Trends in drug abuse and stroke associated with drug abuse in discharges of persons aged 18 to 44 years from Texas hospitals, from January 1, 2000, to December 31, 2003: annual prevalence rate of abuse of various drugs (A), percentage change in prevalence rates from 2000 (B), incidence rates of stroke associated with history of abusing various drugs (C), and percentage change in incidence rates from 2000 (D). All data are given plus or minus the standard error. The P values were obtained using the Cochrane-Armitage test for trend (2-tailed). In C and D, “other strokes” indicate all strokes not associated with alcohol, cocaine, cannabis, amphetamines, opioids, or hallucinogens.
The percentages of hemorrhagic (A) and ischemic (B) strokes attributable to each risk factor after adjusting for all other risk factors (multivariate population-attributable risk percentages) in 2003 are given. These estimates assume that the associations are causal and unbiased. Percentages may not total 100 because of rounding.
Westover AN, McBride S, Haley RW. Stroke in Young Adults Who Abuse Amphetamines or CocaineA Population-Based Study of Hospitalized Patients. Arch Gen Psychiatry. 2007;64(4):495-502. doi:10.1001/archpsyc.64.4.495
The abuse of stimulant drugs is increasing in the western United States. Although numerous case reports and animal studies suggest a link with stroke, epidemiologic studies have yielded conflicting results.
To test the hypothesis that young adults who abuse amphetamines or cocaine are at a higher risk of stroke.
Design, Setting, and Participants
Using a cross-sectional design and from a quality indicators' database of 3 148 165 discharges from Texas hospitals, we estimated the secular trends from January 1, 2000, to December 31, 2003, in the abuse of various drugs and of strokes. We developed separate logistic regression models of risk factors for hemorrhagic (n = 937) and ischemic (n = 998) stroke discharges of persons aged 18 to 44 years in 2003, and for mortality risk in patients with stroke.
Main Outcome Measure
Incidence of stroke using definitions from the Agency for Healthcare Research and Quality's stroke mortality Inpatient Quality Indicator.
From 2000 to 2003, the rate of increase was greatest for abuse of amphetamines, followed by cannabis and cocaine. The rate of strokes also increased, particularly among amphetamine abusers. In 812 247 discharges in 2003, amphetamine abuse was associated with hemorrhagic stroke (adjusted odds ratio [OR], 4.95; 95% confidence interval [CI], 3.24-7.55), but not with ischemic stroke; cocaine abuse was associated with hemorrhagic (OR, 2.33; 95% CI, 1.74-3.11) and ischemic (OR, 2.03; 95% CI, 1.48-2.79) stroke. Amphetamine, but not cocaine, abuse was associated with a higher risk of death after hemorrhagic stroke (OR, 2.63; 95% CI, 1.07-6.50).
Increases in stimulant drug abuse may increase the rate of hospital admissions for strokes and stroke-related mortality.
Evidence has been accumulating for 2 decades supporting a link between abuse of stimulant drugs and strokes in young people.1- 8 Human imaging and postmortem examination, as well as laboratory animal models, suggest that stimulant drugs, such as cocaine and amphetamines, might produce strokes by direct effects on the cerebral circulation, including elevated blood pressure, vasculitis, and cerebral vasospasm.7,9- 18
The support for a link between stimulant abuse and strokes in humans has come primarily from case reports and case series of strokes in young people who have abused these drugs. Such reports led the Food and Drug Administration to issue a public health advisory on the stimulant phenylpropanolamine and a request for companies to remove it from all products19 and to ban ephedra from over-the-counter products in February 2004.20 The ban was struck down in April 200521 and then reinstated by the Tenth Circuit US Court of Appeals in August 2006.22 Despite the many case reports, the causal link has remained controversial because of the possibility of coincidental co-occurrence from the high prevalence of drug abuse and methodologic weaknesses and conflicting findings of the 5 epidemiologic case-control studies and 1 cross-sectional study that have examined the association.23- 28
Concerns over medical complications from cocaine abuse have recently been amplified by police and news reports of a rapid increase in abuse of methamphetamine, illegally produced in backyard drug laboratories from pseudoephedrine and illegally imported from Mexico. To address the problem longitudinally in a large population-representative sample that would address many of the shortcomings of prior studies, we analyzed a database of all patients hospitalized from January 1, 2000, to December 31, 2003, in Texas hospitals covered by a state quality-of-care reporting law.
