Rationale

Only recently have observational studies begun to address whether prescription stimulants are associated with adverse cardiovascular events. Providers, patients, and policy makers need clearer guidance on the best way to balance potential benefits and harms of these rapidly increasingly used medications.

Objective

The aim of this study was to systematically review population-based studies of children and adults that tested the association between exposure to prescription stimulants and adverse cardiovascular outcomes. Methodological challenges that face the field and suggestions for future directions of research are described.

In this review, “prescription stimulants” refer to prescribed medications in the amphetamine-family of drugs, namely comprised of amphetamine, methylphenidate, methamphetamine and their variants. Sometimes stimulants are referred to in the plural as “amphetamines” (as distinguished from the drug amphetamine which has a specific chemical structure).

Methods

For the systematic review, studies were considered using the following criteria: 1) retrospective or prospective population-based study, 2) children or adults as participants, 3) prescription stimulant use as the independent variable, and 4) one or more hard cardiovascular outcomes as the dependent variable and primary outcome. Blood pressure, pulse, and EKG changes—established physiological effects of stimulants—were not considered hard clinical events. PUBMED, MEDLINE, EMBASE, and Google Scholar databases were searched for studies published in English in peer-reviewed journals between January 1, 1990 and April 1, 2012 using MESH terms and keywords: ADHD; Adults; Amphetamine; Amphetamines; Arrhythmias, Cardiac; Cardiovascular Diseases; Cardiovascular System; Central Nervous Stimulants; Cerebrovascular; Cohort Studies; Case–control Studies; Death; Death, Sudden, Cardiac; Dextroamphetamine; Drug Toxicity; Methamphetamine; Methylphenidate; Myocardial Infarction; Stimulant; Stroke; Safety. In addition, we hand-searched potentially relevant studies cited in the reference section of electronically identified articles. Titles and abstracts were screened for inclusion/exclusion, and full text versions were retrieved. Selected population-based studies of children were categorized separately from studies of adults for the purpose of comparisons. Studies were assessed for bias at both the study and outcome levels. Catchment, comparison groups, exposure and outcome ascertainment, statistical power and methodologies were assessed.

Population-based observational studies of prescription stimulant use

Overall, 551 unique records were identified in searches. Most records were excluded based on review of titles and abstracts due to not being topical to prescription stimulants and not having appropriate endpoints. Twenty-seven full-text articles were assessed for eligibility. Most (14) were eliminated due to lack of hard cardiovascular endpoints. Ten studies met the inclusion criteria and were included in the qualitative synthesis (Figure 1). These studies, using large population-based datasets specifically designed to detect a signal of cardiovascular harm associated with medical use of stimulants, found mixed results (Table 1).

Children and adolescents

In Winterstein et al.’s study of 55,383 Florida Medicaid beneficiaries, 3 to 20 years old, no increase in cardiac death or hospitalizations was observed [61]. However, there was a 20% increased risk in cardiac-related emergency room visits among current users of stimulants. A second study by Winterstein et al., compared the risk between methylphenidate and amphetamine salt medication preparations [62]. In 2,131,953 Florida Medicaid beneficiaries, 3 to 20 years old, no difference was found in risk of emergency room visits for cardiac reasons between the two medication groups.

McCarthy et al. studied a UK database of patients, 2 to 21 years old, who were prescribed methylphenidate, dexamphetamine, or atomoxetine to determine whether use was associated with a greater risk of sudden death [63]. In 18,637 patient-years, six of the seven deaths were determined to not be cases of sudden death, and one death was of indeterminate cause. Compared to a reference rate of sudden death, the investigators did not find a significantly increased risk of sudden death associated with use of these medications.

Gould et al. conducted a case–control study of 564 cases of sudden death among youth 7 to 19 years old, matched with 564 deceased passenger victims in motor vehicle accidents. Stimulant use (amphetamine, dextroamphetamine, methamphetamine, methylphenidate) was associated with increased odds of sudden death (odds ratio = 7.4; 95% CI 1.4 to 74.9) [64].

Schelleman et al. (2011) studied youth 3 to 17 years (n = 241,417), derived from a 5-state Medicaid database and a 14-state health insurance database [65]. Hazard ratios for incident exposure to stimulants (amphetamines, atomoxetine, methylphenidate, or combination therapy) compared to non-exposure were not significantly elevated for primary outcomes of sudden death/ventricular arrhythmia, stroke, myocardial infarction, and composite stroke/myocardial infarction. Cardiovascular outcomes were validated via medical records and independent adjudication. However, only 48% of requested records were obtained, and the adjudicators were not reported as blinded. In secondary analyses using claims-based outcomes (not validated) and prevalent (i.e. not incident) stimulant use, significant associations were found between sudden death/ventricular arrhythmia and methylphenidate, atomoxetine, and any ADHD medication.

Recently in 2011, Cooper et al. conducted a retrospective matched cohort study of children and young adults 2 to 24 years (n = 1,200,438) using Medicaid (Tennessee and Washington), health insurance, state death certificates, and the National Death Index [66]. Current use of ADHD medications was not significantly associated with increased risk for serious cardiovascular events (sudden death, stroke, and myocardial infarction). Additional secondary analyses, with adjustments to inclusion criteria and independent and dependent variables did not find a significant relationship between ADHD medication use and cardiovascular events. Analyses were adjusted for baseline and time-varying covariates as well as site-specific propensity scores. The study benefitted from review of medical records (79% availability) to perform end-point validation [67].

Olfson et al. in 2012 studied 6 to 21 year olds (n = 171,126) with ADHD and no known cardiovascular risk factors, comparing stimulant users to nonusers [68]. The investigators intended to evaluate the risk of “severe cardiovascular events” which included sudden death, stroke, acute myocardial infarction, and respiratory arrest. But only one incident event in the entire cohort was discovered and no analysis was conducted due to lack of power. The risk of “less severe cardiovascular events” among persons with ADHD—comprising angina pectoris, cardiac dysrhythmias, and transient cerebral ischemia—was not significantly different between stimulant users and nonusers.

