Classifying Patients by Antipsychotic Adherence Patterns Using Latent Class Analysis: Characteristics of Nonadherent Groups in the California Medicaid (Medi-Cal) Program


Jeonghoon Ahn, Department of Pharmaceutical Economics and Policy, University of Southern California, 1540 East Alcazar Street CHP-140, Los Angeles, CA 90089-9004, USA. E-mail:


Objectives:  This study identifies latent classes defined by varying degrees of adherence to antipsychotic drug therapy and examines the sociodemographic, clinical, and resource utilization correlates associated with membership in each adherence class.

Data and methods:  Patient-level data were drawn from the 1994 to 2003, 100%-sample California Medicaid fee-for-service paid claims data for patients with schizophrenia (N = 36,195). The date of the first antipsychotic medication filled after January 1, 1999 was then used to divide each patient's data into a 6-month preindex (baseline) and a 12-month postindex (follow-up) period. Three categorical adherence indicators—a dichotomous variable of medication possession ratio greater than 0.80, the number of antipsychotic treatment attempts, and time to a change in antipsychotic medications—and two covariates—a categorical variable of duration of therapy and a dichotomous variable of polypharmacy—were used in the latent class model.

Results:  A three-class model returned the lowest values for all the information criteria and was therefore interpreted as follows: The prevalence rates of the latent classes were 1) 14.8% for the adherent; 2) 20.7% for the partially adherent; and 3) 64.5% for the nonadherent. Membership in the nonadherent class was associated with minority ethnicity, being female, eligibility due to welfare status, prior hospitalizations, and a higher number of prior treatment episodes. Membership in the partially adherent class was associated with higher use of outpatient care, higher rates of depot antipsychotic drug use, and polypharmacy.

Conclusion:  Multiple indicators of adherence to antipsychotic medication can be used to define classes of adherence that are associated with patient characteristics and distinct patterns of prior health-care use.