The Texas Health Care Information Council (THCIC) oversees the mandatory reporting of a standardized International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM)–coded discharge database for generation of quality indicators from all state-licensed hospitals, except small rural hospitals exempted by the state statute. These data have been provided to the Agency for Healthcare Research and Quality's Healthcare Cost and Utilization Project for public reporting of quality indicators since 1999. Compared with 501 Texas hospitals listed in the American Hospital Association Guide in 2003, THCIC averaged 434 hospitals reporting during 2003 (86.6%), which included approximately 91.3% (57 277/62 726) of hospital beds and 95.1% (2 439 043/2 565 861) of admissions in the state (excluding Department of Veterans Affairs and military hospitals that do not provide admissions' information to either the THCIC or the American Hospital Association).29,30 We analyzed the annual databases from 2000 to 2003, which together contained coded records for 10 834 435 hospital discharges during this 4-year period. In the target age group of 18 to 44 years, the THCIC database contained 751 568 discharges in 2000, 777 997 in 2001, 806 353 in 2002, and 812 247 in 2003.
Because the system generates publicly reported quality indicators, THCIC electronically edits records submitted by hospitals for consistency and conformity to its data collection procedures. Records failing the edits are returned to the hospital for correction, and then resubmitted. The THCIC then builds the database that each submitting hospital reviews, corrects, and certifies. The THCIC database contains the following 11 diagnosis fields: 1 admitting diagnosis, 1 principle discharge diagnosis, 8 secondary discharge diagnoses, and 1 external cause of injury field. The completeness of documentation of comorbidities for mortality adjustment in state discharge data from record systems providing 9 diagnosis fields (sensitivity, 0.926) has been shown to be virtually the same as that from record systems providing 25 discharge fields (sensitivity, 0.933).31 In the THCIC database, all 9 diagnosis fields were used in 13.5% of records in 2000 and 20.6% of records in 2003, and an average of 4.4 codes per record were recorded in 2000 and 4.9 in 2003.
Acute strokes were defined using the ICD-9-CM codes in the Agency for Healthcare Research and Quality's Inpatient Quality Indicator 17 (Table 1).32 To define strokes, we searched in only the principal discharge diagnosis field, which has been shown to maximize the sensitivity and specificity of stroke ascertainment.33 There were 1887 all-type strokes in 2000, 2097 in 2001, 2133 in 2002, and 2252 in 2003.
Strokes were further categorized as hemorrhagic, ischemic, or “other” only in the 2003 data, the only year in which the distinction was accurately coded by the hospitals. Before a coding rule change by the Centers for Medicare and Medicaid Services in October 2002, hospitals were not required to make the distinction.34 In 2003, 937 hemorrhagic and 998 ischemic strokes occurred. Analyses of secular trends during the 4 years were performed for all-type strokes combined.
We defined all conditions that might predispose to stroke (Table 1)23 from the codes in all available diagnosis fields to maximize the completeness of ascertainment. We defined abuse of a given drug by the ICD-9-CM codes for either abuse or dependence. To capture those most likely to have a close temporal relationship between stroke and substance abuse, patients with diagnoses coded as “in remission” were not categorized as substance abusers for our analysis. Nonspecific codes of drug abuse or dependence not indicating the specific drug (codes 304.8, 304.9, and 305.9) were not used.
To avoid statistical disclosure for subjects with sensitive diagnoses, the THCIC database provides age in 5 broad categories (eg, 18-44 years) and does not release sex. Other coded measures used in the analysis included in-hospital death and the patients' metropolitan statistical area, defined by the US Census Bureau35 and used as a measure of urbanity. The cause of in-hospital death was not available. Patients' race was categorized as Asian, black, white, Hispanic, or “other” (comprising patients with missing race data or those classified as American Indians, Eskimos, Aleutians, or other race); the other category accounted for 6.1% of patients.
