Incorporating adherence in cost-effectiveness analyses of asthma: a systematic review

Abstract

Aims: Non-adherence is associated with poor clinical outcomes among patients with asthma. While cost-effectiveness analysis (CEA) is increasingly used to inform value assessment of the interventions, most do not take into account adherence in the analyses. This study aims to: (1) Understand the extent of studies considering adherence as part of the economic analyses, and (2) summarize the methods of incorporating adherence in the economic models.

Materials and methods: A literature search was performed from the inception to February 2018 using four databases: PubMed, EMBASE, NHS EED, and the Tufts CEA registry. Decision model-based CEA of asthma were identified. Outcomes of interest were the number of studies incorporating adherence in the economic models, and the incorporating methods. All data were extracted using a standardized data collection form.

Results: From 1,587 articles, 23 studies were decision model-based CEA of asthma, of which four CEA (17.4%) incorporated adherence in the analyses. Only the method of incorporating adherence by adjusting treatment effectiveness according to adherence levels was demonstrated in this review. Two approaches were used to derive the associations between adherence and effectiveness. The first approach was to apply a mathematical formula, developed by an expert panel, and the second was to extrapolate the associations from previous published studies. The adherence-adjusted effectiveness was then incorporated in the economic models.

Conclusions: A very low number of CEA of asthma incorporated adherence in the analyses. All the CEA adjusted treatment effectiveness according to adherence levels, applied to the economic models.

Keywords:

• Asthma
• adherence
• cost-effectiveness
• economic evaluation
• systematic review

JEL classification codes:

• I00
• I18

Introduction

Asthma is one of the most common chronic respiratory diseases, affecting ∼300 million people worldwide^1,2. It is characterized by several symptoms. For example, coughing, wheezing, shortness of breath, chest tightness, and expiratory airflow limitation². Asthma symptoms may be absent for weeks or months, and may be responsively resolved to the medications^1,2. The prevalence of asthma has been increasing over the last few decades. In the US, it was ∼8.2%, while in China, Malaysia, Singapore, and Thailand it was 5.1%, 9.1%, 10.6%, and 12.2%, respectively^3,4.

Patient non-adherence is a common and costly problem for asthma treatment⁵. The evidence revealed that 50% of children and adults did not take their prescribed medications. This is associated with uncontrolled symptoms, and the increase in exacerbation rates and asthma-related death⁶. Healthcare utilizations for asthma was very high, being expected to reach up to 2% of the total healthcare expenditure in developed countries². Previous studies indicated the importance of patient adherence for asthma. In brief, increasing adherence was associated with improvement in asthma control and lung function, as well as reductions in exacerbation rate and healthcare utilizations^7–13.

CEA is a well-established framework, that is used to estimate the incremental costs per unit of the incremental benefits provided by an intervention¹⁴. Decision model-based CEA facilitates the use of multiple data sources, including costs of treatment and long-term health outcomes, which are not captured through trial-based CEA. Results of the cost-effectiveness, referred to as an incremental cost-effectiveness ratio (ICER), is used as a supportive document, enabling healthcare professionals and policy-makers to make effective decisions pertaining to health technology assessment (HTA) of the interventions¹⁵.

Previous literature reviews investigated CEA that incorporated adherence in the analyses. A systematic review by Rosen et al.¹⁶ determined the inclusion of patient adherence in the CEA. The authors found that, among 177 included studies, less than one-third (54 studies) integrated sub-optimal adherence in the economic analyses. A study by Hughes et al.¹⁷ evaluated the impact of non-adherence on the cost-effectiveness of different drug therapies. The authors included 22 studies in the review, and showed that non-adherence reduced the efficacy of therapies, but its impact on health care costs was varied. Another study by Cleemput et al.¹⁸ reviewed literature on the economics of therapeutic non-adherence, and identified methodology flaws. Eighteen studies were included, and assessed according to their definition, measurement of adherence, study designs, as well as identification and valuation of costs and outcomes. The results indicated that most studies lacked methodological rigor, and failed to meet qualitative standards. The most updated review, conducted in 2007 by Hughes et al.¹⁹, highlighted the importance of integrating adherence in economic evaluations. Although the methods of incorporating adherence were characterized, there was a great deal of inconsistency in adherence definitions, and the integrating methods from study to study. In addition, the authors only included 10 studies in the analysis. This resulted in limited generalizability of the study findings across therapeutic areas.

Non-adherence is associated with poor clinical outcomes among patients with asthma. Although CEA is increasingly used to inform value assessment of the interventions by healthcare professionals and policy-makers, one important aspect that still lacks clarity is how to incorporate adherence in the analysis. To our knowledge, none of the studies provide an insight regarding the methods of incorporating adherence in CEA of asthma, thus the knowledge on such practices is still limited. This study aims to: (1) Understand the extent of studies considering adherence as part of the economic analyses, and (2) Summarize the methods of incorporating adherence in the economic models. Our findings will provide healthcare professionals and policy-makers with current evidence of the methods used to incorporated adherence in the CEA of asthma.

Methods

Search strategy

A literature search was performed from the inception to February 2018 using four databases: PubMed, EMBASE, NHS EED, and the Tufts CEA Registry. The search filters used for identifying economic evaluations were combined with various search terms including cost-effectiveness, cost-utility, economic evaluation, and asthma²⁰. The bibliographies of retrieved articles were examined for the studies that were not indexed in the aforementioned databases.

