Asthma control among treated US asthma patients in Practice Fusion’s electronic medical record research database

Study population

A retrospective cohort was established which included patients with asthma and a valid ACT measurement in Practice Fusion’s EMR database between January 1, 2015 and December 31, 2018, and with at least 1 prescription for any asthma treatment in the 6 months prior to the 4-week recall period of their first valid ACT measurement. The date of a patient’s first valid ACT measurement was defined as their index date. Patients were required to have activity in the database, defined as an encounter in the database for any reason, at least 6 months (182 days) prior to their index date. In addition, patients were excluded from our sample if they had ≥1 chronic obstructive pulmonary disease diagnosis code(s) reported at any time on or before their index date or had a missing value for their calendar year of birth (Fig. 1).

Fig. 1: Study design.figure 1

ACT asthma control test; Chronic obstructive pulmonary disease.

Data source

Practice Fusion is a cloud-based connected health platform used in 30,000 healthcare practices with 8% market share among small practices (1–3 physicians) in the US, is linked with 90% of US pharmacies, and 600 laboratory and imaging entities. Practices were included in Practice Fusion’s research database if they met any of the following criteria: over 13,000 chart pulls; 1 or more providers with a verified National Provider Identifier and over 500 chart pulls; sent 500 or more electronic prescriptions; sent 500 or more laboratory orders. Practices were excluded if they were used by Practice Fusion for testing and production purposes, did not have at least one Doctor of Medicine, or were located outside the US. Practice Fusion’s research database contained patient-level data on demographics, office visits, insurance, allergies, vitals, medications, laboratory tests, diagnoses, prescriptions, and immunizations. As of December 2018, the cut of Practice Fusion’s research database contained data for 1.9 million asthma patients with ≥1 ICD-9 493.xx, ICD-10 J45.xx, or SNOMED-CT CTV3 H33xx diagnosis codes between 2007 and 2018 with patients across all 50 US states.

The database is certified as statistically de-identified through the removal of all personally identifiable indicators, transformation of dates, generalization of certain demographic and geographic information, standardization of free text and other sensitive fields, and substitution of patient- and provider-related unique identifiers with random values.

Asthma control

The ACT is comprised of five questions, each item response is captured on a 5-point scale (where 1 is the worst scenario and 5 is the best) utilizing a 4-week recall period. ACT scores range from 5 (poor control of asthma) to 25 (complete control of asthma) with higher scores reflecting greater asthma control. ACT scores ≤19 reflect not well-controlled asthma while ACT scores >19 reflect well-controlled asthma15.

Beginning in 2015, Practice Fusion implemented a clinical decision support program that notified providers that an ACT should be conducted when a patient with asthma missing symptom assessments visited them. While the notification indicated that an ACT should be completed, the system did not require clinicians to complete and/or record the ACT results.

We defined a valid ACT as: (1) having complete responses for all 5 questions; (2) not occurring on the same date as another ACT measurement for the same patient; and (3) not occurring within 28 days of another ACT measurement for the same patient. Scores that reflect asthma control as measured by the ACT cannot be calculated if any of the 5 questions are missing responses. The rationale behind this 28-day time gap is that the ACT reflects a 4-week recall period; if two ACT scores are measured on the same day or within 28 days of each other, it is impossible to determine which of these indicate the correct measurement of asthma control.

GINA step

GINA Step was assessed based on the medications prescribed during the 6-month period prior to the 4-week recall period of patients’ ACT record at index date. Asthma treatment was defined as one of the following medications: short-acting β2-agonists (SABA), short-acting muscarinic antagonist (SAMA), inhaled corticosteroids (ICS), ICS and long-acting β2-agonist (ICS/LABA) combination products, leukotriene receptor antagonist, cromolyn or nedocromil (mast cell stabilizers), methylxanthines, biologics (e.g., mepolizumab) or long-acting muscarinic antagonist. Further details on GINA Step definition and asthma treatments are in Supplementary Table 1.

Determination of a patient’s GINA Step required calculation of ICS and ICS/LABA daily doses. The Practice Fusion prescription data includes fields that were generated using MedEx, a natural language processing system which extracts medication information from clinical notes16. There are three MedEx-derived fields that were used for calculation of ICS daily dose: (1) frequency (e.g., once per day); (2) dose amount (e.g., ‘2’ in ‘2 puffs’); (3) dose unit (‘puff’ in ‘2 puffs’). For missing values of frequency, dose, or dose amount, we imputed values from the mode across each National Drug Code. We converted all ICS strength to micrograms (mcg) prior to calculating ICS daily dose17. ICS daily dose was calculated as: (Frequency)*(Dose amount)*(Strength). Finally, the ICS and ICS/LABA dosage levels required for GINA Step calculation were defined for each medication based on generic names or ingredients (Supplementary Table 2).

