Abraham, Jean M. PhD; Nyman, John A. PhD; Feldman, Roger PhD; Barleen, Nathan BS
* Outline the key elements of UPlan Fitness Rewards Program (FRP) incentive-based exercise initiative at University of Minnesota, including the structure of the incentive program.
* Identify the approximate proportions of employees participating and earning the FRP incentives.
* Summarize the findings on how the FRP initiative affected medical expenditures, including factors affecting the study conclusions.
An active lifestyle, including regular physical activity, is an important factor for reducing the risk of being overweight or obese—an outcome that affects 68% of the adult population and almost half of full-time workers in the United States.1,2 Regular physical activity lowers a person's risk of developing or dying from heart disease, Type 2 diabetes, stroke, osteoporosis, and certain forms of cancer,3 and it reduces depression, anxiety, and sleeping problems.4–6 Regular exercise also is associated with lower rates of hospitalization7 and lower health care spending among older insured adults.8–10 Despite these benefits, almost half of the US adult population does not exercise regularly, according to the 2009 Behavioral Risk Factor Surveillance System (www.brfss.cdc.gov, 2010).
Many employers are implementing wellness programs to encourage healthier lifestyles and to address concerns about rising costs. Today, approximately 74% of all firms that offer health insurance as a fringe benefit to their employees offer at least one wellness program.11 Exercise initiatives are among the wellness programs most commonly offered. A key issue for employers is whether investment in an incentive-based exercise program yields a positive return on investment. That is, can a wellness program motivate employees to make sustained changes in behavior, and does this decrease medical expenditures enough to offset program costs?
Although many studies have evaluated the overall impact of employer-based wellness programs, fewer have considered the effectiveness of particular initiatives.12,13 Only one recent study has investigated the impact of an incentive-based exercise program on changes in health care costs. Lu and colleagues14 analyzed the impact of a program sponsored by IBM that offered $150 to US employees who self-reported engaging in at least 20 minutes of physical activity 3 days per week for 10 of 12 weeks. The authors found that average annual health care costs for participants increased by $291 compared with an increase of $360 for nonparticipants (a marginally significant difference). This study has two potential limitations. First, the program offered by IBM was less conventional than other exercise initiatives because it was completely Internet based. Second, employees were asked to self-report physical activity, which may not correspond to actual physical activity.
This study evaluates whether participation and regular exercise in conjunction with the UPlan FRP, a new incentive-based exercise initiative launched by the University of Minnesota in January 2008, is associated with a decrease in medical care expenditures in the initial year of implementation. Section 2 describes the methods, including the data, measures, and empirical model. Section 3 presents the results of the multivariate analyses. Finally, section 4 concludes.
The University of Minnesota is a large public university with major campuses in Minneapolis, St Paul, and Duluth. The university has about 18,000 employees, of whom 17,348 were enrolled in the “UPlan” self-insured health plan option in 2008. The UPlan offers comprehensive benefits through two nonprofit carriers: HealthPartners and Medica. In 2008, 7126 employees were enrolled in HealthPartners and 10,222 in Medica.
Like most employer-sponsored insurance programs, the UPlan has experienced rising costs, leading university administrators to seek ways to contain cost growth by encouraging covered employees and their dependents to manage illnesses more effectively and to adopt healthy behaviors. On January 1, 2008, the University began offering the FRP to eligible employees and dependents across the state. The FRP is a significant addition to a set of existing voluntary health promotion programs, which include online health risk assessments, lifestyle coaching, disease management, and a 10,000 steps walking program.
The FRP provides a monthly credit toward fitness center membership dues if an individual exercises at a participating facility at least eight times per month. The FRP has three eligibility requirements: (1) enrollment in a Medica or HealthPartners UPlan health insurance product; (2) aged 18 years or older; and (3) membership in one of approximately 300 participating fitness centers in the health plans' networks. Participating fitness centers include several chains (eg, YMCA, Lifetime Fitness), independent centers, and university-based recreation facilities.
