PEDIATRICS Vol. 109 No. 5 May 2002, pp. 728-732
Factors Associated With Prescription Drug Expenditures Among Children: An Analysis of the Medical Expenditure Panel Survey
* Department of Pediatrics, UCLA School of Medicine, Los Angeles, California
Division of Cardiology, Department of Pediatrics, Harbor-UCLA Medical Center, Torrance, California
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
Objective.Pharmaceutical costs have reached 14% of total health care costs in the United States and continue to rise. Many studies have looked at factors that influence utilization of hospital and ambulatory care services in the pediatric population. This study examines the factors that influence utilization of prescription drugs in the pediatric population.
Methods.Data from the 1996 Medical Expenditure Panel Survey (MEPS) were used in the analysis. A 2-part multivariate regression analysis was conducted using pediatric (ages 0–17) prescription drug expenditures as the dependent variable. Independent variables were constructed using demographic variables, socioeconomic variables, health status, and medical conditions.
Results.Black children are less likely than white children to use any prescription drug (odds ratio: 0.67). Similarly, uninsured children are less likely than privately insured children to use any prescription drug (odds ratio: 0.62). Among children who had any prescription drug expenditure in 1996, children who are black, Asian, and Hispanic had lower prescription drug expenditures than children who are white. Children who are uninsured had lower expenditures than children who are privately insured. Children in near-poor families had lower prescription drug expenditures than those in high-income families, even after controlling for insurance status. Children who are covered by Medicaid had comparable prescription drug expenditures to children who are covered by private insurance.
Conclusion.Socioeconomic characteristics such as race, insurance status, and family income levels had significant impact on pediatric prescription drug expenditures, even after controlling for the influences of health status and medical conditions.
Key Words: prescription drug • expenditure • children • Medical Expenditure Panel Survey
Abbreviations: MEPS, Medical Expenditure Panel Survey • HMO, health maintenance organization • ADL, activities of daily living • OR, odds ratio
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
The cost of prescription drugs has risen to approximately 14% of total health care expenditures in the United States, accounting for 43% of the total increase in health care costs in 2000.1,2 Because of its economic implications, it is now more important than ever to understand the pattern of prescription drug expenditures.
Previous studies have demonstrated that variations in the pattern of use of hospital and ambulatory care services among children are attributable to predisposing socioeconomic characteristics such as race and gender and enabling factors such as family income and insurance status.3,4 One might hypothesize that prescription drug expenditure for children would be subjected to these factors. However, few studies have examined the relationship between these factors and prescription drug use in the pediatric population.5,6
A study by Hahn6 used the 1987 National Medical Expenditure Survey (the predecessor of Medical Expenditure Panel Survey [MEPS]) to examine the relationship between ethnicity and prescription drug expenditure in children and found disparities in the use of prescription drugs among various racial/ethnic groups. Disparities that are associated with other factors such as insurance status and family income were not examined. In anticipation of the development of cost-containment measures and emphasis on appropriate prescription drug use, which will most likely occur in the upcoming years to slow rising pharmaceutical costs, characterizing these disparities in the pediatric population is of utmost importance. The release of the 1996 MEPS made it possible for such analyses to be conducted using the most recent and comprehensive expenditure data. The characterization of disparities in prescription drug expenditure among children would enable policy makers to be mindful of the vulnerability of certain pediatric population groups when implementing new prescription drug policies and/or cost-containment measures.
The goal of the present study was to isolate the independent effects of race, insurance status, and family income on influencing prescription drug expenditures for children, after controlling for confounding factors such as health status, medical conditions, and other socioeconomic factors.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
Data Source
Data from the Household Component File and the Medical Conditions public use file from the 1996 MEPS database were analyzed in the study. The MEPS was designed to provide nationally representative estimates of expenditures, health services utilization, insurance coverage, insurance type, and sources of payment for the current US civilian noninstitutionalized population.7 The survey is conducted approximately every 10 years by the Agency for Healthcare Research and Quality. The Household Component File collects detailed data on approximately 10 000 families and 24 000 individuals across the nation on demographic characteristics, health status, health conditions, use of medical services, charges and payments, income, employment status, and health insurance coverage. A stratified multistage area probability design was used in the survey. In addition, ethnic minorities and low-income families were oversampled. This was done because of the policy interests of the Agency for Healthcare Research and Quality.
