Identifying Subgroups Associated with Switching Patterns of Asthma Treatment: A Real-World Drug Utilization Study Using the Parametric G-Formula.

Background: Comorbidities can aggravate asthma, influencing treatment needed to control symptoms, or leading to under- or overtreatment when causing asthma-like symptoms. However, little is known about how the presence of comorbidities affects asthma treatment trajectories in clinical practice. The g-formula offers a promising approach to analyze the complex, dynamic trajectories of asthma treatment over time.

Objectives: This study aims to investigate whether the parametric g-formula can approximate observed asthma treatment trajectories and predict subgroup differences in switching behavior.

Methods: This retrospective cohort study, using data from 1994 to 2021 from the IADB.nl community pharmacy dispensing database, included 16- to 45-year-olds initiating inhaled asthma medication in the Netherlands. We used multinomial logistic regression models with the parametric g-formula to predict asthma treatment trajectories for two years after treatment initiation, and their association with various patient characteristics (age, sex, chronic drug treatment for atopic diseases (ATD), cardiovascular diseases (CVD), thyroid diseases, arthritis, diabetes, gastroesophageal reflux disease (GERD), mental health problems (MHP), and immunosuppressants).

Results: The g-formula simulations predicted 76% of 24 506 individuals to switch treatment steps on average 2.3 times, with the first switch occurring after a mean duration of 8.3 months. This aligned closely with the actual observed rates (77%, 2.3 times and 7.9 months, respectively). Comparatively, fewer 45-year-olds than 16-year-olds switched treatment steps (74% vs. 78%, p < 0.001), but they did so earlier (8.1 vs. 8.6 months, p < 0.001) and more frequently (2.4 vs. 2.3 times, p < 0.001). Males switched less frequently. Subgroups dispensing chronic medication for ATC, CVD, MHP, or GERD, showed less switching (fewer switchers and after longer duration: p < 0.05 for ATC, CVD, MHP, and GERD; lower number of switches among switchers: p < 0.05 for CVD and MHP).

Conclusions: The parametric g-formula closely approximated the empirical distribution of treatment trajectories in general and within subgroups. We found decreased switching behavior within subgroups of higher age, male sex, ATD, CVD, MHP, or GERD. The dynamic g-formula provides a good alternative to regression analyses when analyzing processes of a dynamic nature or when interested in full trajectories.