A HealthPartners member can get up to $20 per month if he or she visits a participating fitness center at least eight times per month. If a dual or family gym membership is held, a second adult UPlan family member can earn up to $20 per month by exercising. Medica enrollees can earn up to $40 per month if two individual adult gym memberships are held, but only up to $20 per month with a single, dual, or family gym membership. The “up to” designation stems from a provision that restricts the FRP incentive from exceeding the full amount of the monthly fitness center dues.
Enrollment in the FRP is ongoing throughout the calendar year and is straightforward. Contingent on membership at a participating fitness center, an individual fills out a form and provides evidence of health plan enrollment. As of January 2010, enrollment in the FRP was estimated to be 6905. Approximately half of those who have signed up for the FRP meet the exercise frequency requirement each month to receive the credit.
Several data sources were used. First, we obtained permission from the UPlan to access administrative and claims data for all eligible UPlan enrollees for 2006 to 2008. This includes 2 years prior to FRP implementation (2006 to 2007) and 1 year postimplementation (2008). The eligibility data include an encrypted identifier for each enrollee, date of birth, sex, and health plan enrollment, whereas the medical claims data cover inpatient, outpatient, and pharmacy claims. For the latter, fields include an encrypted enrollee identifier, date of service, diagnosis (International Classification of Diseases, Ninth Revision, Clinical Modification) and procedure codes (Current Procedural Terminology-4), place of service, allowed charges, and amounts paid by the plan and enrollee (eg, co-payments).
We supplemented these data with information provided by the UPlan's wellness vendors. We obtained annual data capturing the presence of chronic conditions among UPlan enrollees. Chronic condition indicators were constructed from medical claims and enrollees' responses to questions on the annual health risk assessment. In addition, we obtained information on enrollees' participation in the university's other voluntary health promotion programs, including disease management, lifestyle management, and the 10,000 steps programs. Finally, we obtained data from the health plans on monthly fitness center utilization for each individual who signed up for the FRP during 2008. These data include an enrollee's fitness center location and the number of times he or she used the facility for each month they signed up.
The study population included all university employees and adult family members covered by the UPlan from 2006 through 2008 for whom complete information from the other sources was available.
Equation 1 is the baseline model for the effect of FRP participation on average monthly medical spending:
The unit of observation is the ith UPlan enrollee in year t. The dependent variable is the ith individual's average monthly medical care expenditures in year t (Mit). Expenditures are the sum of plan-paid and enrollee-paid spending, inflated to 2009 dollars.
In this baseline specification, we define one “cohort” variable to represent whether a UPlan enrollee signed up for the FRP in 2008 (FRP Participant). This variable stands for permanent but unobserved personal characteristics that affect his or her spending in all years—even before signing up for the FRP. These characteristics might include a predisposition toward behaviors associated with wellness (eg, belief in the efficacy of exercise). The interaction of this cohort variable and year 2008 captures the FRP “intervention.” Obtaining a negative and statistically significant estimate of the coefficient φ4 would support the hypothesis that FRP participation leads to a reduction in medical care expenditures in 2008, on average, relative to the reference category of nonparticipants. The model also controls for time (year 2007 and year 2008). The coefficients φ2 and φ3 represent the time trend of medical spending that affects FRP participants and nonparticipants alike.
We include several other enrollee attributes (Xkit) in the model. The βk coefficients estimate the effects of these attributes on medical expenditures. The variables are age (yr), gender (female), whether an individual is an employee (reference is adult family member), enrollment in a Medica health plan (reference is HealthPartners), and indicators for whether the enrollee participated in the UPlan's disease management, lifestyle management, or 10,000 steps program, which existed prior to the launch of the FRP. We also include indicators corresponding to whether or not an enrollee reported having each of the following chronic conditions in a given year: diabetes, asthma, heart disease, arthritis, depression, osteoporosis, musculoskeletal conditions, low back pain, or migraines.
