T H E
S
C I E N C E O F H E A L T H P R O M O T I O NFinancial Analysis; Method, Issues, and Results in Evaluation and Research
Estimating Risk Reduction Required to Break Even
in a Health Promotion Program
Ronald J. Ozminkowski, PhD; Ron Z. Goetzel, PhD; Jan Santoro; Betty-Jo Saenz, MPH;
Christine Eley, MS; Bob Gorsky, PhD
Abstract
Purpose.To illustrate a formula to estimate the amount of risk reduction required to break even on a corporate health promotion program.
Design.A case study design was implemented. Base year (2001) health risk and medi-cal expenditure data from the company, along with published information on the relation-ships between employee demographics, health risks, and medical expenditures, were used to forecast demographics, risks, and expenditures for 2002 through 2011 and estimate the required amount of risk reduction.
Setting.Motorola.
Subjects.52,124 domestic employees.
Measures.Demographics included age, gender, race, and job type. Health risks for 2001 were measured via health risk appraisal. Risks were noted as either high or low and related to exercise/eating habits, body weight, blood pressure, blood sugar levels, cholesterol levels, depression, stress, smoking/drinking habits, and seat belt use. Medical claims for 2001 were used to calculate medical expenditures per employee.
Results.Assuming a $282 per employee program cost, Motorola employees would need to reduce their lifestyle-related health risks by 1.08% to 1.42% per year to break even on health promotion programming, depending upon the discount rate. Higher or lower pro-gram investments would change the risk reduction percentages.
Conclusion.Employers can use information from published studies, along with their own data, to estimate the amount of risk reduction required to break even on their health promotion programs. (Am J Health Promot 2004;18[4]:316–325.)
Key Words:Health Promotion, Health Risk, Wellness, Cost, Return on Invest-ment, Economics, Prevention Research
Ronald J. Ozminkowski, PhD, is the director of Health and Productivity Management Re-search at Medstat in Ann Arbor, Michigan. Ron Z. Goetzel, PhD, is vice president of Con-sulting and Applied Research at Medstat, Ann Arbor, Michigan, and director of the Institute for Health and Productivity Research at Cornell University in Washington, DC. Jan Santoro is a consultant for Employer Services at Medstat in Schaumburg, Illinois. Betty-Jo Saenz, MPH, is manager of Global Rewards—Wellness Initiatives & Education at Motorola, Plan-tation, Florida. Christine Eley, MS, is supervisor of the Wellness Center at Motorola in Tem-pe, Arizona. Bob Gorsky, PhD, is president and senior consultant in health for Health Care and Productivity Management, HPN Worldwide, Inc, in Elmhurst, Illinois.
Send reprint requests to Ronald J. Ozminkowski, PhD, The Medstat Group, Inc, 777 East Eisenhow-er Parkway, 804B, Ann Arbor, MI 48108.
This manuscript was submitted December 30, 2002; revisions were requested June 20, 2003; the manuscript was accept-ed August 20, 2003.
Copyrightq2004 by American Journal of Health Promotion, Inc. 0890-1171/04/$5.0010
INTRODUCTION
Tradeoffs between investments in health promotion and improvements in health can be viewed from an em-ployer’s perspective. Senior business managers often realize that poor health may be caused by modifiable health risk factors that lead to in-creased health care costs. Examples of these risk factors include smoking; poor eating and exercise habits, which may cause overweight or obesi-ty and related problems; and high blood sugar or high cholesterol, eventually resulting in disorders of the eyes, kidneys, coronary arteries, and peripheral vascular systems. These and other risk factors can lead to preventable illnesses or death.1
The relationships between these risks and health care expenditures were described by Goetzel et al.2
The inherent nature of risk means that most employees will not develop serious health problems in the short run, but many will as they age, and a high turnover rate will not make the problem go away. Risky employees who leave the company may be re-placed by risky new employees. Al-though some exceptions exist, such as pre-employment drug testing3or
the use of physical fitness tests for some occupations, risk reduction ef-forts typically take place after hiring. It is not feasible to reduce every health risk in the population to zero, but risks can still be lessened by of-fering employees incentives to partic-ipate in health promotion and well-ness programs and providing reason-able access to such activities either at or near work.4
Table 1
Wellness Initiatives at Motorola and the Risk Factors These Initiatives Address
Initiative Risk Factors Addressed
Awareness communication Health fairs
Monthly health observances
Lifestyle modification/educational classes Back care Cancer prevention Cholesterol prevention Diabetes prevention Disease management High cholesterol High blood glucose
Depression, high blood glucose, high blood pressure, high cholesterol Hypertension prevention Kid aerobics Kid nutrition Shiftwork wellness Stress management Tobacco cessation Weight management
High blood pressure Poor exercise habits Poor nutritional habits High stress
Former tobacco user, current tobacco use Poor nutritional habits, extreme high or low
weight Health promotion/screening
2437 RN telephone line Body composition/body fat analysis Bone density screening
Flu vaccinations
Health screenings and health power profile program
Extreme high or low weight
High blood pressure, high cholesterol Lactation rooms/breastfeeding pumps
Mammograms Massage therapy
Nutritional supplement discount purchase program
Prostrate cancer screening
Special delivery/maternal health program
High stress
Poor nutritional habits
Physical fitness
Aerobic, boxing, pilates, spinning, tai chi, and yoga classes
External wellness reimbursement
Fitness incentives (i.e., Active for Life, Get Fit, Olympics, etc.)
