FormalPara Key Summary Points

Why carry out this study?

Inadequate medication adherence (the extent to which a patient acts in accordance with the interval and dose indicated) and persistence (time from therapy initiation to discontinuation) are widespread issues in asthma treatment that, if properly addressed, could lead to improved symptom control and reduced exacerbations.

Here we compare treatment persistence and adherence among patients using different inhaled corticosteroid/long-acting beta2-agonist (ICS/LABA) combinations in order to evaluate the real-world use of ICS/LABA therapy.

What was learned from the study?

The likelihood of treatment discontinuation in patients treated with fluticasone furoate/vilanterol (FF/VI) within the 12 months following initiation was between 31% and 35% lower than its comparators.

The odds of achieving ≥ 50% and ≥ 80% in the proportion of days covered (a measure of adherence) were greater for FF/VI than its comparators.

Our findings suggest that the FF/VI is associated with a significantly lower likelihood of discontinuation and a higher adherence to treatment versus other ICS/LABA comparators; improvements could be due to factors such as the simplified (once-a-day) dosing regimen associated with FF/VI and improved asthma control.

Introduction

Approximately 5.4 million people receive treatment for asthma in the United Kingdom (UK) [1]. Inhaled corticosteroid/long-acting beta2-agonist (ICS/LABA) combination therapies are recommended for patients whose asthma remains uncontrolled with an ICS alone [2]. ICS/LABA therapies have demonstrated effective control of asthma symptoms, including a reduction in severe exacerbations [3]. Various ICS/LABA combinations are available in the UK for the treatment of asthma, including fluticasone furoate/vilanterol (FF/VI), budesonide/formoterol (BUD/FM) and beclometasone dipropionate/formoterol (BDP/FM) [4,5,6]. Current literature suggests that patients treated with FF/VI may exhibit improved control of asthma symptoms compared with other ICS/LABA combinations [7, 8]. The Salford Lung Study on asthma, an open-label, two-arm, effectiveness randomised controlled trial (RCT), assessed the percentage of responders (patients who achieved an Asthma Control Test score of ≥ 20, or an ≥ 3-point increase from baseline in Asthma Control Test score at 24 weeks), and found that a higher proportion of patients with uncontrolled asthma achieved better control of their asthma when they were initiated with FF/VI (71%) versus patients continuing usual care (56%) [7, 8]. In another study, patients treated with FF/VI were shown to have a lower risk of treatment discontinuation versus BUD/FM or fluticasone propionate/salmeterol (27% and 30% lower, respectively) [9].

Inadequate medication adherence (the extent to which a patient acts in accordance with the interval and dose indicated) and persistence (time from therapy initiation to discontinuation) [10] is a widespread issue, with an estimated 50% of prescriptions filled for chronic diseases not taken as indicated [11]. Adherence to asthma medication is known to reduce over time, data for which may not be captured in a relatively short-term RCT [12]. A systematic literature review found that good adherence was associated with a lower risk of severe asthma exacerbations, suggesting that improved adherence to ICS/LABA therapy may improve asthma symptom control [13].

Particular features of a medicine, such as daily dosing schedule (once-daily versus twice-daily administration), duration of action, and ease-of-use of the device may each impact treatment adherence and persistence among patients using asthma maintenance treatments [14,15,16]. This study described the real-world use of ICS/LABAs by comparing treatment persistence and adherence among patients initiating FF/VI or either BDP/FM or BUD/FM (comparator ICS/LABA combinations, both commonly used in the UK [may be viewed as substitutable]).

Methods

Study Population

The study population was derived from the IQVIA Medical Research Database (IMRD, including patient data gathered via The Health Improvement Network [THIN; the anonymised UK patient data collection scheme from general practitioners], a Cegedim database [17]). Key inclusion criteria were: ICS/LABA-treated patients for whom the first recorded prescription of FF/VI, BUD/FM or BDP/FM occurred during the study period, with a confirmed asthma diagnosis on or before the index date and at least one subsequent prescription of the same FF/VI, BUD/FM or BDP/FM within 180 days post-index date. See Supplementary Methods for exclusion criteria.

