Our study population comprised patients with a recorded doctor diagnosis of and drug treatment for schizophrenia registered with medical practices in the General Practice Research Database between June 1987 and September 2000. The database is described elsewhere.18
Briefly, it contains data from around 400 general practices, dealing with 3.5 million patients in England and Wales. Continuous information has been collected for over 10 years, providing more than 30 million patient years of observation.19
We derived our data from patient registration records, medical records, and prescription records stored in the database. The source of the database in our study is the Epidemiology and Pharmacology Information Core. Our study protocol was approved by the Scientific and Ethical Advisory Group of the database.
Patients were eligible from the time that up to standard data were provided for them in the database—that is, the data recording had reached a standard appropriate for research use. In addition, all previous diagnoses and treatments are recorded in the medical records. Eligibility criteria for our study included a diagnosis of and treatment for schizophrenia recorded at any time. We defined the study period for patients identified as schizophrenic before the inclusion of up to standard data as the date when the practice began submitting up to standard data for the patient to the date when the patient changed practice, died, or had had final data submitted to the database. For all other patients the start of the study period is defined as the earliest date of diagnosis with schizophrenia. To be eligible for our study, patients needed at least three months of up to standard data.
For the cohort analysis we calculated the incidence rates from the number of patients who developed diabetes within three months of using the drug of interest divided by the person time use for the drug of interest. We calculated the person time use from the number of prescriptions divided by 12 (the typical prescription in the General Practice Research Database is written for one month).
Selection of participants
Incident cases of diabetes were defined as the earliest date of a diagnosis of or treatment for diabetes, occurring at least three months after the beginning of the study period. We define the date for diagnosis as the index date. To ensure that the patients with diabetes were incident cases, we checked the medical and prescription records for any diagnosis of or treatment for diabetes before the study began. Patients identified as cases should not have had a prescription for insulin or oral antidiabetic agents within three months of the index date. The use of a three month window avoided the exclusion of patients with diabetes who contributed no more than three months of data. However, 97% of the cases were free of diabetes for six months or more.
For each case we matched six controls with study periods at least as long as that of the case by age at index date (SD 5 years), sex, and index date. Controls that met the matching criteria were selected at random with SAS software. Controls were selected from patients who had been diagnosed as having or treated for schizophrenia but not diagnosed as having or treated for diabetes at any time. Controls were assigned the same index date as the cases to which they were matched. Therefore the calendar time distributions of the index date were the same for both cases and controls.
We classified antipsychotics as conventionals (depot or non-depot), olanzapine, risperidone, and other newer drugs. Non-depot conventional antipsychotics included benperidol, chlorpromazine, droperidol, flupenthixol, fluphenazine, haloperidol, loxapine, methotrimeprazine, oxypertine, pericyazine, perphenazine, pimozide, prochlorperazine promazine, sulpiride, thioridazine, trifluoperazine, trifluperidol, and zuclopenthixol. Depot conventional antipsychotics included flupenthixol decanoate, fluphenazine decanoate, fluphenazine enanthate, fluspirilene, haloperidol decanoate, and pipothiazine palmitate. Other newer antipsychotics included amisulpiride, remoxipride, and sertindole.
We abstracted all prescriptions written by the doctor for the treatment of schizophrenia and diabetes between the start of the study period and the index date. We defined drug use as the receipt of at least one prescription for an antipsychotic within three months of the index date. The selection of a three month window was based on a review of the case reports suggesting a mean time to onset of glucose dysregulation of three months after starting olanzapine.7–14
Patients not taking the drugs of interest were those who did not have a prescription for an antipsychotic within three months of the index date.
We conducted all analyses with SAS version 7.0. Our main study comprised a nested case-control analysis. To account for the matched study design, we modelled the effect of drug use on the risk of diabetes development using conditional logistic regression.20
We used different referent groups to compare the risk of diabetes developing among users of the different antipsychotics. The first group included all patients except those receiving the drug of interest. The second group included patients taking conventional antipsychotics. The third group included patients with no prescription for an antipsychotic within three months of the index date. In addition to the matching variables, we adjusted the analysis for use of other drugs known to affect the risk of diabetes, such as α blockers, β blockers, thiazide diuretics, corticosteroids, phenytoin, oral contraceptives containing norgesterol, and valproate.