Objective. The purpose of this study was to evaluate geographic variation in the prevalence of prescription stimulant use and predictors of use among a nationally representative, commercially insured population 5 to 14 years old.
Methods. Prescription claims activity from January 1, 1999 through December 31, 1999 for a continuously eligible population 5 to 14 years old was evaluated. Age-gender adjusted prevalence rates were estimated for each state. Multivariate logistic regression using hierarchical linear modeling was used to evaluate the impact of age, gender, number of child dependents, and region of the country on stimulant prevalence. The contextual effects of urban or rural residence, median income, percent white, and physician rate per 100 000 residents were also controlled for.
Results. The 1-year prevalence of stimulant treatment for the entire study sample was 4.2%. Multivariate logistic regression indicated that stimulant prescription use was positively associated with age, male gender, fewer child dependents, living in higher income communities, and living in communities with greater percent white. Compared with children living in the Western region of the country, children living in the Midwest and South were 1.55 (99% confidence interval: 1.28–1.87) and 1.71 (99% confidence interval: 1.42–2.06) times more likely to consume at least 1 stimulant medication, respectively. Differences in stimulant prevalence across urban and rural residence were also noted.
Conclusions. Geographic variation in the prevalence of stimulant use exists nationally, despite controlling for important predictors of use including age and gender. Possible reasons for the variation are discussed as are calls for additional research.
Stimulants are one of the most commonly prescribed medications for school-aged children.1 It is widely considered that the primary use of stimulants in children is for the treatment of attention-deficit/hyperactivity disorder (ADHD). Approximately 80% of children receiving psychotherapeutic medications for ADHD are prescribed stimulants,2–5 considered the hallmark of treatment. By far the most frequently prescribed stimulant is methylphenidate (eg, Concerta, Ritalin).2,4–6 Other stimulants used to treat ADHD include amphetamines (eg, Adderall), dextroamphetamine sulfate (eg, Dexedrine) and pemoline (Cylert).
Although stimulants have shown to be highly effective in the management of core ADHD symptoms,7 their use is not without controversy. During the 1990s concerns were expressed over the increased prevalence of use among school-aged children,8,6 the uncertainty surrounding the implications of long-term use in children,7 and studies showing 4- to 10-fold geographic variations in prevalence of use.9–11
Although studies of geographic variation in the use of stimulants are consistent in their findings of significant variation, considerable methodologic differences limit generalizability. For example, studies have been confined to a particular state (eg, Maryland10 and Michigan9,11), or in a particular population (eg, Medicaid,10 commercially insured populations,9 and all residents regardless of coverage status11). Additionally, studies vary in the age of the study sample (0–19,11 <18 years,9 and 5–14 years10), the geographic area (eg, county,11 hospital service area,9 or region10), and the specific drugs evaluated (eg, methylphenidate only,10,11 all ADHD drug therapy9). Studies also vary in the source of data (eg, stimulant use from pharmacy administrative claims data,9,10 and state triplicate prescription claims reports11) and the method of calculating prevalence (eg, age-gender adjusted,9 raw rate10,11).
The purpose of this study is to expand on the existing literature by reporting the prevalence of stimulant treatment in youths 5 to 14 years old in a nationally representative commercially insured population and identifying predictors of use in this population.
The study sample came from a database representing a random sample of Express Scripts, Inc (ESI) members who were commercially insured with integrated mail-order and network pharmacy benefits. ESI provides pharmacy benefit management (PBM) services, including network pharmacy claims processing, mail pharmacy services, and other pharmacy benefit consulting activities to small and large employer groups, third-party administrators, and managed care organizations.
From this database, prescription claims for 187 068 children continuously enrolled during all of 1999 and 5 to 14 years old as of January 1, 1999 were drawn. Excluded were members who were outliers based on prescription cap amount (ie, individual or family cap <$1000; n = 4477), copay amount (ie, brand copay >$60 or generic copay >$30; n = 3786), family size (ie, card holder plus dependents >20; n = 4), and total annual drug spend (ie, actual ingredient cost >$60 000; n = 1). The resulting sample size was 178 800.
For each child, all pharmacy claims adjudicated by ESI during 1999 were evaluated. Claims were adjusted to 30-day equivalent prescriptions; for example, a prescription claim with a 90-day supply was converted to 3 prescription claims, each with a 30-day supply. Children from all 50 states and the District of Columbia were represented in the data.
Period prevalence of stimulant treatment was defined as the proportion of sample children having at least 1 stimulant prescription claim (ie, methylphenidate, amphetamines, dextroamphetamine sulfate, and pemoline) during the year 1999.