For patients identified as initiating BUD/FM treatment, only those identified in the THIN database as prescribed medication at strengths of 200/6, 100/6 and 400/12 µg (flexible or fixed dose) were included in our analysis, with the aim of generating a cohort representative of SYMBICORT (AstraZeneca) for comparison with FF/VI. Other identifiable BUD/FM medications with differing doses were excluded (a step necessitated as the THIN database does not capture generic/branding medication information) but non-distinguishable generics were included in the specified cohort. Please note that the term “BUD/FM”, used hereafter, therefore refers to the specified cohort only. Only one brand of each of FF/VI (RELVAR, GlaxoSmithKline plc.) and BDP/FM (FOSTAIR, Chiesi Ltd) are available in the UK, and so this consideration did not apply to these medications.

Study Design

The primary objective of this study was to compare treatment persistence (time to discontinuation) for patients initiating FF/VI versus patients initiating a comparator ICS/LABA treatment (BUD/FM or BDP/FM, separately). Secondary objectives were to compare adherence to treatment (mean proportion of days covered [PDC]), assess the proportions of patients with ≥ 50% and ≥ 80% PDC, and describe the annualised number of short-acting bronchodilator (SABD) prescriptions for rescue use per patient.

This retrospective new-user active comparator database study was conducted in the IMRD primary care database (Supplementary Fig. S1). The study period was from January 1, 2013 to January 17, 2018, inclusive. The index date was defined as the date of the first recorded prescription of FF/VI or the comparator ICS/LABA (BUD/FM or BDP/FM) within the study time period. Patients were required to have at least 12 months of data recorded prior to the index date to determine the prescription history (no previous ICS/LABA prescription) and variables to be considered for the propensity score (PS) model. A second prescription of the same index treatment was required within 180 days to better identify patients using ICS/LABAs for long-term maintenance versus short-term treatment. This was termed the index second prescription date.

Patient follow-up started on the index date (date of ICS/LABA initiation) for all analyses, except the discontinuation analyses, where follow-up started on the index second prescription date to avoid potential bias relating to any differences in the time between the index and second prescription when patients were not able to experience the outcome of interest [18]. See Supplementary Methods for additional follow-up period information for the discontinuation, adherence and SABD rescue use analyses.

For the regression analysis, the risk of discontinuation and the odds of adhering to medication 12 months after the index date were assessed.

A patient was considered to have switched medication from the index treatment if prescription of either of the other two treatments of interest (FF/VI, BUD/FM or BDP/FM) occurred after the index second prescription date. This definition of switching was considered up until 60 days from the date at which the most recent prescription of the original index therapy was due to end.

Persistence was defined as the time to discontinuation (in days) of the index treatment, this being the event of interest. Discontinuation was defined as a gap of 60 days after ICS/LABA prescription or switch to a different ICS/LABA therapy within 60 days after the end of the previous prescription. Adherence was measured as PDC (number of days in period covered by medication divided by the number of days in period, with 100% indicating full or complete adherence). Additionally, two binary outcomes—patients ≥ 50% adherent and patients ≥ 80% adherent—were created and analysed. Taking ≥ 80% of medication as prescribed has been generally considered as acceptable adherence for clinical outcomes, based on findings from a seminal hypertension study [19]; however, as it is now recognised that this cut-off value may vary for different diseases and classes of medication [20], we also analysed a lower adherence cut-point, chosen to allow comparison with previous asthma studies [21,22,23,24,25]. The protocol was approved by local ethics committees. Appropriate consent/approval was obtained to use the dataset; as this study utilized de-identified retrospective claims data compliant with the Health Insurance Portability and Accountability Act (HIPAA), no institutional review board (IRB) approval was required.

Methods

Prior to comparative analyses, PS matching (PSM) methodology was used to create the following cohorts: patients receiving FF/VI were matched at a ratio of 1 to ≤ 4 with either patients prescribed BUD/FM or BDP/FM, separately. Greedy nearest-neighbour matching without replacement was used with a caliper of 0.01 standard deviation of the logit. Standardised differences and variance ratios in patient characteristics were used to assess whether appropriate balance between matched groups had been achieved. An absolute standardised difference ≤ 0.1 (10%) and a variance ratio of 0–2.0, inclusive, was considered to indicate a close match (good covariate balance) between groups [26]. A comparison of demographic characteristics was conducted between the matched and unmatched cohorts to ensure an unbiased matched population. Prior to persistence and adherence analyses, rules were set regarding multiple prescriptions on the same day, the second prescription and prescriptions lasting more than 365 days. See the Supplementary Methods for additional detail on data cleaning decisions and the considered covariates and confounders.