Study selection

Initially, the titles and abstracts were screened to identify the potential studies. Only decision model-based CEA of the pharmacological interventions for asthma, which included the results of incremental costs per unit of the incremental benefits, were included in the analysis. Decision model-based CEA of non-pharmacological interventions were out of the scope of this review. Outcomes of interest were the number of studies that incorporated adherence in the analyses, and the methods of incorporating adherence in the economic models. Only studies published in English were included. Full texts of relevant studies were assessed by BC and PD, and all disagreements between the investigators were resolved by a third reviewer (NC).

Data extraction and quality assessment

Data extraction was undertaken by BC and PD, using a standardized data collection form. The extracted data included author names, publication years, study countries, study objectives, the characteristics of participants and interventions, comparators, study outcomes, types of economic analysis, study models, perspectives, cycle lengths, time horizons, study results, and adherence data. All studies were assessed for their methodological qualities, and the quality of the reporting studies by BC and PD, using the Consensus on Health Economic Criteria-extended (CHEC-extended) and the Consolidated Health Economic Evaluation Reporting Standards (CHEERS), respectively (Supplementary Table S1E)^21–23.

Data analysis

The number of CEA that incorporated adherence in the analyses were calculated as the percentage of studies considering adherence as part of the economic analyses. All the CEA were reviewed to summarize adherence data, methods of incorporating adherence, and the impact of adherence on the cost-effectiveness results.

Results

The initial search yielded 1,587 articles, of which 344 duplicates were removed. The remaining 1,243 articles were screened through the titles and abstracts. A total of 1,080 articles were excluded because of their irrelevance to asthma and the CEA. This resulted in 163 articles being assessed for their eligibility. A further 140 articles were excluded for the following reasons; non-English articles (n = 13), duplications (n = 17), non-decision model-based CEA (n = 42), as well as the abstracts, reviews, correspondence, and letters to the editor (n = 68). This yielded a total of 23 CEA of asthma, of which four incorporated adherence in the analyses^24–27. A Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram is shown in Figure 1. Results of the initial search are presented in Supplementary Table S2E.

Figure 1. A PRISMA flow diagram describing the study selection process. This figure demonstrates the approach of studies, included in this systematic review. Overall, 1,587 articles were identified through the initial search, of which 163 were assessed for their eligibility. In conclusion, a total of 23 CEA of asthma were identified, and four of the studies incorporated adherence in the analyses.

General characteristics

Twenty-one studies (91.3%) were conducted to carry out the cost-effectiveness of interventions in a single country; US (eight studies)^24,27–33, Colombia (three studies)^25,26,34, UK (three studies)^35–37, Canada (two studies)^38,39, Italy (two studies)^40,41, Australia (one study)⁴², Germany (one study)⁴³, and Sweden (one study)⁴⁴, whereas two studies (8.7%) were conducted in multiple countries; UK, Netherlands, and Spain⁴⁵, and two World Health Organization (WHO) sub-regions, countries in sub-Saharan Africa with very high adult and high child mortality, and countries in South East Asia with high adult and high child mortality⁴⁶ (Figure 2(a)). The study characteristics of 23 CEA are shown in Table 1.

Figure 2. Diagrams demonstrating characteristics of the included studies: (a) Studies conducted to carry out the cost-effectiveness of interventions in a single or multiple countries. (b) Types of decision-analytic models, used in the analyses.

Table 1. Characteristics of the included studies.

