The goal of the analysis was to investigate the drug therapy patterns of non-institutionalized adult (20–64 years of age) Oregon fee-for-service Medicaid enrollees prescribed atypical antipsychotic medications. Using an observational cohort constructed from administrative claims data, patients with a new prescription for an atypical antipsychotic medication (clozapine, olanzapine, quetiapine, risperidone, ziprasidone) between January 1, 2004, and December 31, 2004, were identified. A new prescription (index fill) was defined as a patient’s first claim with no previous claim for any atypical antipsychotic medication for a minimum of 6 months (earliest historical date July 1, 2003). To ensure complete ascertainment of claims and no loss of follow-up due to lost eligibility, patients were required to have continuous fee-for-service Medicaid enrollment for a total of 18 months (6 months prior and 12 months following index fill). However, patients were followed for up to 2 years following their index fill. If atypical antipsychotic therapy continued beyond 2 years, these data were omitted from analysis (i.e., patients were followed for a maximum of 2 years).
Demographic data including age, sex, ethnicity, urban or rural residence, dual Medicare eligibility, diagnostic information, and index prescribing provider type were summarized. Urban and rural classification was based on 2000 census information by the county listed as the patient’s residence. Ethnic determination was based on enrollment data, which we consolidated into one of the following: White, African-American, Native American, Hispanic, Asian/Pacific Islander, Other/Unknown. To evaluate the generalizability of our longitudinal cohort, we identified basic demographic and utilization data for a comparison group that included all patients between the ages of 20 and 64 with any fee-for-service enrollment during the 12 month capture period.
Prescribing provider information was determined based on the patient’s index prescription. For each submitted claim, the dispensing pharmacy is required to submit information regarding the prescribing provider. If a prescribing provider is not an authorized Medicaid provider, however, a pharmacist may enter an emergency prescribing provider default code in order to facilitate timely claims processing. Unfortunately, this exemption is used beyond the initial intention and roughly one third of processed claims have no prescribing provider information attached. Furthermore, institutions such as clinics and hospitals can have valid provider identifiers which may also be entered, though it may be difficult to identify an individual prescribing provider responsible for a specific claim. Data on physician specialty (e.g., psychiatry, internal medicine) are also kept in the Medicaid provider file. For index claims where a prescribing provider was identified, we classified the provider as a nurse practitioner (presumed to be a combination of psychiatric and primary care-based nurse practitioners), general practitioner (e.g., internal medicine, general practice, family practice specialty listed), or psychiatry (either a psychiatrist or a mental health clinic, whose prescribing providers could be psychiatrists or psychiatric nurse practitioners). These prescribing provider classifications may slightly underestimate the proportion of psychiatric providers, but generally reflect the proportions of general practice versus psychiatric prescribing providers who are identified in the claims data.
Patient diagnostic information was abstracted from the Medicaid medical encounter claims dataset. Depression, anxiety disorders, bipolar disorder, schizophrenia, dementia, personality disorder, PTSD, and insomnia were identified using International Classification of Disease – 9th Revision Clinical Modification (ICD9CM) codes on submitted medical claims. Depression was defined by the ICD9CM codes 3090x, 3091x, 311xx, 2969x, 2962x, and 2963x. Schizophrenia was defined by the ICD9CM code 295xx. Bipolar disorder was identified using the ICD9CM codes 2964x, 2965x, 2966x, 2967x, and 2968x. Anxiety disorder was defined by the ICD9CM code 300xx. Dementia was defined as ICD9CM code 290xx. Personality disorder was defined by ICD9CM codes 301xx. Codes 30981 and 308xx were used to identify PTSD. Insomnia was defined as ICD9CM codes 78050, 78051, and 78052. Finally, other psychiatric diagnoses were identified using the remaining ICD9CM codes in the mental disorders category (290xx–319xx) not already specified above. Diagnostic criteria were screened for 6 months before and during the entire patient follow-up.
Patients were followed from index fill for up to 2 years depending on continuation of therapy. For patients with more than 2 years of treatment, we included only the first 2 years of data. For each claim, an interval of treatment was quantified by using the dispensing date and days supply (i.e., begin date = dispensing date, end date = dispensing date + days supply). Follow-up of patients was stopped if they switched to another atypical antipsychotic medication, had no further atypical antipsychotic claims, had a gap in therapy of longer than 31 days, or had continuous therapy beyond 2 years. Although there is not current consensus regarding medication persistence and what would be considered an allowable “gap” in therapy, many have suggested 50% of the previous days supply dispensed is reasonable. To accommodate the small, but significant proportion of patients who receive their prescriptions through the state’s mail order pharmacy which allows a maximum of 90 days supply to be dispensed, an absolute gap of 31 days was selected as the midpoint between 15 day (50% of 30 day supply) and 45 days (50% of 90 day supply).19
Each patient’s therapy was characterized by the length of atypical antipsychotic treatment, augmentation with other mental health medication, as well medication adherence. Augmentation was defined as concurrent use of either an antidepressant (selective serotonin reuptake inhibitors, venlafaxine, mirtazapine, nefazadone, duloxetine, and bupropion) or mood stabilizer (lithium, carbamazepine, gabapentin, lamotrigine, levetiracetam, oxcarbazepine, phenytoin, pregabalin, tiagabine, topiramate, valproate/valproic acid/divalproex, and zonisamide) for at least 60 days at any point.
Adherence was assessed using the medication possession ratio (MPR).20–22
The MPR is a commonly employed method for measuring medication adherence and is calculated by dividing the length of therapy on medication by the total day supply dispensed during the period.19
An MPR of 1 indicates sufficient supply for a dose every day during the treatment period. Subjects with an MPR of less than 0.8 were classified as having poor adherence because they did not have sufficient medication for the treatment period. If the MPR was greater than or equal to 0.8 subjects were considered fully or overly adherent. The MPR was only analyzed for subjects with more than 30 days of therapy to minimize the impact of those subjects with only one fill. This categorization is similar to that used in other studies in which antipsychotic medication adherence measured with medication claims has been associated with an increased risk of admission as well as increased costs of care.23, 24
Finally, atypical antipsychotic medication dosing was evaluated. A daily dose was calculated from the unit strength, dispensed quantity, and days supply fields from each claim. For each individual, the most frequently prescribed daily dose (modal dose) was established and averaged (mean modal dose). For each drug, the mean modal dose was compared to the recommended therapeutic dose according to the labeled indication as well as CATIE protocol specifications.25–30
The daily adult dose for clozapine was defined as 300–900 mg, 10–30 mg for olanzapine, 300–800 mg for quetiapine, 2–6 mg for risperidone, and 80–160 mg for ziprasidone. Patients were considered on a sub-therapeutic dose if their modal dose fell below the recommend range. Demographic and drug therapy characteristics were compared between those receiving sub-therapeutic doses versus those prescribed therapeutic and supra-therapeutic doses. Statistical comparisons were made using the Chi-Square test of proportions, or Fisher’s exact test, for categorical data. Continuous data were compared using Student’s t-test. Finally, a multivariate logistic regression was used to model the association between sub-therapeutic dosing (yes/no) and demographic and drug therapy characteristic variables previously described. Variables were entered into the model using a backwards stepwise procedure with the selection criteria set at a p-value of 0.05. Multicollinearity between predictor variables was assessed using correlation matrices and the Variance Inflation Factor, and was deemed not to be significant. All statistical analyses were conducted using SAS version 9.1.
The research protocol was approved by the Institutional Review Board for the Protection of Human Subjects at Oregon Health & Science University.