Age-Specific Impact of Obesity on Prevalence and Costs of Diabetes and Dyslipidemia

Authors

  • Eric A. Finkelstein PhD,

    Corresponding author
    1. RTI International, Research Triangle Park, NC, USA;
    2. RTI-UNC Center of Excellence in Health Promotion Economics, Research Triangle Park, NC, USA;
    Search for more papers by this author
  • Derek S. Brown PhD,

    1. RTI International, Research Triangle Park, NC, USA;
    2. RTI-UNC Center of Excellence in Health Promotion Economics, Research Triangle Park, NC, USA;
    Search for more papers by this author
  • Justin G. Trogdon PhD,

    1. RTI International, Research Triangle Park, NC, USA;
    2. RTI-UNC Center of Excellence in Health Promotion Economics, Research Triangle Park, NC, USA;
    Search for more papers by this author
  • Joel E. Segel BA,

    1. RTI International, Research Triangle Park, NC, USA;
    2. RTI-UNC Center of Excellence in Health Promotion Economics, Research Triangle Park, NC, USA;
    Search for more papers by this author
  • Rami H. Ben-Joseph PhD

    1. sanofi-aventis, Bridgewater, NJ, USA
    Search for more papers by this author

Eric A. Finkelstein, RTI-UNC Center of Excellence in Health Promotion Economics, RTI International, 3040 Cornwallis Road, PO Box 12194, Research Triangle Park, NC 27709-2194, USA. E-mail: finkelse@rti.org

ABSTRACT

Objective:  To estimate the differential impact of obesity on prevalence and medical costs overall and for three major obesity-related complications (diabetes, dyslipidemia, and their joint occurrence) over the life cycle.

Methods:  The impact of obesity on age-specific medical costs and diagnosed prevalence was estimated using econometric analyses of the 2001–03 Medical Expenditure Panel Survey data. Obesity was measured using body mass index.

Results:  Obesity increases the risks for diabetes and dyslipidemia at all ages. Obesity also increases per person medical costs and the magnitude of the increase is greater at older ages. Although obese individuals represent 49% of the population with diabetes, they are responsible for 56% of total diabetes costs. They also represent 34% of the population with dyslipidemia yet are responsible for 52% of total dyslipidemia costs.

Conclusions:  These results highlight the potential savings over the life cycle resulting from effective interventions that target obesity and/or its comorbid disorders. Targeting individuals with both obesity and comorbidities is particularly important given the high medical costs associated with this subset of the obese population. Effective strategies that improve the comorbidity profile of these individuals may have the best chance of showing a positive financial return.

Introduction

Over the past several decades there has been a rapid rise in the prevalence of obesity. Currently, 31% of US adults are obese, an increase from 15% in 1976–80 and 23% in 1988–94 [1]. These increases in obesity rates have spurred corresponding growth in the prevalence of several diseases, including type II diabetes, cardiovascular disease, hypertension, osteoarthritis, several types of cancer, gallbladder disease, and sleep apnea [2].

Partly as a result of the increase in obesity rates, aggregate health spending has also ballooned. The average obese adult increases annual medical expenditures by roughly $732 (or 37%) per year; complications from obesity now cost the US medical system over $93 billion annually [3]. Nevertheless, focusing on average or aggregate prevalence and cost increases ignores the temporal impact of obesity over the life cycle. As a result of the chronic nature of obesity-related diseases, it is likely that the impact of obesity on both prevalence and annual medical costs varies over the life cycle. For example, Must et al. [4] show that prevalence ratios for select obesity-related diseases are lower among those greater than age 55 years compared to adults less than age 55 years. Contrarily, Finkelstein et al. [3] show that the increase in per person obesity-attributable medical spending is roughly twice as great for Medicare recipients than for the general population.

The goal of this analysis is to further explore the impact of obesity over the life cycle. First, using nationally representative data, we quantify the age-specific prevalence of: 1) diabetes; 2) dyslipidemia; and 3) these two conditions combined among obese and nonobese individuals and compare prevalence ratios over the life cycle for these two groups. We hypothesize that the prevalence of these conditions is likely to be similar among young individuals, regardless of body mass index (BMI, defined as weight in kg/height in m2), but will be significantly greater among obese individuals at older ages. Focusing on these conditions is important as they comprise two of the five risk factors used to define the metabolic syndrome [5] according to National Cholesterol Education Program Adult Treatment Panel III (NCEP-ATP III) guidelines. The metabolic syndrome, in turn, greatly increases the risk of coronary heart disease and other diseases [6]. The remaining risk factors, including high waist circumference and elevated blood pressure and triglycerides were not available in the data used for this analysis so we were unable to measure metabolic syndrome using all of the NCEP-ATP III guidelines.

