Opening the black box of the group‐based trajectory modeling process to analyze medication adherence patterns: An example using real‐world statin adherence data

• Published in: Pharmacoepidemiology and drug safety 2020-03, Vol.29 (3), p.357-362
• Main Authors: Hickson, Ryan P., Annis, Izabela E., Killeya‐Jones, Ley A., Fang, Gang
• Format: Article
• Language: English
• Subjects: Adult; Aged; Bayes Theorem; Bayesian analysis; Female; group‐based trajectory modeling; health behavior; Hospitalization; Humans; Hydroxymethylglutaryl-CoA Reductase Inhibitors - therapeutic use; Male; Mathematical models; Medicare; Medication adherence; Medication Adherence - statistics & numerical data; methods; Middle Aged; Myocardial Infarction; Patient compliance; pharmacoepidemiology; Statins; statistical distributions; United States
• ISSN: 1053-8569; 1099-1557
• DOI: 10.1002/pds.4917

Purpose The rationale for choosing a final group‐based trajectory modeling (GBTM) specification and evaluations of patient adherence patterns within groups are often omitted in the GBTM medication adherence literature. We aimed to (1) reveal the complexity of GBTM and (2) assess model discrimination of patient medication adherence patterns. Methods Medicare administrative claims were used to measure statin medication adherence as a continuous value in the 6 months before an acute myocardial infarction (AMI) hospitalization. Different GBTM specifications beyond default settings were constructed and compared with the Bayesian information criterion. Spaghetti plots were used to compare individual adherence patterns with group averages. Results Overall, 113,296 prevalent statin users met eligibility criteria. Four adherence groups were identified: persistently adherent, moderately adherent, progressively nonadherent, and persistently nonadherent. Spaghetti plots showed the persistently adherent and persistently nonadherent groups had relatively homogeneous adherence patterns that matched predicted trajectories well. Spaghetti plots also showed that, while adherence patterns in the progressively nonadherent group were not as homogeneous, most patients in this group appeared to be discontinuing statin therapy pre‐AMI. Conclusions Subjective decisions are necessary to identify a final trajectory model. Greater transparency and disclosure of these decisions in the medication adherence literature are needed. Individual patient adherence patterns from spaghetti plots provided additional diagnostic information about trajectory models beyond standard model‐fit assessments to determine if group‐average adherence estimates represent homogeneous patterns of medication adherence.