The Impact of Obesity on Health Service Utilization and Costs in Childhood

Authors

  • Leonardo Trasande,

    Corresponding author
    1. Department of Community and Preventive Medicine, Mount Sinai School of Medicine, New York, New York, USA
    2. Department of Pediatrics, Mount Sinai School of Medicine, New York, New York, USA
    Search for more papers by this author
  • Samprit Chatterjee

    1. Department of Health Policy, Mount Sinai School of Medicine, New York, New York, USA
    Search for more papers by this author

(leo.trasande@mssm.edu)

Abstract

Most studies of the economic costs of childhood obesity have focused upon hospitalization for comorbidities of obesity, whereas increased expenditures may also be the result of additional outpatient/emergency room visits or prescription drug expenditures. To quantify the magnitude of increased health-care utilization and expenditures among overweight and obese children, we performed descriptive, bivariate, and multivariable analyses on data from 6- to 19-year olds in the 2002–2005 Medical Expenditure Panel Survey (MEPS), a national probability survey of the noninstitutionalized civilian population in the United States. Compared with normal/underweight children, we found that children who were obese during both years of the MEPS had $194 higher outpatient visit expenditures, $114 higher prescription drug expenditures, and $12 higher emergency room expenditures. Children who were overweight during both years, or overweight in one year and obese in the other had $79 higher outpatient visit expenditures, $64 higher prescription drug expenditures, and $25 higher emergency room expenditures than normal/underweight children. Significantly, increased utilization was noted for outpatient visits, prescription drug use, and emergency room visits. Increased costs and utilization were concentrated among adolescents, though 6–11-year-old children who were obese in both years did have more outpatient visits and expenditures than other children. Extrapolated to the nation, elevated BMI in childhood was associated with $14.1 billion in additional prescription drug, emergency room, and outpatient visit costs annually. Although further research is needed to identify effective interventions, the immediate economic consequences of childhood obesity are much greater than previously realized, and further reinforce efforts to prevent this major comorbidity are needed.

Introduction

Obesity is now known to be a major cause of morbidity among American children. Diabetes, slipped capital femoral epiphysis (1), gallbladder disease (2), and obstructive sleep apnea (3) are among the conditions associated with obesity in childhood. Mental health is also significantly impacted (4). The increased presence of comorbidities in children with obesity is likely to lead to increased health-care utilization and expenditures even during the school age years and adolescence. While the health consequences of obesity are well documented, few studies have quantified the impact of obesity on health-care utilization and expenditures during childhood. Hampl et al. (5) found that children in a large pediatric integrated delivery system diagnosed with obesity during a well-child visit had $172 higher annual health-care expenditures than children with normal BMI. An analysis of the 2001–2003 Medical Expenditure Panel Survey (MEPS), identified that overweight children, defined by the American Medical Association Expert Committee and other national associations as children with a BMI in the 85–94th percentile for age and sex (6), had annual total health-care expenditures $180 higher than children with a normal BMI, whereas obese children, or children with a BMI ≥95th percentile for age and sex, had $220 higher expenditures, on average (7).

Although these analyses suggest that obesity in childhood is associated with increased health-care utilization and costs, they fail to identify the exact components of these additional costs. Earlier studies have identified that increases in in-patient costs of obesity-associated comorbidities correspond to increases in childhood obesity prevalence (8), but increased expenditures associated with elevated BMI in childhood may also be the result of additional outpatient visits or prescription drug expenditures. Estabrooks and Shetterly have found increased mental health and sick visits among overweight children in one integrated nonprofit health-care system (9), but the national impact of elevated BMI on the various types of health-care utilization and expenditures has not been quantified. The MEPS provides not only total health-care expenditure data, but also allows for analysis of the impact of BMI status on prescription drug, emergency room, outpatient visit, and in-patient expenditures in a nationally representative sample. It also permits analysis of the impact of BMI status on the number of drug prescriptions, outpatient visits, hospitalizations, and emergency room visits. We, therefore, analyzed the 2002–2005 MEPS to quantify whether obese and overweight children had higher prescription drug, emergency room, in-patient and outpatient visit expenditures, and utilization than children with a normal BMI.

