#### Learning Objectives

* Briefly summarize the return-on-investment (ROI) model used for this study, including the key variables input and what is measured in the ROI calculation.

* Discuss the findings regarding the financial impact of workplace interventions for obesity, identifying the most important factors affecting the ROI of a given intervention.

* Outline the information you would provide to an employer investigating options for obesity interventions, including advice on what conditions would increase the chances of a positive ROI.

Obesity and its related complications have significant economic consequences for employers. Finkelstein et al^{1} have shown that roughly half of the $90 billion medical cost of obesity is financed by the private sector. A nationally representative 1000-person company incurs roughly $277,000 more in annual costs because of the presence of overweight and obesity.^{2}

These estimates provide an idea of the magnitude of the savings that could be achieved through interventions that successfully reduce obesity prevalence. Nevertheless, these interventions are not without costs. As a result, even successful obesity interventions may ultimately cost employers more money than they save. Many employers may be unwilling to implement obesity and other wellness initiatives that do not show a positive return-on-investment (ROI) in a reasonable time frame.

To allow employers to make more informed decisions concerning the net benefits of select obesity interventions, RTI International, with financial support from the Centers for Disease Control and Prevention (CDC), developed a toolkit that quantifies the costs and benefits of weight-based interventions. The toolkit performs two primary functions. First, using readily available company-specific inputs, the toolkit estimates the direct medical costs and the financial value of increased absenteeism that are directly attributable to obesity.

Second, the toolkit includes a module that estimates the ROI for select obesity interventions. This module requires the user to input information concerning the expected total or per capita costs of the intervention and the expected weight loss among the target weight group. Using the cost data and an algorithm that converts weight loss into annual savings in medical expenditures and reduced absenteeism, the toolkit estimates the number of years before a positive ROI is reached, if ever. This article describes the simulation model used to calculate the ROI. We also present several examples that use the model to evaluate the ROI for published workplace obesity interventions.

#### Simulation Model

The toolkit provides an estimate of the number of employees in the following weight categories using industry- or state-specific prevalence data or user-entered values: normal weight (body mass index [BMI; kg/m^{2}] <25), overweight (BMI 25 to 29.9), obese I (BMI 30 to 34.9), obese II (BMI 35 to 39.9), and obese III (BMI ≥40). Estimates by industry come from the 2005 National Health Interview Survey (NHIS); state-specific estimates are from CDC’s 2005 Behavioral Risk Factor Surveillance System. In the ROI module, users provide the target weight class for a proposed intervention: all overweight and obese employees, all obese employees, or any one of the overweight or obesity sub-categories. Users also need to provide estimates of the costs of implementing the intervention and the average weight loss expected from the intervention.

A prevalence-based approach was used to translate expected weight loss for the target population into expected changes in medical expenditures and absenteeism for the employer’s workforce. First, expected absenteeism and medical expenditures as a function of BMI were estimated. Next, nationally representative estimates of the BMI distribution in the working age population were used to estimate the number of employees at each BMI within each weight category. Then, the weight distribution of the target weight categories is shifted by the average expected weight loss of the intervention to estimate the number of employees at each BMI postintervention. Finally, the financial benefits of the intervention to the company are the difference in predicted absenteeism and medical expenditures for the employees between the initial BMI distribution and the postintervention distribution. Note that we used BMI units, as opposed to BMI categories to allow for financial benefits of weight loss that did not necessarily move people out of a particular BMI category. Details of the estimation approach are described below.

##### Absenteeism

Absenteeism as a function of BMI was estimated using the 2003 through 2005 NHIS Person files and the Adult files. The sample was restricted to those ages 18 to 64 who were employed full time the entire year and were not pregnant (*N* = 34,170). NHIS sampling weights were used to generate nationally representative estimates.