Reference	Year	Country being studied	Participant	Intervention	Comparator	Outcome measure	Type of decision-analytic model	Perspective	Time horizon	Is the intervention cost-effective?
Altawalbeh et al.^28	2016	US	Older adults, ≥ 66 years, with asthma	Inhaled corticosteroids plus long acting beta-agonists	Inhaled corticosteroids plus leukotriene receptor antagonists	(1) Costs (2) QALY (3) ICER	Markov model	Healthcare perspective	20 years	No
Bruggenjurgen et al.^43	2010	Germany	Adults with moderate-to-severe asthma	Combination inhaler: (1) Beclomethasone dipropionate plus formoterol fumarate	Separated inhalers: (1) Beclomethasone dipropionate plus formoterol fumarate	(1) Costs (2) ICER	NR	Healthcare perspective	6 months	Yes
Campbell et al.^29	2010	US	Patients with moderate-to-severe asthma	Omalizumab plus standard therapy	Standard therapy	(1) Costs (2) QALY (3) ICER	Markov model	Healthcare perspective	Lifetime	No
Dewilde et al.^44	2006	Sweden	Patients with severe asthma	Omalizumab plus standard therapy	Standard therapy	(1) Costs (2) QALY (3) ICER	Markov model	Societal perspective	Lifetime	No
Doan et al.^38	2011	Canada	Children with asthma	Bronchodilators (metered-dose inhaler)	Bronchodilators (wet nebulization)	(1) Costs (2) ICER	Decision tree model	Hospital perspective (including emergency department and hospital admission costs)	Two days post- emergency department admissions	Yes
Doull et al.^35	2007	UK	Adults and adolescents, and children with asthma	Combination inhaler: Salmeterol xinafoate plus fluticasone propionate	(1) Fluticasone propionate (same dose) (2) Fluticasone propionate (increased dose) (3) Salmeterol xinafoate plus fluticasone propionate (separated inhalers) (4) Budesonide plus formoterol (combination inhaler)	(1) Costs (2) QALY (3) ICER	NR	Societal perspective	1 year	Yes
Faria et al.^36	2014	UK	Adults and adolescents, and children with severe asthma	Omalizumab plus standard therapy	Standard therapy	(1) Costs (2) QALY (3) ICER	Markov model	Healthcare perspective	Lifetime	No
Fuhlbrigge et al.^30	2006	US	Women, 35 years, with asthma	Inhaled corticosteroids	No inhaled corticosteroids	(1) Costs (2) QALY (3) ICER	Markov model	Societal perspective	10 year	Yes
Gerzeli et al.^40	2012	Italy	Adults with moderate-to-severe asthma	Combination inhaler: (1) Beclomethasone plus formoterol	Combination inhaler: (1) Fluticasone propionate plus salmeterol	(1) Costs (2) QALY (3) ICER (4) Time spent in successful control state	Markov model	Healthcare perspective	Lifetime	Yes
Ismaila et al.^39	2014	Canada	Patients with asthma	Combination inhaler: (1) Salmeterol xinafoate plus fluticasone propionate	(1) Fluticasone propionate (same doses: low, medium, high) (2) Fluticasone propionate (increased doses: from low to medium, medium to high)	(1) Costs (2) QALY (3) ICER	NR	Healthcare perspective	1 year	Yes (only when compared to low, medium doses, and the increased doses)
Marchetti et al.^41	2004	Italy	Adults with moderate to severe asthma	1) Beclomethasone (2) Beclomethasone- extrafine (3) Fluticasone propionate (4) Budesonide	(1) Beclomethasone (2) Beclomethasone-extrafine	(1) Costs (2) QALY (3) ICER	Decision tree and Markov models	Healthcare and societal perspectives	2 months	Yes (beclomethasone-extrafine was the most cost-effective therapy)
Norman et al.^37	2013	UK	Adults and adolescents, and children with severe asthma	Omalizumab plus standard therapy	Standard therapy	(1) Costs (2) QALY (3) ICER (4) Number of exacerbations	Markov model	Healthcare perspective	Lifetime	No
Paggiaro et al.^45	2013	UK, Netherlands, and Spain	Patients with asthma	Combination inhaler: (1) Fluticasone propionate plus salmeterol (medium and high doses)	Combination inhaler: (1) Beclomethasone plus formoterol	(1) Costs (2) QALY (3) ICER	Markov model	Healthcare perspective	6 months	No
Rodriguez-Martinez et al.^26	2013	Colombia	Children with asthma	(1) Budesonide (2) Fluticasone propionate (3) Ciclesonide	Beclomethasone dipropionate	(1) Costs (2) QALY (3) ICER	Markov model	Healthcare perspective	1 year	Yes (beclomethasone dipropionate and fluticasone propionate)
Rodriguez-Martinez et al.^34	2015	Colombia	Children with asthma	Inhaled corticosteroids (daily therapy)	Inhaled corticosteroids (intermittent)	(1) Costs (2) QALY (3) ICER	Markov model	Healthcare perspective	1 year	Yes
Rodriguez-Martinez et al.^25	2016	Colombia	Children with asthma	Budesonide (800 µg) once daily	Budesonide (400 µg) twice daily	(1) Costs (2) QALY (3) ICER	Markov model	Healthcare perspective	1 year	Yes
Shih et al.^27	2007	US	Adults and adolescents with mild-to-moderate asthma	Combination inhaler: (1) Fluticasone propionate plus salmeterol	(1) Fluticasone propionate (2) Non-fluticasone propionate inhaled corticosteroids (3) Leukotriene modifiers	(1) Costs (2) ICER	Decision tree model	Healthcare perspective	1 year	Yes
Simonella et al.^42	2006	Australia	Adults with asthma	(1) Current treatment (89% of the patients were in contact with the healthcare system, and 60% of them used inhaled corticosteroids, half of whom used it regularly) (2) Optimal treatment (all patients were in contact with the health system, except those with very mild disease, were deemed to use inhaled corticosteroids regularly and those with moderate or severe disease were deemed to use combined inhaled corticosteroids plus long acting beta-agonists)	No treatment	(1) Costs (2) YLD (3) ICER	NR	Healthcare perspective	1 year	Yes
Stanciole et al.^46	2012	Two WHO sub-regions: (1) AfrE (2) SearD	Patients with asthma	(1) Inhaled corticosteroids (low dose) (2) Inhaled corticosteroids (medium dose) (3) Inhaled corticosteroids plus long acting beta-agonists (low dose) (4) Inhaled corticosteroids plus leukotriene receptor agonists	Placebo	(1) Costs (2) DALY (3) ICER	NR	NR	Lifetime	Yes (low dose inhaled corticosteroids in AfrE and SearD, and only low dose inhaled corticosteroids plus long acting beta-agonists in AfrE)
Whittington et al.^31	2017	US	Adults with severe asthma	Mepolizumab plus standard therapy	Standard therapy	(1) Costs (2) QALY (3) ICER	Markov model	Healthcare perspective	Lifetime	No
Wu et al.^32	2007	US	Adults with severe asthma	Omalizumab plus standard therapy	Standard therapy	(1) Costs (2) QALY (3) ICER	Markov model	Societal perspective	10 years	No
Zafari et al.^24	2014	US	Adults with asthma	Full-adherence scenario: Inhaled corticosteroids or inhale corticosteroids plus long-acting beta-agonist	Status quo scenario (current status of adherence): Inhaled corticosteroids or inhale corticosteroids plus long-acting beta-agonist	(1) Costs (2) QALY (3) ICER (4) Number of exacerbations	Markov model	Societal perspective	10 years	Yes
Zafari et al.^33	2016	US	Patients with moderate-to-severe asthma	Omalizumab	Standard therapy	(1) Costs (2) QALY (3) ICER (4) Number of exacerbations	Markov model	Healthcare perspective	5 years	No

