The costs of publication of this article were defrayed, in part, by the payment of page charges. This article must, therefore, be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Moderate and Severe Obesity Have Large Differences in Health Care Costs
Article first published online: 6 SEP 2012
2004 North American Association for the Study of Obesity (NAASO)
Volume 12, Issue 12, pages 1936–1943, December 2004
How to Cite
Andreyeva, T., Sturm, R. and Ringel, J. S. (2004), Moderate and Severe Obesity Have Large Differences in Health Care Costs. Obesity Research, 12: 1936–1943. doi: 10.1038/oby.2004.243
- Issue published online: 6 SEP 2012
- Article first published online: 6 SEP 2012
- Received for review March 24, 2004; Accepted in final form October 18, 2004
- health care use;
- health care expenditures;
- body weight class
Objective: To analyze health care use and expenditures associated with varying degrees of obesity for a nationally representative sample of individuals 54 to 69 years old.
Research Methods and Procedures: Data from the Health and Retirement Study, a nationwide biennial longitudinal survey of Americans in their 50s, were used to estimate multivariate regression models of the effect of weight class on health care use and costs. The main outcomes were total health care expenditures, the number of outpatient visits, the probability of any inpatient stay, and the number of inpatient days.
Results: The results indicated that there were large differences in obesity-related health care costs by degree of obesity. Overall, a BMI of 35 to 40 was associated with twice the increase in health care expenditures above normal weight (about a 50% increase) than a BMI of 30 to 35 (about a 25% increase); a BMI of over 40 doubled health care costs (∼100% higher costs above those of normal weight). There was a difference by gender in how health care use and costs changed with obesity class. The primary effect of increasing weight class on health care use appeared to be through elevated use of outpatient health care services.
Discussion: Obesity imposes an increasing burden on the health care system, and that burden grows disproportionately large for the most obese segment of the U.S. population. Because the prevalence of severe obesity is increasing much faster than that of moderate obesity, average estimates of obesity effects obscure real consequences for individuals, physician practices, hospitals, and health plans.
Over the past several decades, obesity rates have been increasing dramatically for all population groups (1, 2, 3). The most important consequences of this growth in unhealthy body weight are increases in obesity-related medical problems, most notably type II diabetes, hypertension, cardiovascular disease, and disability (4, 5, 6, 7, 8, 9). The adverse health effects of obesity have resulted in increased health care use and expenditures (8, 10, 11, 12, 13, 14, 15). The effects of obesity on health care costs already exceed those of smoking or problem drinking (16). In aggregate terms, obesity accounts for ∼6% to 10% of national health care expenditures in the U.S. and 2.0% to 3.5% in other Western countries (17, 18, 19, 20, 21, 22, 23, 24).
Although the overall health care costs associated with obesity are interesting, such aggregate numbers obscure important differences across degrees of obesity. The prevalence of clinically severe obesity among U.S. adults is increasing at a much faster rate than the spread of obesity (25, 26). Between 1986 and 2000, the prevalence of self-reported BMI over 40 (∼100 pounds overweight for a man of average height) quadrupled from ∼1 in 200 adult Americans to 1 in 50, whereas obesity based on self-reported BMI of over 30 roughly doubled during the same period, from ∼1 in 10 to 1 in 5 (25). Because obesity-related medical problems are more common among severely obese than moderately obese individuals, one would expect commensurate differences in health care costs. Yet, few data on the impact of severe obesity on medical expenditures are available.
This paper seeks to fill this gap in the literature by analyzing health care use and expenditures associated with varying degrees of obesity for a nationally representative sample of individuals 54 to 69 years old. On average, obesity (BMI ≥ 30) is associated with health care expenditures that are about one-third above medical costs of otherwise similar individuals of normal weight (16, 17). The question remains, however, whether such averages obscure important differences across weight classes that would be of interest to private health plans and Medicare alike.
Research Methods and Procedures
This study used three waves of the Health and Retirement Study (HRS),1 a national biennial panel survey of community-dwelling adults 51 to 61 years old at baseline (1992) and their spouses, designed to study the issues of retirement, health insurance, savings, and economic well-being of the population approaching the retirement age. The initial 1992 panel included 12, 562 respondents from 7702 households (82% response rate), of whom 9824 participants were age-eligible. Follow-up telephone interviews were conducted in 1994, 1996, 1998, 2000, and 2002, with proxy interviews after death. The survey oversampled Hispanics, blacks, and Florida residents and provided weighting variables to make it representative of the community-based population nationwide (27). More information about the survey design is available online (http:hrsonline.isr.umich.edu).