Analysis

Descriptive statistics were performed for all baseline variables for unmatched and PS-matched cohorts (Supplementary Methods). Kaplan–Meier (KM) curves were used to estimate the time to treatment discontinuation (including switch). Log-rank tests tested observed differences in KM curves between study groups, and Cox proportional-hazard models assessed the association between study groups and persistence at 1 year. Logistic regression analysis was performed to estimate the adherence to treatment (as measured by PDC) over 12 months post-index, whereby censoring criteria included death, exit from the THIN database, end of study time-period and switch to another therapy. In parallel, a chi-square test was performed to compare the proportion of patients persistent 1 year after index. SABD use was analysed descriptively.

Sensitivity analyses included (1) re-analyses of persistence, with discontinuation defined as a gap of 30 and 90 days; and (2) re-analysis of PDC with censoring criteria, including discontinuation (defined as a gap of 60 days).

Results

PSM

In the IMRD, eligible patients with asthma were identified as new users (ICS/LABA naïve in the past) of FF/VI, BUD/FM or BDP/FM, and PSM was applied as described to generate matched cohorts for analysis. In total, 966 patients initiated treatment with FF/VI, 5931 initiated treatment with BUD/FM, and 9607 initiated treatment with BDP/FM during the study (Fig. 1). A total of 945 FF/VI patients were PS-matched to 3272 BUD/FM new users (Table 1, top), and 902 FF/VI patients were PS-matched to 3465 BDP/FM new users (Table 1, bottom). In both comparisons, PSM resulted in an absolute standardised difference of less than 0.1 for all variables, indicating well-balanced cohorts based on the covariates used for the PSM (Table 1), which included age at baseline, gender, comorbidities, and the number of GP visits or hospitalisations, ICS or SABD prescriptions in the prior year.

Fig. 1
figure 1

Patient flow and eligibility diagram (GlaxoSmithKline plc. study 209967). BDP/FM beclometasone dipropionate/formoterol, BUD/FM budesonide/formoterol, COPD chronic obstructive pulmonary disease, FF/VI fluticasone furoate/vilanterol, ICS inhaled corticosteroid, IMRD IQVIA medical research database, LABA long-acting beta2-agonist, PS propensity score, PSM propensity score matching

Full size image
Table 1 Comparison of demographic characteristics of FF/VI versus BUD/FM or BDP/FM cohorts
Full size table

Further data cleaning to exclude potential confounders (as detailed in the Supplementary Methods) excluded 8 FF/VI and 40 BUD/FM patients from the FF/VI versus BUD/FM PS-matched cohort, and 8 FF/VI and 32 BDP/FM patients from the FF/VI versus BDP/FM PS-matched cohort (Fig. 1). The number of ICS prescriptions pre-index for FF/VI versus BUD/FM and BDP/FM was found to be similar across treatment groups (Supplementary Tables S1 and S2).

FF/VI Versus BUD/FM

Unadjusted median time to discontinuation of treatment was 99 days for FF/VI, with a median follow-up time of 393 days, and 116.5 days for BUD/FM with 817 days’ median follow-up time. FF/VI had notably less follow-up time available than comparators, leading to a lower unadjusted median time to discontinuation. The variation in follow-up times was driven by the different launch dates for these medications and their consequent availability—as FF/VI has been available for a shorter period of time, and numbers for FF/VI are skewed towards the end of the inclusion period (as uptake increased over time), the average follow-up time is shorter. KM-adjusted (for variable follow-up time) median time to discontinuation was not available for FF/VI (as the median time was not reached), but was 427 days for BUD/FM (Table 2), with a significant difference in discontinuation over 12 months (log-rank test: p < 0.0001) (Fig. 2). At 12 months, after adjusting for variable follow-up time, KM analysis showed that the proportion of patients persistent on FF/VI was 69% (324/937; 95% confidence interval [CI] 0.65–0.72) versus 53% (1246/3232; 95% CI 0.51–0.55) of patients prescribed BUD/FM (Fig. 2; Table 2). The likelihood of discontinuing treatment within 12 months after initiation was 35% lower for FF/VI than BUD/FM [Cox proportional-hazards model, index year-adjusted, hazard ratio (HR) 0.65; 95% CI 0.56–0.75; p < 0.001] (Table 3).

