Health Promotion International, Vol. 14, No. 1, 43-51,
March 1999
© Oxford University Press 1999
Do employees participating in voluntary health promotion programs incur lower health care costs?
Assistant Professor, Consumer Economics, Department of Health and Human Development, Montana State University, 106 Herrick Hall, Bozeman, MT 59717, 1 Assistant Professor, Health Promotion, Health Promotion Programming, Planning, and Evaluation, Department of Health and Human Development, Montana State University, 122 Hosaeus Complex, Bozeman, MT 59717 and 2 Professor, Agricultural Economics, Agricultural Economics and Economics, Montana State University, Linfield Hall 307C, Bozeman, MT 59715, USA
Address for correspondence: George Haynes, Department of Health and Human Development, Montana State University, 106 Herrick Hall Bozeman, MT 59717, USA
 |
SUMMARY |
|---|
 TOP SUMMARY INTRODUCTIONBACKGROUNDEMPIRICAL MODELANALYSISRESULTSDISCUSSIONREFERENCES |
|---|
During the past two decades there has been a rapid increase in the number of wellness activities within public and private companies. A rationale for implementing worksite wellness programs has been the assumption that wellness programming can contain health-related costs. This investigation examined the relationship between health insurance costs and employee wellness program participation using a sample of 1757 university employees over a 3-year period. Based upon empirical models and analytic techniques that are appropriate for these investigations, the authors suggest that voluntary wellness programs may face a serious adverse selection problem in that relatively unhealthy individuals may self-select into wellness programming. Specifically, the authors show that employees who participate in worksite wellness programming incur higher rather than lower health claims costs.
Key words: cost benefit; cost containment; health promotion; wellness
|
INTRODUCTION |
|---|
|
TOPSUMMARYINTRODUCTIONBACKGROUNDEMPIRICAL MODELANALYSISRESULTSDISCUSSIONREFERENCES |
|---|
During the last two decades, numerous public (state and federal) and private organizations have implemented worksite wellness programs. This growth has been fueled in part by the rapid increase in health-related costs and the relationship between wellness programming (health screenings, exercise, etc.) and the containment of health-related costs. Health researchers demonstrated the foundation for this relationship by showing the association of health status and mortality with lifestyle behaviors (Belloc and Breslow, 1972
; Belloc, 1973; Lalonde, 1974; Fries et al., 1989). Because of this association, many organizations initiated wellness programming as a tool to contain health-related costs (Aldana, 1998). The logic was that health promotion programming would allow participants to enjoy better health, require health care services less frequently and use less costly health care services. Consequently, an outcome of the employee wellness program would be lower health-related costs for the organization. This study examines the impact of employee wellness programming on health insurance costs in an environment where participation in wellness programming is voluntary. |
BACKGROUND |
|---|
|
TOPSUMMARYINTRODUCTIONBACKGROUNDEMPIRICAL MODELANALYSISRESULTSDISCUSSIONREFERENCES |
|---|
Because of the cost containment logic associated with wellness programming, researchers have investigated the effects of wellness programs on health care costs by examining health care expenditures on wellness participants and non-participants (Belloc and Breslow, 1972
; Erfurt and Foote, 1984; Pelletier and Lutz, 1988; Vickery et al., 1988). Other researchers examined specific interventions, including medical self-care (Vickery et al., 1988), hypertension screenings (Erfurt and Foote, 1984), stress management (Pelletier and Lutz, 1988), smoking cessation (Bertera et al., 1990), cholesterol screenings (Wilson et al., 1992), and physical fitness (Browne et al., 1984). Additional research was conducted to evaluate the cost-effectiveness of comprehensive health promotion programs (Gibbs et al., 1985; Bly et al., 1986; Sciacca, 1986; Bertera, 1990; Golaszewski et al., 1992). Typically, comprehensive programs offer employees multiple services such as medical screenings, physical fitness activities and nutritional interventions. The vast majority of these investigations related to specific interventions and comprehensive programming showed that health promotion programming was associated with lower health-related costs (Pelletier, 1996; Aldana, 1998).