Because the FRP is a voluntary program, an important issue to consider is the potential confounding effect of acute health shocks. For example, an enrollee who is diagnosed with a serious acute condition may be less likely to participate in the FRP and more likely to have high medical expenditures. Not controlling for this type of shock can introduce an upward bias into the parameter estimates and lead to an overstatement of the program's effectiveness. Alternatively, onset of an acute condition in 1 year could lead individuals to engage in health improvement activities in the next year, including participation in the FRP.15 To address this concern, we include three binary indicators to capture time-varying acute health shocks: whether an enrollee experienced a hospitalization related to pregnancy, cancer, or another serious, exercise-sensitive condition (eg, diagnoses of blood disorders, vascular, metabolic, chronic kidney disease, back disc, joint trauma, fractures, injury, burns, poisoning, and complications from hospitalization). Finally, the model includes a random error term (ηit).
A limitation of using a single cohort based on FRP participation is that we cannot differentiate among participants in terms of the quantity of FRP-related exercise during the year. To investigate this question, we modify Equation 1 as follows:
Equation (2) includes four “cohort” variables that reflect varying degrees of persistent, regular FRP exercise based on the number of credits that an enrollee earned in 2008. The four cohorts are highly persistent FRP exercisers, earning between 9 and 12 monthly credits (Credits 912); moderately persistent exercisers who earned between 5 and 8 monthly credits (Credits 58); sporadic exercisers who earned between 1 and 4 monthly credits (Credits 14); and non-FRP exercisers, which includes people who did not sign up as well as those who signed up but never earned a credit (reference category). The cohort variables stand for unobserved time-invariant characteristics of enrollees that might influence their medical care consumption in the absence of the FRP program. For example, highly persistent exercisers might be more “health-conscious” or have better diets than other individuals. The coefficients γ1 − γ3 capture these permanent effects, which we hypothesize are negative (more persistent FRP exercisers have permanent unmeasured behaviors associated with lower medical spending).
Enrollees can sign up for the FRP throughout the year. Thus, the number of months an employee is eligible to receive the credit varies among persons, leading to an additional challenge regarding measurement of the cohorts. Because the highly persistent FRP exerciser cohort requires at least 9 months of eligibility, we exclude individuals who signed up for between 1 and 8 months. This consists of approximately 8.1% of the total UPlan population. Finally, as in the baseline model, Equation (2) includes time indicators and a large set of controls that capture demographics, health status, and participation in other UPlan-sponsored wellness programs.
We estimate both models with generalized linear models (GLM), which directly estimate the mean and variance functions on the original scale of the dependent variable.16 GLMs are appealing because the link function directly characterizes how the expectation on the original scale is related to the predictors. Using the Box-Cox test to identify the appropriate scale and the modified Park test to identify the appropriate distribution family, we determined that a log scale and a gamma distribution family are appropriate for the model. All analyses were conducted in STATA 12.0/SE (StataCorp LP, College Station, Texas).
RESULTS AND DISCUSSION
Table 1 provides descriptive statistics for the 19,478 unique UPlan enrollees for years 2006 to 2008. Average monthly expenditures (2009 constant dollars) during this period were $560 (SD $1817). Approximately 16% of UPlan enrollees signed up for the FRP in 2008. Among those who signed up, 38% earned the credit 1 to 4 months; 22% earned the credit 5 to 8 months; and 29% earned the credit 9 to 12 months. Moreover, 46% of those who signed up for the FRP participated in at least one of the other UPlan wellness programs (disease management, lifestyle management, or 10,000 steps) (data not shown).
Table 2 provides parameter estimates and standard errors for the baseline model (column 1) and the model with four exercise cohorts (column 2). In the first column, we observe a positive and significant effect of the FRP participation “cohort” variable, suggesting that individuals who signed up for the FRP have permanent unmeasured behaviors associated with higher medical spending. This finding contradicts our hypothesis that FRP participants would have lower spending, on average. The parameter estimate on the interaction of FRP participant and year 2008 is negative but not statistically significant.