Table 2 Overall time to discontinuation for FF/VI versus BUD/FM (adjusted for variable follow-up time)
Full size table
Fig. 2
figure 2

Figure adapted from Svedsater H, Parimi M, Ann Q, et al. P230 A retrospective database study of persistence and adherence in patients with asthma in the UK (UK-THIN): fluticasone furoate/vilanterol (FF/VI) versus budesonide/formoterol (BUD/FM) Thorax 2019;74:A215

Kaplan–Meier curve for persistence to FF/VI and BUD/FM (time to discontinuation). BUD/FM budesonide/formoterol, FF/VI fluticasone furoate/vilanterol. *p value is for analysis at 12 months.

Full size image
Table 3 Risk of discontinuation for FF/VI versus BUD/FM within 12 months after initiation
Full size table

Mean adherence, measured by PDC, was higher for FF/VI than for BUD/FM (77.7 vs 72.4; p < 0.0001) (Table S3). Median (interquartile range) PDC was also higher for FF/VI than for the BUD/FM cohort: 88.2 (61.4–100.0) versus 77.7 (50.7–100.0). After adjusting for the year of index ICS/LABA prescription, the odds of achieving ≥ 50% PDC were higher for patients initiating treatment with FF/VI compared with those on BUD/FM [779/936 (83.2%) vs 2447/3232 (75.7%); odds ratio (OR) 1.35; 95% CI 1.09–1.67; p = 0.006]. Similarly, the odds of achieving ≥ 80% PDC were higher for patients initiating FF/VI compared with those on BUD/FM [544/936 (58.1%) vs 1562/3232 (48.3%); OR 1.28; 95% CI 1.08–1.52; p = 0.004]. The annualised rate of SABD prescriptions per patient-year was 4.7 for FF/VI and 4.2 for BUD/FM, respectively.

After adjusting for the year of treatment initiation, a Cox regression sensitivity analysis showed that patients who initiated treatment with FF/VI had a 27% lower risk of discontinuing compared with BUD/FM when the definition for discontinuation (or switch) was redefined as a gap of 30 days (HR 0.73; 95% CI 0.64–0.82; p < 0.001) and a 38% lower risk of discontinuing compared with BUD/FM (HR 0.62; 95% CI 0.52–0.74; p < 0.001) when the definition was set to 90 days.

The logistic regression sensitivity analysis of PDC was conducted whereby discontinuation was included as a censoring criterion. After adjusting for year of treatment initiation and for baseline variables in the PS model, FF/VI had a similar adherence pattern compared with BUD/FM (≥ 50% PDC, OR 1.03; 95% CI 0.72–1.47; p = 0.880; ≥ 80% PDC, OR 1.03; 95% CI 0.86–1.23; p = 0.744).

FF/VI Versus BDP/FM

For FF/VI, the median time to discontinuation of treatment was 97 days with a median follow-up time of 402 days, and 91 days for BDP/FM with a median follow-up time of 539 days (not adjusted for variable follow-up). As was the case in the previous comparison, the variation in follow-up times was caused by the different launch dates for these medications and their consequent availability. When adjusted for variable follow-up time by KM analysis, the median time to discontinuation was not reached for FF/VI, but was 537 days for BDP/FM (Table 4), with a significant difference in discontinuation over 12 months (log-rank test: p < 0.0001) (Fig. 3). After adjusting for variable follow-up, KM analysis demonstrated that the proportion of patients persistent on FF/VI at 12 months was 69% (321/894; 95% CI 0.66–0.73) versus 57% (1172/3433; 95% CI 0.55–0.58) of BDP/FM patients (Fig. 3; Table 4). The likelihood of discontinuing treatment within 12 months after initiation was 31% lower for FF/VI than BDP/FM (Cox proportional-hazards model, index year-adjusted, HR 0.69; 95% CI 0.60–0.80; p < 0.001) (Table 5).