Abbreviations. AfrE, Countries in sub-Saharan Africa with very high adult and high child; mortality DALY, disability-adjusted life years; ICER, incremental cost-effectiveness ratios; QALY, quality-adjusted life years; SearD, Countries in South East Asia with high adult and high child mortality; YLD, years-lived with disability.

Study participants

Among 23 CEA, six studies investigated the CEA in all patients with asthma; moderate-to-severe (two studies [8.7%])^29,33, severe (one study [4.3%])⁴⁴, and at varying levels of asthma severity (three studies [13.0%])^39,45,46. Seven studies investigated CEA in adults with asthma: moderate-to-severe (three studies [13.0%])^40,41,43, severe (two studies [8.7%])^31,32, and at varying levels of asthma severity (two studies [8.7%])^24,42, while four studies (17.4%)^25,26,34,38 only investigated CEA in children at varying levels of asthma severity. Other groups of study participants are shown in Table 1.

Pharmacological interventions

Six studies compared the use of standard therapy plus monoclonal antibodies; omalizumab (five studies [21.7%])^{29,32,36,37,44}, and mepolizumab (one study [4.3%])³¹, to standard therapy alone. One study (4.3%)³³ compared omalizumab to standard therapy. Two studies (8.7%)^40,45 compared between the combination inhalers; beclomethasone dipropionate plus formoterol, and fluticasone propionate plus salmeterol. One study (4.3%)³⁵ compared the combined salmeterol xinafoate plus fluticasone propionate with fluticasone propionate, salmeterol xinafoate plus fluticasone propionate (separated inhalers), and budesonide plus formoterol (combination inhaler), whereas one (4.3%)³⁹ only compared the combined salmeterol xinafoate plus fluticasone propionate with fluticasone propionate. One study (4.3%)²⁷ compared the combined fluticasone propionate plus salmeterol with fluticasone propionate, non-fluticasone propionate inhaled corticosteroids, and leukotriene receptor antagonists (LTRA). One study (4.3%)⁴¹ compared beclomethasone, beclomethasone-extrafine, budesonide, and fluticasone propionate with beclomethasone dipropionate or beclomethasone dipropionate-extrafine. Other pharmacological interventions are shown in Table 1.

Types of decision-analytic model, perspectives, and time horizons

Eighteen studies reported models used in the analyses. The majority of studies (15 studies [83.3%])^{24–26,28–34,36,37,40,44,45} applied the Markov models, while two studies (11.1%)^27,38 used decision tree models. One study (5.6%)⁴¹ applied both Markov and decision tree models (Figure 2(b)). Twenty-two studies reported the perspectives. Most studies (16 studies [72.7%])^{25–29,31,33,34,36,37,39–43,45} used the healthcare perspective, five studies (22.7%)^{24,30,32,35,44} used the societal perspective, and one (4.5%)³⁸ used the hospital perspective, which included all the costs pertaining to the emergency department and hospital admissions. The two most commonly reported time horizons were lifetime (30.4%)^{29,31,36,37,40,44,46} and 1 year (30.4%)^{25–27,34,35,39,42}. Other time horizons are shown in Table 1.

Incorporating adherence in cost-effectiveness analyses

Among 23 CEA, four studies (17.4%) incorporated adherence in the analyses. A study by Shih et al.²⁷ estimated the cost-effectiveness of fluticasone propionate plus salmeterol administered in a single-inhaler, compared with fluticasone propionate, non-fluticasone propionate inhaled corticosteroids, and LTRA. Results showed that combined fluticasone propionate plus salmeterol was the most cost-effective strategy. A study by Rodriguez-Martinez et al.²⁶ compared budesonide, ciclesonide, and fluticasone propionate with beclomethasone dipropionate. The study revealed that beclomethasone dipropionate was associated with the lowest cost, while fluticasone propionate gained the highest number of QALY. Beclomethasone dipropionate was the most cost-effective therapy when the willingness-to-pay threshold was less than £21,129.22/QALY, or otherwise, it was fluticasone propionate. Another study by Rodriguez-Martinez et al.²⁵ compared once-daily budesonide with a twice-daily dose, and demonstrated that a once-daily dose was a dominant strategy. A study by Zafari et al.²⁴ investigated the cost-effectiveness of improving adherence to controller medications, comparing between the hypothetical scenario, in which all patients were fully adherent to the medications (full-adherence), and status quo scenario (the current status of patient adherence). The authors showed that full-adherence was cost-effective.