Adults

Holick et al.’s population-based study of adults 18 years and older (adults ≥ 65 years old were included) compared use of atomoxetine (a non-stimulant used in the treatment of ADHD that is associated with increases in blood pressure [69]; n = 21,606) to use of stimulant ADHD medications (n = 21,606) [70]. Propensity scoring was used to match atomoxetine and prescription stimulant users. Use of atomoxetine was not associated with a either a greater risk of stroke or transient ischemic attack (TIA), compared to stimulant ADHD medications. In a secondary analysis, atomoxetine and stimulant users, matched to a general population cohort (n = 42,993), had a significantly increased risk of TIA, but not stroke.

The study by Habel et al. represents the largest and most comprehensive study of ADHD medications and cardiovascular outcomes in adults to date (n = 443,198) [71]. In this retrospective cohort study of adults 25 through 64 years old, each ADHD medication user was matched to two nonusers. Medical charts, autopsy reports and death certificates were obtained and adjudicated for myocardial infarction (MI), sudden cardiac death (SCD), and stroke outcomes where available. Adjustment for confounding included the use of a cardiovascular risk score (CRS), which summarized cardiovascular risk factors. Unexpectedly, current use of ADHD medications compared to nonuse was significantly protective against serious cardiovascular events (MI, SCD, or stroke; adjusted rate ratio 0.83, 95% CI, 0.72-0.96). The authors submitted a more detailed report to the AHRQ, using only MI and SCD as outcomes, with the same conclusion that the results did not support an association between ADHD medication use and the risk of MI and SCD [72].

The most recent study of adults (18 years and older), by Schelleman et al. (2012), matched methylphenidate users (n = 43,999) to nonusers (n = 175,955) and found an increased risk of sudden death or ventricular arrhythmia among users (adjusted hazard ratio 1.84, 95% CI 1.33 to 2.55) [73]. No statistically significant difference in risk was found for stroke, myocardial infarction, and a combined endpoint of stroke/myocardial infarction. In a secondary analysis, the risk of all-cause death was significantly increased for methylphenidate users compared to nonusers (adjusted hazard ratio 2.38, 95% CI 2.20 to 2.56).

Considerations for future studies: Methodological challenges and proposed strategies

In the following section, we summarize the main methodological challenges and outline considerations and strategies for future studies. Low Absolute Rate of Cardiovascular Events. Randomized clinical trials (RCTs) are not feasible because of the low incidence rate of stroke, acute myocardial infarction and cardiovascular death, the need for very large numbers of patients, and ethical concerns. Due to power considerations, future work will need to rely on very large population-based cohorts. Hard Clinical Outcomes. Future investigations should be designed with hard clinical endpoints such as death, myocardial infarction, and stroke as the primary outcomes of interest. While prior literature has demonstrated that small increases in blood pressure lead to increased cardiovascular morbidity and mortality on a population-basis [45], there is no proven association between stimulant-induced soft endpoints (e.g. increases in blood pressure and heart rate) and hard cardiovascular endpoints. Study Populations. Because the baseline rates of sudden death, stroke, and myocardial infarction in children are so low, increases in risk may not have a large absolute impact. However in adults a modest increase in risk could have significant clinical impact. Thus an important strategy is to use an enriched population, such as older adults, as greater than 80% of deaths due to heart disease occur in those 65 years or older [74]. Measurement of Prescription Stimulant Use. The relationship of time and dose of stimulants to the risk of adverse cardiovascular events is unknown. It may be that there are both acute and chronic risks with prescription stimulant use. It stands to reason that a higher dose of stimulants incurs more risk than a lower dose, but this has not been proven [27]. Sensitivity analyses, where modeling of use is varied, can help address this problem. Confounders/Selection Bias/Control Groups. Accounting for confounding and selection bias is the greatest challenge for observational studies seeking to study stimulants and the risk of cardiovascular events. Confounding. Observational studies attempt to account for confounding by including known confounders as covariates in a multivariate analysis. Of particular concern is confounding by indication, when a variable is present in the non-exposed, but also an indication for the exposure of interest (e.g. treatment of obesity with prescription stimulants). Confounding by contraindication, where a patient avoids treatment due a contraindication, is also a potential problem. One way to address confounding by indication or contraindication is to limit the study population, by excluding persons with the problematic indication(s). But this comes at the expense of generalizability. Selection Bias. Persons that use prescription amphetamines differ in important ways from persons that are not prescribed amphetamines. Some of these ways are known and measured, and can normally be adjusted for in an analysis, such as comorbid medical risk factors for cardiovascular events. But groups differ in ways that are unmeasured. These unmeasured factors are known in some cases (e.g. poor diet), and unknown in others. Propensity scoring and instrumental variable techniques are two more sophisticated statistical approaches used to adjust for biases in observational data. To date, five studies have used propensity scoring [66, 68, 70, 71, 73]. Studies that do not account for selection bias are extremely difficult to interpret, as it becomes unclear whether the risk is due to the independent variable (prescription stimulant use), or the differences in the populations that are compared. Even with studies that do use propensity scoring, such as the study by Habel et al., it can be difficult to determine whether such techniques adequately accounted for unmeasured confounders. For example, inability of these techniques to fully control for healthy user effects probably explains why Habel et al. found a lower risk-adjusted rate of adverse cardiovascular events among stimulants users. Use of propensity scoring and instrument variables offers the promise of adjusting for selection biases. As the ability to control for confounding and selection bias increases, so does the confidence in the results of such studies.