Intramural sport leagues On-site wellness centers Walks/runs events
Poor exercise habits Poor exercise habits Poor exercise habits Poor exercise habits Poor exercise habits Poor exercise habits Supportive environment
Blood pressure machines Book/video lending library
Cafeteria natural balance menu selections Community activities
Fitness tracking software
High blood pressure Poor nutritional habits Poor exercise habits Indoor/outdoor walk trail
Relaxation room
Vending machine healthy selections
Poor exercise habits High stress
Poor nutritional habits The purpose of this paper is to
es-timate how much risk reduction may be required each year to break even financially on a corporate investment in health promotion. A case study at Motorola is used to address this ques-tion. Data on demographics, job type, health risks, and medical ex-penditures at Motorola were used to estimate cost savings that may be as-sociated with risk reduction strate-gies. The methods described here il-lustrate how risks can be forecasted using projections of employee demo-graphics and job type. They also show how the combination of demo-graphic and risk forecasts can be used to estimate medical expendi-tures associated with programs de-signed to reduce health risks. This information can then be used to forecast the monetary impact of health promotion and wellness pro-grams for Motorola’s workforce.
Although the case study reported here pertains to Motorola, the meth-ods used to forecast financial impli-cations can be applied elsewhere.
Health Promotion Programs at Motorola
As noted in Table 1, Motorola of-fers many programs designed to re-duce employee health risks, along with other programs that are expect-ed to enhance morale, expect-educate em-ployees, or offer services that might eventually reduce health care expen-ditures paid by the company and its employees. These programs cost Mo-torola about $240 per employee an-nually. An additional $42 per employ-ee is spent on scremploy-eening services, such as administering a health risk appraisal (HRA) and collecting blood pressure, cholesterol, or other biometric values. From a business perspective, Motorola management must determine whether the $282 to-tal cost per employee per year should be increased, decreased, or stay the same in the future.
One can address this question by showing, in absolute terms, how much risk reduction must result from participation in Motorola’s health promotion programs in order to break even on their investment. Pro-gram managers and senior business managers may consider this
informa-Table 2
Definition of Risk Factors in Motorola and Health Enhancement Research Organization (HERO) Studies
Risk Factors Included
At Risk Criteria
Used in Original HERO Study Used in Motorola Study
Poor exercise habits 0 days per week of aerobic exercise 0 days per week of aerobic exercise Poor eating habits Based on total score on 5 questions Based on total score on four questions* High body weight Based on weight being more than 20% higher
or lower then age/gender recommendations
Based on body composition of$30% body fat
Currently smoke cigarettes Yes Yes
Former smoker Yes Not asked; default taken from HERO study†
High total cholesterol $240 $240
High blood glucose $115 Not collected; default taken from HERO study‡
High blood pressure Diastolic$100 or systolic$160 Diastolic$100 or systolic$160 Being at high stress Frequently at high stress and unable to respond
well to it
Based on one question: stress is out of control ‘‘very often’’ or ‘‘always’’
Being depressed Frequently depressed and unable to respond well to it
Based on one question: feel down/depressed/blue ‘‘very often’’ or ‘‘always’’
Heavy alcohol use d/wk drink$5 servings52 d d/mo drink$5 servings5$5 d * Based on average daily consumption of fruits, vegetables, complex carbohydrates, and fats.