The survey was fielded with interviews and questionnaires; separate instruments were used for adults and children (age 0–17 years). Children aged 0 to 17 years (N = 6856 observations) were used for the present study.
The Medical Conditions public use file contains 76 426 records. Each record represents 1 household-reported medical condition during 1996. Each condition can be identified using International Classification of Diseases, Ninth Revision, Clinical Modification codes. A confidential personal identification number links the condition to the person.
Dependent Variable
The total prescription drug expenditure is the dependent variable used in the analysis. Expenditure is defined as the amount paid for health care services from all payment sources, including out-of-pocket expenses. Over-the-counter medication is not included in the prescription drug expenditure. The data were collected jointly by a household questionnaire and the pharmacy component survey, cross-referenced by medical records and insurance provider data.
Independent Variables
The independent variables were constructed using Andersen’s Behavioral Model of Health Services Utilization,8 which is the most frequently used model for modeling health services utilization.3,6,9–11 It is a social model that attempts to capture all socioeconomic and behavioral factors that may influence a person’s decision to use health care. It consists of 3 main components: 1) predisposing social characteristics, 2) enabling factors, and 3) the need for health care. Predisposing social characteristics are demographic factors, such as age, gender, and race, that have known correlation with health care utilization. Enabling factors are factors, such as family income, insurance status, education status, and region of residence, that are associated with known patterns of health care utilization. The need for health care consists of factors, such as health status, medical conditions/diagnosis, functional limitations, and disabilities, that determine the health care needs of an individual.
Predisposing social characteristics include age, gender, and race (represented by indicator variables Asian, Native American, Hispanic, black, and other; white is the reference group).
Enabling factors include family income level (based on the 1996 US Census Poverty Level, represented by the following indicator variables: <125% of poverty line as poor; 125%–199% as low-income; 200%–399% as middle income; and 400% and greater as wealthy; wealthy group is the reference group), maternal education level (represented by indicator variables high school diploma, bachelor’s degree, master’s or above; less than high school education is the reference group), region (represented by indicator variables south, northeast, midwest; west is the reference group), metropolitan statistical area status, and insurance status (represented by indicator variables private non–health maintenance organization [HMO], HMO, Medicaid, other public insurance, and uninsured; private non-HMO is the reference group).
The need for health care include activities of daily living (ADL) limitations, instrumental ADL limitations (both ADL and instrumental ADL limitations are specified to be attributable to physical and mental disabilities), presence of any general limitation, resisting illness well, less healthy than other children, catches colds frequently, perceived general health status, perceived mental health status (proxy-reported health status was used in the pediatric population; proxy-reported health status is the health status of a child as reported by a proxy such as a parent or guardian), pharyngitis, chickenpox, gastroenteritis, tinea, otitis media, allergic rhinitis, pneumonia, influenza, asthma, and eczema (these conditions were selected because they are common pediatric conditions that may lead to the use of prescription drugs).
All independent variables were constructed using Andersen’s behavioral model. The primary independent variables of interest are race, insurance status, and family income level. Other independent variables control for confounding.
Statistical Analysis
The Rand Health Insurance Experiment 2-part regression model was conducted using STATA statistical software for windows (version 7; College Station, TX) for analysis.12,13 The 2-part model was used because health care expenditure distribution (ie, prescription drug expenditure distribution) tends to be skewed rather than normally distributed with a significant number of people who have no expenditure at all. Ordinary linear regression is not adequate in handling such a skewed distribution, and the 2-part model is generally accepted as the model of choice.10–17
Part I: Logistic Regression
Any child with prescription drug expenditure greater than 0 dollars for 1996 is defined as a user (1 = yes, 0 = no). A logistic regression analysis was performed on user as the dependent variable to predict the probability of use (Pr(Any Use)). Independent variables consisted of all of the independent variables listed in the Independent Variables section.
