This was a retrospective cohort study that used data assembled from the Florida Medicaid fee-for-service program, which represented a total of 2 131 953 children and adolescents during the 10-year study period. Details on patient characteristics and drug use pattern have been reported elsewhere.9
We included all beneficiaries who were between 3 and 20 years of age, enrolled between July 1994 and June 2004, and had at least 1 inpatient or outpatient claim for ADHD, defined as International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) code 314.xx. A new-user design wherein patients entered the cohort when they were newly started on methylphenidate or amphetamine salts (index date) was used.10
The index date had to be preceded by a 6-month period of continuous eligibility without any claim for methylphenidate or amphetamine. Patients with claims for atomoxetine (n
= 3702), methamphetamine (n
= 979), or pemoline (n
= 1123) at any time during the study period were excluded. Patients were followed until the outcome of interest occurred, eligibility ended, a diagnosis for malignant neoplasm occurred, they turned 21 years of age, switched from 1 drug class to the other, or started to use drugs from both categories concomitantly, whichever came first. Thus, only the initial treatment period with the second of the 2 ADHD drug categories was considered in an attempt to avoid bias when patients were switched to 1 drug category because the first category was not tolerated.
Use of central nervous system stimulants was determined from prescription drug claims for methylphenidate, amphetamine, and dexamphetamine, including all immediate- and sustained-release forms. Because prescriptions are typically filled for a 30-day supply, we assumed that each prescription was active during the month it occurred and the subsequent month. For each cohort member, each month of follow-up was classified according to stimulant use into 2 categories: (1) time when a stimulant prescription was active was assigned to the current use period; and (2) time after any month of current use with no active claim was classified as former use. The category former use was established because, within each drug cohort, the individual characteristics of this group should be similar to those of current use. Thus, the risk for cardiac events between former use periods of methylphenidate and amphetamine salts should be similar if the selection of a specific stimulant was not driven by patient cardiac risk factors (confounding by indication).
Consistent with our previous report, we defined cardiac events as all ED visits with the principal diagnosis for the following diseases or symptoms: acute myocardial infarction (ICD-9-CM 410.xx, 411.8x), stroke (430.xx–436.xx), hypertensive disease (401.xx–405.xx, excluding malignant causes 40x.0), angina (413.xx), aortic or thoracic aneurysm (441.0x, 441.1x), arrhythmias (426.89, 427.xx), syncope (780.2x), or tachycardia or palpitation (785.0x, 785.1x).
Patients with congenital anomalies of the heart and other hereditary diseases that are often associated with cardiac adverse events were identified by presence of an inpatient or outpatient claim at any time in the study period for the following conditions: hereditary hemolytic anemia (ICD-9-CM 282.xx), hemophilia (286.0x–286.4x), anomalies of bulbus cordis and cardiac septal closure (745.xx), other congenital anomalies of heart (746.xx), congenital anomalies of circulatory system (747.0–747.4xx), Down syndrome (758.0x), gonadal dysgenesis (758.6x), and fragile × syndrome (759.83). Preexisting heart problems was defined as presence of any inpatient or outpatient claim within 6 months before the index date for diseases of the circulatory system (390.xx–459.xx), syncope (780.2x), tachycardia or palpitation (785.0x, 785.1x), and chest pain (786.50). Using the same prediagnosis/pretreatment time period, we used 2 variables as proxy for general health status, namely hospital admission for any cause and whether Medicaid eligibility was based on disability (designated by Medicaid as “SSI” = Supplemental Security Income). To account for concurrent use of other drugs that have been associated with cardiac effects, we ascertained drug claims data for appetite suppressant drugs, monoamineoxidase inhibitors, bronchodilators (β-agonists, ipatropium bromide, or theophylline analogs), antidepressants, and antipsychotics. No attempt was made to find claims for oral decongestants because most were available over-the-counter and an attempt to account for their use on the basis of claims data would be prone to misclassification.
Beneficiaries with conflicting information on self-reported race/ethnicity, gender, and age were assigned the respective category that was most often used in the Medicaid database. For race/ethnicity, specific designation of a race was given priority over the category “other.” Age was categorized according to population-based distributions of mortality and morbidity statistics into <5, 5 to 9, 10 to 14, and ≥15 years. Last, because drug use patterns were expected to shift during the 10-year study period, the year of the index date was included as covariate as well.
Risk between current users of methylphenidate versus amphetamine salts (mixed amphetamine salts or dexamphetamine) and former users of drugs in these categories was compared using a time-dependent Cox proportional hazard model to adjust for group differences in age; gender; race; year of the index date; SSI; congenital anomalies; history of circulatory disease; history of hospital admission; and use of antidepressants, antipsychotics, and bronchodilators as covariates. Three binary independent variables above versus below age thresholds of 5, 10, and 15 years were used. Likewise, 3 independent variables for race/ethnicity—white versus nonwhite, black versus nonblack, and Hispanic versus non-Hispanic—were used. Age and use of antidepressant, antipsychotic, and asthma drugs were assessed for every month of follow-up and included in the regression model as time-dependent variables. All variables were entered in a stepwise forward manner and included when they showed an independent significant association with the clinical end point as dependent variable (P < .05). Data management and analysis were conducted with SAS 9.1 (SAS Institute, Cary, NC).