The dependent variable was annual days of work missed. We used a negative binomial model with log link to account for the fact that the dependent variable was measured in positive integers. The model included BMI, BMI squared, age, age squared, gender, race or ethnicity, education, household income, smoking status, occupation group, whether paid hourly, years of work experience, family size, alcohol consumption, and functional limitations. Expected days of work missed for each half BMI unit were predicted using mean values for the other covariates. A half BMI unit is equivalent to 3.5 pounds for a man of average height and 3 pounds for an average height woman.^{3} Appendix Table 1 presents the regression results for the absenteeism analyses. These results are similar to those presented in Finkelstein et al.^{2}

We estimated the financial costs of absenteeism by multiplying the predicted days of work missed at each half BMI unit by the average hourly cost of an employee and by 8 hours in a workday. Data on wages and benefits were drawn from the March 2005 Current Population Survey. Earnings for the sole or principal job from calendar year 2004 were scaled up to 2007 values using the Bureau of Labor Statistics Employment Cost Index. Industry-specific average hourly wages were calculated from average weekly earnings from the sole or primary job by dividing by 40 hours. Current Population Survey sampling weights were used to generate nationally representative estimates of average wages.

Estimates of total compensation, defined as wages plus the value of benefits offered to employees, were obtained by scaling average hourly wages by a benefits multiplier. The benefits multiplier is the percentage of an employee’s total compensation that is received as benefits. The national estimate for this share is 33% based on data from the Statistical Abstract of the United States. Average hourly wages were a weighted average of wages for those with benefits and wages for those without benefits, with weights equal to the percent of full-time workers with benefits in the 2005 NHIS.

##### Medical Expenditures

Data were pooled from the 2001 through 2003 Medical Expenditure Panel Survey Consolidated Data Files. The estimation sample included nonpregnant adults ages 18 to 64 who work full time as defined by working greater than or equal to 35 hours per week (*N* = 27,927). Estimates were adjusted to be nationally representative using Medical Expenditure Panel Survey sampling weights for years 2001 through 2003. All expenditure data were inflated to 2007 dollars using the Medical Care Consumer Price Index.

We used a two-part model of total expenditures to estimate the costs associated with obesity.^{4,5} The first part was a logit model for the probability of any expenditures; the second part modeled expenditures for people with nonzero expenditures using a generalized linear model with log link and gamma variance function. Expected annual per capita expenditures were predicted by multiplying the probability of any expenditures by expected expenditures conditional on having any expenditures. All regressions included BMI, BMI squared, age, age squared, gender, race or ethnicity, education, household income, smoking status, Census region, and year. Expected medical expenditures for each half BMI unit were predicted using mean values for the other covariates. The regression results are reported in Appendix Table 2. These results are again similar to those presented in prior studies.^{2,6}

##### BMI Distribution

Data from the 2003–2004 National Health and Nutrition Examination Survey (NHANES) were used to estimate the number of employees at each half BMI unit within weight category. Interventions that result in weight loss smaller than half BMI units generate no savings in the model. We used the sample of nonpregnant adults aged 18 to 65 (*N* = 3955), which was representative of the working population with respect to BMI. NHANES sampling weights were used to generate nationally representative estimates.

There were not enough observations at each half BMI unit to use point estimates directly from the data. Therefore, we assumed an empirical distribution for measured BMI and used this distribution to estimate the share of people at each half BMI unit. To sample from a normal distribution, we transformed measured BMI in the NHANES data to 1/ln (BMI); we could not reject the null hypothesis that the distribution of the transformed BMI was normally distributed. We then simulated the national BMI distribution using 10,000 draws from the normal distribution using the mean and standard deviation of the transformed BMI observed in NHANES and then retransformed each of the draws to BMI. Figure 1 shows histograms for BMI in NHANES and in our simulated distribution. The Figure makes clear that the simulated distribution simply fills out the NHANES BMI distribution.

This empirical distribution was used to calculate the proportion of people at each half BMI unit within each weight category for all BMI ranges above 25. As an example, the empirical distribution derived from NHANES indicates that 11% of working age adults in the overweight category has a BMI between 25.0 and 25.5. Suppose the toolkit reports 200 overweight (BMI 25 to 29.9) employees for a sample company. In this case, 22 employees would be assigned to be in the 25.0 to 25.5 range. The BMI distribution for the sample company would be constructed in a similar manner for the entire overweight or obese range in half BMI units based on the number of employees reported for each weight category and the empirical distribution estimated from NHANES.