Part II: Multivariate Linear Regression
Only children who were designated as user were selected and modeled in the multivariate regression analysis using ln(Total Prescription Drug Expenditure) as the dependent variable, along with all independent variables specified above. This regression model predicts the amount of use given that one is a user, which is designated E(Expenditure/Any Use).
Combining the 2 Parts
The 2-part model was put back together to obtain the predicted expenditure or E(Expenditure) by the following equation: E(Expenditure) = Pr(Any Use) multiplied by E(Expenditure/Any Use). Smear factors were used to retransform the estimates. (As ln(Total Prescription Drug Expenditure) was used in the multivariate regression as the dependent variable, the results obtained by the multivariate regression is distorted by a factor, when converting the results from the log [ln] scale back to the original scale, smear factor adjusts for the distortion.)18,19 Clustering was adjusted at the household level using STATA (version 7).20,21 All regression procedures were weighted using person-level weights provided by MEPS to reflect the US pediatric population.
Ordinary linear regression was also performed using Total Prescription Drug Expenditure as the dependent variable, along with all independent variables specified earlier (as with the 2-part model). The estimates were compared with the 2-part model estimates for any large discrepancies.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
Weighted descriptive statistics for the sample are summarized in Table 1. The mean age was 8.5 years (standard deviation: 5.2). The mean family income was $46 381. The proportion of children with any prescription drug expenditure was 57%. Forty-eight percent of the population were female. Unadjusted differences in prescription drug expenditure by race, insurance status, and family income level are presented in Table 2. As described in Table 2, children who were white tended to have higher unadjusted prescription drug expenditures than black, Asian, Native American, and Hispanic children. Other patterns were also notable: the wealthy seemed to spend almost twice as much as the poor or low-income children on prescription drugs. Similarly, those who have any insurance coverage spent nearly 3 times as much as the uninsured. The range of prescription drug expenditures was wide, ranging from $0 to $27 840 in 1996 for the sample population. This also accounts for the wide standard deviations in Table 2. Table 3 summarizes the results from the logistic regression analysis on any prescription drug use. The full logistic regression model controlled for all confounding variables as discussed earlier, namely, demographic variables, region, maternal education, health status, and medical conditions. The model has a pseudo-R2 of 0.21 and a P value of .0001. The most notable results in Table 3 are the odds ratio (OR) for blacks compared with whites and the OR for uninsured compared with privately insured reference group. The OR is 0.67 for blacks and 0.62 for uninsured. Both are statistically highly significant (P < .001).
|
|
|
Among children who had any prescription drug expenditure, Table 4 summarizes the results of the multivariate regression analysis. The multivariate linear regression (second part of the 2-part model) has an adjusted R2 of 0.19 and a P value of < .001. Again, after controlling for confounding variables, we noted that blacks, Asians, and Hispanics spent less on prescription drugs than did whites (ß coefficients are negative). In addition, it was not surprising to find that uninsured children spent less than those who had private insurance (ß = -0.36). Similarly, those who were poor or of low income spent less on prescription drugs than those who were wealthy (ß = -0.16, -0.24, respectively). All were statistically significant at P < .05.
|
Summarizing the multivariate linear regression results, the major predictors of prescription drug expenditures are race (black and Asian), family income level (poor and low-income), and insurance status (uninsured). It is important to note that low-income children were more likely to be vulnerable in prescription drug expenditures than were poor children because those who have low incomes but do not qualify for Medicaid probably have a harder time paying for prescription drugs than those who are poor and eligible for Medicaid.
Finally, combining the 2-part logistic regression and multivariate linear regression analyses (multiplying the probability of prescription drug use Pr(Any Use) with the amount of spending E(Expenditure/Any Use)), we predicted the following prescription drug expenditures adjusted for confounders (Table 5)13,14,22–25: blacks spent an average of $44.03 per child in 1996 compared with $79.12 for whites; poor and low-income children spent an average of $55.01 and $46.80, respectively, compared with $97.16 for the wealthy in 1996; and uninsured children had a predicted average expenditure of $26.41 compared with $72.45 for privately insured children. Although, not as precise, ordinary least square multivariate analysis showed similar trends in prescription drug expenditure as predicted by the 2-part model.