Predictor variables included in multivariate modeling were age, gender, number of child dependents, whether the member had a deductible, and region of the country. The analysis controlled for the contextual effects of urban or rural residence, median income, percent white, and physician rate per 100 000 residents. Member age and gender were obtained from ESI eligibility files. Median income, percent white, and percent urban were obtained from the 1990 US Census files and matched at the zip code level to member eligibility records. Percent urban represents the percent of the population within a given zip code that is living in an urbanized area (UA). The Census Bureau defines an UA as a densely settled territory with a population of 50 000 or more. Percent white was measured at the zip code level and calculated as the sum of residents reporting ‘white’ as their race divided by the total population.
Physician rates were calculated using the indirect standardized method with weighting based on the number of physician visits per person per year obtained from the US Bureau of the Census as of July 1, 1998.12 Physician supply data for 2001 were obtained from the American Medical Association. These data provide number of physicians desegregated by physician specialty at the zip code level. All physician specialties except those “unspecified” were summed to obtain the total number of physicians. The physician supply rates per 100 000 residents were analyzed in relation to county, age and gender counts from the 1990 US Census. The resulting physician rates were then linked to each member by zip code using the algorithm from zipinfo.com to link patient zip code to the appropriate county.
The number of child dependents was assigned at the member level and defined as the number of eligible members within the household 18 years of age or younger. Finally, children were linked to US Bureau of the Census region using the member’s state of residence obtained from member eligibility files.
Both descriptive and multivariate analyses were conducted. Age-gender adjusted prevalence rates were calculated for each state using the direct standardization method with the overall study population used as the reference sample.13 Members were stratified into 4 different age/gender categories: girls and boys ages 5 to 10 years old and 11 to 14 years. Because of the potential to overreport or underreport use in states with small sample sizes, age-gender adjusted rates are presented only for those states with at least 1000 children 5 to 14 years old.
Three measures of variation were examined: the weighted coefficient of variation (CV), the systematic component of variation (SCV), and the interquartile ratio (IQ). The weighted CV is the ratio of the standard deviation of the prevalence rates to the mean rate among the states, weighted by the population in each state.14 Although the limitations of the CV are recognized, it is included to allow for comparisons to another study. The SCV estimates variance across states that cannot be explained by the variation within the state. A detailed explanation of the SCV is provided by McPherson and colleagues.15 The IQ represents the ratio of the rate in the state ranked at the 75th percentile divided by the rate in the state ranked at the 25th percentile. For all 3 measures, the higher the value, the greater the variation.
Multivariate logistic regression was conducted using hierarchical linear modeling (HLM). HLM or multilevel modeling controls or adjusts for the effects of group level characteristics on individual level outcomes, thereby providing more precise estimates of the standard errors.16 In this analysis, HLM was used to control for the clustering effect of members within a particular client group, where members would have similar medical benefits. Variables were deemed significant at the α = 0.01 level. HLM models were estimated using SAS system software version 8.1. Because of non-normal distribution, the number of child dependents was categorized as number of children = 1 to 3 or 4 or more children with number of children 1 to 3 as the reference category. Percent urban was dummy coded into 4 levels: 100% rural (percent urban = 0%; reference category), mostly rural (percent urban = >0% and <50%), mostly urban (percent urban = ≥50% and <100%), and 100% urban. Percent white and median income were both log transformed. Because of the curvilinear relationship between age and prevalence of stimulant treatment, an age-squared term was added to the model.
Members were located in all 50 states and the District of Columbia with just over 43% located in the South, 32% in the Midwest, 16% in the Northeast, and 9% in the Western region of the country.
The average age was ∼10 years with boys representing >51% of the sample (Table 1). Approximately 65% of all stimulant claims were for methylphenidate, approximately one-third were for amphetamines, and <2% were for pemoline. This represents a somewhat lower market-share for methylphenidate and higher market-share for amphetamines than previously reported.4,6,17 Among those with at least 1 stimulant medication claim, 13% had only 1 stimulant claim. The annual median number of stimulant claims was 4 (range: 1–22).
Prevalence of Use
For the entire sample, the unadjusted prevalence of stimulant medication use was 4.3%. As presented in Fig 1, the prevalence of use increased in both girls and boys with peak use occurring at age 11 and declining slightly thereafter. The state with the highest age-gender adjusted prevalence rate was Louisiana at 6.5 and the lowest was the District of Columbia at 1.6. Higher use is primarily concentrated in the South and Midwest regions of the country (Fig 2).
