(178) Lasso Logistic Regression and Cluster Analysis in Predicting Adherence and Drug Patterns among New Users of Monotherapy for Antihypertensive Drugs: a long-term inception cohort study

Background: In hypertension management, patient drug adherence is essential for successful monotherapy. However, not much is known about real-world long-term adherence and drug patterns in antihypertensive monotherapy.

Objectives: Our study aims to assess adherence rates and drug patterns among patients after start with anti-hypertensive monotherapy.

Methods:

Design: A retrospective inception cohort study.
Setting: Pharmacy records were extracted from the University of Groningen IADB.nl dispensing database from January 1, 1996, to December 31, 2020. Among the 33,427 adult starters with antihypertensive drug monotherapy, Cohort 1 contained 13,155 patients with a follow-up period exceeding 3 years (1080 days), while Cohort 2 comprised 2,086 patients with a follow-up duration surpassing 10 years (3600 days).
Exposures: Antihypertensive drug monotherapy included Angiotensin-converting enzyme inhibitors (ACEIs), Angiotensin II receptor blockers (ARBs), Beta-blockers (BBs), Calcium Channel Blockers (CCBs), and thiazides.
Main outcome measures: Primary outcomes included adherence rates, calculated as the proportion of days covered by antihypertensive monotherapy during follow-up. Binary adherence classified levels as high (≥0.8) or low ( <0.8). Secondary outcomes encompassed drug patterns, such as discontinuation, continuation, switch, and add-on.
Statistical analysis: Adherence and drug patterns were expressed using descriptive statistics. Cluster analysis was performed to categorize patients with similar risk factors. Lasso logistic regression was then employed to construct prediction models.

Results: Adherence rates were 84.2%, 92.1%, and 94.9% in all patients, cohort 1, and cohort 2 patients receiving antihypertensive drug monotherapy, respectively. In cohort 1, drug pattern rates were 45.4% for discontinuation, 37.8% for continuation, 11.6% for switch, and 31.1% for add-on. Thiazides exhibited the highest rates in discontinuation, switch and add-on, while CCBs showed the highest continuation rate. In cohort 2, drug pattern rates were 4.1%, 92.1%, 0.6%, and 4.1%, respectively. CCBs had the highest rates of continuation, switch, and add-on, while ARBs showed the highest discontinuation rate. ACEIs, ARBs, and Thiazides demonstrated a lower likelihood of low adherence compared to BBs in cohort 1, while most monotherapies displayed greater ease in terms of discontinuation, switching, and adding compared to BBs in cohort 1 or 2. Similar results were observed across different distinct risk clusters within both cohorts.

Conclusions: The statistical clustering analysis of adherence and drug patterns among patients on monotherapy offers valuable insights into medication options.