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
The results of our analyses revealed that race/ethnicity, insurance status, and family income level influence prescription drug expenditure in children, based on the data from the 1996 MEPS. These influences exist even after controlling for need variables such as health status, medical conditions, and other socioeconomic variables.
Our results were consistent with other studies, which looked at the disparity in health services use among different racial/ethnic, insurance, and income groups.3,6,25–31 Similar to Hahn’s study, our study demonstrated that black children are particularly vulnerable to underexpenditure when it comes to prescription drug spending.6 Hahn did not examine the effects of insurance status on prescription drug expenditures. We found that uninsured children are especially vulnerable to underexpenditure. These findings are consistent with other pediatric inpatient and ambulatory service expenditures and adult prescription drug studies.3,4,29,32–35 On the basis of a recent literature review, our findings on the effects of race/ethnicity, insurance status, and family income level on prescription drug expenditures provide a more comprehensive picture than currently available studies. We believe that such findings can provide more complete and accurate information for policy makers when making decisions about prescription drugs, especially in the context of the recent literature focusing on the unmet health care needs of children.29,34,35
We also found that maternal education is not a statistically significant factor in influencing prescription drug expenditure in children after health status and medical conditions are included in the model (P = .14 for postcollege education, P = .57 for college, and P = .63 for high school education compared with less than high school education reference group). Past studies have demonstrated that maternal education can be an important factor in health care expenditure in children; however, those studies were not as effective in controlling for health status and medical conditions as can be done by a study using the MEPS database.6
It is important to acknowledge 3 limitations in the current study. 1) There may still be omitted-variable bias. For example, the race variables white, black, Hispanic, Asian, Native-American, and other race do not adequately capture the cultural and personal preferences of people who are categorized into these groups. Race and ethnicity in themselves may lead to prejudices that influence prescribing pattern and decreased prescription drug spending; however, there are preferences and beliefs associated with race/ethnicity that play an important role in being prescribed a medication and in getting a medication. 2) Another major category of variables that can significantly influence prescription drug expenditure but is not included in our analysis is the group of variables that specify prescription drug benefits within different insurance types. These variables are not available in the MEPS Household Component File for incorporation into our analysis. 3) This study does not address the question of whether those groups that had lower prescription drug expenditure were underprescribed or appropriately prescribed. This study also does not address the question of where the disparity occurs. Is it at the physician level (ie, prescribing pattern) or at the individual level (ie, preferences, compliance, etc.)? Undoubtedly, there is disparity in prescription drug expenditures as presented earlier. However, the underlying causes of disparity remain to be identified.
|
CONCLUSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
Socioeconomic characteristics such as race/ethnicity, insurance status, and family income levels have significant individual effects on pediatric prescription drug expenditures even after adjusting for the influences of health status, medical conditions, and other socioeconomic factors. Future studies should attempt to examine the underlying causes of the disparity in prescription drug use among children.
|
ACKNOWLEDGMENTS |
|---|
Dr Chen is a Robert Wood Johnson Clinical Scholar at UCLA School of Public Health.
We thank the UCLA Clinical Scholars Program staffs for being extremely helpful throughout the study period.
|
FOOTNOTES |
|---|
Received for publication Oct 17, 2001; Accepted Jan 9, 2002.
Reprint requests to (A.Y.C.) UCLA Robert Wood Johnson Clinical Scholars Program, 911 Broxton Ave, Ste 301, Los Angeles, CA 90024. E-mail: alexycchen{at}yahoo.com
The views expressed in this article represent those solely of the authors and do not necessarily represent those of the Robert Wood Johnson Foundation.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
- Thornton D. By the numbers. The rising costs of prescription drugs. Healthplan.1999; 40 :71 –72[Medline]
-
Charatan F. US spending on prescription drugs rose by 19% in 2000.
