STRENGTHS AND LIMITATIONS OF THIS STUDY

Long follow-up time using electronic health record (EHR) analysis of controller medications prescribed from 12 months before enrolment into the Person Empowered Asthma Relief (PREPARE) study through 15 months after enrolment.

Even if data were missing, we were able to use all available data due to the type of analysis we used—linear mixed effect model.

EHR data have inherent missingness and can have gaps due to infrequent visits and prescribing activities.

Some individuals were on dual inhaled corticosteroid (ICS) inhalers not counting the study-prescribed beclomethasone, and these individuals were assigned based on the highest potency of any of the non-study-prescribed ICS inhalers.

In addition, not every site within PREPARE provided data, but we had data from the majority of sites.

Introduction

25 million adults have asthma in the USA.1 Asthma exacerbations cause the largest number of lost days (one-third of all days) from school or work for children and young adults.2 3 Despite new medications,4 5 new drug regimens6–11 and the ongoing evolution of treatment guidelines,12–14 the number of people with asthma exacerbations in the previous year has decreased only slightly over the past 20 years, from 51.6% to 46.0%.15 16 Asthma is also a disease with a high degree of disparities in outcomes. Black individuals have exacerbation and death rates that are 2–2.5 times higher than white individuals and Asian individuals,17 while Hispanic individuals, particularly Caribbean Hispanic individuals,18 19 have 2 times the rate of exacerbations and 1.5 times the death rate.20 21 Thus, there is an ongoing need to expand and improve treatment approaches for individuals with asthma.

The paradigm concerning the use of inhaled corticosteroids (ICS) only as a controller, that is, as a once or twice a day medication, has been evolving. The use of ICS-formoterol as both controller and rescue therapy (single maintenance and reliever therapy) is endorsed by multiple guideline groups across the world, including the US, GINA and UK guidelines.12 14 22 Previous randomised controlled trials have also demonstrated that as-needed, stand-alone, ICS use in conjunction with short-acting beta-agonist (SABA) for acute asthma symptom relief can improve asthma outcomes.7 23–29 The Person Empowered Asthma Relief (PREPARE) trial used a stand-alone ICS combined with participants’ usual controller and rescue therapy in an approach called Patient Activated Reliever-Triggered ICS (PARTICS).30

The PARTICS intervention in PREPARE decreased asthma exacerbations, improved asthma control and quality of life and reduced reliever use in African American/black (AA/B) and Hispanic/Latinx (H/L) adults with moderate to severe asthma, a group with disproportionate asthma morbidity that has been difficult to reduce.30 This research extended the general concept of using ICS with all rescue beta-agonist use, whether long-acting or short-acting. It also demonstrated effectiveness, through changes in validated assessments, in AA/B and H/L adults, integrated with rescue nebuliser therapy, required no changes in a participant’s current controller therapy and appears to have fewer insurance barriers than an ICS-formoterol-based approach. Participants on PARTICS also reported lower SABA use, both as a metered dose inhaler and via nebuliser, and fewer controller refills as well as a self-reported reduction in controller ICS dosage.30

The objective of this paper is to report on a prespecified PREPARE trial subanalysis using electronic health record (EHR) data to determine if treatment assignment (the PARTICS intervention vs usual care) had an impact on controller therapy through clinicians’ written prescriptions. The analysis compares changes in controller intensity (eg, low, medium or high ICS dose) as determined by EHR data over a 28-month period (12 months prior to enrolment, month of enrolment and 15 months after enrolment) between the two arms of the study.

MethodsBrief description of the PREPARE study

The PREPARE study was a pragmatic, open-label, patient-level randomised trial designed to observe the effects of adding ICS to rescue therapy among AA/B and H/L adults with moderate-to-severe asthma.31 In the PREPARE study, the Intervention/PARTICS group (hereafter, ‘PARTICS group’) received additional ICS medication (donated by TEVA, beclomethasone 80 µg) and was asked to use 1 puff of beclomethasone for every 1 puff of usual rescue inhaler and 5 puffs of beclomethasone for every 1 rescue nebulisation, in addition to usual care. The control group received usual care (hereafter, ‘UCare group’), as described previously.31 Exacerbations were tracked using patient-reported outcomes that were adjudicated using EHR data and patient interviews. The substudy detailed here covers the prescriber side of asthma controller medications, as recorded in the EHRs of the primary asthma treating clinician (primary care, pulmonologist or allergist) for a subset of enrollees. To determine the effect, we analysed PREPARE participants’ asthma controller medications over 28 months (12 months before enrolment into PREPARE, month of enrolment into PREPARE and 15 months after enrolment).