Using the same data, we then explore the impact of obesity on age-specific per person medical costs in the general population, and among individuals with each of these three conditions. We then combine the prevalence and per person cost data to estimate aggregate medical costs for diabetes and dyslipidemia at select age intervals, and the percent of aggregate costs that accrue to those who are obese. Again, we hypothesize that as a result of the chronic nature of obesity-related diseases, the impact of obesity on annual medical costs will be small among young obese adults and will increase with increasing age. We further suggest that the impact of obesity on annual medical costs will be greater among those with precursors to the metabolic syndrome, as excess weight may simultaneously exacerbate these conditions and complicate treatment. We conclude with a discussion of the implications of these results for obesity prevention and treatment.

Methods

Data

We used the 2001–03 Medical Expenditure Panel Survey (MEPS) data for all analyses. MEPS is a nationally representative survey of the US civilian, noninstitutionalized population administered by the Agency for Healthcare Research and Quality. MEPS includes data on demographics, self-reported medical conditions, and detailed medical expenditure data reported by households. The household data are augmented with data from medical providers and insurers and professional coders convert verbatim self-reported narratives into fully specified International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes. MEPS follows each respondent over 2 years using an overlapping panel design. To obtain sufficient sample size for the disease subgroups, we pooled 3 years of data. We excluded pregnant women and those with missing data. Our final analysis sample includes 66,300 adults. Survey weighting variables were used throughout all analyses to generate nationally representative estimates for the civilian, noninstitutionalized population.

Obesity status was based on self-reported height and weight, which were converted into BMI in the MEPS data set. The analysis focused on BMI because other measurements of obesity, such as waist circumference for measurement of abdominal obesity, were not available. Normal weight included a BMI greater than or equal to 18.5 and less than 25, underweight included a BMI less than 18.5, overweight included a BMI greater than or equal to 25 and less than 30, and obese included a BMI greater than or equal to 30. To preserve respondent confidentiality, MEPS recodes almost all fully specified ICD-9-CM codes into 3-digit ICD-9 code categories. MEPS also group related fully specified ICD-9-CM codes into clinical classification categories. We used the clinical classification categories for diabetes with and without complications (codes 49 and 50, or ICD-9-CM 250.xx, 790.2, 790.21, 790.22, 790.29, 791.5, 791.6, V45.85, V53.91, V65.46) and “disorders of lipid metabolism” (code 53, or ICD-9-CM 272.0, 272.1, 272.2, 272.3, 272.4) to identify individuals with diabetes and/or dyslipidemia, respectively.

Adjusted Prevalence Model

The age-specific prevalence of diagnosed diabetes, dyslipidemia, and their joint occurrence in the MEPS data was calculated using a logistic regression to control for factors that may influence differences in the prevalence of obesity. These adjustments help to isolate the age-specific impact of obesity on the conditions of interest and minimize confounding by sociodemographic factors. We included covariates for age, age-squared, age-cubed, sex, race/ethnicity (white, African American, Hispanic), region (north-east, midwest, south, west), rural residence, income (below the federal poverty level (FPL), 100% to 199% of FPL, 200% to 399% of FPL, and 400% or greater than FPL), level of education (less than college degree, bachelor’s degree, postgraduate degree, other degree), marital status (single, married, divorced, widowed), and insurance status (covered or not covered). Predicted prevalence rates by age were then generated for a representative cohort using the regression coefficients and the mean values for all independent variables other than age.

Prevalence ratios for ages 20, 30, 40, 50, 60, and 70 years were computed by dividing the predicted proportion obese with each of the three conditions by the predicted proportion nonobese with each of the three conditions. Bootstrapped 95% confidence intervals around the prevalence ratios were estimated using 1000 simulations.

Cost Model

We estimated age-specific total annual medical costs for persons with obesity and three sets of obesity-related complications (diabetes, dyslipidemia, and their joint occurrence) using regression-based econometric techniques. Because medical cost data are highly skewed [7], simple linear regressions often fit the data poorly and naïve application of parametric models, such as the common two-part ordinary least squares model, may be affected by a variety of misspecification problems [8]. We followed the procedures in [7] and [9] to test the distribution of the cost data and select the most appropriate model. A two-part generalized linear regression model with a gamma distribution and a log link was found to be most appropriate for the cost analysis.