The annual prevalence rates of abuse of specific drugs were defined as the number of discharges with the recorded drug abuse per 100 discharges (with the standard error of a proportion). The percentage change in drug abuse prevalence was the prevalence rate in the given year minus the prevalence rate in the baseline year 2000, divided by the prevalence rate in 2000 (with the standard error of a percentage change36). The incidence rate of acute stroke was the number of discharges with codes meeting the Agency for Healthcare Research and Quality's-Inpatient Quality Indicator definition for stroke per 100 000 discharges (with the standard error of a proportion). The significance of the change in rates during the 4 years was tested with the Cochrane-Armitage test for trend, which tests the null hypothesis that during the 4 years the slope of the prevalence was 0 (not changing over time). The adjusted odds ratios (ORs) for the association of risk factors with acute hemorrhagic or ischemic stroke and their 95% confidence intervals (CIs) were obtained by multivariate logistic regression. The multivariate population-attributable risk percentage was calculated from the adjusted ORs and the risk factor prevalences by the method of Bruzzi et al.37 Statistical analyses were performed with SAS statistical software, version 9.1, for Windows (SAS Institute Inc, Cary, NC).
In patients hospitalized in Texas hospitals, cocaine was reported to be the second most frequently abused drug, after alcohol, and amphetamines were the fifth most frequent (Figure 1A). While the rates of abuse of alcohol and hallucinogens did not increase during the 4 years, the rates of abuse of cocaine, cannabis, opioids, and amphetamines increased significantly (Figure 1A). Of these increases, the rate of increase was greatest for amphetamines (Figure 1B). The prevalence of amphetamine abuse in 2003 was higher in hospitals in rural (non–metropolitan statistical area) counties than in suburban or urban counties (OR, 1.40; 95% CI, 1.26-1.55; P<.001); the prevalence of cocaine abuse was lower in rural counties than in urban or suburban counties (OR, 0.72; 95% CI, 0.67-0.76; P<.001).
The incidence rate of stroke among amphetamine abusers, cannabis abusers, and hospitalized patients without any associated abuse of alcohol, cocaine, cannabis, amphetamines, opioids, or hallucinogens trended upward from 2000 to 2003 (Figure 1C). The rate of increase was greatest in stroke associated with amphetamine abuse (Figure 1D).
Multivariate logistic regression models from 2003 identified different patterns of association with hemorrhagic and ischemic stroke (Table 2). Amphetamine abuse was strongly associated with hemorrhagic stroke, but not with ischemic stroke. The strength of its association with hemorrhagic stroke was more than twice that of cocaine or tobacco use, but less than that of cerebrovascular anomalies, intracranial tumors, and hypertension. Combinations of abused drugs did not significantly contribute to the logistic regression models. Alcohol abuse has been associated with hemorrhagic stroke in prior studies,38 and trended toward significance in this model. Atrial fibrillation or flutter was a strong risk factor for ischemic stroke in univariate analysis (OR, 1.72; 95% CI, 1.04-2.85) (eTable), but it did not remain significant in the multivariate model because of strong collinearity with the “miscellaneous cardiac” variable (malignant neoplasm of the heart, acquired mural thrombus following myocardial infarction, heart valve disorder, prosthetic heart valve, and atrial septal defect).
After controlling for amphetamine and cocaine abuse, a summary indicator of “any illicit drug use” was not independently associated with hemorrhagic stroke (OR, 1.42; 95% CI, 0.93-2.18). The associations between amphetamine abuse and hemorrhagic stroke (P=.14 by the Breslow-Day test for homogeneity) and cocaine abuse and hemorrhagic (P=.55) or ischemic (P=.37) stroke did not vary by race.
Among persons aged 18 to 44 years in 2003, in-hospital death occurred in 3763 admissions (0.46%). An analysis of in-hospital death in the 2003 data showed that hemorrhagic stroke carried a much higher risk of death than that in all other hospitalized patients (OR, 58.3; 95% CI, 49.6-68.5; P<.001); in contrast, the increased mortality risk from ischemic stroke was much less (OR, 11.7; 95% CI, 8.8-15.6; P<.001). In patients with hemorrhagic strokes, only amphetamine abuse, coagulation defects, and hypertension were strong independent predictors of in-hospital death; in contrast, in patients with ischemic stroke, only acute myocardial infarction was significantly associated with death (Table 3). Repeating the analysis in patients with all types of strokes combined showed that the increased risk of death from amphetamine abuse (OR, 3.92; 95% CI, 1.79-8.59; P<.001) was greater than the increased risk from coagulation defects (OR, 3.06; 95% CI, 1.89-4.95; P<.001), and 3 times higher than that from hypertension (OR, 1.29; 95% CI, 0.97-1.73; P = .08).
If the associations are causal and unbiased, in 2003 in Texas, 14.4% of hemorrhagic strokes and 14.4% of ischemic strokes in hospitals were accounted for by abuse of drugs, including amphetamines, cocaine, cannabis, and tobacco (Figure 2).