Extensive background information in the HRS survey provided a rich set of control variables for our analysis. The study contained detailed information about demographics, family structure, income, health status, cognition, functional limitations, health care use and expenditures, employment status and history, housing, insurance, and expectations of the survey participants. The HRS data were particularly useful for this study because the study included a wide range of items to assess health status and functioning of individuals. In addition, the HRS focused on the preretirement age group—a population at high risk for development of chronic health problems and disability.
Although the survey started in 1992, total medical expenditure data were collected from 1996 on. We pooled the data from the survey waves in 1996, 1998, and 2000, so that the initial sample included 23, 770 observations on 8762 unique individuals. We imposed a number of sample restrictions. First, we excluded individuals with large weight changes (>10%) between two waves (10% of the original sample). In the HRS, major weight loss was typically accompanied by substantial deterioration in health and increases in health care costs. Moreover, individuals reducing weight by >10% between baseline and the next period were more likely to die in the following 4 years than individuals with stable weight (odds ratio of 3.2 at p < 0.01). Thus, we focused on stable-weight individuals because using current weight for older adults might have obscured estimation of the links between obesity and health care outcomes. Second, we excluded individuals with a BMI under 18.5 (1.0% of the original sample) because this group was too small to provide precise estimates. The final exclusion (1.4% of the original sample) was for data problems such as discrepancies in medical spending and use data (e.g., reported health care use but no health care expenditures). After all exclusions, 7971 individuals remained eligible for analysis, contributing 19, 648 observations.
The main dependent variables were health care costs and three measures of health care use. Health care costs were measured as the total medical expenditures, inflated to 2002 dollars using the medical care component of the consumer price index. We used health care expenditure data from the HRS files developed by RAND (available to registered HRS users at http:hrsonline.isr.umich.edudataindex.html). In the HRS, average annual total medical spending was estimated to be $4119 for adults 56 to 68 years old (the age group of the 1998 study), which is consistent with $4375, the estimate of total medical expenditures for a comparable age group in the Medical Expenditure Panel Survey (28).
The three health care use measures were based on a 2-year recall and included the reported number of outpatient visits, an indicator for any inpatient stay, and the number of inpatient days over the period of 2 years. Use of services directly reflecting dependence or disability was too low to include them as outcome variables: special health facilities (4% of the sample), home health care (3%), and nursing home care (<0.5%).
The primary explanatory variable was a measure of weight class based on self-reported height and weight: normal (BMI 18.5 to 24.9), overweight (BMI 25.0 to 29.9), moderate obesity (BMI 30.0 to 34.9), severe obesity (BMI 35.0 to 39.9), and extreme obesity (BMI ≥ 40.0).
Other explanatory variables included in all models were age (5-year increment groups of 60 to 64 and 65 to 69; 54 to 59 was the reference), race/ethnicity (black, Hispanic, other; white was the reference), insurance status (public, private; uninsured was the reference), marital status (married, divorced/separated, widowed; never married was the reference), educational achievement (college, some college, high school; less than high school was the reference), annual family income on the log scale (inflated to 2002 dollars using the consumer price index), region (Northeast, Midwest, West; South was the reference), and survey wave indicators. In addition, we used current tobacco smoking and heavy alcohol drinking (three and more drinks per day) to account for the effects of behavioral risk factors on health care use and costs. Because our intention was to evaluate the effects of varying degrees of obesity on health care use and costs, including the change in use and costs because of the increased risk of many chronic diseases related to obesity, health conditions or general health were not considered as predictors in the model. Table 1 reports descriptive statistics of the analysis sample by weight class.
|Normal weight: 18.5 ≤ BMI <25||Overweight: 25 ≤ BMI <30||Moderate obesity: 30 ≤ BMI <35||Severe obesity: 35 ≤ BMI <40||Extreme obesity: BMI ≥ 40||Total sample|
|Share of respondents||0.34||0.42||0.18||0.04||0.02||1.00|
|Mean age (years)||61.3||61.2||61.2||61.1||60.6||61.2|
|Less than college||0.22||0.19||0.19||0.18||0.16||0.20|
|Less than high school||0.17||0.20||0.24||0.30||0.29||0.21|
|Annual family income in 2002 ($)||60, 930||61, 563||51, 979||42, 432||48, 069||58, 558|
|Number of outpatient visits||7.0||7.8||8.7||11.9||15.0||8.0|
|Number of inpatient days||1.06||1.19||1.31||2.00||2.82||1.24|
|Probability of any inpatient stay||0.15||0.18||0.21||0.25||0.32||0.18|
|Total health care costs per year in 2002 ($)||4089||4270||5075||5967||8224||4503|
|Number of observations||6416||8308||3556||932||436||19, 648|
We pooled the HRS data across the three waves with total medical expenditure data to increase the sample size. All analyses accounted for the complex sampling design of the HRS using information on the survey weights, strata, and primary sampling units as implemented in survey data estimation commands in Stata 7.0 (Stata Corporation, College Station, TX). The Huber/White nonparameteric correction was used to adjust SEs for repeated observations on the same individuals.