Adherence data

Adherence data, including the definitions, therapeutic levels, and data sources, were varied from study to study (Table 2). Shih et al.²⁷ used patients’ refill patterns that were adapted from observational studies and claims data^47–51. Rodriguez-Martinez et al.²⁶ applied an assumption of decreasing adherence over time for a twice-daily dose of inhale corticosteroids (ICS), assessed by counting the remaining doses in the inhaler device based on a randomized controlled trial (RCT)⁵². Subsequently, the authors applied a difference in adherence between once-daily and twice-daily administrations which was adapted from a randomized, single-blind, clinical trial⁵³. Likewise, Rodriguez-Martinez et al.²⁵ applied an assumption of decreasing adherence over time for a once-daily dose of ICS, assessed by using a device that recorded the date and time of an inhaler actuation based on an observational study⁵⁴, and used the difference in adherence levels between once-daily and twice-daily administrations, adapted from a randomized, single-blind, clinical trial⁵³. Zafari et al.²⁴ calculated adherence levels using the proportion of days covered (PDC) of patients that were extrapolated from a RCT⁵⁵.

Table 2. Adherence data.

Reference (year)	Adherence data
	Definitions	Levels	Data sources
Shih et al. (2007)^27	Refill patterns among patients with mild-to-moderate persistent asthma	(1) Salmeterol/fluticasone propionate: 34% (2) Fluticasone propionate: 19% (3) Non-fluticasone propionate inhale corticosteroids: 19% (4) Leukotriene modifiers: 38%	Observational studies/patients claim data^47–51
Rodriguez-Martinez et al. (2013)^26	The reported frequency of administration by counting the remaining doses in an inhaler, and applied the assumptions regarding the decrease adherence over time	(1) Budesonide and ciclesonide (once-daily): 88% (3 months), 62% (9 months) and 61% (12 months) (2) Beclomethasone dipropionate and fluticasone propionate (twice-daily): 77% (3 months), 54% (9 months) and 53% (12 months)	Randomized controlled trial^52 and randomized single-blind trial^53
Zafari et al. (2014)^24	Proportion of days covered by patients	(1) Patients who were not using any controller medication: 0% (2) Patients who received low-dose controller therapy: 25% (3) Patients who received medium or high dose of controller therapy: 75%	Randomized controlled trial^55
Rodriguez-Martinez et al. (2016)^25	The reported frequency of administration using a device that recorded the date and time of an inhaler actuation, and applied the assumptions regarding the decrease adherence over time	(1)Budesonide (once-daily): 47.6% (2 months), 39.6% (4 months), 35.9% (6 months), 28.5% (8 months), 29.7% (10 months) and 28.5% (12 months) (2)Budesonide (twice-daily): 58.3% (2 months), 48.5% (4 months), 43.9% (6 months), 34.9% (8 months), 36.4% (10 months) and 34.9% (12 months)	Observational study^54 and randomized single-blind trial^53

Methods of incorporating adherence

Only the method of incorporating adherence by adjusting treatment effectiveness according to adherence levels was demonstrated in this review. Two approaches were used to derive the associations between adherence and effectiveness. The first approach was to apply a mathematical formula that assumed the effectiveness slowly decreased at the first following an exponential curve, as adherence fell below 100%, and increased the rate when it was below 30% following a linear curve. This mathematical formula was derived based on the consultation with an expert panel, involved pulmonologists and allergists as demonstrated below, $$\%\text{Treatment}\hspace{0.17em}\hspace{0.17em}\text{effectiveness}=\%\text{adherence}\hspace{0.17em}\hspace{0.17em}\text{rate} \text{If}\hspace{0.17em}\hspace{0.17em}\text{adherence}\hspace{0.17em}\hspace{0.17em}\text{rate}\le 30\%\%\text{Treatment}\hspace{0.17em}\hspace{0.17em}\text{effectiveness}=1–\hspace{0.17em}\text{exp}\hspace{0.17em}(-5*(\%\text{adherence} \text{rate}–0.2287)) \text{If}\hspace{0.17em}\hspace{0.17em}\text{adherence} \text{rate}>30\%$$

The authors assumed an exponential decline with the constant rate equal to 5, and applied a modifying factor of 0.2287 to confirm the intersection between non-linear and linear functions at an adherence rate of 30%. The adherence-adjusted effectiveness was taken into account as an input parameter, and incorporated in the economic model.

This approach was developed by Shih et al.²⁷, and used in the other two studies by Rodriguez-Martinez et al.^25,26. In the Shih et al.²⁷study, the effectiveness measures included the proportion of patients that were free of symptoms, and from the use of rescue medication, adapted from the RCT. A decision tree model was used to follow the patients at 3-month intervals, throughout the year study period, starting at the initiation of their medications. Patients were assumed to have the opportunity of switching to another therapy, or withdrawing from the study. At the end of the study, they were in one of the following health states: (1) Free of symptoms, (2) Experienced mild symptom, but had not needed rescue medication, (3) Experienced mild symptom that required the use of rescue medication, and (4) Experienced one or more exacerbations. In the Rodriguez-Martinez et al.^25,26studies, the effectiveness was the proportion of patients at risk of exacerbation, adapted from a systematic review and meta-analysis of RCT. The Markov model, which consisted of three health states: (1) No symptoms, (2) Suboptimal control, no exacerbation, and (3) Exacerbation, was applied to the studies, using a cycle length of 1 week over a 12-month period.