The research that has taken place within worksite settings varies based upon whether participation was voluntary (Evans et al., 1989), mandatory (Cady et al., 1985; Steinhardt et al., 1991) or facilitated through large incentives (Browne et al., 1984; Gibbs et al., 1985; Sciacca, 1986; Bertera et al., 1990). These policies are significant because the employee's reasons for participating will differ based upon the organization's participation policy.
Similarly, researchers have incorporated diverse techniques in the development of participant classification systems. For example, many investigators compared employees who participate in worksite-based wellness programs with employees who do not participate. The criteria used to classify an employee as a participant varied across studies. In some assessments, participation was based upon the completion of health screenings (Gibbs et al., 1985). In others, the selection criteria was based on the individual's worksite assignment (Bly et al., 1986); length of employment and a willingness to follow exercise prescriptions (Golaszewski et al., 1992); or, participation in selected health enhancement activities (Baun et al., 1986). Still other investigators conducted cost analyses based upon employee health status (Steinhardt et al., 1991; Kingery et al., 1994). These studies exhibit little consistency with respect to the methods used to group employees' participation in wellness programming.
Finally, there is little consensus about the statistical procedures that should be used in cost-effectiveness studies of worksite health promotion programs. Numerous authors and investigators have discussed the difficulties associated with these studies (Murphy et al., 1987; Tze-ching et al., 1991; Lynch et al., 1991, 1992; Kingery et al., 1994). It has been recognized that health care cost data are highly skewed and violate normality assumptions with respect to the error term. Consequently, the use of standard statistical procedures that rely on the normality assumption is problematic. This study uses a non-linear tobit statistical algorithm to address this problem.
|
EMPIRICAL MODEL |
|---|
|
TOPSUMMARYINTRODUCTIONBACKGROUNDEMPIRICAL MODELANALYSISRESULTSDISCUSSIONREFERENCES |
|---|
Subjects
The sample used in this investigation was comprised of 1757 Montana State University (MSU) employees which represent more than 5000 employee life-years of health insurance costs data. Approximately 44% (785) of the subjects were participants in the Montana State University Employee Program (MSUEWP). The MSUEWP participants were well educated with 60% of the subjects having at least a bachelor's degree. The group was primarily female (60%) with a group mean age of 45.9 and just over 50% were classified as non-professional clerical/ service workers. The MSUEWP non-participants were also well educated with 65% of the subjects having at least a bachelor's degree. The group was primarily male (61%), with a group mean age of 45.5 and was predominately comprised of faculty or university administrators (57%). The population characteristics of the subjects are detailed in Table 1
.
|
Data
The amount of health care costs incurred by the individual is affected by many socio-economic and demographic characteristics, as well as participation in wellness programs. This study integrated data from three sources: health insurance claim files administered by Mutual of Omaha, human resource files administered by the Vice President for Administration at MSU, and employee wellness participation files administered by the Director of the MSUEWP.
Health claims information supplied by Mutual of Omaha contained data on the amount of health insurance claims using morbidity and mortality information (i.e. ICD-9 codes) for each employee for 3 years from July 1, 1988 through June 30, 1991. The first year of data (for the period July 1, 1988 through June 30, 1989) was utilized as an indicator of previous health care utilization of each individual in the sample. The remaining health insurance claim information (for the period July 1, 1989 through June 30, 1991) was used as the measure of total health insurance claim amounts.
The MSU human resources database was used to determine the gender, age, job classification, level of education, number of dependants, and earnings for employees. This data set included information on all individuals who were MSU employees during the period July 1, 1988 through June 30, 1991. Any employee who was not employed at MSU during the 3-year evaluation period was excluded from the analysis.
Employee Wellness office files were used to determine whether an employee participated in health promotion programs. Each individual who participated in at least one health promotion activity during the period from 1988 to 1991 was categorized as an active participant. Wellness program health promotion activities included blood, mammogram and colon screening, the completion of a Health Risk Appraisal and the completion of a liability form for participating in exercise classes. However, the data did not provide a measure of intensity of participation, e.g. how much an employee exercised. In addition, employees who participated in health enhancement activities independently, using a health club or other providers of wellness activities outside the university, were considered non-participants even though they were actively participating in health enhancement programming.