EHR data acquisition

Each site was asked to provide EHR data for each participant within the PREPARE study. The DARTNet Institute provided a set of instructions for each site including the variables needed from the EHR. 16 of the 19 participating sites provided data from their EHR or data warehouse. Requested information included care site information, patient visits, medications, diagnostic codes and procedures. Each participating site provided a pilot data pull that was standardised to the Observational Medical Outcomes Partnership Common Data Model V.5.4 (OMOP CDM V.5.4) and reviewed for data quality. Data quality concerns were communicated back to the sites; in some cases, new data were submitted. After all patients from a given site had exited the study, that site repulled their data and sent it to DARTNet for data standardisation. (Note: One site, site 6, did not provide data for their patients from before their enrolment into PREPARE.)

EHR data standardisation

DARTNet used its standard procedures (python scripts and SQL coding) to transform the data received from each site into the OMOP CDM V.5.4.32 Briefly, data were loaded into a receiving database. As data were moved into the final OMOP data model, known taxonomies were auto-mapped to source OMOP concept IDs and a ‘standard’ OMOP concept ID. Source data using idiosyncratic codes were hand-mapped and processed through to the OMOP data model. Data quality reports were evaluated for all relevant sets of conditions, medications, procedures and measures. Final data were sent to the American Academy of Family Physicians (AAFP) for cleaning and analysis.

EHR data cleaning

Once the standardised data were received from DARTNet, the individual site files were combined for analysis. DARTNet worked with the AAFP to review and modify, if needed, existing ‘value sets’ (ie, list of codes associated with clinical concepts—either specific medications, conditions or procedures). Value sets were created that segregated asthma medications as specifically needed for this analysis. Medications were categorised on a monthly basis as active using the following rules:

Start date of medication: The month the prescription was written was considered the start date/month.

End date of medication:

If an end date was provided, then it was used if it was 12 months or less from the time the prescription was written.

If no end date was provided, then an end date of 12 months after the prescription was written was used.

Dosage/quantity and refills provided were also used to determine if a 12-month end date was appropriate or a shorter time span should be used due to how the prescription was written.

Once each asthma medication for each patient had been assigned months, all medications were assigned to types of asthma medications (eg, low/medium/high-dose ICS, long-acting beta-agonists, long-acting muscarinic antagonist). Using those assigned types, all medications for each patient were collapsed into one record for each month (patient-month). The medication types for each month were assigned to a controller step based on the highest ICS dose (ie, low, medium, high) that they were prescribed along with additional controller medications.

Outcome groupings

Each patient-month was assigned to a controller step based on a six-step classification scheme. The hierarchy of controller step levels (located in online supplemental materials) was based on guideline steps (hereafter referred to as step) outlined in the National Asthma Education and Prevention Programme guidelines.14 If a patient had no medication used within the classification scheme, then the patient was assigned a ‘0’ for the month indicating the patient had no asthma controller medications prescribed to him/her within a given month. The six-step classification scheme (plus ‘0’ for no medication) was used as the dependent variable in the statistical models.

Statistical methods

This analysis was a secondary data analysis. Descriptive and inferential statistics were completed as appropriate. To start, changes in asthma controller medication were determined by the starting and stopping of prescriptions according to the EHR data rules. A change was the movement up or down in the step hierarchy. Then to model the EHR data, linear mixed effect model (LME; overall model and a spline version) was used with the assigned asthma controller medication step in each month as the dependent variable; independent variable(s) varied depending on the model. A series of LMEs was completed to determine the individual demographic variables that contributed (p<0.05) to the model. The dependent variable for all LMEs was the six-step classification scheme described above. If a demographic variable contributed significantly (p<0.05) to an LME containing month, group, site and that demographic variable, then the variable was included in the final models discussed below. We also included a COVID-19 correction variable to account for any possible changes in asthma controller medications that occurred due to the COVID-19 pandemic (ie, 0=no COVID-19 pandemic present; 1=COVID-19 pandemic present). For each variable included in the LMEs, the final category (eg, site 16 out of sites 1–16) was used as a reference category. LMEs require a reference category for each categorical variable as a comparison for the other categories. Due to the nature of the data and the autocorrelation present, the covariance matrix used was a heterogeneous first-order autoregressive to account for the previous month’s influence. LMEs were used due to the ability to cluster to the person/patient, the acceptance of potentially missing data, and that it also works for data that violate any distributional assumptions.33–35 An overall LME was completed for all 28 months of data (located in online supplemental materials) and an LME with spline was completed for before and after enrolment data. For the final LMEs, the random variable was the intercept, clustered to the patient. The fixed variables were patient-month, group (PARTICS/UCare), site, decade of birth and COVID-19 correction. An alpha of 0.05 was used. All analyses were completed by using SAS V.9.4.