BMJ.2001; 322
:1198
[Free Full Text] -
Butler JA, Winter WD, Singer JD, Wenger M. Medical care use and expenditure among children and youth in the United States: analysis of a National Probability Sample.
Pediatrics.1985; 76
:495
–507
[Abstract/Free Full Text] -
Wood DL, Hayward RA, Corey CR, Freeman HE, Shapiro MF. Access to medical care for children and adolescents in the United States.
Pediatrics.1990; 86
:666
–673
[Abstract/Free Full Text] - Hong SH, Shepherd MD. Outpatient prescription drug use by children enrolled in five drug benefit plans. Clin Ther.1996; 18 :528 –545[CrossRef][Medline]
-
Hahn BA. Children’s health: racial and ethnic differences in the use of prescription medications.
Pediatrics.1995; 95
:727
–732
[Abstract/Free Full Text] - Cohen JW, Monheit AC, Beauregard KM, et al. The Medical Expenditure Panel Survey: a national health information resource. Inquiry.1996; 33 :373 –389
- Aday LA, Andersen RM. Equity of access to medical care: a conceptual and empirical overview. Med Care.1981; 19 :4 –27[CrossRef][Medline]
- Hakkinen U. The production of health and the demand for health care in Finland. Soc Sci Med.1991; 33 :225 –237
- Manning WG, Bailit HL, Benjamin B, Newhouse JP. The demand for dental care: evidence from a randomized trial in health insurance. J Am Dent Assoc.1985; 110 :895 –902[Abstract]
- Van Doorslaer E, Wagstaff A, Van Der Burg H, et al. Equity in the delivery of health care in Europe and the US. J Health Econ.2000; 19 :553 –583[CrossRef][Medline]
- Brook RH, Ware JE Jr, Rogers WH, et al. Does free care improve adults’ health? Results from a randomized controlled trial. N Engl J Med.1983; 309 :1426 –1434[Abstract]
- Manning WG, Newhouse JP, Duan N, et al. Health insurance and the demand for medical care: evidence from a randomized experiment. Am Econ Rev.1987; 77 :251 –277[Medline]
- Duan N, Manning W, Morris C, Newhouse J. A comparison of alternative models for the demand for medical care. J Bus Econ Stat.1983; 1 :115 –126
- Grootendorst PV. A comparison of alternative models of prescription drug utilization. Health Econ.1995; 4 :183 –198[Medline]
- Newhouse JP, Manning WG, Morris CN, et al. Some interim results from a controlled trial of cost sharing in health insurance. N Engl J Med.1981; 305 :1501 –1507[Abstract]
- Van Doorslaer E, Wagstaff A, Calonge S, et al. Equity in the delivery of health care: some international comparisons. J Health Econ.1992; 11 :389 –411[CrossRef][Medline]
- Duan N. Smearing estimate: a nonparametric retransformation method. J Am Stat Assoc.1983; 78 :605 –610[CrossRef]
- Manning WG. The logged dependent variable, heteroscedasticity, and the retransformation problem. J Health Econ.1998; 17 :283 –295[CrossRef][Medline]
- Zeger SL, Liang KY. Longitudinal data analysis for discrete and continuous outcomes. Biometrics.1986; 42 :121 –130[CrossRef][Medline]
- Norton EC, Bieler GS, Ennett ST, Zarkin GA. Analysis of prevention program effectiveness with clustered data using generalized estimating equations. J Consult Clin Psychol.1996; 64 :919 –926[CrossRef][Medline]
- Newhouse JP, Manning WG, Duan N, et al. The findings of the rand health insurance experiment—a response to Welch et al. Med Care.1987; 25 :157 –179[CrossRef][Medline]
- Van Doorslaer E, Wagstaff A, Bleichrodt H, et al. Income-related inequalities in health: some international comparisons. J Health Econ.1997; 16 :93 –112[CrossRef][Medline]
- Wagstaff A, Van Doorslaer E, Van Der Burg H, et al. Equity in the finance of health care: some further international comparisons. J Health Econ.1999; 18 :263 –290[CrossRef][Medline]
-
Wells KB, Manning WG, Duan N, Newhouse JP, Ware JE Jr. Use of outpatient mental health services by a general population with health insurance coverage.