The second approach of deriving the associations between adherence and effectiveness was to extrapolate the relationships from previous published studies. This approach was used in a study by Zafari et al.²⁴. First, they calculated adherence levels based on an actual dose of ICS, taken by patients in a RCT⁵⁵, resulting in the PDC values of 0%, 25%, and 75%. Second, an association between each of the 25% decreasing PDC and relative risk (RR) of 1.26 for the rates of exacerbation was adapted from a retrospective cohort study⁵⁶. The authors combined those PDC values with this RR, and then estimated the RR of exacerbation to be ∼2 for the patients with PDC of 25%, and 1.2 for PDC of 75%. For those who did not use any medication, or having PDC of 0%, the authors applied the RR of 1.53, obtained from a systematic review and meta-analysis of RCT which compared the clinical outcomes of using ICS vs no controller medication⁵⁷. Lastly, those RR associated PDC were adjusted based on follow-up periods of the studies, resulting in the RR of 1.40, 1.36, and 1.09 for patients with PDC of 0%, 25%, and 75%, respectively. The adherence-adjusted RR were then applied to the model. In the Zafari et al.²⁴ study, the authors developed the Markov model, which used a cycle length of 1 week throughout the 10-year time horizon, in which the patients transitioned between the following health states: (1) Controlled asthma, (2) Partially controlled asthma, (3) Uncontrolled asthma, (4) Exacerbation, and (5) Death.

Impact of adherence on cost-effectiveness results

Out of four CEA, two studies (50.0%) assessed the impact of adherence on cost-effectiveness results. Shih et al.²⁷ performed one-way sensitivity analysis by varying adherence levels for all the ICS to be 70%, and assumed the associations between adherence and effectiveness of the clinical outcomes; proportion of patients that were free of symptoms, and free of rescue medication use to be fully exponential or linear. Results showed that single-inhaler salmeterol and fluticasone propionate remained cost-effective. Another study by Zafari et al.²⁴ varied the RR of exacerbation associated PDC, and determined that the full-adherence scenario was cost-effective, as long as each of the 25% increases in the PDC reduced the exacerbation rates by at least 1.1-fold.

Quality of studies

According to the CHEC-extended, all studies clearly identified the description of the interventions, study designs, time horizons, perspectives, costs, outcomes, discounting, input parameters’ uncertainty, and study conclusions. Most studies clearly described their research questions (20 studies [87.0%])^{24–28,30–42,44,45}, potential conflicts of interest (19 studies [82.6%])^{25,26,28–33,35–40,42–46}, study populations (18 studies [78.3%])^{24–34,36–39,41,43,44}, ethical issues (16 studies [69.6%])^{24–28,30–34,36,37,40,42,45,46}, and generalizability of the study findings (nine studies [39.1%])^{24–28,32,36,38,40} (Table 3). For the quality of reporting the studies, estimated by CHEERS, all studies provided the explicit statements of background and objectives, comparators, choice of health outcomes, measurement and valuation of the outcomes, estimating resources and costs, currency, price date and conversion, analytical methods, incremental costs and outcomes, study findings, limitations, generalizability, and current knowledge. Most studies reported their settings and locations (22 studies [95.7%])^{24–30,32–46}, study perspective (22 studies [95.7%])^24–45, source of funding (22 studies [95.7%])^{24–40,42–46}, study assumptions (21 studies [91.3%])^{24–41,44–46}, and measurement of effectiveness (20 studies [87.0%])^{24–39,41,42,44,45} (Table 4).

Table 3. Quality assessment of the study methodology.

Reference (year)	Checklist items
	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20
Altawalbeh et al. (2016)^28	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y
Bruggenjurgen et al. (2010)^43	Y	Y	N	Y	N	Y	Y	Y	Y	Y	Y	Y	NA	Y	Y	Y	Y	N	Y	N
Campbell et al. (2010)^29	Y	Y	N	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	N
Dewilde et al. (2006)^44	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	N
Doan et al. (2011)^38	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	NA	Y	Y	Y	Y	Y	Y	N
Doull et al. (2007)^35	N	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	N
Faria et al. (2014)^36	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y
Fuhlbrigge et al. (2006)^30	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	Y
Gerzeli et al. (2012)^40	N	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y
Ismaila et al. (2014)^39	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	N
Marchetti et al. (2004)^41	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	N	N	N
Norman et al. (2013)^37	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	Y
Paggiaro et al. (2013)^45	N	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	Y
Rodriguez-Martinez et al. (2013)^26	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	N	Y
Rodriguez-Martinez et al. (2015)^34	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	Y
Rodriguez-Martinez et al. (2016)^25	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y
Shih et al. (2007)^27	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	NA	Y	Y	Y	Y	Y	N	Y
Simonella et al. (2006)^42	N	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	NA	Y	Y	Y	Y	N	Y	Y
Stanciole et al. (2012)^46	N	Y	N	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	Y
Whittington et al. (2017)^31	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	Y
Wu et al. (2007)^32	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y
Zafari et al. (2014)^24	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	N	Y
Zafari et al. (2016)^33	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	Y

Abbreviations. N, no; NA, not applicable; Y, yes.

Table 4. Quality assessment of the reported studies.

