as a visual aid to complement our analytical model. The formula used to calculate treatment effectiveness is as follows:

Figure 2. Treatment effectiveness by adherence level.

Health resource utilization and model costs

Healthcare resource utilization and associated costs were catalogued, encompassing prescription refills for maintenance therapy, outpatient visits, and rescue therapy for patients across all asthma control states. Exacerbation care was differentiated between non-admitted and admitted patients, with healthcare utilization tailored to each scenario, including outpatient attendance, community asthma care services, and medication costs for non-admitted patients, and outpatient service, hospitalization, and ambulance services for admitted patients. These costs were aligned with recent national unit cost data, ensuring an accurate economic analysis. All costs are presented in 2022 GBP (Great British Pounds).

The model acknowledges variations in patient treatment steps, as defined by the National Institute for Health and Care Excellence (NICE, 2021), to tailor treatment intensity and minimize unnecessary medication use while maintaining asthma control. This approach is based on an observational study by (Bloom et al., 2020), which provided insights into treatment step distribution within the population.

Outpatient attendance rates were estimated based on GINA 2023 recommendations (Asthma, 2023), with well-controlled patients assumed to visit every six months, partially controlled patients every three months, and uncontrolled patients monthly. The unit costs for medication refills were sourced from the NICE website and NHS Reference Costs 2021/22 (NHS, 2023). This study's scope specifically addresses adherence and its effectiveness in relation to pharmacological management, excluding the variable of outpatient visit frequency as influenced by medication adherence levels from the analysis. Consequently, the model differentiates the impact of adherence levels solely on maintenance therapy-related healthcare resource utilization. In contrast, resource use for rescue therapy and exacerbation treatment remains consistent across adherence levels, reflecting the expectation that patients with poor adherence will still seek emergency care in severe conditions.

Health state utility values and cost-effectiveness analysis

Health state utility values were adopted (Briggs et al., 2021), with HRQoL measured using the EQ-5D-3L instrument, adjusted for age and sex based on the Health Survey for England (Craig et al., 2013). Outcomes were quantified in quality-adjusted life years (QALYs), incorporating utility data for both non-admitted (61%) and admitted (39%) patients.

Results are presented related to the impact of changes to the adherence rate on the mean values of the key outputs of the decision analytic model, specifically number of exacerbations, life years, QALYs, and total costs to the NHS and PSS. Consistent with best practice for conducting comparative cost-effectiveness analysis in the UK, defined by the NICE methods guide, all outputs are reported over a 20-year time horizon but with costs and QALYs are discounted at a rate of 3.5% per annum (NICE, 2022).

To assess the robustness of our model's outcomes against uncertainties in parameter estimates, a probabilistic sensitivity analysis (PSA) was conducted. This analysis adheres to the methodological framework proposed by Briggs et al. (Briggs et al., 2006), which recommends a comprehensive approach to examining the impact of parameter uncertainty on model results. Specifically, the model was configured to execute 1,000 Monte Carlo simulations for each of the 10 levels of medication adherence reported, encapsulating the variability in key parameters. Distributional assumptions for these parameters were selected based on best practices outlined in the cited guidelines, ensuring that the analysis reflects realistic variations in model inputs, a full list of the distributions used, and the informative sources and assumptions, is available in the Appendix. For each of the four key model outputs the mean cohort values are reported alongside a 95% confidence interval drawn from the Monte-Carlo simulation.