After obtaining Tulane/VHA IRB and VHA Research and Development (R&D) approvals for this study protocol, the VISN 16 data warehouse was used to evaluate antipsychotic medication patterns and health outcomes. We examined metabolic monitoring in a retrospective cohort study in patients with schizophrenia receiving care from the Veterans Health Administration, US Department of Veterans Affairs (VA) which is an integrated health system providing comprehensive health care services. Patients' demographic information, inpatient care, outpatient care, outpatient pharmacy records, vital signs, and laboratory results during the period of October 2002 through September 2005 in VISN 16 were extracted from the VISN 16 data warehouse. Data format and content which was stripped of patient identifiers was in compliance with the Health Insurance Portability and Accountability Act (HIPAA) requirements.
The baseline metabolic monitoring services in the study were required to occur within 180 days prior to a new SGA treatment episode (i.e., the index date). The evaluation method had intentionally excluded the first 30 days after the new episode of SGA was prescribed to avoid potential misattribution of pre-existing metabolic syndrome to the index drug. It was previously shown that rapid weight gain can occur early on - within the first 2-3 weeks of treatment - particularly with certain SGA such as olanzapine [19, 20]. Such rapid and early weight gain could lead to increase the probability that some patients who did not have a pre-existing metabolic syndrome might have met the metabolic syndrome criteria within the first 30 days after a new treatment episode. Due to a retrospective study design, it was hard to ascertain whether lab testing belonged to baseline monitoring or follow-up care during the first month following the prescribed index SGA was filled by patients.
Treatment episode with any SGA reflected 3 different types of treatment episodes: switch, new start, and augmentation. The switch episode was defined by the discontinuation of the previous antipsychotic agent within 60 days after the index date. A new start episode was defined by receiving a new episode of antipsychotic agent after a medication break of at least 60 days. An augmentation episode was defined by the concurrent use of the new antipsychotic agent and the previous antipsychotics for longer than 60 days. Only was the first antipsychotic episode included if there were multiple episodes for a patient.
Additional file 1 presents the process leading to the analytical sample. We initially identified 8,816 patients who had at least 1 record of antipsychotic prescription after October 2003 in order to ensure that all patients at least had 1 year before the earliest possible index date (October 2002-September 2003). All patients also had a diagnosis of schizophrenia based on the codes of the International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM: 295.xx) for either at least 1 inpatient service or at least 2 outpatient visits. Patients were excluded if they were diagnosed with a bipolar disorder (ICD-9-CM code: 296.0x, 296.4x, 296.5x, 296.6x, 296.7x, 296.8x, 296.9x) or any type of dementia (ICD-9-CM code: 294.1x, 290.4x). We excluded non-treatment episodes (n = 2,354) who had been on only 1 antipsychotic drug without any break in therapy for more than 30 days during the whole study period. These excluded patients may not present as the target candidates for the VISN 16 antipsychotic monitoring program because the baseline MetMon required a new treatment episode. Therefore, a total of 6,462 patients had at least 1 new treatment episode, starting from the index date when the patient started a new antipsychotic agent. We further excluded patients who only had 1 filled prescription (n = 393), received more than 2 index antipsychotic drugs (n = 677), or switched to an FGA agent (n = 683).
The analytical sample of 4,709 patients was used to identify 2 comparison cohorts. One cohort (n = 3,568) included patients who received the baseline MetMon (at least 1 claims record or medical record of blood glucose testing or lipid panel) within 6 months before the index date (MetMon+) and another cohort (n = 1,141) included those who did not (MetMon-). All treatment episodes were further classified by type of episodes (switch, new start, and augmentation).
The MetMon+ and MetMon- groups were compared on several variables assessed during the pre-existing year (i.e., during the year prior to starting a new antipsychotic episode). These variables included the socio-demographics (age, gender, race, and insurance coverage other than VHA), illness characteristics including the Charlson Comorbidity Index (CCI) based on ICD-9-CM codes  and ICD-9-CM diagnosis of substance (drug and/or alcohol) dependence disorders, antipsychotic medication use patterns in the pre-existing year including the duration and medication possession ratio (MPR) of antipsychotic agents, numbers of SGAs, and use of SGA+FGA combination. We also compared the 2 cohorts on outpatient visits (psychiatric or nonpsychiatric) during 1 year prior to the index date. For inpatient care records, the count of hospital admission as a confounder was examined during the 180 days before receiving the baseline MetMon services because glucose and lipid testing may be more likely to have been performed during a recent hospitalization over the same timeframe of 6 months than otherwise.
For patients who underwent the baseline MetMon, we also examined 5 cardio-metabolic risk factors: (1) presence of serum high density lipoprotein (HDL) cholesterol level <40 mg/dl (male) or <50 mg/dl (female); (2) fasting plasma glucose (FPG) level of at least 110 mg/dl; (3) serum triglyceride (TC) level of at least 150 mg/dl; (4) blood pressure of at least 130/85 mm/Hg; and (5) waist circumference of more than 102 cm (>40 inches) for men and >88 cm (>35 inches) for women. Since waist circumference was rarely reported in the medical records, we replaced it with a body mass index (BMI) ≥ 28.8 because it was found to be correlated with a waist circumference of 102 cm for men and 88 cm for women [22, 23]. In addition, we further reported a recent BMI in 4 categories: underweight (<18.5), normal (18.5-24.9), overweight (25.0-29.9), and obese (≥ 30.0), which is the weight categorization used by the National Heart, Lung, and Blood Institute . For those without the baseline MetMon, we reported their most recent values of blood pressure and BMI prior to the index date. Using the lab testing-based indicators, we created a dichotomous variable reflecting whether a patient met the criteria for metabolic syndrome as defined by the National Cholesterol Education Program (NCEP) (at least 3 out of 5 parameters) or not (2 or less indicators) in 2 methods. To meet the metabolic syndrome using method 1, it was required to include patients who had all 5 indicators with no missing lab tests. Furthermore, while keeping the same sample as method 1, method 2 was used to expand the lab testing-based indicators to the composite indicators, which include all 3 types of records: lab results, diagnosis codes (diabetes, dyslipidemia, hypertension), and medication uses (antidiabetics, lipid-lowering drugs, and antihypertensives). Because the NCEP criteria has 2 lipid-related indicators (HDL and triglyceride), to be conservative, either dyslipidemia diagnosis or lipid-lowering drugs for a patient was counted as only 1 positive lipid indicator rather than 2.
The variables, in which the 2 cohorts were shown to significantly differ in the year prior to the index date through t-tests or chi-square tests, were then fit into a parsimonious logistic regression to identify predictors for undergoing the baseline MetMon. Specifically, a step-wise progression method was used to select the most parsimonious multivariate model. All analyses were performed using the SAS version 9.13. The statistical significance was set at the level of 0.05.