Each regression was run on the age 18 to 85 years population and included the same set of sociodemographic independent variables used in the adjusted prevalence equations. All cost estimates were adjusted to 2004 dollars using the medical component of the consumer price index.

To compute the costs attributable to obesity, the econometric model included binary terms for underweight, overweight, obesity, and an interaction term between age and the obesity binary variable (normal weight is the omitted reference group). This interaction allows the impact of obesity to vary nonlinearly across ages. After estimation, predicted costs were generated for a representative obese and normal weight person of each age using the regression coefficients and the mean values for the independent variables.

The cost models for diabetes, dyslipidemia, and their joint occurrence used the same econometric specification as for the costs of obesity versus normal weight. For each condition, a single model was used to estimate costs for each condition among obese persons and among nonobese persons, and costs for the general population without the condition of interest. Each model included the set of independent variables listed earlier, plus two indicator variables, one for the condition of interest interacted with obesity (BMI ≥ 30) and another for the condition interacted with nonobesity (BMI < 30) and corresponding interactions between each of these variables and age. The age interaction terms allow the impact of each condition (with and without obesity) to vary nonlinearly across ages. After estimation, predicted costs were generated as above.

Cost ratios for ages 20, 30, 40, 50, 60, and 70 years were computed by dividing the predicted per person medical costs for an obese person with each of the three conditions by the predicted per person medical costs for a nonobese person with each of the three conditions. Bootstrapped 95% confidence intervals around the cost ratios were estimated using 1000 simulations.

Lastly, using the results of the regressions and the full MEPS sample, we predicted total expenditures attributable to diabetes and, separately, to dyslipidemia in 10-year age intervals. We also calculated the percentage of these costs that accrue to obese individuals. These estimates were generated by predicting total costs for each age interval with the disease indicator variables (with and without obesity) set to one and then again with these variables set to zero. The difference between these two estimates represents the costs attributable to the disease among obese and nonobese individuals.

Results

Prevalence

Figure 1 depicts the prevalence estimates. The diagnosed prevalence of each of the three conditions (diabetes, dyslipidemia, and their joint occurrence) is shown in a separate panel. A solid line indicates the prevalence of each condition among persons who are not obese and a dashed line indicates the prevalence of the condition among obese individuals. Dotted lines indicate the overall prevalence of each condition among the population, and thus, represent the weighted average of the other two lines.

Figure 1.

The impact of obesity on the age-specific prevalence of diabetes, dyslipidemia, and both conditions.

The prevalence of all three conditions increases with age, as expected. Nevertheless, none of the conditions are prevalent at more than 5% until at least age 35 years, even among the higher risk obese population. The prevalence of diabetes rises steadily after age 30 years for each segment of the population, although the rate of increase is more rapid for obese individuals. Consequently, prevalence approaches 30% among the obese by age 65 years, whereas it is only about 12% among the nonobese population and 16% among the overall population. Both the higher prevalence and the more rapid rise among the obese are consistent with our hypothesis. Results for dyslipidemia are similar although the prevalence of dyslipidemia is relatively higher among the nonobese, especially at older ages. As a result, the difference in dyslipidemia prevalence between the obese and nonobese peaks at an eight percentage point difference around age 65 years.

Regardless of obesity status, the prevalence of both diabetes and dyslipidemia is less than half that of the prevalence of either diabetes or dyslipidemia. The impact of obesity among the joint occurrence of these conditions is substantial and increases with age, consistent with our hypothesis. Given the higher prevalence and the smaller impact of obesity on dyslipidemia, this result is likely dominated by obesity’s impact on diabetes.

The first three columns of Table 1 show prevalence ratios for all three conditions. Among those aged 70 to 79 years, for example, the prevalence of diabetes is 2.4 times greater among the obese than among those who are not obese. For diabetes, dyslipidemia, and their joint occurrence, the prevalence ratios are greatest among young adults, ranging from 4.27 for diabetes, 2.42 for dyslipidemia, and 6.94 for both conditions. Nevertheless, the overall prevalence of these conditions remains low—less than 2%—among younger ages. Beyond age 30 years the prevalence ratios decrease monotonically, with the largest decrease seen for the joint occurrence of diabetes and dyslipidemia (from 6.94 at age 20 to 2.54 at age 70). The relative impact of obesity likely decreases at older ages because individuals are more likely to develop diabetes, dyslipidemia, or both, independent of obesity, at older ages.