In the analysis of health care expenditure data, we used a linear estimation model on the logarithmic scale to address the skewness of expenditure data among health care users (i.e., there is a small number of people with extremely high expenditures) (29). We tested whether the single equation log-linear model would provide a better fit to our data than the two-part model by using a split-sample sign test (29). The log-linear model produced a lower mean squared forecast error than the two-part model in ∼350 of 400 replications. The average mean squared forecast error across all replications was significantly lower in the log-linear model than in the two-part model (p < 0.01). The fact that we had almost invariably positive expenditure data (3.5% of the sample were zero outcomes) may explain better performance of the log-linear model over the two-part model in model comparison tests.
To convert estimates of log health care expenditures back to the dollar scale, we used a nonparametric homogenous smearing retransformation (29). The homogenous smearing retransformation provides a biased estimate of the response of interest when the log scale residuals are heteroscedastic in a way that depends on the covariates (30, 31). This issue is of less concern in our data because there is no heteroscedasticity by the main variable of interest, i.e., weight class. For the analyses of health care use data, we estimated a linear model for the continuous outcomes (i.e., number of outpatient visits and hospital days) and a probit model for the dichotomous outcome (i.e., indicator of any inpatient stay). Poisson (count) models were considered for outcomes such as the number of outpatient visits and inpatient days, but they provided the same qualitative results.
Based on specification tests for interactions between weight class and sociodemographic covariates, we estimated all models separately for men and women but not by age, race/ethnicity, or educational attainment. We checked the robustness of our results to the exclusion of outliers in health care expenditures [as identified by the method of Hadi (32, 33) to detect multiple outliers in multivariate data at p < 0.05].
Health Care Expenditures
Table 2 reports the average predicted values of annual health care expenditures (adjusted for differences in sociodemographics and health risk behaviors) by weight class among men (top) and women (bottom) 54 to 69 years old. Increasing weight class was strongly associated with higher health care costs for both men and women (significantly increased above normal weight at p < 0.05, adjusted for differences in sociodemographics). For men, the health care costs in 2002 dollars increased from $3915 per annum in the normal weight class to $8017 in the extreme obesity class (p < 0.01), with similar increases for women ($3991 for normal weight vs. $8440 for extreme obesity, p < 0.01).
|Normal weight: 18.5 ≤ BMI <25||Overweight: 25 ≤ BMI <30||Moderate obesity: 30 ≤ BMI <35||Severe obesity: 35 ≤ BMI <40||Extreme obesity: BMI ≥ 40|
|Average total health care expenditures per year in 2002 ($)||3915 (3604 to 4254)||4561* (4298 to 4840)||4738 (4319 to 5198)||6179* (5112 to 7470)||8017 (5636 to 11, 404)|
|Increase above the preceding weight class ($)||646||177||1441||1837|
|Relative to the preceding weight class||1.17 (1.14 to 1.19)||1.04 (1.00 to 1.07)||1.30 (1.18 to 1.44)||1.29 (1.10 to 1.53)|
|Average total health care expenditures per year in 2002 ($)||3991 (3758 to 4239)||4365* (4106 to 4640)||5085* (4636 to 5577)||5723 (4853 to 6749)||8440* (6854 to 10, 392)|
|Increase above the preceding weight class ($)||374||720||638||2717|
|Relative to the preceding weight class||1.09 (1.09 to 1.10)||1.16 (1.13 to 1.20)||1.13 (1.05 to 1.21)||1.47 (1.41 to 1.54)|
There was, however, a strong nonlinearity in the effects of varying degrees of obesity on health care costs for both men and women. For men, health care expenditures increased most dramatically between moderate and severe obesity ($1441, p < 0.05). In comparison, the costs grew by $646 (p < 0.01) with a movement from normal to overweight, and the difference in health care costs of overweight men and moderately obese men was particularly small ($177, p = 0.48). The costs between severely and extremely obese men were not significantly different ($1837, p = 0.20).