Model
An empirical model was developed to examine the relationship between health insurance claims and participation in the MSUEWP. The dependent variable (CLM) was the log of the dollar value of health care claims for the 2-year period July 1, 1989–June 30, 1991. The log transformation was conducted to decrease the variance and more normally distribute the health care claims scores.
The set of explanatory variables includes the individual's participation in any employee wellness activity (EWP), previous health status (PREVIOUS), employment classification (CLASS), education (ED), earnings (LEARN), age (AGE and AGE2), gender (GENDER) and number of dependants (DEPEND). The model is specified as follows:
CLM = a0 + a1EWP + a2PREVIOUS + a3CLASS + a4ED + a5LEARN + a6AGE + a7AGE2 + a8GENDER + a9DEPEND + e
where:
CLM = log amount of all health claims (7/1/1989-6/30/1991);
EWP = participation in at least one employee wellness activity;
PREVIOUS = previous health insurance utilization (7/1/1988-6/30/89);
CLASS = faculty classification (non-professional service and clerical and faculty/administrator);
ED = level of education (dummy variables are used for no college, bachelor's, master's and terminal degree);
LEARN = log of maximum earnings during any 1 year from 7/1/1988 through 6/30/1991;
AGE = age from the individuals date of birth to June 30, 1991;
AGE2 = age squared;
GENDER = male gender;
DEPEND = dummy variable for having any dependants; and,
e = error term.
Each explanatory variable identified an important characteristic of the population which influences the demand for health care services, and consequently, claims submitted to the health insurance program. Classification as an MSUEWP participant was set as a zero-one dichotomous variable, which was set equal to 1 if the individual participated in health promotion activities. Previous research would suggest that the expected sign on the coefficient associated with wellness programming to be negative. That is, employee wellness participants have lower health insurance claims than non-wellness participants.
Past health insurance utilization is often an important predictor of future utilization. Therefore, the investigators utilized information through the Mutual of Omaha health insurance data to assess the employee's previous health insurance utilization. The previous utilization variable PREVIOUS, a 0–1 dummy variable, identifies individuals who submitted health insurance claims in fiscal year 1989. Individuals submitting health claims in fiscal year 1989 were assumed to be more prominent users of health insurance benefits than individuals who did not submit health care claims.
The job classification, educational status, income, and dependent variables were all needed because these variables are important in determining health care utilization. Employees were divided into two employment groups: faculty/ administrators and non-professional clerical/ service employees. Educational status was coded using dummy variables for no college, bachelor's, master's, and terminal degrees. Earnings were measured by the highest salary paid over the 2-year period from July 1, 1989 through June 30, 1991. The log of the earnings was used to help decrease the variance and more normally distribute the scores.
Finally, the age, age squared, and gender variables were included in the model. Age is a significant predictor of health care utilization because as an individual grows older the use of curative health services increase. An age squared term was added to the model to accommodate the fact that older individuals realize a higher rate of increase in health care costs than younger individuals. Gender was also included because females, particularly females in their child-bearing years tend to have more health-related costs than males. Based upon this rationale, the investigators included these dependent and exploratory variables to develop the empirical model used in this investigation.