ICS/LABA includes fixed-dose ICS/LABA combination medications and ‘open’ ICS/LABA combinations. For patients that had individual ICS and LABA prescriptions, we considered them as ‘open’ ICS/LABA combinations only if the ICS medication and LABA medication were prescribed within 30 days of each other. Patients with separate ICS and LABA prescriptions more than 30 days apart were considered as ‘ICS only’ in the GINA Step calculation. For patients that had multiple ICS or ICS/LABA prescriptions in the eligible period, we used only the prescription(s) that were closest to the patient’s index date for calculation of the ICS daily dose.

GINA steps were defined according to GINA asthma treatment guidelines in 201817. The GINA 2019 treatment guidelines include the addition of as-needed low-dose ICS-formoterol for Step 113. Given that this additional criteria for Step 1 did not align with our observation period, we chose to define GINA Step according to the guidelines clinicians would have followed at the time they prescribed asthma medications in our study. We assumed that any oral corticosteroid (OCS) use was not used continuously (e.g., supply ≤28 days) by the patient and thus had no impact on GINA Steps. This decision was made given the difficulty in calculating a consistent day supply for OCS from the Practice Fusion prescription data. Treatment with SABA, SABA-SAMA, or SAMA was classified simply as SABA. Treatment with SABA only was defined as GINA Step 1. However, SABA use was allowed in all other steps. All individuals that were missing key information required for the GINA Step determination or had combinations of prescriptions that did not clearly meet definitions for a GINA Step were classified as ‘Undefined’.

Covariates

Age in years was calculated as the difference between the calendar year of a patient’s index date and their birth year. Ethnicity was defined as ‘Hispanic’, ‘Non-Hispanic’ or ‘Missing’. Race was defined as ‘White’, ‘Black/African American’, ‘Other’ and ‘Unknown’. ‘Unknown’ race was assigned to individuals that had conflicting responses for race at any time in the database (e.g., patients may have multiple race information) or did not have any documentation of race in the Practice Fusion database. ‘Current’ smoking status was assigned to patients with a status of ‘current smoker’ on the smoking status record closest to their index date. ‘Former’ was assigned to patients with smoking status of ‘former smoker’ at any time on or before their index date. ‘Non-smoker’ was assigned to patients with only records of ‘non-smoker’ at any time in the database on or prior to their index date. Finally, we used the value for body mass index (BMI) in kg/m2 that was recorded on the individual’s index date or a prior record closest (e.g., least number of days) to the index date. Visit type was categorized according to the specialty of the provider with whom the patient had an appointment for the encounter on their index date: ‘Primary Care’ includes ‘Internal Medicine’, ‘General Medicine’, and ‘Family Medicine’; ‘Specialist’ includes ‘Allergy and Immunology’, ‘Pulmonary Disease’, and ‘Emergency Medicine’; ‘Other’ includes all other specialties.

Statistical analysis

Descriptive frequencies both overall and by patient asthma control status at index date were calculated for each GINA Step. We used Poisson regression with robust variance to directly estimate the prevalence ratio (PR) and 95% confidence intervals (95% CI) of not well-controlled asthma by patient GINA Step at index date, adjusting for age, race, Hispanic ethnicity, smoking status, BMI, and the visit type at index date. Given that not well-controlled asthma was quite common in our study population (e.g., >10%), odds ratios derived from logistic regression would violate the rare disease assumption and consequently would overestimate the strength of associations and not approximate the relative risk18. However, Poisson regression models with robust variance can directly estimate the PR and are a suitable alternative to logistic regression modeling in cross-sectional studies with a dichotomous outcome19. The primary exposure of interest was GINA Step at index date with Step 1 as the reference group. The dependent variable (outcome) was not well-controlled asthma at index date, defined as an ACT score ≤19. We used a Directed Acyclic Graph to identify covariates for confounding control in the regression model. The final model included age (in years), BMI, race, Hispanic ethnicity, smoking status, and type of visit at index date.

Ethics

The data used in this study are data collected from routine activity as part of patients’ interactions with the healthcare system through their provider’s medical records software. The original data collection is for administration and healthcare delivery purposes but is aggregated and deanonymized for research purposes. The analysis used fully deidentified retrospective data, and as such, this is not classified as research involving human participants as defined by 45 CFR 46.102(f) under the US Department of Health and Human Services Policy for Protection of Human Subjects (https://www.hhs.gov/ohrp/regulations-and-policy/regulations/2018-req-preamble/index.html). Therefore, institutional review board approval and informed consent were not required.

Reporting summary

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

留言 (0)

沒有登入
gif