The Effect of Participation in an Incentive-based Wellness Program on Self-Reported Exercise

Abstract

Employers are increasingly trying to promote healthy behaviors, including regular exercise, through wellness programs that offer financial incentives. However, there is limited evidence that these types of programs affect exercise habits within employee populations. In this study, we estimate the effect of participation in an incentive-based wellness program on self-reported exercise. Since 2008, the University of Minnesota's Fitness Rewards Program has offered a $20 monthly incentive to encourage fitness center utilization among its employees. Using 2006-2010 health risk assessments and university administrative files for 2,972 employees, we conducted a retrospective cohort study utilizing propensity score methods to estimate the effect of participation in the Fitness Rewards Program on self-reported exercise days per week from 2008-2010. On average, participation in the program led to an increase of 0.59 vigorous exercise days per week (95% Confidence Interval: 0.42, 0.78) and 0.43 strength-building exercise days per week (95% Confidence Interval: 0.31, 0.58) in 2008 for participants relative to non-participants. Increases in exercise persisted through 2010. Employees reporting less frequent exercise prior to the program were least likely to participate in the program, but when they participated they had the largest increases in exercise compared to non-participants. Offering an incentive for fitness center utilization encourages higher levels of exercise. Future policies may want to concentrate on how to motivate participation among individuals who are less frequently physically active.

Introduction

A 2012 survey conducted by the RAND Corporation found that 35% of employers with 50 or more employees use financial incentives to encourage participation in wellness programs.¹ Through these programs, employers aim to promote behaviors that are associated with worker well-being and lower healthcare costs.^2–4 Of particular interest is the capacity for these programs to promote regular exercise^5,6, which is associated with increased physical functioning⁷ as well as decreased risks for type 2 diabetes⁸ and cardiovascular disease⁹.

There is limited evidence that fitness-based wellness programs that offer incentives promote regular exercise. Randomized control trials, typically conducted outside of employer settings, have shown that offering monetary rewards for regular attendance increased participants’ fitness center utilization.^10–12 Likewise, observational studies have presented evidence of increased physical activity following the launch of a program.^1,13–17 However, any conclusion regarding the effect of incentive-based wellness programs on exercise is constrained by methodological limitations in the existing literature.^1,5 Participants in randomized control trials typically received an incentive for only one month with fitness center utilization returning to pre-intervention levels when the incentive period ended.¹² Observational studies have been able to examine longer time periods, but have lacked either a comparison group of non-participants or the data to control for pre-intervention exercise trends. Furthermore, many studies have included concurrent incentives for multiple health behaviors making it difficult to distinguish the effect of participation in the exercise component.^13–15

Launched in 2008, the University of Minnesota's Fitness Rewards Program (FRP) offers employees a monthly incentive if they utilize a fitness center at least eight times. We analyze whether participation in the FRP, defined as enrollment in the program, subsequently had an effect on self-reported exercise. Additionally, we investigate if the effect of participation differed by pre-FRP exercise levels. Given the nature of the FRP, we restrict our analysis to exercise that is likely to be done at fitness centers, specifically vigorous exercise and strength-building exercise. We are able to address many of the methodological concerns identified above by leveraging employee-submitted Health Risk Assessments (HRAs) from 2006-2010 that captured respondents’ exercise levels before and after the program was launched, comparing participants and non-participants, and examining a context that uses a specific fitness-based incentive.

Methods

Study Setting

The University of Minnesota is a large, public university with major campuses in Minneapolis, St. Paul, and Duluth. Approximately 95% of the university's employees enroll in its ‘UPlan’ medical insurance program. On January 1^st, 2008, the university launched the FRP to promote regular exercise among the UPlan population. The FRP offers a $20 credit in each month that a participant utilizes a fitness center at least 8 times. Eligibility is dependent on being at least 18 years old and enrolled in the UPlan. Participating fitness centers include university centers, national and local chains, and independent facilities. Individuals are responsible for signing up for the FRP through their fitness centers. The FRP utilizes fitness center membership card swipes to track enrollee visits.