Column (2) reports the results for the four-cohort model that examines the influence of regular FRP exercise on medical care spending. Individuals who signed up for the FRP and received the credit 1 to 4 times in 2008 have permanent unmeasured behaviors associated with higher medical spending relative to non-FRP exercisers. However, we find no differences between FRP exercisers who earned the credit at least 5 times during the year and non-FRP exercisers.
For the interactions of the cohort variables and year 2008, we obtain a negative and significant estimate (P = 0.032) for the cohort that earned the monthly credit 9 to 12 times during 2008. This supports the hypothesis that highly persistent regular exercise associated with the FRP resulted in lower average monthly expenditures in 2008. The magnitude of the parameter estimate, given the log-scale, suggests that highly persistent exercisers spent approximately 18% ($86 per month) less on medical care in 2008 relative to non-FRP exercisers.a
In both models, older enrollees and females have higher average monthly spending, whereas employees spent less than adult dependents. Not surprisingly, we found positive and significant associations between an enrollee's health status (pregnancy, diabetes, heart disease, depression, and cancer) and medical spending.
We performed several types of sensitivity checks. First, we tested for preperiod trends by including interaction terms between the cohort variables and the year 2007 indicator. We did not find evidence that persistent FRP exercisers and non-FRP exerciser cohorts were on diverging trends prior to the program's implementation. Second, we tested the sensitivity of our cohort definitions by changing the categories in Equation 2 to 1 to 3, 4 to 7, and 8 to 12 months of credit. The results are very close in magnitude and the highly persistent FRP exerciser effect remains statistically significant.
Third, we checked the sensitivity of the results to the presence of outliers by reestimating the models after excluding “high-spending” enrollees in the top 1% of the spending distribution during any of the 3 years (n = 488 unique individuals). Table 3 shows that this exclusion dramatically affects the results. In the baseline model using one cohort (column 1), the estimated effect of the interaction of FRP participant and year 2008 is −0.001 and is statistically insignificant. In the multicohort specification, the parameter estimate on the Month credits 9 to 12 × year 2008 interaction is still negative, but the estimate becomes smaller and it is no longer statistically significant.
We examined FRP participation for the 488 outlier individuals. Sixty-four of these individuals (13%) signed up for the FRP in 2008, which is slightly less than the overall population sign-up rate of 16%. Next, we summarized average monthly medical spending for these individuals, stratified by the year in which an individual was in the “high spending” category and FRP participation status (Table 4).
Among individuals who were outliers in 2006, we observed little difference in average spending between FRP participants and nonparticipants in 2008. In contrast, among individuals who were outliers in 2007, we observed a large difference in average spending between FRP participants and nonparticipants in 2008. Individuals who participated in the FRP had an average monthly spending of $3999 in 2008 compared with $6212 for nonparticipants. A similar pattern was observed for individuals who had outlier expenses in 2007 and became highly persistent FRP exercisers in 2008 (2008 spending = $3457), compared with 2007 outliers who did not become highly persistent FRP exercisers in 2008 (2008 spending = $6025).
These results call into question whether the findings reported in the main specification reflect an effective program or some other phenomenon. One possible explanation is that the result reflects regression to the mean, which in our context, refers to the tendency of individuals who had very high medical expenditures in the pre-period to have lower costs in 2008. However, there were high-cost people in both the FRP regular exerciser cohort and the other cohorts. Thus, we would expect regression to the mean to affect both groups and difference away in the model. A more plausible explanation of the result is that some high-cost people in 2007 recovered or improved more quickly than others for reasons that are unobserved by the investigators and that this recovery was also associated with their desire to join the FRP. In this case, the main finding from the baseline specification could represent an omitted variable bias such as a stronger preference or ability to exercise regularly in conjunction with the FRP.