Methods and Procedures

The MEPS is a national probability survey of the noninstitutionalized civilian population in the United States and is conducted annually by the Agency for Healthcare Research and Quality. Further details about the sampling and data collection approach are provided elsewhere (10). We utilized the household component of the MEPS, which collects data about health status, health system utilization, and insurance coverage (11). This analysis involved the data already collected and de-identified and was also deemed exempt from review by the Institutional Review Board of the Mount Sinai School of Medicine.

The MEPS utilizes an overlapping panel design, with each panel undergoing six interviews over a 2-year span. To represent the longitudinal impact of obesity on health-care expenditures as accurately as possible, we analyzed data from only those participants who had completed the 2-year study. We, therefore, included participants from panels 7–9 in our analysis, and excluded data from panels 6 and 10, even though these participants were part of the 2002 and 2005 MEPS, respectively. We also excluded participants who were identified as pregnant in either year.

We analyzed expenditure and utilization data for 19,613 children of 6–19 years of age for whom anthropometric data were available in both years of the MEPS. The MEPS reports calculated BMI for children of 6–17 years of age based upon parentally reported weight and height, and for >17-year olds based upon self-reported anthropometrics. During the second and fourth interviews, parents are asked to provide height and weight of their children, while >17-year olds are asked to report their height and weight during their third and fifth interviews. For children who aged from 17 to 18 during a given survey year, this can result in a child having parentally reported and self-reported BMI for that year. We categorized these children by their self-reported BMI. In categorizing BMI, we applied the American Medical Association Expert Panel definitions of overweight (85–94th percentile for age and gender) and obese (≥95th percentile for age and gender) (6).

After adjusting all expenditures to 2005 dollars using the Healthcare Consumer Price Index from the Bureau of Labor Statistics (12), we aggregated the health-care expenditures and utilization for each participant over the 2-year follow-up period. We decided to compare expenditures and utilizations across three nonoverlapping groups: (i) children who were obese in both years of the MEPS; (ii) children who were overweight in both years of the MEPS, or overweight in one year and obese in the other; and (iii) children who had normal or underweight BMI in at least 1 year of the MEPS. Sample size considerations prevented us from examining, as an independent subgroup, children who moved from normal BMI to overweight, overweight to obese, or from an elevated BMI category to a normal BMI category. Across the three comparison groups, we compared in-patient, prescription drug, outpatient visit, and emergency room facility expenditures. We also examined differences in in-patient hospitalizations, drug prescriptions, outpatient visit, and emergency room visits by BMI status. For each of these expenditures and utilization variables, we performed separate regression analyses to assess the impact of obesity among children who were in the age groups of 6–11 and 12–19 years upon entering the MEPS.

Analyses of health-care expenditure and utilization data present significant challenges, especially because zero outcomes are frequent, and the frequency of distribution of these outcomes is positively skewed. Although some researchers have used Heckman-type selection models to deal with this issue (13), the assumptions required for this modeling are rarely met (14). Two-part models are almost always the preferred choice, with a logit or probit model to predict expenditure/nonexpenditure or use/nonuse, and the second part, estimated independently of the first, to predict level of use or cost (15,16). These two-part regression analyses account for the impact that obesity has on the presence/absence of expenditures/utilization and further weigh the results obtained from a linear regression using only those observations with nonzero expenditures/utilization.