Separately from the FRP, the university offered employees $65 to submit annual HRAs from 2006-2010. Employees completed HRAs online and answered questions regarding health behaviors including exercise frequency. We obtained de-identified HRAs as well as FRP and UPlan administrative files from the university's data warehouse after approval by the Office of Human Resources.

Study Sample

The study population included employees who were FRP-eligible in 2008 and who submitted a HRA in both 2006 and 2007. The latter criterion was needed to establish pre-FRP trends in exercise. In order to assess whether the FRP had a lasting effect over our time period, we further restricted our sample by excluding employees who were first-time participants after 2008. We identified 23,655 FRP-eligible employees in our enrollment data, of which we excluded 19,925 employees who did not submit an HRA in 2006 and 2007. Next, we excluded 534 employees who were first-time participants after 2008. Lastly, we excluded 234 employees because of missing data on one or more measures. Our final sample included 2,972 FRP-eligible employees, of which 1,044 (35%) participated in the FRP. The overall population participation rate was 28%. The sample was similar to the overall population in terms of age and health, although employees in the sample were more likely to be female (see Appendix Table B1).

Measures

Exercise

We obtained two self-reported measures of exercise from the HRAs, which we used to compare participants and non-participants before and during the program. These measures pertain to exercise we reasonably expected to be done at fitness centers. Our first measure was vigorous exercise days per week. Specifically:

How many days per week do you participate in 20 minutes or more of vigorous exercise? Examples include brisk walking, running, fast cycling, swimming, aerobics, racquetball, and stair/ski/rowing machine.

Our second measure was strength-building exercise days per week. Specifically:

How many days per week do you do strength-building exercises such as curl-ups, push-ups, or using weight-training equipment?

For both questions, respondents indicated the number of days on a 0 to 7 scale. Although the two questions give different examples of exercise, some HRA respondents may have considered their exercise to fall under both the vigorous and strength-building categories and thus we do not assume that the two measures are mutually exclusive.

We also obtained a self-reported binary indicator for having a physical condition that limited the ability to get enough exercise and a binary indicator for whether a respondent reported that he/she had started to “get more exercise” in the 6 months prior to submitting their 2008 HRA to help control for pre-FRP exercise trends.

Participation

We considered an FRP-eligible employee to be a participant if he/she enrolled in the FRP, regardless of whether he/she ever visited a fitness center in conjunction with the program. We allowed this measure to remain constant, such that we compared individuals who ever participated to individuals who never participated. Among our sample of participants, 15% dropped out in either 2009 or 2010. Reasons for discontinuing participation were not collected.

Health status

We used a risk-adjustment algorithm that uses individuals’ ICD-9 diagnoses and prescribed medicines from medical claims to calculate employees’ health risk.¹⁸ Larger risk scores signify greater health risk. Additionally, we obtained self-reported BMI from the HRAs.

Demographic attributes

We obtained age, sex, and number of child dependents (i.e. children enrolled in family's health plan), and campus work location from the UPlan enrollment files.

Statistical Analysis

Initial examination of our data revealed substantial differences between the pre-FRP characteristics of participants and non-participants (Table 1). For each characteristic, we considered a standardized difference of the mean between FRP-participants and non-participations that was greater than 0.1 in absolute value to be a meaningful difference.^19,20 In the year prior to the launch of the FRP, participants averaged 0.62 more vigorous exercise days per week and 0.55 more strength-building exercise days per week than non-participants. Participants were also younger and had lower BMIs than non-participants. Furthermore, the baseline characteristics in Table 1 were likely to affect our exercise outcome measures and such we considered them to be potential confounders. Not controlling for these potential confounders would likely lead to an overestimation of the FRP's impact, particularly because participants were more likely to be frequent exercisers prior to the program relative to non-participants. To control for potential confounding, we used inverse probability of treatment weighting to balance the pre-FRP characteristics of participants and non-participants.¹⁹

Table 1.