For women, the increase in health care expenditure between classes of obesity was also nonlinear, yet the rate of change between classes of obesity was different. For example, the largest growth in health care expenditure for women was predicted between severe and extreme obesity ($2717, p < 0.01). It was notably above the increases in health care costs of women between the moderately obese and severely obese classes ($638, p = 0.50), and the difference in health care costs between overweight and moderately obese women ($720, p < 0.05). Finally, the smallest expenditure increase was predicted between normal-weight and overweight women ($374, p < 0.05).
As the results described above show, there was a difference by gender in how costs changed with obesity class. For men, the increase in health care costs between overweight and moderate obesity was the smallest of all changes across weight classes, whereas for women, this effect exceeded the increase in costs between normal weight and overweight. At the same time, health care expenditures did not change significantly for women when they progressed from moderate to severe obesity, whereas that effect was quite large for men.
Health Care Use
Table 3 presents the results for health care use. The table shows the average predicted values of the use measures over a 2-year period (adjusted for differences in sociodemographics and health risk behaviors) by weight class for men (top) and women (bottom) 54 to 69 years old. Overall, higher obesity class was associated with increased use of outpatient health care services for both men and women. Paralleling expenditures, the effects of increasing weight class on outpatient visits were nonlinear. For both men and women, the largest increase in outpatient care use was seen between moderate and severe obesity [a 38% increase for men (p < 0.01) and a 34% increase for women (p < 0.01)]. The smallest difference in the number of outpatient visits for men was between overweight and moderate obesity (4%, p = 0.50), whereas for women the same classes differed by 15% (p < 0.05). There was no significant difference for either men or women in the number of outpatient visits between the severe and extreme obesity classes.
|Normal weight: 18.5 ≤ BMI <25||Overweight: 25 ≤ BMI <30||Moderate obesity: 30 ≤ BMI <35||Severe obesity: 35 ≤ BMI <40||Extreme obesity: BMI ≥ 40|
|Number of outpatient visits||6.7 (4.8 to 8.7)||7.6* (5.8 to 9.4)||7.9 (6.0 to 9.9)||10.9* (8.4 to 13.5)||13.6 (9.4 to 17.9)|
|Relative to the preceding weight class||1.13 (1.08 to 1.21)||1.04 (1.03 to 1.05)||1.38 (1.36 to 1.40)||1.25 (1.12 to 1.33)|
|Any inpatient stay (%)||18.2 (13.9 to 23.2)||19.5 (15.2 to 24.4)||21.7 (16.8 to 27.4)||23.6 (16.9 to 31.7)||32.7 (21.2 to 46.0)|
|Relative to the preceding weight class||1.07 (1.05 to 1.09)||1.11 (1.10 to 1.12)||1.09 (1.01 to 1.16)||1.39 (1.25 to 1.45)|
|Number of inpatient days||1.4 (0.6 to 2.2)||1.5 (0.7 to 2.2)||1.5 (0.7 to 2.3)||2.2 (0.9 to 3.5)||3.2 (1.1 to 5.4)|
|Relative to the preceding weight class||1.04 (1.00 to 1.19)||1.00 (1.00 to 1.04)||1.47 (1.29 to 1.52)||1.45 (1.22 to 1.54)|
|Number of outpatient visits||7.1 (5.7 to 8.5)||8.4* (6.9 to 9.9)||9.7* (8.2 to 11.3)||13.0* (10.7 to 15.3)||15.8 (12.2 to 19.5)|
|Relative to the preceding weight class||1.18 (1.16 to 1.22)||1.15 (1.14 to 1.19)||1.34 (1.30 to 1.35)||1.22 (1.14 to 1.27)|
|Any inpatient stay (%)||13.3 (10.1 to 17.3)||16.6* (12.8 to 21.0)||19.9* (15.4 to 25.2)||25.7* (19.8 to 32.5)||32.6 (24.4 to 41.7)|
|Relative to the preceding weight class||1.24 (1.22 to 1.27)||1.19 (1.18 to 1.20)||1.29 (1.28 to 1.29)||1.27 (1.23 to 1.28)|
|Number of inpatient days||0.9 (0.4 to 1.4)||1.0 (0.5 to 1.5)||1.2 (0.7 to 1.7)||1.9† (1.1 to 2.6)||2.8 (1.8 to 3.8)|
|Relative to the preceding weight class||1.12 (1.08 to 1.29)||1.20 (1.13 to 1.40)||1.58 (1.53 to 1.60)||1.47 (1.46 to 1.58)|
Overall, there was a strong positive association between increasing weight class and the probability of any inpatient stay for women but not for men. For women, any inpatient stay was 19% more likely between overweight and moderate obesity classes (p < 0.05), whereas the difference between severe and extreme obesity was 27% (p < 0.10). No significant differences across weight classes were found for men.
