OBJECTIVES: Patient adherence to medications can be improved by proper identification of patients at risk for non-adherence. This study aims to classify patients to latent non-adherence casemix classes varying in degree of adherence, and to examine the socio-demographic and clinical correlates of class memberships. METHODS: We used patients (n=36,195) with a schizophrenia diagnosis from the 100%-sample Medi-Cal fee-for-service paid claims data (1999~2003). The date of the first antipsychotic medication (index date) was used to define a 6-month pre-index and a 12-month post-index periods. Latent class analysis was applied to four categorical adherence indicators: a dichotomous variable derived from medication possession ratio (MPR; cut-off=0.8); number of treatment attempts (1, 2, 3, 4, >4), duration of uninterrupted therapy (<30, 30~60, 60~120, 120~240, 240~365, >365 days); and time to first switching of medication (no switching, <30, 30~90, 90~180, 180~365 days). Determinants and consequences of non-adherence were examined by profiling each latent class in terms of covariates and utilization outcomes. RESULTS: Models with up to five classes were explored, leading to a final selection of three classes: adherent (prevalence 14.9%), partially adherent (28.1%) and non-adherent (57.0%) based on the smallest classification error (2.14%). Non-adherent class was associated with minority ethnicity, more suicide attempts, more hospitalizations and inpatient rehabilitations than other classes. Partially adherent class displayed higher outpatient care use, higher depot antipsychotic drugs, and higher rate of polypharmacy. Adherent class displayed the exact opposite characteristics as non-adherent class. Total costs in the 12-month follow-up period were $9,370 (adherent), $12,619 (partially adherent) and $8,675 (non-adherent). The adherent class had a lower average cost than the non-adherent class in all the cost categories but drug costs were $1,939 higher (all p-values <0.05). CONCLUSION: Multiple indicators of adherence to antipsychotic medication capture heterogeneity in non-adherence behaviors, which can be modeled using latent classes, so define non-adherence casemix.