Study design and data sources

This was an observational study (HO-17-17252) of adults in the US with asthma treated with FDC ICS/LABA identified from the Optum Research Database (ORD). The ORD is a geographically diverse, de-identified research database comprising administrative claims, containing both medical and pharmacy information. It is built from a variety of geographic regions and employer groups, and thus it preserves a level of diversity and also represents the overall trend in commercial health plan coverage31. In 2018, ~19% of the US commercially enrolled population was represented in the ORD. The study was approved by the New England Institutional Review Board (IRB# 120190029) on 8 March 2019.

The sample identification period was 12 months (1 March 2018 to 28 February 2019), after which eligible patients were invited to participate via mail following the Dillman method32, with a survey fielding period of 8 weeks. As part of the Institutional Review Board (IRB) submission, Optum requested a waiver of documentation of informed consent. The study packet contained an IRB-approved informed consent statement, which did not require a signature. The consent form asked patients to return the study survey if they elected to participate in the study. Consent was implied when patients returned study materials, and signed consent was not obtained. Patients may have withdrawn consent at any time. Optum did not begin recruitment until IRB approval of all components of the study was obtained. Respondents were sent a $25 post-paid incentive for their study participation. Cross-sectional survey data from respondents was linked with medical and pharmacy claims data for the 12-month period prior to and including the survey completion date (baseline period); where available, data were also linked for the 6-month period following survey completion (follow-up period) (Supplementary Fig. 3).

Sample identification

Eligible patients identified from the ORD were ≥18 years at the time of survey fielding, with ≥1 International Classification of Disease 10th edition Clinical Modification (ICD-10-CM) diagnosis code for asthma in any position during the 12-month sample identification period, 12 months of continuous enrollment including both commercial medical and pharmacy benefits, and ≥2 pharmacy claims for FDC ICS/LABA labeled for use with asthma (≥1 of these claims during the most recent 6 months of the sample identification period). Patients were excluded if they had an ICD-10-CM diagnosis code (i.e., ≥1 medical claim) for chronic obstructive pulmonary disease, cystic fibrosis, or interstitial lung diseases, or claims-based evidence of a diagnosis or treatment for lung cancer prior to or at the time of the survey fielding.

Patients identified by these claims-based sample criteria were grouped into one of three FDC ICS/LABA medication dose cohorts (patients with claims for only low and/or low-medium daily dose treatments [low-dose cohort], medium and/or medium-high daily dose treatments [medium-dose cohort], or only high daily dose treatments [high-dose cohort]; Supplementary Table 3). A random sample of 750 patients from each of these three medication dose cohorts was selected to receive the survey to ensure a range of asthma severity levels were included. Patients were excluded from the analytic sample if they did not return a complete survey (including evaluable ACT and ACQ-6), did not report a healthcare provider diagnosis of asthma, and/or did not report current asthma maintenance treatment. In addition to the claims criteria noted above, patients were also excluded from the final analytic sample if they did not have 12 months of continuous enrollment prior to and including the survey completion date (baseline). Patients who were disenrolled during the follow-up period were excluded from analyses that included the follow-up period.

Measures

To assess the prevalence of suboptimal asthma control in patients treated with FDC ICS/LABA, risk factors associated with suboptimal asthma control, and the impact of suboptimal asthma control on patients and HCRU, the following measures were collected from patient surveys and administrative claims (from the ORD).

Patient survey assessmentsACT-assessed asthma control

The five-item ACT was used to determine the patient’s level of asthma control14. The recall period is 4 weeks. Each of the items of the ACT is scaled on a 1–5 point scale with higher values indicative of better asthma control. The item response values of the ACT are summed to produce a single score that ranges from 5 (poor asthma control) to 25 (complete control of asthma). The minimum clinically important difference (MCID) is 3 points or greater. An ACT score was computed only if the patient provided a response to all five items. No processes are available for computing a summary score when one or more item responses are missing. In addition to the total score, the count and percent for all responses to each of the five items was reported.