Characteristics of Fitness Rewards Program Eligible Employees

	FRP Participants (N=1,044)	FRP Non-participants (N=1,928)	Standardized Difference of the Meana
Exercise History
Mean Vigorous Exercise Days per Week in 2006 (SD)	2.45 (1.88)	1.83 (1.95)	0.322
Mean Vigorous Exercise Days per Week in 2007 (SD)	2.76 (1.89)	2.13 (1.97)	0.327
Mean Strength-building Exercise Days per Week in 2006 (SD)	1.38 (1.50)	0.90 (1.51)	0.323
Mean Strength-building Exercise Days per Week in 2007 (SD)	1.51 (1.50)	0.96 (1.52)	0.418
Percentage Stated They Had Started to “Get More Exercise” in 6 Months Prior to 2008 Health Risk Assessment	26.1	19.0	0.172
Percentage with an Exercise Limitation	8.7	11.6	−0.097
Health Status in 2006
Mean Risk Scoreb (SD)	0.99 (1.33)	1.01 (1.46)	−0.011
Mean Body Mass Index (SD)	26.4 (5.5)	27.1 (6.1)	−0.117
Demographic Characteristics in 2006
Mean Age (SD)	44.2 (10.6)	45.7 (10.3)	−0.148
Percentage Female	69.3	66.6	0.060
Mean Number of Child Dependents (SD)	0.23 (0.56)	0.20 (0.53)	0.019
Campus Location in 2006
Percentage East Bank, Minneapolis	54.8	53.1	0.035
Percentage West Bank, Minneapolis	12.0	13.4	−0.043
Percentage St. Paul	13.2	16.0	−0.079
Percentage Duluth	13.0	10.7	0.072
Percentage Other campus	7.0	6.6	0.014

Notes: Our study was conducted from 2006-2010. Our sample included 2,972 University of Minnesota employees who were eligible for Fitness Rewards Program. All measures are self-reported from annual health risk assessments except for number of child dependents, risk score, and campus location, which were obtained from University of Minnesota medical claims and administrative data.

Abbreviations: FRP, Fitness Rewards Program; SD, standard deviation.

^aA standardized difference of the mean < 0.1 in absolute value indicates negligible difference in the mean of a variable between FRP participants and non-participants

^bWe calculated each employee's risk score using a risk-adjustment algorithm designed by the University of California-San Diego for the Chronic Illness and Disability Payment System that takes into medical claim diagnoses and prescription medications.¹⁸

We estimated the propensity of being a participant using a probit regression model. We controlled for pre-FRP exercise levels by including vigorous exercise days per week (data from 2006 and 2007), strength exercise days per week (2006 and 2007), an indicator for a self-reported exercise limitation (2007), and an indicator for having started to get more exercise in the 6 months prior to submitting the 2008 HRA. We controlled for health status using the health risk score (2007) and BMI (2007). Demographic controls included age (2007), female, number of children (2007), and campus work location (2007). Because we expected that the propensity of FRP participation would be non-linear in many of these control variables we included several of them as polynomial constructs in our model (cubic for the exercise days per week measures and quadratic for health risk score, BMI, and age). The full model specification is available in Appendix Table B2.

We used the results of the probit to predict the probability of being a participant, e. Then, we calculated the inverse probability treatment weight, w, for each FRP-eligible employee i as w_i = FRPparticipant_i/e_i – (1 – FRPparticpant_i)/(1 – e_i), where FRPparticipant is equal to 1 if the employee ever participated in the FRP and equal to 0 otherwise. By weighting all covariates by w we generated a sample where participants and non-participants had similar distributions of pre-FRP characteristics. To check this property, we calculated the standardized difference of the mean for each weighted covariate (see Appendix A for further detail). A standardized difference with absolute value less than 0.1 indicates adequate similarity of the mean of a weighted covariate between participants and non-participants. ^19,20

We estimated the average treatment effect of participation, in each year y, from 2006 through 2010 by weighting each exercise days per week measure by w and taking the difference between participants and non-participants. The average treatment effect for each exercise measure was

$$1∕n({\sum }_{i=1}^{n}FRPparticpan{t}_i\ast ExrDay{s}_i∕e_i-{\sum }_{i=1}^n(1-FRPparticpan{t}_i)\ast ExrDay{s}_i∕(1-e_i)),$$

where n denotes the number of FRP-eligible employees. Not all FRP-eligible employees submitted a HRA in each year from 2008 through 2010. Additionally, for 2008, we did not include participants who submitted their HRA prior to their first month of participation in the average treatment effect calculations. Our sample sizes were 2,727 in 2008, 2,618 in 2009 and 2,592 in 2010.