Finally, we found very few significant increases in the number of hospital days due to increasing weight class. For men, there were no statistically observable differences in the number of inpatient days associated with an increase in weight class. For women, the only significant increase in the number of inpatient days was found for the movement from moderate to severe obesity (58%, p < 0.10).
Finally, our sensitivity analyses suggested that the results were robust to the exclusion of outliers. In the sample without outliers, we found that the increases in health care expenditures associated with varying degrees of obesity were significant for all weight classes. Although the general patterns were the same, as expected, the relative changes were smaller for the higher degrees of obesity than those presented in Table 2. For example, severe obesity was associated with 51% higher health care expenditure relative to normal weight for men in the sample without outliers and 58% in the primary analysis sample.
The estimates presented in this paper showed that there were large differences in obesity-related health care costs by degree of obesity. We found a strong nonlinearity in the effects of varying degrees of obesity on health care expenditures and use of outpatient health care services for both men and women. There were, however, differences by gender in how health care use and costs changed with obesity class.
Overall, a BMI of 35 to 40 based on self-reported height and weight was associated with twice the increase in health care expenditures above normal weight (about a 50% increase) than a BMI of 30 to 35 (about a 25% increase), and a BMI of over 40 doubled costs (∼100% higher health care costs than normal weight). Widely cited average effects of obesity on health care costs obscure major differences across degrees of obesity and will underpredict future health care costs because the prevalence of more severe obesity is growing at a much faster rate than the spread of obesity in general (25, 26). We found that the average increase in health care costs associated with a BMI of 30 and higher (averaging across all obesity classes) for individuals 54 to 69 years old was 33%, almost identical to estimates in the literature (36% to 37%) based on other data sets and methodological approaches (16, 17).
There are some study limitations that should be noted. First, self-reported height and weight generally underestimate BMI, so our prevalence of severe obesity was lower than in objectively measured data (e.g., the National Health and Nutrition Examination Survey). This leads to higher estimated effects than if objectively measured BMI were used because “true” BMI is higher. In addition, the age range covered in the HRS sample is limited. Yet, this age group is arguably the most relevant for analyses of our type because the onset of the health effects of obesity is most likely to occur toward the end of middle age. Along the same lines, the HRS sampling frame excludes the institutionalized population, which may bias our results downward. The health care expenditure data are imprecisely measured. We find, however, that average expenditures in the sample are similar to estimates for the same age group from the Medical Expenditure Panel Survey. Finally, our estimates are essentially cross-sectional in nature and cannot be interpreted as causal effects.
Despite these limitations, the results from our study provide important information regarding the large increases in health care use and costs associated with varying degrees of obesity and nonlinearity of these effects. From a policy perspective, the population we studied—individuals nearing retirement age and those recently retired—deserves special attention in the analysis of obesity effects for implications of their transition into Medicare and Social Security programs. Obesity imposes an increasing burden on the health care system, and that burden grows disproportionately large for the most obese segment of the U.S. population. Because the prevalence of severe obesity is increasing much faster than that of moderate obesity, average estimates of obesity effects obscure real consequences of obesity in the U.S.
Nonstandard abbreviation: HRS, Health and Retirement Study.
- 5World Health Organization. (1998) Obesity: Preventing and Managing the Global Epidemic. Technical Report Series 894. World Health Organization: Geneva, Switzerland.
- 111998) What is the economic case for treating obesity? Obes Res. 6(Suppl 1): 2S–7S.(
- 172003) National medical spending attributable to overweight and obesity: how much, and who's paying? Health Aff. (Suppl W3): 219–226., , (
- 231995) The impact of obesity on health status: some implications for health care costs. Int J Obes Relat Metab Disord. (Suppl 6): S13–S16.(
- 27Accessed August 18, 2004) Technical Description of the Health and Retirement Survey Sample Design: University of Michigan, Ann Arbor (1995). http:hrsonline.isr.umich.edudocssamplehrsHRSSAMP.pdf., (
- 28MEPS, Center for Cost and Financing Studies, Agency for Healthcare Research and Quality. (Accessed March 9, 2004) Medical Expenditure Panel Survey, 1996-2000. http:www.meps.ahrq.gov.
- 291983) A comparison of alternative models for the demand for medical care. J Bus Econ Stat. 1: 115–126., , , (
- 321992) Identifying multiple outliers in multivariate data. J R Stat Soc Ser B. 54: 761–771.(
- 331994) A modification of a method for the detection of outliers in multivariate samples. J R Stat Soc Ser B. 56: 393–396.(