Controlling for other risk factors, we found that amphetamine abuse was associated with twice the risk of hemorrhagic stroke as cocaine abuse. In contrast, amphetamine abuse was not associated with increased risk of ischemic stroke, while cocaine abuse was associated with an increased risk. Amphetamine abuse, but not cocaine abuse, was associated with increased risk of death after a hemorrhagic stroke. The public health implications of these findings are heightened by growing news accounts suggesting a recent increase in methamphetamine abuse, particularly in the southwestern, western, and midwestern states.39,40 This concern was supported by our finding that, among hospitalized patients in Texas from 2000 to 2003, the rate of amphetamine abuse was increasing faster than that of any other drug, including cocaine, and the rate of strokes among amphetamine abusers was increasing faster than the rate of strokes among abusers of any other drug.
Animal studies support the biological plausibility of a causal link between cocaine and amphetamine abuse and strokes. Intravenous methamphetamine, in rhesus monkeys, causes microhemorrhaging, thrombosis, infarction, poor vascular filling, and fragmentation of small arterioles and capillary beds.12,13 Methamphetamine also has been shown to exacerbate ischemic brain injury in mice.41 Cocaine causes vasoconstriction11 and disruption of cerebrovascular autoregulation in the presence of increased blood pressure.42
Past controlled epidemiologic studies in humans, beset by the difficulties of studying the medical effects of drug abuse, have not convincingly supported the link between stimulant use and stroke. Of the 4 studies providing affirmative evidence, all used the case-control design. Kaku and Lowenstein23 linked drug abuse in general to stroke, but did not separate out stimulants from nonstimulants. Petitti et al,24 using rigorous means to document exposure and outcome measures in a managed care population of women, found a strong association between stimulant use and stroke, but had relatively few strokes and did not separately quantify the effects of cocaine and amphetamines on hemorrhagic and ischemic stroke. Kernan et al27 found no overall association between phenylpropanolamine—a stimulant in many over-the-counter products—and hemorrhagic stroke, but in subgroup analysis found a strong association in women but not in men. Reanalysis of the same case-control series by Morgenstern et al28 found no overall association of ephedra-containing products with stroke, but subgroup analysis found a nonsignificant trend toward association in the group using the highest daily dose. These 2 studies21,43 proved controversial. Two studies by Qureshi et al25,26 reported the lack of association between crack cocaine use and stroke. A case-control study,25 which excluded approximately half of the stroke cases for lack of clear drug abuse history, found paradoxical protective effects for known stroke risk factors, including smoking, alcohol abuse, and diabetes mellitus. An analysis of the Third National Health and Nutrition Examination Survey found no association between cocaine use and self-reported history of nonfatal stroke.26 To our knowledge, no study has assessed the link between amphetamine abuse and stroke in the context of the recent increase in methamphetamine abuse in the southwestern, western, and midwestern states.
Our finding of an increasing secular trend in the prevalence rate of amphetamine abuse and the incidence rate of amphetamine-associated strokes in a hospital patient population can be explained by either an increase in amphetamine abusers in the community or an increasing intensity of use leading to more complications. National prevalence surveys of drug abuse show that the rate of methamphetamine abusers is highest in the western, southwestern, and midwestern states,40 but apparently did not increase during the early years of this decade.44 This suggests that the increased rate in our hospital population is because of the increased intensity of methamphetamine use. This interpretation is supported by 2 recent reports. First, among methamphetamine abusers, the percentage meeting abuse or dependence criteria of illicit drugs during the past 12 months has increased precipitously (from 27.5% in 2002 to 59.3% in 2004).45 Second, the American Association of Poison Control Centers Toxic Exposure Surveillance System46 reported a statistically significant increase in total methamphetamine-related deaths (from 13 to 23) from 2002 to 2003; deaths from cocaine abuse were virtually unchanged (from 52 to 53), and heroin-related deaths decreased (from 40 to 23).
Although case reports47- 50 have suggested a link between cannabis use and stroke, to our knowledge, this is the first controlled epidemiologic study to report a significantly increased OR for this link. Human immunodeficiency virus infection was associated with ischemic stroke in univariate analysis, but did not remain significant after controlling for the other known causes. This is consistent with prior studies51- 54 of human immunodeficiency virus and stroke.