We estimated confidence intervals for the average treatment effects using bootstrapping. We drew random samples (equal to the full sample size) with replacement and re-estimated our propensity score model and average treatment effect calculations 200 times. We use the 2.5^th and 97.5^th percentiles of each of the 200 average treatment effects as a 95% confidence interval.

We were also interested in whether the effect of participation differed for employees by pre-intervention exercise frequency. We stratified our propensity score into quintiles with the 1^st quintile being the least likely to participate and the 5^th quintile being the most likely to participate. Because pre-intervention exercise had a substantial and significant effect on participation, these quintiles are subgroups of similar employees with increasing mean levels of exercise days per week (e.g. Quintile 1 was mostly infrequent exercisers, Quintile 5 was mostly very frequent exercisers). We conducted a t-test on the difference of each exercise days per week measure by participation status for each quintile.

Sensitivity analyses

If specific ranges of the distribution of propensity scores do not include both participants and non-participants then treatment weights may become unstable, leading to biased results.²¹ In the event that ranges without adequate support existed, we re-estimated our average treatment effects dropping observations from any troublesome ranges. The 2008 cohort may systematically differ from later cohorts by including ‚first-movers’ of either frequent exercisers, who knew they could obtain the incentive with little or no increase in exercise, and/or highly motivated infrequent exercisers. To test for this, we re-estimated our analysis using participants who had their first participation in 2009 (N=282).

All analyses were conducted during 2015 using Stata Version 13 (StataCorp LP; College Station, TX).

Results

Among our sample, 35% were participants. Strength-exercise days per week in 2007 (pre-intervention) had the most influence on participation. On average, each strength-exercise day per week was associated with a 9.7 percentage point (95% Confidence Interval (CI): 7.1, 12.3) increase in the probability of participation. Each year of age was associated with a 0.3 percentage point (95% CI: 0.1, 0.5) decrease in the probability participation. Health status had no effect on participation. See Appendix Table B2 (probit coefficients) and Appendix Table B3 (average marginal effects) for further results regarding factors influencing participation.

After applying the inverse probability treatment weights, participants and non-participants had similar mean pre-FRP characteristics (Table 2). Participants and non-participants were within 0.02 days per week for all pre-FRP exercise measures. The range of standardized differences over all baseline characteristics was −0.016 to 0.012, well below the 0.1 absolute value threshold. This result suggests that differences in the mean observable baseline characteristic of FRP participants and non-participants are inconsequential after propensity scored weighting.

Table 2.

Characteristics of Fitness Rewards Program Eligible Employees, Weighted by the Inverse Probability of Participation

	FRP Participants (N=1,044)	FRP Non-participants (N=1,928)	Standardized Difference of the Meana
Exercise History
Mean Vigorous Exercise Days per Week in 2006 (SD)	2.07 (1.99)	2.05 (1.96)	0.009
Mean Vigorous Exercise Days per Week in 2007 (SD)	2.35 (1.96)	2.35 (1.96)	−0.002
Mean Strength-building Exercise Days per Week in 2006 (SD)	1.06 (1.55)	1.06 (1.53)	0.003
Mean Strength-building Exercise Days per Week in 2007 (SD)	1.19 (1.53)	1.19 (1.54)	0.006
Percentage Stated They Had Started to “Get More Exercise” in 6 Months Prior to 2008 Health Risk Assessment	21.9	21.8	0.001
Percentage with an Exercise Limitation	10.6	10.7	−0.004
Health Status in 2006
Mean Risk Scoreb (SD)	0.98 (1.34)	0.99 (1.41)	−0.008
Mean Body Mass Index (SD)	26.7 (5.9)	26.8 (5.9)	−0.006
Demographic Characteristics in 2006
Mean Age (SD)	45.9 (10.5)	46.1 (10.5)	−0.016
Percentage Female	67.9	67.3	0.012
Mean Number of Child Dependents (SD)	0.21 (0.53)	0.21 (0.52)	0.004
Campus Location in 2006
Percentage East Bank, Minneapolis	54.0	53.8	0.005
Percentage West Bank, Minneapolis	13.0	12.8	0.007
Percentage St. Paul	15.3	15.0	0.008
Percentage Duluth	10.9	11.3	−0.011
Percentage Other campus	6.8	6.8	−0.001