† Not historically asked since focus of screening has been controllable risks moving forward. ‡ Not included in screenings done during applicable time period.
tion as they negotiate with potential program vendors and decide how much they want to spend for health promotion services.
METHODS Design
A case study design was used to es-timate the amount of risk reduction required to break even on Motorola’s health promotion programs. Using information published in Leutzinger et al.,5data on demographics and job
type were collected from personnel files and used to forecast changes in health risks among Motorola employ-ees. Risks were measured in a binary fashion (yes or no for being at high risk) and focused on exercise habits, eating habits, blood pressure, blood sugar levels, cholesterol levels, de-pression, stress, smoking and drink-ing habits, and seat belt use.
The health risk projections were subsequently used to project changes in medical expenditures over time. The expenditure projections were based on relationships between health risks and treatment costs, which were published by Goetzel et al.2The risk-expenditure
relation-ships were then used to find the an-nual percentage reduction in health
risks required to allow Motorola to break even financially, given a $282 per employee per year investment in all of its health promotion and well-ness programs.
Sample
The analyses pertain to the 52,124 full-time, active, domestic employees who were employed at Motorola in 2001 (the base year). Their average tenure with the company was 9 years. Analyses also applied to Motorola employees expected to be employed each year until 2011, the end point for the study. Given market trends and normal attrition, the total em-ployee count was expected to drop by 2% each year from 2001 until 2011.
Measures
The main outcome of interest in this study was the annual percentage reduction in health risks required to break even on Motorola’s investment in health promotion over the period from 2001 to 2011. To calculate this outcome, several measures were re-quired for each year of the study pe-riod. These included
(1) Demographics: The average age of Motorola employees; the
per-cent of women; the perper-centage of African-American, Hispanic, and other non-White employees; and the percentages of employ-ees with sales and professional jobs.
(2) Health risks: These were defined to be as similar as possible to those reported in Goetzel et al.,2
because the forecasting process was designed to take advantage of the risk-expenditure relation-ships reported there. Table 2 provides the risk definitions that were used, along with those used in the Goetzel et al.2study.
Baseline demographic and risk data for the year 2001 were provided by Motorola. The risk data had been collected from an HRA process. De-mographics and risks for subsequent years in the study period were fore-cast using processes described below.
For 2001, a comprehensive health screening (HRA process) was offered at most worksites during work hours. Motorola used a standard HRA devel-oped in 1997 by Wellness, Inc, and HPN WorldWide, Inc. This HRA, known as the Health Power Profile (HPP), requested self-reported infor-mation on demographic factors and most of the risks noted above.
Readi-Table 3
Demographic Projections by Year
All Motorola Employees 2001 2003 2005 2007 2009 2011
Average age Female gender (%) Race was Black (%) Race was Hispanic (%)
39.14 27.50 4.61 9.69 40.43 27.50 4.61 9.69 41.77 27.50 4.61 9.69 43.15 27.50 4.61 9.69 44.58 27.50 4.61 9.69 46.05 27.50 4.61 9.69 Race was other non-White (%)
Had sales job (%) Had professional job (%)
17.44 0.56 75.23 17.44 0.56 75.23 17.44 0.56 75.23 17.44 0.56 75.23 17.44 0.56 75.23 17.44 0.56 75.23
ness to change poor habits was also assessed.
The Motorola HRA was completed using personal digital assistants (PDAs). In addition, the HPP elec-tronically merged data on cholesterol levels, blood pressure, and height and weight measures collected at the screening. The reliability and validity of the HRA data are not yet known.
Analyses
Five major steps were completed to conduct analyses for this project. These are described below. Key as-sumptions are noted for each step.
Step 1: Obtain Baseline (Year 2001) De-mographic Data From Motorola and Forecast Demographics for the Following 10 Years.Motorola provided informa-tion on age, gender, race, and job type for its workforce for 2001. Two assumptions were made to generate forecasts for these variables for the following 10 years, which constituted the rest of the study period: (1) The average age of the workforce was the only variable Motorola expected to change over the next 10 years. Thus, the percentages of employees expect-ed in other demographic and job type categories were held constant at year 2001 levels for the remainder of the study period. (2) Data provided by Motorola staff showed that the av-erage age of the Motorola workforce increased by 1.0164% from 2000 to 2001, so this rate of increase was ex-pected to occur annually for the re-mainder of the study period as well. Results from the demographic and job type projections are shown in Ta-ble 3.