First author (year)	Checklist items
	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20	21	22	23	24
Altawalbeh et al. (2016)^28	Y	Y	Y	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	Y	Y
Bruggenjurgen et al. (2010)^43	Y	N	Y	Y	Y	Y	Y	Y	N	Y	N	Y	Y	Y	N	N	Y	N	Y	Y	N	Y	Y	N
Campbell et al. (2010)^29	N	Y	Y	Y	Y	Y	Y	N	N	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	Y	N	Y	Y	N
Dewilde et al. (2006)^44	N	Y	Y	Y	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	Y	N	Y	Y	N
Doan et al. (2011)^38	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	Y	N	Y	Y	N
Doull et al. (2007)^35	N	N	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	Y	N	Y	N	Y	Y	Y	N
Faria et al. (2014)^36	N	N	Y	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y
Fuhlbrigge et al. (2006)^30	N	N	Y	Y	Y	Y	Y	N	N	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y
Gerzeli et al. (2012)^40	Y	Y	Y	N	Y	Y	Y	N	N	Y	N	Y	Y	Y	Y	Y	Y	N	Y	Y	N	Y	Y	Y
Ismaila et al. (2014)^39	N	Y	Y	Y	Y	Y	Y	N	N	Y	Y	Y	Y	Y	N	Y	Y	N	Y	Y	N	Y	Y	N
Marchetti et al. (2004)^41	N	N	Y	Y	Y	Y	Y	N	N	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	Y	Y	Y	N	N
Norman et al. (2013)^37	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	N
Paggiaro et al. (2013)^45	N	N	Y	N	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	N	N	Y	Y	Y
Rodriguez-Martinez et al. (2013)^26	N	N	Y	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	N	Y	Y	Y	Y
Rodriguez-Martinez et al. (2015)^34	Y	N	Y	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y
Rodriguez-Martinez et al. (2016)^25	Y	N	Y	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	N	Y	Y	Y	Y
Shih et al. (2007)^27	N	Y	Y	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y
Simonella et al. (2006)^42	N	N	Y	N	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	N	N	Y	N	Y	Y	N	Y	Y	Y
Stanciole et al. (2012)^46	N	N	Y	N	Y	N	Y	N	N	Y	N	Y	Y	Y	N	Y	Y	N	Y	Y	Y	Y	Y	Y
Whittington et al. (2017)^31	N	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	Y	Y
Wu et al. (2007)^32	N	N	Y	Y	Y	Y	Y	N	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y
Zafari et al. (2014)^24	N	N	Y	Y	Y	Y	Y	N	N	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y
Zafari et al. (2016)^33	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	N	Y	Y	N	Y	Y	Y

Abbreviations. N, no; Y, yes.

Discussion

Patient non-adherence is associated with asthma-related clinical outcomes and healthcare utilizations; however, one important aspect that still lacks clarity is how to incorporate adherence in the economic analysis. This study investigated the extent of studies that considered adherence as part of the economic analyses, and the methods of incorporating adherence in the CEA of asthma. Our findings show that very low numbers of the CEA incorporated adherence in the analyses. Only the method of incorporating adherence by adjusting treatment effectiveness according to adherence levels was demonstrated in this review. Two approaches were used to derive the associations between adherence and effectiveness. The first approach was to apply a mathematical formula, developed by an expert panel, and the second was to extrapolate the associations from previous published studies.

Incorporating adherence by adjusting treatment effectiveness according to adherence levels was the only method exploited in the economic analysis for asthma, while different methods were observed for other diseases. A literature review by Hiligsmann et al.⁵⁸ revealed that recent CEA of the interventions for osteoporosis^59–61 integrated the probabilities of patients that can be at risk of discontinuation over time. The patients were assumed to have a risk of stopping therapy in each cycle. In addition, offset time of the treatment that was similar to the treatment duration was also applied to the analysis. During this time, the treatment effectiveness is assumed correspondingly. This approach is based on two implicit assumptions. First, patients did not receive any medications after stopping the therapy. Second, the effectiveness of interventions throughout the offset time was estimated by the authors, i.e. the RR of fracture reduction linearly declined to zero by the end of time. Some limitations are recognized by using this method. First, many patients, in fact, can restart their medications any time after they discontinued the therapy. The evidence revealed that one-third of patients restarted their medications within 6 months after discontinuation. Second, the information of treatment effectiveness during the offset time lacked supporting data. It is, therefore, difficult to estimate the effectiveness of interventions in the long-term.

The approach recommended by Hiligsmann et al.⁵⁸ is to apply real-world estimates among the patients who complied with the medications. Using this approach, patients were classified into two groups; (1) Compliant patients (medication possession ratio [MPR] ≥ 80%), and (2) Poorly compliant (MPR < 80%), were assigned the probabilities of being adhered or not based on the real-world adherence data. The associations between adherence levels and the RR of fracture reduction were also assumed accordingly. This approach is in line with what we found in a study by Zafari et al.²⁴. Both studies classified adherence levels into various groups. However, the difference is that, in the Zafari et al.²⁴ study, patients who did not use any medication were applied the RR, obtained from a systematic review and meta-analysis of RCT, while in the Hiligsmann et al.^62,63 studies, the authors used this information that was derived from a real-world database. Even though a systematic review and meta-analysis of RCT demonstrates the highest quality of evidence compared to other types of study design⁶⁴, using adherence data extrapolated from this will not be able to provide real-world data, since the pooled estimates were calculated based on the RCT. Therefore, using adherence data adapted from a real-world database would provide more accurate findings, but the quality of evidence is needed to be confirmed whether it is sufficiently high to synthesize the information that meets healthcare requirements.