To estimate the impact of BMI status on annual health-care cost and utilization in children, we used a two-part regression model. Both parts of the model controlled for insurance status (categorized by any private insurance during the 2-year period, only public insurance during the 2-year period, and uninsured the entire 2-year period), family income as a percentage of the federal poverty line (based upon reported status during the second year of the MEPS), census region (Northeast, Midwest, South, and West), gender, and race/ethnicity (non-Hispanic white, non-Hispanic black, Hispanic, and other). Interaction among these covariates were assessed by comparing patterns of residuals against the categorical variables for each part of the two-part model. No interaction effects were observed.

The first part of our regression model identified the impact of BMI status and other predictors on the presence or absence of health-care expenditures or utilization (using logistic regression), whereas the second part identified the incremental expenditure or utilization associated with BMI status (using generalized linear models). For the latter part of the model, model fitting was performed using different combinations of normal, γ, and inverse Gaussian distributions of the dependent variable and identity, log, and power(−1) links to the independent variable. The results of these models were compared to identify which maximized goodness of fit by Akaike Information Criteria. For each health-care expenditure variable, a γ distribution with a log link emerged as the best choice. A truncated Poisson regression with identity link was utilized for all of the health-care utilization variables.

To complete our two-part regression analysis, we ran separate bootstraps (each with 1,000 iterations) for each health-care expenditure or utilization variable for the aggregate sample of children under the group of 6–19 years of age, using Stata 10.0 (StataCorp LP, College Station, TX) to apply appropriate weighting that accounts for the MEPS sampling design. To assess the impact of BMI status on different age groups, we ran separate bootstraps for children under the groups 6–11 and 12–19 years of age, separating age groups in the same way as the Centers for Disease Control and Prevention reports obesity prevalence (17). Bonferroni corrections were utilized to account for multiple comparison (using a P = 0.025 threshold to account for two simultaneous comparisons) (18), and bias-corrected 97.5% confidence intervals (CIs) are presented to represent the results of the bootstrapping. Due to the infrequency of hospitalization, the sample size of the MEPS did not permit convergence of the two-part regressions for in-patient hospitalizations or expenditures. However, results of one-part regression analysis for in-patient hospitalizations did converge for all children as well as for younger and older subgroups, whereas expenses only converged in one-part regression analysis for all children and older subgroups.

When we identified statistically significant differences associated with BMI status in the two-part regression analysis at the P = 0.025 level across all 6–19-year-old children, we extrapolated from the increment in visits or expenditures identified from this nationally representative sample to calculate the annual expenditures and health-care utilization or that can be associated with elevated BMI nationally. We multiplied the per capita increment we estimated for each elevated BMI category through the two-part regression model by the corresponding number of children, which we obtained by multiplying the percentage of the MEPS sample in each elevated BMI category and the appropriate population estimate from the US Census Bureau data of the year 2005 (19). Because we aggregated expenditures and utilization for each participant for 2 years, we then divided by two to estimate on an annual basis the national health-care expenditures and utilization associated with elevations in BMI. To identify the percentage of health-care costs and utilization associated with elevated BMI, we divided the increment in costs associated with each elevated BMI category by the mean expenditure or utilization within the appropriate subgroup of children.

Results

Table 1 presents demographic and self-reported anthropometric data on participants. Consistent with recent population prevalence data from the Centers for Disease Control and Prevention (17), one-fifth (20.54%) of participants were obese in both years, whereas another fifth (20.38%) were overweight in both years, or obese in one year and overweight in the other. A plurality of the sample resided in the South, and approximately one-third of the sample was either Hispanic or non-Hispanic black. One-third (37.17%) of the sample families lived <200% of the federal poverty line. Significant differences in health-care expenditures and utilization were identified through bivariate and one-part multivariable regression analysis of the demographic characteristics we analyzed, and are presented in Supplementary Appendix 1 online to the manuscript. Supplementary Appendix 1 also presents results of the one-part regression analyses that quantify separately the impact of elevated BMI on presence/absence of expenditures/utilization from the impact of elevated BMI on the magnitude of expenditures/utilization (given the presence of expenditures).