A strength of our study is that it was done in a database representative of hospitalized conditions in all but the smallest rural hospitals in Texas. Because approximately 80% of patients who experience strokes are hospitalized,24,55 and approximately 80% of deaths within 30 days after hospital admission take place before discharge,56 our study represents the state's stroke population reasonably well. Our findings confirmed the popular view that methamphetamine abuse is more common in rural populations, whereas cocaine abuse is more common in urban populations. The large size of the THCIC database provides unusually ample statistical power to test associations with hemorrhagic and ischemic strokes. The THCIC database, although subject to misclassification of measurements, has several characteristics that are likely to have maximized the accuracy of the data. Specifically, these include state-mandated public reporting overseen by a state agency, a standardized coding protocol, electronic auditing, and rigorous edit criteria required by law, and the fact that the number of discharge diagnoses collected did not limit the sensitivity of condition reporting.31
Whereas the exact temporal link between last drug abuse and stroke was not quantified in the database, by not counting diagnoses of substance abuse in remission, we narrowed the measure of drug abuse to active drug users. Because case reports8,57,58 have documented strokes delayed by several days to months after last use, our measures of active drug abuse should be sufficient for demonstrating epidemiologic links. In some cases, a secondary diagnosis may be the consequence of stroke rather than a risk factor. For example, while atrial fibrillation is a known risk factor for ischemic stroke,59 it also can be a rare complication of ischemic stroke.60 In this study, we were unable to distinguish primary from secondary or recurrent strokes; consequently, the incidence rates and population-attributable risk percentages refer to all strokes.
Still, the major concern is misclassification of variables in a database of ICD-9-CM–coded discharge diagnoses. The occurrence of acute strokes has been shown to be accurately captured by ICD-9-CM principal discharge diagnosis in 2 validation studies.33,61 The specificity of drug abuse histories is thought to be high when it is recorded. In a multihospital study,62 when measured against urine toxicology test results, the sensitivity of self-report of cocaine use was 72%. A similar study63 in an obstetrical unit found self-report sensitivities of 58% to 70% for different drugs of abuse. One third of trauma patients in the US National Trauma Data Bank received a urine toxicology screen.64 The sensitivity of major comorbid conditions in hospital ICD-9-CM codes is generally greater than 65%.65 The prevalence rates of amphetamine and cocaine abuse in our study were similar to those measured in the National Survey of Drug Abuse,66 the California study67 of women in labor, and the study by Petitti et al.24
Misclassification of drug abuse history is an unavoidable hazard of studies of drug abuse. More to the point, however, is whether and how much such misclassification biases estimates of its associations with stroke and similar outcomes to which it predisposes. Latkin et al68 found, in a study of human immunodeficiency virus, that failure of subjects to disclose drug abuse was strongly related to a measured scale of social desirability concerns, but the level of social desirability concerns was not associated with the rate of human immunodeficiency virus infection, thus indicating a nondifferential information bias. Magder et al69 directly examined whether incomplete reporting of cocaine abuse biased logistic regression analyses of the association of cocaine abuse and stroke. Both studies concluded that the point estimates of the ORs are not seriously affected, while their CIs are broadened. If misclassification of drug abuse history behaves similarly in Texas hospitals, the ORs of our logistic regression models should be relatively unaffected by misclassification of drug abuse histories, and the broadening of their CIs is offset by the large sample sizes available for analysis. Finding strong associations of stroke with certain drugs of abuse, such as amphetamines, cocaine, and tobacco, but not with others, including opioids and hallucinogens, adds further confidence to the associations.
Correspondence: Robert W. Haley, MD, Division of Epidemiology and Preventive Medicine, Department of Internal Medicine, The University of Texas Southwestern Medical Center at Dallas, 5323 Harry Hines Blvd, Dallas, TX 75390-8874 (Robert.Haley@UTSouthwestern.edu).
Submitted for Publication: June 14, 2006; final revision received August 24, 2006; accepted September 5, 2006.
Financial Disclosure: None reported.
Funding/Support: This study was supported in part by grant R25 MH68338-01 from the National Institute of Mental Health.
Role of the Sponsor: The funding body had no role in data extraction and analyses, in the writing of the manuscript, or in the decision to submit the manuscript for publication.
Additional Information: The online-only eTable is available.
Acknowledgment: We thank John Rush, MD, for reviewing the final manuscript and making helpful suggestions; Nicole Johnson, RN, MBA, and Sylvia Cook, MA, for important background information about the database; and John Hedl, PhD, for providing the THCIC database.