Step 2: Obtain Baseline Risk Data for 2001 and Forecast the Percent of Em-ployees at High Risk at Motorola for the Following 10 Years.This part of the process involved three major activi-ties and two assumptions. First, de-mographic and job type information from Step 1 were combined with published results described below in order to estimate the log odds of be-ing at high risk for each risk factor of interest for each year of the study period. Second, those log odds were transformed into the predicted prob-abilities of being at high risk using the math noted in the Appendix. Third, these predicted probabilities were further tailored to Motorola, us-ing a set of adjustment factors. These activities can be conducted for any company; they are predicted on the following assumptions:
● The risk projections used for this step were based in part on rela-tionships between demographics and risk published by Leutzinger et al.5for six Health Enhancement
Research Organization (HERO)-member organizations. These were assumed to be useful for Motorola as well, once the adjustment fac-tors described below were applied.
● Motorola did not collect data on blood glucose level or former smoking patterns. Motorola’s base year values for these two risks were assumed to be the same as those observed in the HERO data set re-ported by Goetzel et al.2
● To conduct the activities noted above, HPP risk data from 2001 were provided by Motorola staff, but forecasts for risks had to be
produced for the following 10 years. These forecasts were based on the demographic forecasts made in Step 1 and on informa-tion about relainforma-tionships between demographics and risk that were published in Leutzinger et al.5
Leutzinger et al.5conducted
analy-ses on the HERO database and pub-lished logistic regression intercept and coefficient values that estimated the probability of being at high risk for each of the 11 health risk factors mentioned above. To apply the 11 re-gression models of Leutzinger et al.5
(one for each risk factor), the user was required to input information about average employee age; the centage of female employees; per-centages of African-American, His-panic, and other non-White ees; and the percentages of employ-ees with sales and professional jobs. For Motorola, the process involved multiplying these model input values by their respective regression coeffi-cients, adding the results, and then adding the regression intercept term. This math is summarized in the Ap-pendix for any particular year and risk factor of interest.
Technically, the results obtained from this mathematical process pro-vided predicted probabilities one would expect for the organizations represented in the HERO database, if those organizations had Motorola’s employee characteristics. Thus, an additional adjustment was required to obtain results pertinent to Motoro-la. For each risk factor, the actual percentage of Motorola employees who were at high risk during the base year (2001) was obtained from the Motorola HRA data. These per-centages were then divided by the predicted probabilities obtained from the exercise described above for that year. This produced an adjustment factor that was multiplied by the pre-dicted probabilities in future years in order to obtain a final set of proba-bilities that were scaled appropriately and therefore were more pertinent to Motorola.
The application of the adjustment factor can be illustrated by example. Using Motorola’s HRA data, the actu-al probability of being at high risk
for high cholesterol among nonpar-ticipants was .1601 (or 16.01%) in 2001. Using the logistic regression exercise described above, the predict-ed value for a HERO member with Motorola’s employee characteristics was estimated to be .1398 (or 13.98%) for that year. Taken togeth-er, these estimates yielded a choles-terol adjustment factor of 1.1455 (i.e., .1601/.139851.1455). This ad-justment factor was then multiplied by the predicted probabilities for high cholesterol that were estimated for subsequent years in order to ob-tain results pertinent to Motorola. Similar analyses were completed for each risk factor.
Step 3: Obtain the Medical Expenditure Figure per Employee for the Base Year and Forecast Medical Expenditures per Employee for the Study Period.This part of the process involved two ma-jor activities. First, information on demographics and risks that were ob-tained in Steps 1 and 2 were used to forecast expenditures for Motorola for the study period. This forecasting process required information about relationships between risk and medi-cal expenditures that was found in Goetzel et al.2Second, the resulting
expenditure projections were multi-plied by adjustment factors in order to report results pertinent to Motoro-la.
One major assumption was also employed for Step 3: The relation-ships between health risks and medi-cal expenditures that were reported in Goetzel et al.2for HERO-member
companies were assumed to be appli-cable for Motorola as well, as long as suitable adjustment factors were found in order to ensure that results pertained to Motorola instead of those HERO-member companies.