The values for the 3 measures of variation were CV = 30.7, SCV = 82.6, and IQ = 1.5. Although there are no tests to determine the statistical significance of these values, Diehr and colleagues14 estimated 95th percentile tables for the CV and SCV using an iterative resampling process for various sampling schemes and prevalence rates.14 According to their tables for a similar prevalence rate (ie, 5000/100 000 population), the CV and SCV measures found in this study fall outside the 95th percentile.
Approximately 90% of members matched to 1990 census data at the zip code level. Model findings indicate that stimulant use was positively associated with age, male gender, living in communities with greater percent white, and higher income (Table 2). In addition, compared with children living in the Western region of the country, children living in the Midwest and South were 1.55 (99% confidence interval [CI]: 1.28–1.87) and 1.71 (99% CI: 1.42–2.06) times more likely to have at least 1 stimulant claim, respectively. No significant difference in prevalence was found between children living in the Western region and children living in the Northeast. Compared with children in smaller families, children in families with 4 or more children were 26% less likely to consume at least 1 stimulant medication. Compared with children living in rural areas, children in mostly rural and mostly urban were 1.15 (99% CI: 1.01–1.32) and 1.14 (99% CI: 1.03–1.27) times more likely to have at least 1 stimulant claim. Prevalence of stimulant use was not significantly different between children living in 100% rural and 100% urban areas. Children with a deductible as part of their prescription benefit were 16% less likely to have a claim for a stimulant medication. Physician supply rate was not statistically significant at the α = 0.01 level.
At the time of our analysis, 2000 census data were not available, thus we relied on 1990 census data for income and percent urban. Although we recognize that demographic shifts occurred in the population over this time period, Geronimus and Bound18 found little difference in goodness of fit or coefficient estimates when comparing models using socioeconomic data from 1970 and 1980 census data.
In the calculation of family size we assumed that all children in a particular household would obtain their benefits from the primary cardholder or employee. In those situations in which both parents work or in situations of shared custody, there is the possibility that not all children within a household would be included on the primary cardholder’s ID. However, we expect that any errors in measurement would be random and not related to the dependent variable.
Studies have found prescription claims data to be a reliable and valid source of data.19,20 However, limitations in their use should be recognized. As in any analysis involving the use of administrative claims data, a claim does not indicate whether the patient actually took the medication. Conversely, members may receive a prescription without a claim being submitted, as in the case where the cost of the prescription is less than a member’s copay. However, we believe that these instances are rare given that the lowest average ingredient cost per prescription across all generic stimulants was $19 and the lowest ingredient cost across all single-source branded products was $40; well above the average generic and brand copays for members but less than the typical deductible.
Geographic Variation and Stimulant Use
Our findings indicate that among commercially insured children, geographic variation in the use of stimulant medications exists nationally, even after adjusting for 2 important individual predictors of use, age and gender. The degree of variation nationally is not as dramatic as the 10-fold variation in the use of methylphenidate found at the county level in Michigan.11 However, both the prevalence range, the CV and IQ found in this study is similar to that reported across hospital service areas in Michigan (prevalence range = 1.6–6.3; CV = 28.7; IQ = 1.6).9
Gender and Stimulant Use
HLM model findings suggest that boys were 3 times more likely to consume at least 1 stimulant medication than girls. The greater likelihood of stimulant use among boys compared with girls is supported in the literature, where the most recent studies reported male to female ratios of 2.6:1 to 3.8:1.2,4 Our results support findings of a narrowing of the male to female gap in stimulant use. Safer and Krager21 reported a decrease in the male to female ratio of stimulant use from 8:1 to 5:1 during the 1980s in Baltimore County schools. Follow-up studies in the 1990s also saw narrowing of the male to female gap, particularly among middle school students.8 Some contend that the narrowing represents greater use of stimulant medications among girls rather than a reduction among boys.21 One plausible explanation for greater use among girls may be increased stimulant treatment among children with attention deficit without hyperactivity, which affects girls proportionately more than boys.22 In their study of Baltimore public school students during the 1970s and 1980s, Safer and Krager23 did find an increase from 7% to 18% in the percent of those on stimulant treatment who were of the inattentive type.