|
ANALYSIS |
|---|
|
TOPSUMMARYINTRODUCTIONBACKGROUNDEMPIRICAL MODELANALYSISRESULTSDISCUSSIONREFERENCES |
|---|
The researchers utilized a non-linear regression technique (tobit) that accounts for the fact that a high percentage of the employees submit no health insurance claims each year. In evaluations of health insurance costs, a large number of the dependent variable observations take on a zero value and hence, the error term in the estimated model is not normally distributed due to the truncation of the health care expenditures at zero. Tobit statistical estimation procedures (Tobin, 1958
) explicitly recognize that the dependent variable is not normally distributed at the limit value, but still use all observations, including those clustered at the limit value, to obtain efficient (minimum variance) estimates of the parameters of the relationship between the dependent variable and the explanatory variables. Thus, tobit estimation procedures were used to explore the relationship between health care claims and the explanatory variables described in the empirical modeling section of this manuscript. The details of the tobit statistical estimation are discussed elsewhere (McDonald and Moffitt, 1980; Roncek, 1992). |
RESULTS |
|---|
|
TOPSUMMARYINTRODUCTIONBACKGROUNDEMPIRICAL MODELANALYSISRESULTSDISCUSSIONREFERENCES |
|---|
The results of this investigation differ from those reported in most studies but were extremely robust for this sample. For example, the results of the detailed data on health care expenditures by ICD-9 classification show that wellness program participants had higher costs than non-participants (Table 2
). Specifically, wellness participants had higher average claims for mental disorders, genitourinary illness, and skin disease. In addition, wellness participants had a higher probability of submitting any claim for infections and parasitic diseases, endocrine (nutritional and metabolic) illness, mental disorders, nervous system problems, circulatory, respiratory, genitourinary, complications of pregnancy/child birth, skin diseases, musculoskeletal/connective tissue and three general classifications (i.e. ill-defined conditions, injury or poisoning and supplemental factors). Furthermore, wellness participants had a higher probability of submitting a high-cost claim (claim over $200) for mental disorders, genitourinary problems, complications of pregnancy/childbirth, connective tissue and ill-defined conditions.
|
The non-linear regression results based upon 1757 subjects and over 5000 life-years were obtained by using the tobit procedure for estimating empirical model are reported in Table 3
and are congruent with the results described in Table 2
. The coefficients associated with the level of education (as indicated by bachelor's, master's and terminal degrees) and the presence of dependants in the household were not statistically significant. However, all other variables, including previous utilization, employment classification, earnings, age, gender, and employee wellness participation were all statistically significant. [When reading Table 3, it is important to note that the parameter estimations only provide information related to the direction of the change (positive score relates to higher costs and a negative score relates to lower costs) and does not directly provide information related to the magnitude of the change.]
|
The results summarized in Table 3
show that those individuals submitting health care claims during the fiscal year prior to this study had significantly higher health care claims. While some of these claims may be carryovers from a previous time period for the same illness, it is important to note that submitting a health care claim in the previous year is an important determinant of the amount of future health care expenditures. In addition, individuals classified as faculty/ administrator incur lower health care claims than non-professional clerical/service employees. Moreover, individuals earning higher salaries have significantly lower health care claims than those earning lower salaries. This relationship may exist because individuals with greater resources (flexible scheduling, money, etc.) are better able to take appropriate actions to reduce the probability of experiencing poor health. As a result, these individuals are likely to have lower health care costs. However, the reason for this relationship cannot be determined through this investigation.
The coefficients associated with the variables AGE and AGE2 indicate that health care claims decline until the individual turns approximately 40 years of age. Thereafter, costs begin to rise. This result provides support to the notion that adolescents incur higher health care costs than young and maturing adults and that elderly adults incur higher health care costs than young and maturing adults. Also, males in this study incur lower health care claims than females, in part because females in their 20s and 30s incur costs associated with pre- and postnatal care.
Most importantly, the results indicate that participants in the MSUEWP have higher average claims than non-participants. Specifically, the regression results imply that, over the estimation period, health care claims for participants were approximately $40 higher than those for non-participants. This estimate is lower than the simple difference between average costs of health care claims submitted by EWP participants and non-participants because other factors, such as personal and demographic characteristics explain some of this difference.
|
DISCUSSION |
|---|
|
TOPSUMMARYINTRODUCTIONBACKGROUNDEMPIRICAL MODELANALYSISRESULTSDISCUSSIONREFERENCES |
|---|
This study has examined the association of participation in the MSUEWP on health insurance claims costs. Participation in employee wellness activities offered at MSU is strictly voluntary. Hence, this study has considered a group of individuals who actively decided whether or not to participate in MSUEWP activities. This section discusses the study's important results, limitations, and identifies potential avenues for future research.