ACQ-6 assessed asthma control

The six-item ACQ was used to determine the patient’s level of asthma control15,16,17. The recall period is 1 week. Each of the items of the ACQ is scaled from 0 (totally controlled) to 6 (severely uncontrolled). All items are equally weighted and the ACQ score is the mean of the six items. The MCID is 0.5 or greater. An ACQ score was computed only if the patient provided a response to all six items. No processes are available for computing a summary score when one or more item responses are missing. In addition to the total score, the count and percent for all responses to each of the six items were reported.

Asthma-related quality of life

The mini-AQLQ is a self-administered 15-item questionnaire that covers four domains: symptoms, activity limitation, emotional function, and environmental stimuli33. Each question is answered on a scale from 1–7, with lower scores indicating greater impairment. The recall period is 2 weeks prior to the test. An overall score and scores for each of the four domains are calculated. The overall score is the mean of all items, while the domain scores are the mean of the specific domain items. The MCID is 0.5 or greater.

General health status

The EQ-5D-3L was used to provide a descriptive profile and index value for health status34. The EQ-5D-3L contains two components. The first consists of five questions comprising dimensions of health (mobility, self-care, usual activities, pain/discomfort, and anxiety/depression) across three levels (no problems, some/moderate problems, and extreme problems). The second component consists of a visual analog scale (EQ VAS). The EQ VAS records the respondent’s self-rated health on a vertical scale with values from the worst health state imaginable (0) to the best health state imaginable (100). Levels of problems for each health state, EQ VAS scores and an EQ-5D-3L index (utility) score were calculated.

Full details of patient survey measures are included in Supplementary Table 4.

Patient survey measures

For the study’s primary objective, participants were asked to assess their asthma control during the past 4 weeks using ACT14. ACT scores were used to categorize patients’ asthma as poorly controlled (ACT score <16), somewhat controlled (ACT score 16–19), or controlled (ACT score ≥20). In some analyses, ACT scores for the poorly and somewhat controlled (score of <16 and 16–19, respectively) groups were combined in a “suboptimal control” group, creating a binary ACT variable, with controlled (score ≥20) patients acting as the reference group. Participants also completed the ACQ-615,16,17 based on their symptoms in the previous week, with total scores of ≥1.50 indicating uncontrolled asthma, >0.75–<1.50 indicating partially controlled asthma, and ≤0.75 indicating controlled asthma17. The ACQ-6 was used rather than the ACQ-7 in this study due to the absence of lung function in the ACQ-6, which is also absent from the ACT. Participants were also asked to complete the mini-Asthma Quality of Life Questionnaire (mini-AQLQ, recall period: 2 weeks)33 and the EuroQol Group 5 Dimension Health Status Measure, 3-level (EQ-5D-3L, recall period: today)34. The survey also collected data on sociodemographic characteristics, smoking behavior, current weight and height, and asthma treatment history.

Claims measures

Claims data were used to assess geographic region, age, and sex (if missing from the patient survey), Charlson Comorbidity Index score35 (CCI; an assessment tool used to predict mortality by classifying or weighting comorbidities), asthma-related comorbidities, asthma treatment, MPR, the proportion of days covered (PDC), asthma exacerbations (hospitalization-, emergency department-, or corticosteroid-defined exacerbation, based on medical and pharmacy claims and asthma diagnosis code), HCRU, and costs. An updated CCI of 12 comorbidities was used35. Asthma-related comorbidities were defined as ≥1 claim with a diagnosis code for angina, cataract, myocardial infarction, pneumonia, upper respiratory tract infection, allergy, allergy/upper respiratory tract infection combination, type-2 diabetes mellitus, or a diagnosis or treatment code for depression or anxiety. ICS/LABA dose category was assigned according to the latest claim for an ICS/LABA medication prior to completing the survey based on the average daily dose and was medication-specific (Supplementary Table 5). MPR was calculated by summing the number of days supplied for an ICS/LABA for all but the last fill in the observation period and dividing by the number of days between the first and last refill. PDC was calculated by dividing the number of days on which medication was available (based on filled prescriptions) by the number of days during the observation period. Healthcare costs were computed as the combined health plan and patient-paid amounts.