##### Return-on-Investment Calculation

The financial benefits of an intervention were measured by the decrease in predicted absenteeism costs and medical expenditures after shifting the weight distribution of the target population (ie, weight categories) by the expected weight loss generated by the intervention. To capture short-term and long-term intervention effectiveness, the calculation was done separately for the average weight loss for the first year and for subsequent years. In both calculations, the original, preintervention BMI distribution was used as the reference. As an example, the user can enter that an intervention results in 7% weight loss from baseline in year one, but 3.5% weight loss from baseline in year two and in subsequent years. In this example, 50% of the weight loss is regained after the first year.

In the ROI calculation, the benefits of the intervention were compared with the costs of implementing the intervention. Intervention costs are provided by the user. Intervention costs were also examined separately for the first year and subsequent years given that some interventions (eg, those that require investments in infrastructure) may have large fixed costs and smaller costs in subsequent years. Because the focus is on ROI for employers, intervention costs were adjusted for any cost sharing by the employees (eg, coinsurance rates) but increased to include the value of any workdays missed due to the intervention (eg, weight loss surgery could be associated with several weeks of missed work). The costs of interventions that included cost sharing by employees were adjusted as follows: (cost of intervention per participant) × (1 − employee cost sharing %) × (number of participants). Interventions that included workdays missed added the following to the intervention costs: (number of days missed due to intervention per participant) × (average daily earnings) × (number of participants). This approach assumes that the employer bears the full cost of any sick leave incurred by the employee.

ROI was calculated by comparing the present value of the savings from the intervention (total medical savings + total absenteeism savings) to the present value of the company’s intervention-related costs. The present value calculations discount the value of future savings and costs at a 3% annual discount rate, which can be set by the user in the toolkit. The toolkit reports the number of years required to reach the break-even point, the point in which cumulative savings exceed cumulative costs, separately for medical costs only and for medical and absenteeism costs combined.

#### Results

We calculated the annual reduction in medical expenditures and absenteeism costs for a company with 1000 employees representative of the working U.S. population with respect to weight, wages, and benefits. Using national obesity prevalence data, this company would be expected to have 321 overweight, 191 obese I, 73 obese II, and 51 obese III employees. Based on the NHIS data, 79% are expected to receive health insurance benefits from their employer, and their average compensation, including the value of all benefits, is $34.15 per hour.

Evidence suggests that among obese individuals, weight loss of 5% to 10% of body weight from baseline reduces the risk and complications of type 2 diabetes, hypertension, hyperlipidemia, and other conditions.^{7,8} Table 1 presents estimated cost savings as a function of the percentage of baseline weight lost for each target overweight or obesity category. Each column represents a different target population for the intervention. For example, the column for overweight reports the total reductions in annual costs per person after shifting the weight distribution of overweight employees, measured in half BMI units, by the reported average weight loss. Across all overweight and obese employees, 5% weight loss results in an average reduction in annual costs (medical plus absenteeism) of $90 per person. Nevertheless, there is substantial variation in savings across weight categories: 5% weight loss among the overweight (obese III) would reduce annual costs by $60 ($160) per person. Table 1 also reveals that, beyond 5% weight loss, weight loss and cost savings are roughly linear even though our functional form did not impose a linear relationship. Table 2 presents analogous results expressed in pounds, as opposed to percentages.

Table 1 Image Tools |
Table 2 Image Tools |

Table 3 reports the savings separately for medical expenditures and absenteeism by weight loss percentage across all overweight and obese employees. Medical expenditures account for between two thirds and three fourths of the savings.

#### Representative Interventions

In this section, we model the ROI for several published interventions that provide data on weight loss and implementation costs. The ROI module in the toolkit allows for estimates tailored to the unique characteristics of a particular organization.