Notes: Our study was conducted from 2006-2010. Our sample included 2,972 University of Minnesota employees who were eligible for Fitness Rewards Program. Each measure is weighted by FRPparticpant_i/e_i –(1-FRPparticpant_i)/(1-e_i) where for employee i FRPparticpant is a binary indicator for ever participating in the Fitness Rewards Program and e is the predicted probability of ever participating in the Fitness Rewards Program. All measures are self-reported from annual health risk assessments except for number of child dependents, risk score, and campus location, which were obtained from University of Minnesota medical claims and administrative data.

Abbreviations: FRP, Fitness Rewards Program; SD, standard deviation.

^aA standardized difference of the mean < 0.1 in absolute value indicates negligible difference in the mean of a variable between FRP participants and non-participants

^bWe calculated each employee's risk score using a risk-adjustment algorithm designed by the University of California-San Diego for the Chronic Illness and Disability Payment System that takes into medical claim diagnoses and prescription medications.¹⁸

In 2008, after the start of the FRP, participants averaged 0.59 (95% CI: 0.42, 0.78) more vigorous exercise days per week and 0.43 (95% CI: 0.31, 0.58) more strength-building exercise days per week than did non-participants (Figure 1). The average treatment effect decreased in 2009 to 0.49 vigorous exercise days per week (95% CI: 0.35, 0.68) and 0.38 strength-building exercise days per week (95% CI: 0.25, 0.50). In 2010, the average treatment effect of participation for vigorous exercise days per week further decreased to 0.38 days (95% CI: 0.24, 0.57), while the average treatment effect for strength-building exercise days per week remained similar to its 2009 level at 0.36 days (95% CI: 0.23, 0.54).

Figure 1.

Difference in Exercise between Fitness Rewards Program Participants and Non-Participants, Inverse Probability of Participation Weighted

We found a general pattern that participants in quintiles with lower propensities to participate experienced a greater increase in exercise relative to non-participants (Figure 2A for vigorous exercise and Figure 2B for strength-building exercise). Because mean pre-FRP exercise days per week increases with each quintile, this result suggests that for participants, less pre-FRP exercise is associated with a greater increase in exercise during the program. In each figure, all values above 0.3 days were significant with p<0.05 and values under 0.3 days were not significant.

Figure 2.

Difference in Exercise between Fitness Rewards Program Participants and Non-Participants, Stratified by Propensity to Participate

Quintile 1 (least likely to participate) and Quintile 2 (second least likely to participate) had the largest differences in exercise days per week between participants and non-participants. Prior to the program in 2007, employees in theses quintiles averaged 1.33 vigorous exercise days per week and 0.23 strength-building exercise days per week. In 2008, participants in Quintile 1 had 1.00 vigorous exercise days per week (95% CI: 0.42, 1.57) more than non-participants, the maximum difference for any quintile across all years. Likewise, the maximum difference for strength-building days was for Quintile 2 in 2008, in which participants had 0.76 strength-building exercise days per week (95% CI: 0.42, 1.10) more than non-participants. Although these differences decreased in 2009 and 2010, participants in these quintiles generally experienced the largest gains in exercise with participation. However, only 20% of employees in these quintiles participated.

Quintile 4 (second most likely to participate) and Quintile 5 (most likely to participate) had the smallest differences in exercises days per week between participants and non-participants. At baseline in 2007, employees in these quintiles averaged 2.88 vigorous exercise days per week and 2.33 strength-building exercise days per week. For vigorous exercise, modest increases in exercise existed in 2008 with participation, but no significant effect existed in 2009 or 2010 for Quintile 4 and in 2010 for Quintile 5. For strength-building exercise days per week, there was no significant difference between participants and non-participants in any year with the exception of Quintile 4 in 2010. Of employees in these quintiles, 52% participated.