Table 4 Overall time to discontinuation for FF/VI versus BDP/FM (adjusted for variable follow-up time)
Full size table
Fig. 3
figure 3

Figure adapted from Svedsater H, Parimi M, Ann Q, et al. P229 A retrospective database study of persistence and adherence in patients with asthma in the UK (UK-THIN): fluticasone furoate/vilanterol (FF/VI) versus beclometasone dipropionate/formoterol (BDP/FM) Thorax 2019;74:A214

Kaplan–Meier curve for persistence to FF/VI and BDP/FM (time to discontinuation). BDP/FM beclometasone dipropionate/formoterol, FF/VI fluticasone furoate/vilanterol. *p value is for analysis at 12 months.

Full size image
Table 5 Risk of discontinuation for FF/VI versus BDP/FM within 12 months after initiation
Full size table

Mean adherence, measured by PDC, was higher for FF/VI than for BDP/FM (78.2 vs 71.0; p < 0.0001) (Table S4). Median (interquartile range) PDC was also higher for FF/VI than for BDP/FM: 89.2 (61.6–100.0) versus 75.9 (50.5–98.0). After adjusting for year of index ICS/LABA prescription, the odds of achieving ≥ 50% PDC was greater for FF/VI than for BDP/FM [747/893 (83.7%) vs 2600/3433 (75.7%); OR 1.50; 95% CI 1.23–1.83; p < 0.001]. Similarly, the odds of achieving ≥ 80% PDC were greater for FF/VI than for BDP/FM [526/893 (58.9%) vs 1571/3433 (45.8%); OR 1.57; 95% CI 1.35–1.83; p < 0.001; index year-adjusted]. The annualised rate of SABD prescriptions per patient-year was 4.6 for FF/VI and 4.7 for BDP/FM, respectively.

In the Cox regression sensitivity analysis, after adjustment for year of initiation within the regression model, patients who initiated treatment with FF/VI showed a 32% lower risk of discontinuing compared with BUD/FM when the definition for discontinuation (or switch) was set as a gap of 30 days (HR 0.68; 95% CI 0.60–0.76; p < 0.001). For a 90-day definition, patients who initiated FF/VI showed a 27% lower risk of discontinuing (HR 0.73; 95% CI 0.62–0.87; p < 0.001).

The logistic regression sensitivity analysis of PDC was conducted whereby discontinuation was included as a censoring criterion. After adjusting for baseline variables in the PS model and for year of initiation within the regression model, FF/VI had an improved adherence pattern compared with BDP/FM (≥ 50% PDC, OR 1.58; 95% CI 1.15–2.19; p = 0.005 and ≥ 80% PDC, OR 1.38; 95% CI 1.17–1.62; p < 0.001).

Discussion

This study indicated that patients who initiated FF/VI were less likely to discontinue treatment (defined as a ≥ 60-day gap in prescription or switch to another index therapy in this period) and showed greater adherence to treatment (PDC measurement) compared with patients who initiated BUD/FM or BDP/FM. These results were supported by discontinuation sensitivity analyses (defined as a 30- or 90-day gap). Mean PDC was higher for FF/VI than for both BUD/FM and BDP/FM. Similarly, the odds of achieving ≥ 50% and ≥ 80% PDC were greater for FF/VI than for both BUD/FM and BDP/FM. The favourable persistence and adherence observed with FF/VI could be due to a number of factors. Firstly, FF/VI is associated with improved asthma control versus other ICS/LABAs and may, therefore, lead to increased patient satisfaction [7]. Secondly, FF/VI is administered once daily via the ELLIPTA inhaler (GlaxoSmithKline plc. group of companies), which has been shown to result in significantly fewer critical errors [14] and is generally preferred to comparator devices [15]. In this study, both comparator ICS/LABAs contained a mix of flexible maintenance and reliever therapy [MART] and fixed dosing regimens. MART may lead to increased adherence (and therefore inflated PDC values) [27] if taken as maintenance plus reliever. Furthermore, ICS/FM combinations are often used only as relief medications in asthma [28], which may bring adherence down; hence, findings in this study may represent a conservative estimate of the PDC difference for FF/VI versus BUD/FM or BDP/FM. It is important that we do not know exactly how many patients are prescribed each regimen and we do not know for sure if and to what extent these effects influence adherence.