Table 1.  Demographic and anthropometric characteristics of 6–19-year-old children, 2002–2005 MEPS
inline image

On completion of the full two-part regression analysis, which incorporated both the impact of obesity on presence/absence of expenditures and on the magnitude of expenditures (given the presence of expenditures), we found that children who were obese during both years of the MEPS had $194 (97.5% CI: $116–338; Table 2), or 32.2% higher, outpatient visit expenditures; $114 (97.5% CI: $34–182), or 35.0% higher, prescription drug expenditures; and $12 (97.5% CI: $3–32), or 9.6% higher, emergency room expenditures than children with a normal/underweight BMI in either year. Children who were overweight during both years, or overweight in one year and obese in the next had $79 (97.5% CI: $27–163), or 13.1% higher, outpatient visit expenditures; $64 (97.5% CI: $11–271), or 19.5% higher, prescription drug expenditures; and $25 (97.5% CI: $13–148), or 19.9% higher, emergency room expenditures than children with a normal/underweight BMI in either year. Extrapolated to the nation, $7.9 billion (97.5% CI: $2.9–16.6 billion) in additional outpatient expenditures, $5.1 billion (97.5% CI: $1.8–15.3 billion) and $1.1 billion (97.5% CI: $461 million to $5.2 billion) in prescription drug expenditures were associated with obesity. In aggregate, the annual direct medical expenditures that were associated with elevated BMI in childhood amount to $14.1 billion (95% CI: $5.9–32.7 billion).

Table 2.  Results of two-part multivariable regression analysis of health-care expenditures
inline image

When expenditures were analyzed among the 6–11-year olds in the full two-part model, no significant difference in per capita emergency room or prescription drug expenditures was identified, but younger children who were obese in both years had $118 (97.5% CI: $6–274) higher outpatient expenditures than school age children who had a normal/underweight BMI in at least 1 year. Adolescents who were overweight during both years, or overweight in one year and obese in the other had $126 higher per capita outpatient visit expenditures than children with a normal/underweight BMI in either year, whereas the 12–19-year olds who were obese in both years had $218 higher outpatient visit expenditures. In this group, those who were overweight during both years, or overweight in one year and obese in the other had $237 higher per capita prescription drug expenditures (P < 0.025), whereas the adolescents who were obese in both years had $111 higher per capita prescription drug expenditures (P < 0.05 but P > 0.025). No significant differences were noted in emergency room expenditures among adolescents, though CIs for both elevated BMI groups approached significance.

Differences in health-care utilization were also detected among children with elevated BMI. On completion of the full two-part regression analysis, we found that children who were obese during both years of the MEPS had 1.96 (97.5% CI: 1.42–3.21; Table 3), or 38.3% more, outpatient visits and 1.51 (97.5% CI: 0.67–2.53), or 29.7% more, prescriptions than other children. Children who were overweight or obese during both years did not have significantly more prescriptions or outpatient visits. Extrapolated to the US population, obesity was associated with 56.6 million (97.5% CI: 41.0–92.6 million) additional outpatient visits and 43.6 million additional prescriptions (97.5% CI: 19.3–73.0 million), annually. Children who were obese in both years had 0.030, or 10.3% more, emergency room visits over the 2-year period of the MEPS, whereas children who were overweight during both years, or overweight in one year and obese in the other had 0.032, or 11.0% more emergency room visits than other children. In aggregate, 1.79 million emergency room visits (97.5% CI: 470,000–2.22 million) annually could be associated with elevations of BMI status.

Table 3.  Results of two-part multivariable regression analysis of health-care utilization
inline image

Older children in both elevated BMI subgroups had significantly more outpatient visits and prescriptions over the 2-year period than other children. Although the 6–11-year olds who were overweight during both years, or overweight in one year and obese in the other did not have more outpatient visits or prescriptions, the 6–11-year olds who were obese during both years of the MEPS had significantly more outpatient visits than other children, over the 2-year period at the P = 0.05 level. Increases in emergency room visits were not detected among age subgroups.