Important results
The contribution this investigation has made to the study of wellness participation and health care costs containment is two-fold. First, based upon 1757 subjects and over 5000 life-years the investigators found a robust association that is contrary to the majority of studies that have been conducted in this area. That is, wellness participants are significantly associated with higher health insurance costs than non-wellness participants. The investigators did not find these results surprising because in an environment where the individual voluntarily chooses to participate in worksite wellness programming, this result is expected. That is to say, individuals having more health-related problems are more likely to participate in employee-sponsored wellness programs than other individuals. This relationship would be enhanced in established programs because problematic employees are more likely to be referred into worksite programs.
Second, the investigators used a theoretical framework and modeling procedures that were appropriate for this type of investigation. Specifically, the use of a tobit analysis and exploratory variables such as previous utilization and select demographic variables were based upon a conceptually sound framework and represent relevant empirical and analytical contributions to this area of study. The model used by the researchers demonstrated that the overwhelming impact of higher health care claims can be determined by previous utilization followed by income, a non-professional service/ clerical job classification, female gender, participation in the MSUEWP, and age (older employees).
Limitations
The primary limitation in this study was that the investigators did not have a satisfactory proxy which controls for possible adverse-selection bias and an accurate measure for wellness participation. Specifically, the fact that participation in the MSUEWP is voluntary may introduce an adverse-selection bias problem, in that individuals with adverse health conditions (and higher health care costs) ‘self-select’ into the MSUEWP. The control of possible adverse-selection within voluntary wellness programs needs to be addressed in future studies.
In addition, better measures are needed to classify the health behaviors of wellness and non-wellness participants. Specifically, the classification procedure used in this study did not allow the investigators to measure intensity of participation in health-promoting behaviors. In fact, if employees were not using the MSUEWP, implicitly they were assumed not to be engaged in any other health enhancement activity. Furthermore, no assessment was made about the intensity of participation for the wellness participants and non-participants. It is quite possible that the intensity of participation is a critical factor in the evaluation of health insurance costs and wellness participation. In future studies, information about the individual's health promotion activities, both within and outside the work environment needs to be considered as an important exploratory variable.
Further research is needed to more thoroughly assess whether wellness programs generate economic gains or losses. This research should incorporate conceptually sound modeling and analytical techniques. Furthermore, investigators should use realistic assessments of the health status and health care costs of program participants had they not joined the wellness program because, in this context the right question is not whether participants use more health care services than non-participants. Rather, it is whether or not wellness programs reduce the health care costs of participants below those that they would have incurred in the absence of the program.
|
REFERENCES |
|---|
|
TOPSUMMARYINTRODUCTIONBACKGROUNDEMPIRICAL MODELANALYSISRESULTSDISCUSSIONREFERENCES |
|---|
Aldana, S. (1998) Financial impact of worksite health promotion and methodological quality of the evidence. The Art of Health Promotion, 2, 1–8.
Baun, W., Bernacki, E. and Tasi, S. (1986) A preliminary investigation: effect of a corporate fitness program on absenteeism and health care cost. Journal of Occupational Medicine, 28, 18–22.[Web of Science][Medline]
Belloc, N. (1973) Relationship of health practices and mortality. Preventive Medicine, 2, 67–81.[Medline]
Belloc, N. and Breslow, L. (1972) Relationship of physical health status and health practices. Preventive Medicine, 1, 409–421.[Medline]
Bertera, R. (1990) The effects of workplace health promotion on absenteeism and employment costs in a large industrial population. American Journal of Public Health, 80, 1101–1105.
Bertera, R., Oehl, L. and Telepchak, J. (1990) Self-help versus group approaches to smoking cessation in the workplace: eighteen month follow-up and cost analysis. American Journal of Health Promotion, 76, 187–192.
Bly, J., Jones, R. and Richardson, J. (1986) Impact of work site health promotion on health care costs and utilization: evaluation of Johnson and Johnson's Live for Life Program. Journal of the American Medical Association, 256, 3235–3240.