Age and Stimulant Use
Throughout the 1980s, the relationship between age and stimulant prevalence typically saw a sharp increase from age 5 to peak use around age 8.21 This was followed by a sharp decrease in use such that by age 13 and 14, prevalence of use had dropped below 1%. By the 1990s, the prevalence of methylphenidate peaked roughly 1 year later, at age 9 and 10, and the drop in use among early teens was not as pronounced.11
Two differences in the relationship between age and stimulant use are of note in this study: a later age of peak use and a less dramatic decrease in use for children in the early teen years. Our findings show peak use at age 11 with only a slight decrease in use at age 14. These findings are consistent with the growing awareness of the chronic nature of ADHD and its persistence into adolescence.24,25 Also supporting persistency was the finding that among those on stimulant treatment in 1999, half had 4 or more claims during the year and 13% had only 1 stimulant claim. The annual number of claims is much higher than previously reported, in which over half of children age 3 to 17 had only 1 stimulant prescription over a 1-year period of time.26
Income and Stimulant Use
Our findings suggest that commercially insured children living in more affluent areas are more likely to use a stimulant than children from lower income areas. Although studies have shown a negative relationship between the prevalence of ADHD and income,27 the impact of income on stimulant use has been mixed (ie, no relationship,11 positive relationship,28 and a negative relationship17,29). In interpreting these findings it should be noted that the sample from which this study was drawn did not include those in the lowest income categories (ie, Medicaid) and therefore the distribution of income would not necessarily reflect that of previous studies. In addition, the interpretation of the relationship between income and stimulant use in this study is as a contextual effect rather than at the individual level. Therefore, the findings could represent community norms and expectations rather than the effect of income on demand.
Race and Stimulant Use
The relationship between race and stimulant treatment has consistently shown higher rates of use among whites than nonwhites, with rates of use 2 to 9 times greater among white children.30,31,4,10,17 Reasons for the lower use among minority children are varied, with some pointing to differences in ADHD knowledge among minority parents32 as well as evidence of unmet health needs among minority children.33
Our findings support the contention that race plays a role in the likelihood of receiving ADHD medications with the finding of a positive relationship between stimulant use and the percent of a population that is white. As with the interpretation of income, the interpretation of the impact of race on stimulant use is from a contextual effect rather than the impact of the individual’s race on the likelihood of stimulant treatment.
Child Family Size and Stimulant Use
Our findings indicate that children in households with 4 or more children under the age of 18 are less likely to consume a stimulant medication than children from families with fewer than 4 members under the age of 18. Although studies have found no significant relationship between ADHD cohorts and controls with respect to family size,27,34 the relationship between family size and stimulant use, to date, has not been evaluated. Studies of overall prescription use support the findings of a negative relationship between family size and prescription use,35,36 however, given the conflicting findings and the lack of a clear theoretical explanation for such a relationship, the role of family size and stimulant use should be further examined.
Urban Versus Rural Residence and Stimulant Use
Studies evaluating the impact of rural/urban residence on the likelihood of stimulant use report no difference in the likelihood of a child receiving stimulant treatment based on urban or rural residence.11,33 In our study, children living in areas with some proximity to urbanized areas are more likely to receive stimulant treatment. The reasons for these differences are unclear and require additional research.
Recent evidence comparing geographic variation across multiple therapy classes in adults and children has found that stimulant use in children exhibits the greatest variation.9,37 Of the 9 therapy classes (use of 7 classes in adults and use of 2 classes in children) in the Dartmouth Atlas of Health Care in Michigan, the greatest variation across hospital service areas was found in the use of ADHD drugs in children. In a nationally representative commercially insured sample, among the 24 therapy classes evaluated for adults and 5 therapy classes evaluated for children in the Express Scripts Prescription Drug Atlas, variation in stimulant use among child beneficiaries was second only to variation in cough/cold/allergy prescription use among children.
Although the reasons for regional variation in stimulant treatment among children are unknown, numerous factors may play a role in explaining this variation including differences in state controlled substance laws, anti-Ritalin campaigns, direct-to-consumer advertising, physician practice style and the values, beliefs, and expectations of adult caregivers (eg, parents, teachers, school counselors).
With respect to the impact of physician practice styles and geographic variation, two theories are worth noting in the discussion of stimulant treatment variation: the professional uncertainty theory38 and the enthusiasm hypothesis.39 The professional uncertainty theory contends that variation in medical procedures is greater for those procedures where greater uncertainty surrounds the diagnosis or treatment of the disease. In the use of stimulant therapy for the treatment of ADHD, evidence suggests that physicians vary in the information they use to diagnosis a child with ADHD,5 thereby introducing variation in the use of these agents.
The enthusiasm hypothesis contends that differences in procedure rates across geographic areas are mainly attributable to variability in the prevalence of physicians who are “enthusiasts” about the use of the procedure or service. The number of enthusiasts would increase because of statewide training programs that increase the awareness of the condition and the treatment alternatives. This too may explain variation in the use of stimulant medications where enthusiasts or, in the case of anti-Ritalin campaigns, opponents exist not only within the medical community but also among adult caregivers including teachers, school counselors, and parents.