##### Recommended Worksite Strategies

CDC’s Community Guide reviewed the literature on nonmedical or surgical worksite strategies for obesity control and prevention and recommended a combination of nutrition and physical activity interventions.^{9} Example components included nutrition education, financial incentives, and on-site exercise facilities among others. The average weight loss pooled across seven qualifying studies in the review was 4.9 pounds (range: 4.4 to 26.4 pounds).

Two studies provided cost data for the worksite interventions.^{10,11} Erfurt et al^{10} report costs per person, updated to 2007 dollars, between $4.71 and $60.77 across four study sites, which varied in intensity of health education programs following a wellness screening. Based on the results presented in Table 2, if the lower-intensity interventions could achieve and sustain weight loss of 5 pounds at an annual cost that is less than $30 per person, these interventions would be cost saving in every year.

Brownell et al^{11} report results from worksite-based weight loss competitions. Costs per kilogram lost, updated to 2007 dollars, were reported to be between $3.75 and $9.72, with lower costs for repeated competitions. Table 2, when converted from pounds to kilograms, suggests an average savings of $13 per kilogram for small weight loss and approximately $20 per kilogram for larger weight loss. If these costs were to remain accurate for today’s worksites, and the savings are sustained once the competition ends, a strong assumption, then these strategies would also be cost saving.

##### Weight Watchers

Weight Watchers is a nonmedical, commercial weight loss program that focuses on behavioral weight control methods, and includes weekly group meetings and weigh-ins. In a clinical trial of Weight Watchers, maximum weight loss was about 5% at 6 months and about 3% at 1 year among those that completed treatment.^{12} Targeted toward all overweight and obese employees, 5% weight loss maintained for a full year would reduce medical and absenteeism costs by $90 per person per year. Three percent weight loss would reduce medical and absenteeism costs by $40 per person per year. The estimated cost to join Weight Watchers (in Philadelphia) was $35 for the first week and $12 per week thereafter, or roughly $647 per person per year. Based on the simulation results, it would not be profitable for employers to cover the full costs of Weight Watchers. Nevertheless, assuming that average weight loss is in the 3% to 5% range, a company that offered a subsidy of approximately 10% of the Weight Watchers cost should expect to break even.

##### Prescription Drug Coverage

As an example of coverage for drug treatment, clinical trial results for orlistat have demonstrated 12-month weight loss in the range of 2.5 to 3.5 kilograms (5.5 to 7.7 pounds) versus placebo.^{13,14} Using a daily dosage of 120 mg, the current retail price of prescription orlistat is $2.50 per day, or $912.50 per year.^{15} Assuming 20% co-pay for employees, the annual costs to insurers would be approximately $730 per year. Across the range of average weight loss and weight classes, anywhere from $30 to $120 of those costs would be expected to be recouped in lower medical expenditures and reduced absenteeism. On average, prescription drug coverage does not have a positive ROI.

##### Workplace Redesign

Although little evidence exists to model the expected savings associated with investments in infrastructure aimed at reducing rates of obesity, the simulation model can be used to quantify the weight loss outcomes that would be required for these investments to be cost saving. Consider a $200,000 investment in infrastructure for our 1000 person company (roughly $200 per employee). This investment could be for walking trails, changes to the cafeteria, or other investments aimed at improving weight outcomes for overweight and obese employees. Simulation results reveal that, assuming all costs are incurred in the first year, these investments would need to result in roughly 4% weight loss in the first year and sustained thereafter for a positive ROI to be realized within 5 years.

#### Discussion

The results of the simulation model indicate that even moderately effective broad-based workplace interventions would need to be relatively inexpensive for them to generate a positive ROI. This results because, although the aggregate costs of overweight and obesity are large, the potential savings associated with moderate weight loss among the general overweight or obese population is relatively modest. Behavioral interventions are generally considered successful if they can generate 5% weight loss over a sustained period of time. Whereas this reduction is expected to have both short-term and long-term health benefits, our simulation results reveal that these benefits equate to only $90 per year across all overweight or obese individuals or $160 per year for those in the highest obesity range. Although not unsubstantial, these savings are much less than the costs for many worksite programs, suggesting that they are unlikely to be cost saving for the employer to finance without significant cost sharing on the part of employees.