The two major activities required for Step 3 are explained below.
In general, the exercise used to forecast medical expenditures for Motorola was similar to the one used to forecast risk. In the original HERO study, Goetzel et al.2reported
regression coefficients and intercept values one may use to predict medi-cal expenditures. The input data re-quired for these calculations includ-ed the same demographic and job
type data used for the risk calcula-tions described above. Additionally, one must input the percentages of employees expected to be at high risk for each risk factor obtained from Step 2. Once demographics and risks were estimated for Motoro-la for each year in the 2001 through 2011 study period, these were insert-ed into the expenditure regression model equations described in the Ap-pendix to predict medical care ex-penditures for each year.
As with the risk calculations men-tioned above, the output from the expenditure estimation exercise yielded expenditure predictions that, technically, pertained to HERO-mem-ber companies with Motorola’s de-mographic and risk features. These were then multiplied by adjustment factors that were used for each part of the two-part regression model in order to scale them according to Mo-torola’s experience, thus making the expenditure predictions relevant to Motorola. The adjustment factor for the first part of the model that esti-mated the probability of having any expenditures was 1.32. This was ob-tained by dividing the actual proba-bility of using any medical care in 2001 at Motorola (i.e., .91) by the projected probability obtained from the application of the Goetzel et al.2
equations (.69).
The adjustment factor for the sec-ond part of the expenditure model that addressed the magnitude of health care expenditures when they occurred was 1.22. This was obtained by dividing average health care ex-penditures in 2001 ($2457.76) by the projected figure obtained from the Goetzel et al.2equations ($2014.56).
Both of these adjustment factors are noted in the equations contained in the Appendix. Their use adjusts for the underestimate that would have resulted by relying solely on the Goetzel et al.2predictive equations. Step 4: Find the Break-even Risk Reduc-tion Percentage for Motorola’s Health Promotion Programs.To find the amount of risk reduction that must occur each year in order to break even on Motorola’s $282 annual per employee investment, the analyses described in Steps 1 through 3 were
repeated several times, each time re-ducing the value of the risk factor percentages entered into the regres-sion model equations. (The amount of risk reduction was restricted in some cases to prevent risks from fall-ing below zero.) Reducfall-ing the risk factor percentages yielded lower medical expenditure estimates for each year. Subtracting these lower ex-penditure estimates from those ob-tained from the first iteration (before risks were reduced) yielded a set of dollar estimates for the benefits of risk reduction. One such estimate was obtained for each year. These benefit estimates were then com-pared with the cost of the health promotion and wellness programs in those years. When total benefits were exactly equal to total program costs for the 10-year study period, a break-even point was achieved.
Step 5: Conduct Sensitivity Analyses.
When conducting the break-even analysis, program cost and benefit figures were discounted each year. Discounting was required to account for the changing value of a dollar over time. Even after adjusting for in-flation, a dollar gained or spent to-day is more valuable than a dollar gained or spent 1 year from now, since the former can be invested to yield more than a dollar 1 year from now. Discounting adjusts for this phe-nomenon. Since there is no univer-sally accepted discount rate,6–9
analy-ses for the Motorola project were conducted several times, varying the discount rate from 0% to 12% per year.
Analyses for the Motorola project were also repeated several times, varying the estimates used to account for the cost of the health promotion program. Earlier it was noted that all of Motorola’s health promotion pro-grams cost $282 per employee per year, but not every program was meant to address health risk, so this $282 figure overestimated the cost of programs to reduce health risks. The actual cost of each program was un-known, because each was not tracked separately by the company. To ac-count for this, analyses were repeated several times using program cost esti-mates ranging from $150 to $400 per
Table 4
Health Risk Projections* by Year
All Employees 2001 Risk (%) 2003 Risk (%) 2005 Risk (%) 2007 Risk (%) 2009 Risk (%) 2011 Risk (%)
Poor exercise habits Poor eating habits
Deviate from ideal body weight Current smoker Former smoker 14.36 68.35 27.49 13.42 31.10 14.76 66.48 27.94 13.42 31.10 15.19 64.59 28.42 13.42 31.09 15.63 62.68 28.91 13.43 31.09 16.10 60.76 29.44 13.43 31.09 16.59 58.82 29.98 13.43 31.08 High cholesterol
High blood glucose High blood pressure High stress Depression Heavy alcohol use
12.60 4.90 4.78 6.98 5.55 0.46 13.36 5.38 5.17 6.95 5.54 0.43 14.19 5.92 5.60 6.93 5.52 0.41 15.09 6.53 6.08 6.90 5.51 0.39 16.06 7.23 6.62 6.87 5.49 0.36 17.11 8.03 7.23 6.84 5.47 0.34 * Risks were forecast using regression-based models published by Leutzinger et al.5and the adjustment factors described in the text.
employee. Values well above $282 were too high for the Motorola case, but were included to give the reader a sense of how robust the results would be under varying assumptions about program cost.