Sensitivity Analyses

Only 11 non-participants with low propensity scores fell below the minimum propensity score of the participants. Our results were not sensitive to their exclusion. Our analysis produced slightly smaller, although not significantly different, average treatment effects when using participants with first participation in 2009.

Discussion

The adoption of employer-based wellness programs that incentivize healthy behaviors is becoming more commonplace. However, few studies have rigorously tested whether these programs have an impact on the behaviors of employee populations, particularly for the subset of the high-risk individuals who likely stand to benefit most from participation. In this study, employees who participated in the incentive-based FRP had approximately one-half more exercise days per week than non-participants after adjusting for differences in pre-intervention exercise trends, health status, and demographic attributes. This result aligns with previous studies which found participation in incentivized wellness programs to be associated with a higher likelihood of being physically active^15,16 and with reporting 3 or more days of exercise per week^13,14. However, previous studies did not estimate a specific increase in number of exercise days. Although the participation effect modestly decreased over time, the effect of the program remained positive and significant over three years.

Individuals who reported exercising frequently prior to the FRP showed only modest increases in response to participation. Because these individuals were likely already active enough to meet the credit threshold of 8 visits per month, the introduction of the program did not require them to make changes to their exercise habits to obtain the monetary reward. However, people who reported exercising infrequently prior to the FRP made the largest gains in exercise frequency as a result of participation. This observation is quite promising given that this segment of the population typically includes persistently inactive individuals who may be at higher-risk for medical conditions.²² In our sample, these employees had the highest health risk scores. However, these employees were also the least likely to participate in the program. Future research should explore ways to motivate this segment of employees.

Employers may seek to modify incentives designs and/or use approaches alongside financial incentives to encourage participation. For example, team-based incentives in randomized control trials have elicited higher fitness center utilization rates relative to individual-based incentives.²³ Team-based incentives rely on peer influence and social pressure to encourage individuals to exercise in order to improve the team's chance of receiving a reward.²⁴ Providing employees with information regarding their exercise frequency in relation to their peers and the incentive threshold may also foster participation. In an intervention aimed at improving the food choices of consumers of a hospital cafeteria, Levy and colleagues (2014) found that providing both a financial incentive to select healthy foods and feedback regarding consumers’ individual choices, the average choices of all consumers, and the choices of the healthiest eaters elicited healthier food choices relative to only providing an incentive.²⁵

This study had several limitations. Foremost, although we used several variables to control for pre-intervention exercise levels, there is the potential that unobserved motivation for exercise could contribute to the difference in exercise between participants and non-participants. In such a case, our average treatment effect would overestimate the effect of the program. Second, our sample included only 13% of all FRP eligible-employees. The effect of participation could differ for the full population if employees who submitted HRAs had different responses to participation relative to employees who did not submit HRAs. Third, self-reported exercise days per week may not accurately reflect measures of objective exercise.²⁶ This issue could bias our results if any reporting bias systematically differs between participants and non-participants. However, previous research suggests this case is unlikely. Senso and colleagues (2014) found no difference in the reporting bias between treatment and control groups of physical activity maintenance trials.²⁶ Fourth, we analyzed only the employees of one larger, self-insured employer, which may affect the generalizability of our results. However, the FRP was a relatively simple intervention that did not require large investments by the university. Employers who are not self-insured may be able to find opportunities to work with health insurance companies to administer a similar program.

Conclusions

Participants of an incentivized, fitness-based wellness program maintained higher levels of exercise than non-participants over a 3-year period, after adjusting for pre-intervention differences. Although less frequent exercisers were less likely to participate, they had the largest increase in exercise days per week when they did participate. Future policies may want to concentrate on how to structure incentives to motivate participation among individuals who are least likely to participate in wellness programs.

Highlights.

• We estimated the effect of participation in an incentive-based program on exercise.

• Participants had about one-half more exercise days per week than non-participants.

• Infrequent exercisers were unlikely to enroll, but had large gains when they did.

• Offering an incentive encourages higher levels of exercise.