Table 1.  Adjusted prevalence and cost ratios for diabetes, dyslipidemia, and both conditions, comparing obese to nonobese
Age (years)Prevalence ratioCost ratio
DiabetesDyslipidemiaDiabetes and dyslipidemiaDiabetesDiabetes and DyslipidemiaDyslipidemia
  1. Prevalence and cost ratios are adjusted for age and other sociodemographic variables, as noted in the text. Bootstrapped 95% confidence intervals are shown in parentheses.

20–294.27 (3.64–5.01)2.42 (2.01–2.89)6.94 (4.82–10.23)1.19 (0.90–1.67)1.44 (1.14–1.83)1.60 (1.09–2.33)
30–393.94 (3.50–4.45)2.17 (1.89–2.47)5.75 (4.34–7.76)1.17 (0.93–1.52)1.39 (1.16–1.66)1.48 (1.10–1.99)
40–493.58 (3.29–3.90)1.92 (1.75–2.10)4.73 (3.87–5.86)1.15 (0.98–1.40)1.34 (1.19–1.54)1.38 (1.12–1.71)
50–593.15 (2.98–3.33)1.68 (1.58–1.77)3.84 (3.36–4.40)1.13 (1.00–1.29)1.29 (1.20–1.42)1.28 (1.12–1.49)
60–692.71 (2.58–2.85)1.47 (1.41–1.53)3.08 (2.77–3.41)1.10 (1.01–1.20)1.25 (1.19–1.34)1.19 (1.07–1.34)
70–792.4 (2.25–2.56)1.33 (1.26–1.40)2.54 (2.22–2.92)1.08 (0.98–1.18)1.21 (1.13–1.32)1.11 (0.95–1.30)

Costs

Figure 2 depicts the results of the cost analysis. In the upper left panel of the figure, we show per person costs for obese and nonobese individuals at each age. The incremental cost associated with obesity is small among young adults but increases dramatically in subsequent years, in part because of the increase in the prevalence of diabetes, dyslipidemia, and other complications. By age 65 years, average costs among obese individuals are nearly $1700 greater than costs among those of normal weight.

Figure 2.

The impact of obesity on age-specific medical costs overall and for diabetes, dyslipidemia, and both conditions.

As with prevalence, costs for each of the three conditions (diabetes, dyslipidemia, and their joint occurrence) are shown in a separate panel. For each condition, a solid line indicates annual costs among persons who are not obese (BMI < 30) and a dashed line indicates annual costs among obese (BMI 30+) individuals. Dotted lines indicate the overall annual costs among individuals without the condition. At all ages, costs for those with diabetes—whether obese or not obese—are substantially higher than costs for persons without diabetes. At younger ages, the nearly $2500 to $3000 additional medical costs per year are likely a result of prescription drugs, insulin, and increased rates of routine care. Among older ages, the gap between costs for those with and without diabetes widens to over $4000 per year, likely as a result of diabetes complications that are known to increase with age. Obesity increases the costs of diabetes care by about $700 per year relative to non-obese persons, but surprisingly, the impact of obesity is nearly constant with age. Although not entirely consistent with our hypothesis, a likely explanation may be that persons with diabetes are simply at greater risk of costly complications independent of their weight status. In other words, the incremental impact of obesity is smaller because many of the risks promoted by obesity are already elevated by the presence of diabetes.

The impact of dyslipidemia on per person medical costs is lower than that of diabetes at all ages, ranging between $900 and $1800 per year at younger ages and increasing slightly at older ages. Nevertheless, the impact of obesity on costs for those with dyslipidemia is larger in both absolute and relative terms than its impact on costs for those with diabetes. The cost impact of dyslipidemia alone, and of obesity on dyslipidemia, increases gradually with age, and at a slightly increasing rate.

Considering the joint impact of diabetes and dyslipidemia as precursors or proxies for the metabolic syndrome, the story is different. As expected, diabetes and dyslipidemia increase medical costs substantially. The impact of obesity on diabetes and dyslipidemia costs is large—nearly $2000 at younger ages—but decreases with age. The age-related impact of obesity is partly attenuated among this group because even those at younger ages are at significantly greater risk for costly adverse health outcomes. Although it is inconsistent with our hypothesis, it is plausible that, as for diabetes, living with both conditions makes one at risk for a variety of costly medical complications that increase over time. As a result, the marginal impact of obesity is smaller at higher ages although the absolute difference in costs remains over $1000 per year until around age 70 years.

Table 1 also shows cost ratios for all three conditions. Unlike for prevalence, there is only a small decrease in the ratios at higher ages. The decrease is largest for those with diabetes and dyslipidemia (1.60–1.11) and smallest for those with diabetes (1.19–1.08). Thus, the relative impact of obesity on per person medical costs at different ages is far more constant than the relative impact of obesity on disease prevalence. As medical costs for all individuals increase with age, the impact of obesity on costs nearly keeps pace with this overall trend.