Strengths of this study include the IMRD database, which is representative of patients with asthma in the UK whose disease is predominately managed by their general practitioner [29]; therefore, this study describes real-world findings generalizable to a large population of patients with asthma in the UK. Additionally, the study period provides data relevant to the current asthma treatment landscape in the UK. The use of a new-user active comparator design compared initiators of a therapeutically equivalent drug class and minimised biases due to the inclusion of patients tolerant to prevalent treatments [30]. The large number of patients identified in IMRD (N = 16,504) enabled a matching between FF/VI and comparator ICS/LABAs at a ratio of 1 to > 1, improving the statistical power to detect differences between the matched groups wherever such a difference existed.

Confounding by indication is an important consideration in comparative effectiveness studies resulting from differences in reasons for prescription that may lead to systematic bias in favour of or against FF/VI. PSM seeks to account for potential confounding by indication and the bias caused by unbalanced comparison cohorts. The observed low standardised differences across cohorts support the balance of measured covariates of interest and confirms using PSM for cohort matching and control of potential confounders. Additionally, real-world data from routine clinical practice may provide a more realistic estimate of adherence (based on frequency of prescription) than RCT data, as the complex behavioural factors involved in patient decision-making are less altered, and longer time frames can capture changes in adherence over time [12].

As well as general limitations in conducting retrospective cohort studies using secondary data collected for patient management purposes, the specific limitations of this study include the lack of spirometry and asthma control measure data (i.e. specific advice given to patients, for example smoking cessation and correct inhaler training) were not available in the structured data used in this analysis. As the data recorded in IMRD is restricted to those input by and for the use of the general practitioner, bias may be introduced when these data are used for research purposes. The utility of findings from research leveraging electronic medical record data is limited by the completeness of these data, which may have been influenced by the longer availability of BUD/FM and BDP/FM (available in the UK from June, 2007, and January, 2008, respectively) [31, 32], compared with FF/VI (available in the UK from January, 2014) [33]. Additionally, patient behaviour is seldom directly captured by electronic databases, thus compliance cannot be directly assessed. Time to discontinuation and PDC are proxies to estimate patient persistence and adherence, respectively, with PDC a commonly used method endorsed by the Pharmacy Quality Alliance [34]. Exacerbations in the pre-index period were also identified using a proxy, therefore potentially resulting in incomplete capture of data. No matching method can remove the potential for residual confounding due to unmeasured variables. Several severity proxies were used for PSM, as variables directly related to patients’ severity of asthma were not recorded in IMRD (e.g. daily dose, exacerbations and healthcare resource use).

Our results are broadly comparable with those of three database studies performed in the USA and Japan [21, 35, 36]. In their retrospective cohort study, Stanford et al. found that patients initiating once-daily FF/VI were more likely to be adherent (measured by PDC) and less likely to discontinue treatment than patients initiating twice-daily BUD/FM; however, persistence was low for both treatments [35]. Averell et al. similarly found that adherence (measured by PDC) was higher and discontinuation less likely for FF/VI compared with BUD/FM and fluticasone propionate/salmeterol (FP/SAL) [9, 21]. Moreover, a Japanese cohort study showed that adherence and PDC were significantly higher with FF/VI than with twice-daily FP/SAL treatment [36]. The results of these previous studies add further weight to the finding that FF/VI is associated with a significantly lower likelihood of discontinuation and a higher adherence to treatment versus ICS/LABA comparators. These studies shared some methodological aspects (non-interventional, retrospective cohort studies) and mostly utilised the IMRD for their data; however, Atsuta et al. used data from the Japan Medical Data Center Claims Database [36].

The driving forces behind changes in adherence status could warrant further study. Of particular interest would be whether these changes can be attributed to patient behaviour or to the medication itself. As many of the patients studied were using multiple drugs for multiple conditions, a future study, which is adequately designed for such an evaluation, could compare patient adherence to these other medications with patient adherence to ICS/LABA. This could provide valuable data for evaluating whether patient adherence status is consistent across medication types.

Conclusion

Patients in the UK with asthma initiating FF/VI in our study were more likely to exhibit greater persistence and adherence to treatment than those initiating BUD/FM (200/6, 100/6 and 400/12 µg) or BDP/FM during the same study period.