Browne, D., Russell, M., Morgan, J., Optenberg, A. and Clarke, A. (1984) Reduced disability and health care costs in an industrial fitness program. Journal of Occupational Medicine, 26, 809–816.[Web of Science][Medline]
Cady, L., Phillip, T. and Karwasky, R. (1985) Programming for increasing health and physical fitness of fire fighters. Journal of Occupational Medicine, 27, 111–114.
Erfurt, J. and Foote, A. (1984) Cost-effectiveness of work site blood pressure control programs. Journal of Occupational Medicine, 49, 892–1000.
Evans, G. F., Harris, P., McNeill, A. W. and McKenzie, B. J. (1989) Montana State University: an employee wellness program financed by a self-insured group health plan. American Journal of Health Promotion, 4, 25–33.
Fries, J. F., Green, L. W. and Levine, S. (1989) Health promotion and the compression of morbidity. The Lancet, 3, 481–484.
Gibbs, J., Mulvaney, D., Henes, C. and Reed, R. (1985) Work site health promotion: five-year trend in employee health care costs. Journal of Occupational Medicine, 27, 826– 830.[Web of Science][Medline]
Golaszewski, T., Snow, D., Lynch, W., Yen, L. and Solomita, D. (1992) A benefit-to-cost analysis of a work site health promotion program. Journal of Occupational Medicine, 34, 1164–1172.[Web of Science][Medline]
Kingery, P., Ellsworth, C., Corbett, B., Bowden, R. and Brizzolara, J. (1994) High-cost analysis: a closer look at a case for work-site health promotion. Journal of Occupational Medicine, 35, 1341–1347.
Lalonde, M. (1974) A new perspective on the health of Canadians: A working document. Ottawa: Government of Canada.
Lynch, W., Teitebaum, H. and Main, D. (1991) The inadequacy of using means to compare medical costs of smokers and nonsmokers. American Journal of Health Promotion, 6, 123–129.[Medline]
Lynch, W., Teitebaum, H. and Main, D. (1992) Comparing medical costs by comparing high-cost cases. American Journal of Health Promotion, 6, 206–213.[Medline]
McDonald, J. and Moffitt, R. (1980) The uses of tobit analysis. Review of Economics and Statistics, 62, 318–321.[Web of Science]
Murphy, R., Gasparoto, G. and Opatz, J. (1987) Methodological challenges to program evaluation. American Journal of Health Promotion, 1, 33–40.
Pelletier, K. R. (1996) A review and analysis of the health and cost-effective outcome studies of comprehensive health promotion and disease prevention programs at the work site: 1993–1995 update. American Journal of Health Promotion, 10, 380–388.[Web of Science][Medline]
Pelletier, K. and Lutz, R. (1988) Healthy people-healthy business: a critical review of stress management programs in the workplace. American Journal of Health Promotion, 2, 5–19.[Medline]
Roncek, K. (1992) Learning from tobit coefficients: extending a comparative analysis of political protest. American Sociological Review, 57, 503–507.
Sciacca, J. (1986) A study of health care utilization by type of participation in a work site health promotion program. Unpublished doctoral dissertation, Purdue University, Lafayette, IN.
Steinhardt, M., Greenhow, L. and Stewart, J. (1991) The relationship of physical activity and cardiovascular fitness to absenteeism and medical care claims among law enforcement officers. American Journal of Health Promotion, 5, 1341–1346.
Tobin, J. (1958) Estimation of relationships for limited dependent variables. Econometrica, 26, 24–36.[Web of Science]
Tze-ching Yen, L., Edington, D. and Witting, P. (1991) Association between health risk appraisal scores and employee medical claims costs in a manufacturing company. American Journal of Health Promotion, 6, 46–54.[Medline]
Vickery, D. M., Golaszewski, T. J., Wright, E. C. and Kalmer, H. (1988) The effect of self-care interventions on the use of medical services within a Medicare population. Medical Care, 26, 580–588.[Web of Science][Medline]
Wilson, M., Edmunson, J. and Dejoy, D. (1992) Cost effectiveness of worksite cholesterol screening and intervention programs. Journal Occupational Medicine, 34(6), 642–649.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||