Hosp Community Psychiatry.1986; 37
:1119
–1125
[Abstract/Free Full Text] - Ali S, Osberg JS. Differences in follow-up visits between African American and white Medicaid children hospitalized with asthma. J Health Care Poor Underserved.1997; 8 :83 –98[Medline]
- Blixen CE, Havstad S, Tilley BC, Zoratti E. A comparison of asthma-related healthcare use between African-Americans and Caucasians belonging to a health maintenance organization (HMO). J Asthma.1999; 36 :195 –204[Medline]
- Ozminkowski RJ, White AJ, Hassol A, Murphy M. Minimizing racial disparity regarding receipt of a cadaver kidney transplant. Am J Kidney Dis.1997; 30 :749 –759[Medline]
-
Rosenbach ML, Irvin C, Coulam RF. Access for low-income children: is health insurance enough?
Pediatrics.1999; 103
:1167
–1174
[Abstract/Free Full Text] - Hafner-Eaton C. Physician utilization disparities between the uninsured and insured. Comparisons of the chronically ill, acutely ill, and well nonelderly populations. JAMA.1993; 269 :787 –792[Abstract]
-
Wells K, Klap R, Koike A, Sherbourne C. Ethnic disparities in unmet need for alcoholism, drug abuse, and mental health care.
Am J Psychiatry.2001; 158
:2027
–2032
[Abstract/Free Full Text] - Lillard LA, Rogowski J, Kington R. Insurance coverage for prescription drugs: effects on use and expenditures in the Medicare population. Med Care.1999; 37 :926 –936[CrossRef][Medline]
-
Poisal JA, Murray L. Growing differences between Medicare beneficiaries with and without drug coverage.
Health Aff (Millwood).2001; 20
:74
–85
[Abstract/Free Full Text] -
Mouradian WE, Wehr E, Crall JJ. Disparities in children’s oral health and access to dental care.
JAMA.2000; 284
:2625
–2631
[Abstract/Free Full Text] - Simpson G, Bloom B, Cohen RA, Parsons PE. Access to health care. Part 1: children. Vital Health Stat 10.1997; 196 :1 –46
PEDIATRICS (ISSN 1098-4275). ©2002 by the American Academy of Pediatrics
This article has been cited by other articles:
 |
|
 |
|
  J. Wang, J. M. Noel, I. H. Zuckerman, N. A. Miller, F. T. Shaya, and C. D. Mullins Disparities in Access to Essential New Prescription Drugs between Non-Hispanic Whites, Non-Hispanic Blacks, and Hispanic Whites Med Care Res Rev, December 1, 2006; 63(6): 742 - 763. [Abstract] [PDF]
|
|
|
|
 |
|
 C. R. Gresenz, J. Rogowski, and J. J. Escarce Dimensions of the Local Health Care Environment and Use of Care by Uninsured Children in Rural and Urban Areas Pediatrics, March 1, 2006; 117(3): e509 - e517. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. A. Mohanty, S. Woolhandler, D. U. Himmelstein, S. Pati, O. Carrasquillo, and D. H. Bor Health Care Expenditures of Immigrants in the United States: A Nationally Representative Analysis Am J Public Health, August 1, 2005; 95(8): 1431 - 1438. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. A. Lesesne, S. N. Visser, and C. P. White Attention-Deficit/Hyperactivity Disorder in School-Aged Children: Association With Maternal Mental Health and Use of Health Care Resources Pediatrics, May 1, 2003; 111(5): 1232 - 1237. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. Silverstein, A. E. Sales, and T. D. Koepsell Health Care Utilization and Expenditures Associated With Child Care Attendance: A Nationally Representative Sample Pediatrics, April 1, 2003; 111(4): e371 - 375. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