ResultsDemographics

After cleaning EHR data obtained from each site, there were 713 participants (out of 1201 participants randomised in PREPARE) included in these analyses. The remaining patients in the PREPARE study (488) did not have their data included because the sites did not provide EHR data. The breakdown of participants by site is in table 1. Of these, 350 (49.1%) were in the UCare group and 363 (50.9%) were in the PARTICS group. Most participants were female and non-Hispanic/black (table 2 for comparisons with the full study cohort). As the University of Puerto Rico, a major H/L enrolling site, was not able to provide EHR data, the cohort of participants included in this analysis has a lower per cent of H/L participants than the full study (p<0.001; table 2).

Table 1

Counts of participants used in the EHR data by site (n=713)

Table 2

Demographics and comparisons with main study (n (%))

Changes in asthma controller medication prescriptions

As noted in table 3, asthma controller medication prescription changes, without regard to the change direction, occurred over time (before and after enrolment; comparison of two time points) for both the UCare and PARTICS groups. Changes included addition or withdrawal of inhalers or medications, biologics and leukotriene inhibitors, as well as changes in dosage of all medications. However, changes were statistically similar so both groups changed controller medications with similar frequency over the various time periods included in table 3.

Table 3

Changes in asthma medications over time (n (%))

Modelling of EHR data

After models with the individual demographic variables were completed (model included in supplemental materials for reference), each model contained the assigned step each month as the dependent variable and the month, the group (PARTICS or UCare), the site, the patients’ decade of birth and COVID-19 correction as independent variables. The patients’ decade of birth and COVID-19 correction contributed significantly to the individual models which necessitates inclusion in the final model. After the final overall model (located in the online supplemental materials), a spline was introduced to determine the differences between before and after enrolment and to determine the effect of the intervention using prescribers’ written prescriptions. The next two models detail the LME with spline accounting for before enrolment (table 4) and after enrolment (table 5).

Table 4

Final before enrolment LME results

Table 5

Final after enrolment LME results

Model for before enrolment (12 months prior through 1 month before enrolment)

The UCare and PARTICS patients were at approximately the same step before enrolment. The months before enrolment did show a significant decrease in controller intensity over time (months were inputted as negative values). (Note: The data for before enrolment took place entirely before the start of the COVID-19 pandemic.) The assigned steps differed significantly between the sites.

Four sites (sites 2, 3, 5 and 7) had significantly higher steps of asthma controller medications compared with patients from site 16 (reference site). One site (site 15) had significantly lower steps of asthma controller medications than patients from site 16. The assigned steps differed significantly with the patients’ decade of birth. Patients born in the 1940s/1950s had significantly higher levels of asthma controller medications than patients born in the 1990s/2000s (table 4).

Model after enrolment (1 month after through the 15 months after enrolment)

The UCare patients were at a significantly higher asthma controller medication step overall than the PARTICS patients across all time periods for the time after enrolment. However, the months after enrolment did show a significant decrease in controller intensity over time for both groups, which could be due to the COVID-19 pandemic. When compared with the model before enrolment, the difference between the UCare and PARTICS patients with the changes over time indicates that the PARTICS patients had significantly greater decrease in the intensity of their controller therapy compared with UCare patients (tables 4 and 5).

The assigned steps differed significantly between the sites. Seven sites (sites 2, 3, 4, 5, 7, 11 and 14) had significantly higher steps of asthma controller medications than patients from site 16 (reference site). One site (site 15) had significantly lower steps of asthma controller medications than patients from site 16. The assigned steps differed significantly with the patients’ decade of birth. Patients born in the 1940s/1950s had significantly higher levels of asthma controller medications than patients born in the 1990s/2000s. Lastly, in the months when COVID-19 was not present (before April 2020), both groups of patients had a significantly higher asthma controller medication level than after the COVID-19 pandemic began (table 5).