Sample size and statistical analyses

The sample size was estimated based on the value and desired precision of the proportions required for the study’s primary outcome measure (i.e., the proportion of patients “poorly”, “somewhat”, and “controlled” as assessed by the ACT). Using normative data for asthma control per the ACT in a US asthma patient population, it was estimated that approximately 60% of US patients with asthma are “controlled”36. Assuming a proportion of 50% of patients in the controlled and uncontrolled groups, a final target sample size of n = 385 assured a 95% confidence interval (CI) of having a precision of ±0.05 or better for all proportions observed. Based on an estimated 20% survey response rate (calculated as per American Association for Public Opinion Research (AAPOR) formulas37) and 15% attrition in the 18-month claims observation period, a sampling frame of 2250 patients (with 750 each in low-, medium-, and high-ICS-dose strata) was estimated to reach the minimum target final evaluable sample size.

Two analytic cohorts were created to address study objectives. The overall study cohort comprised participants with survey data and claims data for the 12-month baseline period. The follow-up study cohort included only the subset of patients who were also continuously enrolled for the 6-month period following the survey.

To identify risk factors associated with suboptimal control, two multivariate analyses were conducted based on control measured by the ACT and by the ACQ-6 in the overall study cohort. Factors associated with suboptimal asthma control as measured by the ACT were determined by a generalized linear model (binomial distribution, logit link), with suboptimal asthma control modeled as the binary dependent variable (poorly and somewhat controlled vs controlled). A wide range of covariates were considered for inclusion in the model, including age (continuous), sex (male or female), race (White, Black/African American or Other), marital status (married or living with a partner, or single, never married, separated, divorced and/or widowed), the highest level of education completed (2-year college degree or higher [yes/no]), place of residence (urban/city, suburban or rural), number of years with asthma (continuous), smoking status (non-smoker or ever-smoker), BMI category (calculated from a patient report of height and weight; patients were assigned to one four categories: normal [18.5–<25 kg/m2], overweight [25–<30 kg/m2], obese [≥30 kg/m2]), CCI score (0, 1, or ≥2), count of unique asthma-related comorbid conditions (continuous), PDC for ICS-containing medications (≥0.80 or <0.80), SABA utilization (>6 or ≤6 canister fills), and asthma-related outpatient visits (yes/no). Race was self-reported by respondents and was categorized as American Indian or Alaskan Native, Asian, Black, or African American, Native Hawaiian or Pacific Islander, White, and/or Other race. Ethnicity was self-reported by respondents and was categorized as Hispanic/Latino (yes vs no). For missing values, data were imputed using the most frequent (prevalent) values in order to retain the majority of patients in the regression analysis. Missing values for race were categorized as “White” and for urban/rural residence as “suburban”; missing values for other variables were not imputed. Factors associated with suboptimal asthma control as measured by the ACQ-6 were assessed by a proportional odds model (cumulative logit model), using the same dependent variable and list of covariates. In both ACT and ACQ-6 multivariate models, variables that demonstrated at least marginal statistical significance (p < 0.1) in univariate analyses were included as covariates in the final multivariate model, along with relevant clinical variables (age, sex, race, marital status, and level of education, CCI score, count of unique asthma-related comorbid conditions, BMI, and smoking status) regardless of their statistical significance.

To explore the impact of asthma control on future risk of asthma outcomes, a third multivariate model used asthma control as the independent variable to describe the future risk of “control” measured using exacerbations and high SABA use (a proxy for poor control) in the follow-up period, using data from the follow-up cohort. Factors associated with the composite measure of any asthma exacerbation or high SABA use (>4 SABA canister fills per year, i.e., >2 SABA canister fills during the 6-month follow-up period) in the follow-up period were determined using a generalized linear model (binomial distribution, logit link), developed in a stepwise manner adjusting for four sets of covariates of poorly/somewhat controlled asthma: ACT suboptimal control (yes/no), demographics (age [continuous], the highest level of education [2-year college degree or higher (yes/no)]), comorbidities (non-smoker or ever-smoker, BMI category [<30 or ≥30 kg/m2]), and ICS/LABA dose category (low, medium, or high).

Analyses were conducted using the SAS version 9.4 statistical software package (SAS Institute Inc., Cary, NC, USA). Study outcomes were analyzed descriptively unless otherwise specified.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.