Figures 3 and 4 combine the prevalence and per person cost estimates to show the aggregate predicted costs of diabetes and dyslipidemia, respectively, and the percentage of these costs that accrue to obese individuals. Summing across the age categories in Figure 3 reveals that the aggregate costs attributable to diabetes for individuals between the ages of 20 and 80 years are $53.4 billion, with 81% of these costs accruing among individuals who are over age 50 years. Although obese individuals represent 49% of the population with diabetes, they are responsible for 56% of the total costs. Nevertheless, their share of costs rises until age 50 years, and then decreases for subsequent age categories.

Figure 3.

Total age-specific attributable costs for the noninstitutionalized population with diabetes, by obesity status.

Figure 4.

Total age-specific attributable costs for the noninstitutionalized population with dyslipidemia, by obesity status.

Age-specific results for dyslipidemia are shown in Figure 4. The aggregate costs attributable to dyslipidemia are $39.7 billion, with 81% of these costs accruing among individuals who are over age 50 years. Although obese individuals represent 34% of the population with dyslipidemia, they are responsible for 52% of the total costs. The share of these costs that accrue to obese individuals varies between 40% and 60%.

Discussion

Consistent with prior studies, we show that the burden of obesity, diabetes, and dyslipidemia is large. We also find that the joint occurrence of these conditions is especially costly. In addition, this analysis highlights several important features that are not apparent from aggregate analyses that do not show results by age. We show that, although the prevalence of diabetes and dyslipidemia is low among younger ages, the likelihood of these conditions is substantially greater among obese individuals. Moreover, not only does the prevalence of these conditions increase with age, but the rate of increase of these conditions is more rapid among obese individuals.

For the general adult population, the increase in costs associated with obesity is relatively small at younger ages but increases dramatically as individuals reach their mid-30s and beyond, in part because of the increase in diabetes, dyslipidemia, and other diseases that obesity promotes. For those with diabetes and/or dyslipidemia, obesity significantly increases costs at every age. As a result, obese individuals are responsible for 56% of the $53.4 billion annual cost of diabetes and 52% of the $39.7 billion annual cost of dyslipidemia.

From a policy perspective, these results highlight the potential savings that would accrue over the life cycle from effective interventions that target obesity and/or its comorbid disorders. Nevertheless, they also reveal why employers and/or insurers may be reluctant to finance the costs of such treatments. With roughly one-third of the adult workforce currently obese, and costs of obesity relatively small among young adults (see the upper left panel of Fig. 2), employers may be reluctant to provide broad coverage for obesity treatments. If so, it may be to their advantage to target the subset of obese individuals who have the most to gain from effective treatments. Focusing on obese individuals with diabetes and/or dyslipidemia may be an effective strategy. This strategy offers two primary benefits. First, by limiting the pool of eligible recipients, this will reduce the total outlays required to fund the treatment. Second, as Figure 2 reveals, these individuals are associated with greater costs at every age. As a result, effective strategies that improve the comorbidity profile of these individuals may have the best chance of showing a positive financial return.

This analysis has several limitations. First, the disease prevalence and BMI data are self-reported. We are only able to show the impact of diagnosed (and self-reported) conditions, and therefore miss the impact of undiagnosed conditions; this may be important for conditions such as diabetes or dyslipidemia, which often go undiagnosed [10,11]. This should affect only the prevalence estimates, as the prevalence ratios should remain unbiased. Nevertheless, it is possible that patients with obesity may be in greater contact with the medical system to help control their weight, which could lead to additional testing for risk factors such as diabetes. The inaccuracies of self-reported height and weight are well known in the literature, although the impact on our results is unclear. Second, although our analysis by age illuminates several features not apparent in an aggregate analysis, our approach is based on cross-sectional regressions. Thus, we are unable to establish a firm casual and temporal impact of obesity on these conditions versus an association that varies with age. Additionally, the MEPS sample is nationally representative of the noninstitutionalized population, but will miss any important features of cost or prevalence in the institutionalized population. All of these limitations combine to suggest that there may be some degree of bias in the estimates.

In summary, obesity and its comorbid disorders are responsible for an increasingly large share of aggregate health spending. Effective interventions that reduce the prevalence of these conditions have the potential to improve both the health of the adult population and the financial health of their employers and insurers.

Source of financial support: This project was funded by sanofi-aventis.

Ancillary