(SLA-005) Adherence Trajectories to Antipsychotic Medications Among South Carolina Medicaid Beneficiaries with Schizophrenia or Schizoaffective Disorder

Background: Schizophrenia is a chronic mental disorder that requires a lifetime of treatment. Medication nonadherence leads to poor medical outcomes; the ability to predict a patient’s adherence trajectory cannot be understated in forestalling negative effects of nonadherence.

Objectives: To identify latent adherence classes among patients diagnosed with schizophrenia and evaluate predictors of class membership and trajectories.

Methods: This study is based on Medicaid claims from 2012 to 2020 in South Carolina (SC). Patients who filled any antipsychotic prescriptions were selected from an identification window from 2012 to 2018, with an index date defined as the day of the latest filled prescription. Exclusion criteria included patients with characteristics that could potentially affect prescription fulfillment or the availability of continuous data, such as pregnancy, bipolar disorder, inpatient care and dual eligibility with Medicare and Medicaid. Any remaining patients were observed with baseline characteristics (i.e., sex, age, race, and type of drug administration route), and adherence measured quarterly using the proportion of days covered (PDC) over a 2-year follow-up period.

A latent class mixture modeling (LCMM) framework was employed, consisting of two submodels: a membership model that categorized patients into an adherence class and a longitudinal model that characterized the adherence trajectory over time within each adherence class. Patient characteristics and adherence measurements were used as predictors. The outcome was latent adherence classes of patient medication adherence.

Results: A total of 2,870 eligible patients were included in the study, with 1,707 males and 1,163 females. Four distinct adherence trajectories were identified: (1) best adherent (56% of the population), (2) increasing adherence over time (8%), (3) early drop-off (12%), (4) worst adherent (24%). The impact of predictors varied across classes. For example, patients using long acting injectables were significantly less likely to fall into the worst adherent class in comparison to oral antipsychotics (OR=0.47; 95% CI= 0.38 – 0.60, p < 0.001). These same predictors also impacted how trajectories behaved. In the worst adherent class, adherence decreased as time elapsed (β=-0.03, 95% CI=-0.03 – -0.03, p < 0.001).

Conclusions: Four distinct trajectories for adherence were found among SC Medicaid beneficiaries. Significant patient characteristics that determined class membership and adherence trajectory varied across classes. There are limitations to the generalizability of these findings. However, LCMM can be applied to other populations and help inform strategies that seek to improve medication adherence such as medical interventions.