It is important to note that this study did not attempt to determine if the higher rates of use represent overuse or the lower rates represent underuse, although we believe, as has been reported in previous studies,5,40,41 that both may be occurring. Future research should explore the reasons for this variation to reduce the risk to children from unnecessary drug therapy as well as the negative health and emotional consequences to children with untreated medical conditions.
- Received December 26, 2002.
- Accepted June 7, 2002.
- Reprint requests to (E.R.C.) Express Scripts, Inc, 13900 Riverport Dr, Maryland Heights, MO 63043. E-mail:
All authors are employees of Express Scripts, Inc (ESI), a pharmacy benefit management (PBM) company, and, as part of their employment, hold stock options.
- ↵Teitelbaum F, Martinez R, Parker A, et al. 2000 Drug Trend Report. Maryland Heights, MO: Express Scripts, Inc; 2001
- ↵Robison LM, Sclar DA, Skaer TL, Galin R. National trends in the prevalence of attention-deficit/hyperactivity disorder and the prescribing of methylphenidate among school-age children: 1990–1995. Clin Pediatr.1999;38 :209– 217
- ↵Safer DJ, Malever M. Stimulant treatment in Maryland public schools. Pediatrics.2000;106 :533– 539
- ↵Wolraich ML, Lindgren S, Stromquist A, Milich R, Davis C, Watson D. Stimulant medication use by primary care physicians in the treatment of attention deficit hyperactivity disorder. Pediatrics.1990;86 :95– 101
- ↵Safer DJ, Zito JM, Fine EM. Increased methylphenidate usage for attention deficit disorder in the 1990s. Pediatrics.1996;98 :1084– 1088
- ↵Wennberg J, Wennberg D. Practice variations and the use of prescription drugs. In: Dartmouth Atlas of Health Care in Michigan. Detroit, MI: Blue Cross, Blue Shield of Michigan and Blue Care Network. Available at: http://www.bcbsm.com/atlas/practice/shtml. Accessed February 2, 2001
- ↵Woodwell DA. National Ambulatory Medical Care Survey: 1998 Summary. Advance Data From Vital and Health Statistics, Report Number 315. Hyattsville, MD: National Center for Health Statistics; 2000
- ↵Hennekens CH, Buring JE. Epidemiology in Medicine. 1st ed. Boston, MA: Little, Brown and Company; 1987
- ↵McPherson K, Strong PM, Epstein A, Jones L. Regional variations in the use of common surgical procedures: within and between England and Wales, Canada and the United States of America. Soc Sci Med [A].1981;15(3 pt 1) :273– 288
- ↵Geronimus AT, Bound J. Use of Census-based aggregate variables to proxy for socioeconomic group: Evidence from national samples. Am J Epidemiol.1998;148 :475– 486
- ↵Kirking DM, Ammann MA, Harrington CA. Comparison of medical records and prescription claims files in documenting prescription medication therapy. J Pharmacoepidemiol.1996;5 :3– 15
- ↵Safer DJ, Krager JM. Hyperactivity and inattentiveness: school assessment of stimulant treatment. Clin Pediatr.1989;28 :216– 221
- ↵American Academy of Pediatrics, Subcommittee on Attention-Deficit/Hyperactivity Disorder and Committee on Quality Improvement. Clinical practice guideline: treatment of the school-aged child with attention-deficit/hyperactivity disorder. Pediatrics.2001;108 :1033– 1044
- ↵Sherman M, Hertzig M. Prescribing practices of Ritalin: The Suffolk County, New York study. In: Greenhill L, Osman B, eds. Ritalin Theory and Management. New York, NY: Mary Ann Liebert Inc; 1991:187–193
- ↵Eng HJ, Lairson DR. Prescribed medicines: Expenditure and usage patterns for selected demographic characteristics. J Pharm Marketing Management.1988;3 :19– 36
- ↵Rabin DL. Use of medicines: A review of prescribed and nonprescribed medicine use. Med Care Rev.1972;29 :668– 699
- ↵Motheral BR, Cox ER, Mager D, Henderson RR, Martinez R. Express Scripts Prescription Drug Atlas: A Study of Geographic Variation in the Use of Prescription Drugs, Maryland Heights, MO: Express Scripts, Inc; 2001
- ↵Chassin MR. Explaining geographic variations: the enthusiasm hypothesis. Med Care.1993;31(suppl) :YS37– YS44
- Copyright © 2003 by the American Academy of Pediatrics