Discussion

This analysis further confirms that obesity contributes significantly to health-care expenditures and utilization among American children, and for the first time aggregates the national health and economic burden associated with obesity during childhood. Although earlier studies have focused largely on obesity-associated hospitalization costs, increases in outpatient visits and drug prescription appear to contribute heavily to the economic costs of obesity in children. Compared with the $14.1 billion in increased prescription drug, emergency room, and outpatient expenditures that were associated with obesity each year, costs for obesity-associated hospitalizations were $237.6 million in 2005 (L. Trasande, Y. Liu, G.E. Fryer and M. Weitzman, unpublished data).

Although past studies have identified obese and overweight adolescents as the population for which elevated BMI has quantifiable health consequences, our study is the first to identify increases in health-care expenditures and utilization among younger obese children (though the increases for utilization were only significant at the P = 0.05 level). Additional outpatient visits in this younger subgroup may reflect additional efforts to prevent obesity in adolescence, but may also reflect additional visits for comorbidities of obesity. The MEPS does not report BMI for children <6 years of age, and so we could not examine the impact of elevated BMI on health-care utilization or expenditures among children in the age group of 2–5 years, or children in the 6–23 months of age, an age group among which obesity prevalence has increased especially quickly over the past 20 years (20).

Although analysis of a nationally representative sample such as the MEPS permits extrapolation to quantify the health and economic consequences of childhood obesity, anthropometric data were available for under 20,000 children. Sample size considerations limited our capacity to examine the impact of the shift from normal BMI to overweight, overweight to obese, or incremental increases/decreases of Z-scores over time within an elevated BMI category. Our efforts to control for insurance status, age, socioeconomic status, gender, and race/ethnicity also limited our power to detect differences within age subgroups, especially with relatively infrequent events such as emergency room visits, or to control for other confounders such as health status. Further study using a larger data set, such as the one from a large national insurer, would also yield more accurate measurements of the impact of elevated BMI on health-care utilization. A larger data set would also permit more detailed comparison of frequency or number of hospitalizations among children in different BMI categories than we could execute, and identification of the medical conditions or diagnoses that contributed to increases in health-care utilization and expenditures among children with elevated BMI. Studies have not assessed the validity of parentally or self-reported BMI in children, and analysis of a larger data set with anthropometrics measured by a health-care provider would confirm whether this potential bias is significant.

The increases in health-care expenditure and utilization also represent a small subset of burden that can be attributed to childhood obesity. We could not quantify indirect costs such as lost school and workdays associated with comorbidities of childhood obesity. Because children who are obese are at a greater risk of adult obesity (21,22,23,24), a portion of economic costs associated with adult obesity may also be attributable to childhood obesity.

Effective interventions to prevent childhood obesity still remain elusive (25), in large part because obesity represents an interplay among individual behaviors; community structure, lifestyle and the “built environment;” and environmental factors that may have the capacity to disrupt energy balance (26). Although prevention programs may be more expensive than the immediate costs saved, the long-term benefits obtained through prevention of adult cardiovascular disease, diabetes, stroke, and possibly cancer are likely much greater than the costs we identified in this analysis. Indeed, increases in per capita annual health-care expenditures among overweight and obese adults can reach as high as $2,000 in elderly populations (7). Potential interventions should be assessed with a full cost-benefit analysis that takes into account the short- and long-term benefits. Nonetheless, the immediate economic consequences of childhood obesity are much greater than previously realized, and further reinforce efforts to prevent this major comorbidity.

SUPPLEMENTARY MATERIAL

Supplementary material is linked to the online version of the paper at http:www.nature.comoby

Acknowledgments

We thank Justin Trogdon and Eric Finkelstein of Research Triangle Institute for technical guidance in bootstrapping the regressions used in this manuscript.

Disclosure

The authors declared no conflict of interest.

Ancillary