Using a pre-post design and rigorous econometric methods, we investigated whether an employer-based exercise incentive program is associated with a decrease in medical expenditures in the initial year of implementation. The results suggest positive benefits of lower medical expenditures among highly persistent FRP exercisers. However, the results are subject to two caveats. First, we do not have information on exercise duration or intensity. If employees who exercise more frequently also spend more time at each exercise session, this could bias the estimated coefficients toward finding overly large effects of exercise frequency on medical care expenditure.
Second, and more importantly, the results are sensitive to the treatment of enrollees who were outliers with respect to medical spending. Excluding the outlier observations from the analysis leads to a statistically insignificant finding of FRP participation on medical care expenditures, suggesting that these programs are unlikely to be cost-saving in the short run. Supplemental analyses of these outlier observations suggested that the observed decline in expenditures among highly persistent FRP exercisers in the four-cohort model may reflect a simultaneous improvement in health status among a small number of UPlan enrollees that coincided with engaging in highly persistent FRP exercise.
As exercise-focused employer wellness initiatives are implemented, additional research is needed to explore plausible causal mechanisms by which short-term physical activity can lead to expenditure reduction and health improvement.
1. Flegal K, Carroll M, Ogden C, Curtin L. Prevalence and trends in obesity among US adults, 1998–2008. JAMA. 2010;303:235–241.
2. Finkelstein E, Fiebelkorn I, Wang G. The costs of obesity among full-time employees. Am J Health Promot. 2005;20:45–51.
4. Haskell W, Lee I, Pate R, et al. Physical activity and public health: updated recommendation for adults from the American College of Sports Medicine and the American Heart Association. Circulation. 2007;116:1081–1093.
5. Sherrill D, Kotchou K, Quan S. Association of physical activity and human sleep disorders. Arch Intern Med. 1998;158:1894–1898.
6. Griffiths A. The benefits of employee exercise programmes [review]. Work Stress. 1996;10:5–23.
7. Lambert EV, da Silva R, Fatti L, et al. Fitness-related activities and medical claims related to hospital admissions—South Africa, 2006. Prev Chronic Dis. 2009;6:A120.
8. Pronk N, Goodman M, O'Connor P, Martinson B. Relationship between modifiable health risks and short-term health care changes. JAMA. 1999;282:2235–2239.
9. Martinson B, Crain A, Pronk N, O'Connor P, Maciosek M. Changes in physical activity and short-term changes in health care charges: a prospective cohort study of older adults. Prev Med. 2003;37:319–326.
10. Wang F, McDonald T, Champagne L, Edington D. Relationship of body mass index and physical activity to health care costs among employees. J Occup Environ Med. 2004;46:428–436.
11. Kaiser Family Foundation and Health Research and Educational Trust. Employer Health Benefits 2010 Annual Survey. Available at: http://ehbs.kff.org/pdf/2010/8085.pdf
. Accessed December 2010.
12. Baicker K, Cutler D, Song Z. Workplace wellness programs can generate savings. Health Aff. 2010;29:1–8.
13. Goetzel R, Ozminkowski F, Villagra V, Duffy J. Return on investment in disease management: a review. Health Care Financ Rev. 2005;25:1–19.
14. Lu C, Schultz A, Sill S, Petersen R, Young J, Edington D. Effects of an incentive-based online physical activity intervention on health care costs. J Occup Environ Med. 2008;50:1209–1215.
15. Martinson BC, O'Connor PJ, Pronk NP, et al. Smoking cessation attempts in relation to prior health care charges: the effect of antecedent smoking-related symptoms. Am J Health Promot. 2003;18:125–132.
16. McCullagh P, Nelder JA. Generalized Linear Models. 2nd ed. London, UK: Chapman and Hall; 1989.
a Since the interaction term is dichotomous, the marginal effect is calculated as exp(b)-1. Cited Here...