Use of the Markov model captures the entire cohort of patient adherence in the economic analysis, but not that of individuals. It is noteworthy to highlight the method of incorporating adherence that is exploited in a study by Slejko et al.⁶⁵, where the authors applied a microsimulation modeling technique to determine the real-world adherence scenario of patients with statin therapy for the primary prevention of cardiovascular (CV) disease. A Markov model was modified to simulate individual adherence to statins, by integrating three additional health states in the existing Markov structure. These health states represented the different levels of adherence that were measured as the PDC; (1) PDC < 20%, (2) 20% ≤ PDC < 80%, and (3) PDC ≥ 80%. They assigned transitional probabilities between the PDC levels, and applied the associations between changes in adherence to statins and the risk of CV events, according to pharmacy claims data particularly reflecting patient adherence history. The microsimulation technique identifies individual patients by tracking their characteristics and disease backgrounds, and then used this record information to adjust the transition probabilities, effectiveness, utility values, and costs, to reflect the patient history over the study period. The use of microsimulation models has the potential to provide more accurate data than the cohort-based models. However, the drawbacks of this are the difficulty in obtaining relevant input parameters, and more detail required for the data set in the modeling approach. In addition, there is greater variance in the results because of the random variation in individual outcomes⁶⁶. With respect to some advantages of the technique, a microsimulation modeling might be considered one of the methods used to integrate adherence for the future economic analyses of asthma.

Out of four CEA that were included in the current systematic review, two performed a one-way sensitivity analysis to assess the impact of varying adherence levels on the cost-effectiveness results. One study showed the findings were robust across all change of adherence, while another demonstrated that the results were cost-effective as long as each of the 25% increases in adherence reduced the exacerbation rates by at least 1.1-fold. This reveals that adherence alters the results when it potentially affects treatment effectiveness of the interventions. To date, knowledge on when adherence should be considered in the economic model is still unclear, since the majority of studies do not take into account adherence in the analyses. Further research is warranted to understand more on the effects of adherence on cost-effectiveness findings.

Another aspect worth mentioning is the application of adherence in coverage and reimbursement decision-making. It is important to consider patient adherence as part of the economic analysis, since varying adherence levels can alter the cost-effectiveness results. Among the CEA included in the current review, various levels of adherence were used across the studies. This is because the adherence data were varied from study to study. In addition, some CEA assumed the constant rates of adherence throughout the study periods, while some modified the rates over time. The findings of this may not be sufficient to draw a valid conclusion on which adherence level should be used as a point of reference to incorporate in the analysis. Therefore, incorporating which level of patient adherence in the CEA depends on the medications and therapeutic areas under specified context of the study. In real-world practice, it may be difficult to confirm that the patients adhere with their prescribed medications at a certain level. An effort in improving adherence should be considered as one of the routine components in diagnostic and management of asthma care. Healthcare professionals and policy-makers may consider to provide additional interventions to improve adherence to ensure such patients achieve their adherence targets satisfactorily. There is a call for future research pertaining to adherence strategies that are feasible for implementation in the clinical settings.

Among the included CEA which incorporated adherence in the models, RCT was used as a source of adherence data in two studies, while observational studies were applied in the other two studies. Although RCT minimizes the potential biases and confounders that may arise from study methodology and the clinical heterogeneities, it restricts the characteristics of participants, types of intervention, and the outcomes of interest. This raises concerns on generalizability of the study findings that may be limited by restrictions. Adherence data would ideally be derived from the observational studies or patient claims data. Many factors, i.e. age, comorbidities, and the numbers of medications, are associated with patient adherence⁶⁷, and, thus, having an impact on economic consequences. However, it is vital to ensure the quality of observational studies to obtain accurate estimates based on real world evidence under specified contexts.

The aim of this work was to conduct a systematic review that complied with a PRISMA guideline, while some limitations are acknowledged in this review. First, the majority of included studies fail to report the structural assumptions and validation methods of their economic models, as well as the values, ranges, and probability distributions among input parameters. Caution should be exercised when interpreting the study findings, because biases arising from these could affect their reported outcomes, and thus limit the generalizability of the results. This suggests that further research with rigorous methodology pertaining to this area is warranted to prevent the potential for bias and imprecise results. Second, some of the non-English articles were identified through our search results. However, only the studies published in English were included in this review, because of the lack of experts in other languages involved in our study. This may be one of the reasons why a limited number of studies incorporating adherence have been identified. In addition, although a number of non-decision model based CEA were identified from the search results, our primary objective was to summarize the methods of incorporating adherence in the economic models, by only using the model-based CEA. We believe that the current review provides the most updated evidence relating to the methods of incorporating adherence in the CEA of asthma, based on justified assumptions and study methodology.

Conclusions

This systematic review gathered all relevant evidence pertaining to the CEA of asthma, and summarized the methods of incorporating adherence in the economic models. A very low number of CEA of asthma incorporated adherence in the analyses. All the CEA adjusted treatment effectiveness according to adherence levels, applied to the models. Our findings will provide healthcare professionals and policy-makers with current evidence of the methods used to incorporate adherence in the CEA of asthma.

Transparency

Declaration of funding

The authors received no specific funding for this work.

Declaration of financial/other interests

BC, NC, and PD declare no potential conflicts of interest with respect to the research, organization, and publication of this work. BC received a scholarship from Naresuan University Graduate School for postgraduate study. JME peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Previous presentations

This research work was selected as the best podium research presentation at the International Society for Pharmacoeconomics and Outcomes Research 8^th Asia-Pacific conference, held in Tokyo, Japan, on September 8–11, 2018.

Acknowledgements

The authors wish to thank Peter Barton from the writing clinic at Naresuan University Language Center for editing this work.