Discussion

This study shows, through prescribers’ written prescriptions, there was a greater decrease in asthma controller medication level over time (after enrolment) in the PARTICS patients compared with the UCare patients with both groups starting at the same asthma controller medication level. The decrease in intensity reduces, to some degree, the expected increase in total ICS exposure that would result from adopting the PARTICS strategy. The results in this substudy complement and confirm the results in the main effects paper that that shows that self-reported ICS-containing controller refills were lower for participants assigned to PARTICS versus usual care.30 Between this substudy and the main effects paper, the participants’ reported and prescribers’ EHR data indicate decreased asthma controller medication use over time for the PARTICS patients compared with the UCare patients, even with correcting for COVID-19. When combined with the patient self-reported levels of controller ICS use and rescue ICS fulfilment data from the main effects paper, this substudy supports that PARTICS improves outcomes at the population level, as the intensity of ICS controller therapy dropped as well as self-reported controller ICS use. Treating clinicians were free to adjust asthma medications as deemed necessary during the study. Primary asthma treating clinicians were not provided with the monthly Asthma Control Test scores that were collected for research purposes, but patients may have been more aware of their asthma symptoms through repeated completion of this instrument. For many patients, the primary asthma treating clinician was not the research prescribing clinician, but the treating clinician would have been aware of the study medication. While it is not possible to determine if the drop in controller ICS dose, at the population level, was in response to the added as-needed ICS, the increase in controller ICS dosage for the control patients would suggest changes were more likely to be based on asthma control and not perceived total steroid burden.

While the results of this substudy confirm the results of the main effects paper, there are limitations. EHR data have inherent missingness and can have gaps due to infrequent visits and prescribing activities. In particular, assumptions concerning the length of time a prescription was active were made to determine an active medical regimen. While this may have affected the exact level for a given individual, the assumptions were evenly applied across both groups. Further, some individuals were on dual ICS inhalers not counting the study-prescribed beclomethasone, and these individuals were assigned based on the highest potency of any of the non-study-prescribed ICS inhalers. Again, while this may have affected the ICS level of an individual, the approach was evenly applied at all times across both groups. Thus, neither of these processes would bias the results for either of the study arms. In addition, not every site within PREPARE provided data, but we had data from the majority of sites. This reduction in ICS controller intensity in the PARTICS group was accompanied by a significant reduction in annualised exacerbation rates in this group,30 though total ICS use considering both controller and rescue use likely increased to some degree.

The results show that there were significant changes over time (after enrolment) as well as significant differences between the PARTICS and UCare groups, participating sites, participants’ decades of birth, and the COVID-19 correction. In the spline regressions, the UCare and PARTICS patients were comparable before enrolment, and differences occurred after enrolment in PREPARE. This effect is consonant with our reported reduction in exacerbations in the PREPARE cohort during the COVID pandemic.36 The significant effect due to COVID-19 could be due to a variety of factors including changes in lifestyles that occurred such as working from home and a reduction in air pollution due to fewer cars on the road.36 This effect could be further studied using air pollution data around the patents’ dwellings.

Select sites provided their patients with significantly higher (or lower) asthma controller medications than the reference site (site 16). The differences could be accounted for by different overall participant asthma severity across sites,31 clinicians’ prescribing patterns, organisational policies or environmental factors. Though all clinicians providing asthma care for PREPARE participants completed the Asthma IQ programme37 to help standardise care, this intervention would not affect baseline (or prebaseline) prescribing patterns, which were in place prior to the training. Further exploration into the reasons for the differences is warranted.

This substudy shows that clinicians’ prescribing patterns did change over time, and the PARTICS patients were prescribed lower doses of asthma controller medications by the end of the study.

Data availability statement

Data are available on reasonable request. Data are available on reasonable request to the corresponding author.

Ethics statementsPatient consent for publication

Not applicable.

Ethics approval

This study involves human participants and was approved by Partners Human Research Committee: 2016P001839/BWH. This was a secondary data analysis of the original PREPARE study and informed consent was obtained as part of the main study. The data received were deidentified. Participants gave informed consent to participate in the study before taking part.

Acknowledgments

We acknowledge all study participants for their time and dedication. We also thank the PREPARE Operations Team, Nancy Maher and the PREPARE site principal investigators for their contributions.