Positive ROI is a high threshold for interventions to meet. In fact, most medical or surgical interventions improve health but also increase costs. Interventions that reduce excess weight can still be cost effective even if they are not cost saving on net. The focus on ROI in the simulation model is in response to a request for this information from employers. As with insurers and government, not all employers may chose to use ROI as a criterion. The goal of the simulation model is to provide this information to employers so that they can make informed decisions concerning whether or not to pursue a particular weight loss intervention strategy.

This analysis has many limitations that, if addressed, might lead to different conclusions concerning the net-benefits of select obesity interventions. All reported numbers are estimates and could differ from actual company costs and benefits. The uncertainty in the estimates arises from the combination of several data sources and the fact that the parameters of the statistical analysis are themselves estimates.

The regression coefficients are based on self-reported BMI data. Unfortunately, no nationally representative data set includes measured height or weight, annual medical expenditures, and measures of absenteeism. Therefore, we were forced to rely on BMI values based on self-report. In prior analyses, we have used adjustment factors to convert self-reported height and weight to more closely match measured height and weight, but because that essentially shifts the entire BMI distribution, it has little influence on the marginal effects generated from the regression analyses.

The simulations are based on analyses of cross-sectional data. Therefore, they inherently assume that, for example, someone with a BMI of 40 who loses 10% body weight, regardless of the method of weight loss, will have the same cost profile as the average person with a BMI of 36. Whether all costs of obesity are fully reversible via weight loss, and whether certain weight loss strategies (eg, diet and exercise vs medications, vs surgery) generate more savings than others remain open questions.

The model also assumes that, for any BMI, a given weight loss has the same mean effect for all age, race, and gender strata. If a given weight loss has a larger or smaller effect for some population subsets than for others, then our estimates would be biased. Nevertheless, for tractability reasons and because it is unlikely that an employer would target interventions based on age, gender, or race (after targeting a specific BMI class), we did not incorporate this level of detail into the model. Nevertheless, interventions that successfully target high risk or costs employees, such as disease management programs for employees with comorbidities, have a greater potential for cost savings than those that target the average individual in a particular BMI group. In these cases, our estimates are likely to be conservative.

By design, the analysis addresses the costs and benefits of weight loss interventions only. Interventions that prevent weight gain would avoid future obesity-attributable costs; these “avoided” costs are not included in the analysis. In addition, the analysis focuses solely on reductions in medical expenditures and absenteeism. Obesity has also been linked to increased worker’s compensation and disability costs, reduced productivity, and increased life insurance costs.^{16} A fuller accounting of these costs may increase the ROI for select interventions.

The model also assumes a stable cohort of employees. Employers with high rates of employee turnover are likely to realize smaller gains from interventions as participants leave the company. We recommend that employers compare the expected turnover rate of the target population to the model’s predicted ROI to determine whether the firm is likely to accrue smaller benefits than are predicted by the model.

In reality, not all firms bear the total costs of increased medical expenditures and sick leave resulting from obesity. Some of these costs may be passed back to employees in the form of lower wages or increased health insurance premiums. For example, firms that do not offer sick leave, or that combine sick leave and vacation together, may not face the full costs of absenteeism. For these firms, ROI estimates that focus solely on medical expenditures may be more relevant.

Finally, the examples presented above were based on a hypothetical, representative company. Organizations whose employee and wage profile are substantially different from the average are not expected to realize the same ROI. To generate ROI estimates for these populations, CDC will make the toolkit publicly available on a CDC web page. Interested users are encouraged to quantify obesity-related costs for their organization and the potential for cost savings for select weight-based interventions.

#### Acknowledgment

This study was supported by the CDC Foundation through an unrestricted grant from Sanofi Aventis.