RESULTS
Demographic Differences
The demographic projections made for the Motorola population are presented in Table 3. Results are provided for the base year (2001) and every other year in the study pe-riod.
Table 3 shows that, overall, the av-erage age of the Motorola popula-tion in 2001 was 39. About 28% were women, 5% were African-American, 10% were Hispanic, and 17% were other non-Whites. Less than 1% had sales jobs, and about 75% had profes-sional jobs. As mentioned earlier, these demographic characteristics were assumed to remain the same for each year of the study period, with the exception of employee age, which increased over time.
Risk Projections
Table 4 reports projections of the percent of Motorola employees ex-pected to have each risk factor over the study period. Base year risks ranged from less than 1% for heavy alcohol use to over 68% for poor eat-ing habits. About 5% to 7% of em-ployees said they were at high stress or suffered from depression. About
5% had high blood pressure, and 5% were expected to have high blood glucose. About 13% of the employees had high cholesterol levels, and 14% reported poor exercise habits. About 40% were either current or former smokers, and 28% were overweight or obese.
Because demographics were not expected to change much over time, employee health risks also remained relatively stable during the study peri-od. The health risk changes reported in Table 4 were influenced by the ag-ing of the Motorola population, but not by changes in other demograph-ic or job type measures.
Break-even Risk Reduction Percentage
Table 5 shows the expected pat-tern of medical expenditures for Mo-torola under two scenarios. The first or reference scenario is based on the demographic and risk factor projec-tions described above. This scenario shows expected medical expenditures for Motorola over the study period (the third column in the table) as-suming an aging workforce with the associated risk profile. The expendi-ture figures in the third column were calculated by multiplying the number of Motorola employees in any given year by the per employee expendi-ture value for that year. The per em-ployee expenditure figure for any given year was obtained from enter-ing the Motorola demographic and risk factor percentages into the
re-gression formulas, doing the requi-site math to calculate expenditures based on those input data, and then applying the adjustment factor as de-scribed above.
The second scenario shown in Ta-ble 5 reflects the break-even risk re-duction requirement (the last col-umn in the table). The break-even scenario reports the total medical ex-penditures for Motorola that one would expect if risks were reduced enough to break even on the $282 per employee per year investment in health promotion. After discounting (using a 3% rate), the analyses showed that savings in medical ex-penditures would be equal to the in-vestment Motorola makes in health promotion and wellness programs if these programs could reduce health risks by 1.15% per year.
Results From the Sensitivity Analyses
Table 6 presents the results from the sensitivity analyses. The table shows that the break-even level of risk reduction varied from .44% per year to 2.79% per year, depending on the assumptions made about pro-gram cost and the discount rate.
DISCUSSION
Published, peer-reviewed findings on relationships between demograph-ics, health risks, and medical expen-ditures,2,5along with data on
Table 5
Break-even Risk Reduction Requirement With an Investment of $282 per Employee per Year*
Year
Number of Employees
Reference Case: Total Expenditures With Demographics and Risk
Shifting as Forecast ($)
Total Medical Expenditures of Risks Are Reduced by Break-even Amount of 1.15%/ Year From 2002–2011 ($)
2001 (base year—no addi-tional risk reduction
52,124 128,108,282.24 128,108,282.24 2002 2003 2004 2005 2006 51,082 50,060 49,059 48,078 47,116 128,136,196.51 128,164,110.79 128,271,889.23 128,379,667.67 128,570,384.67 125,101,352.81 122,094,423.39 119,341,370.36 116,588,317.33 114,586,588.71 2007 2008 2009 2010 2011 46,174 45,250 44,345 43,458 42,589 128,761,101.67 129,038,354.22 129,315,606.77 129,683,585.17 130,051,563.56 112,584,860.09 111,577,625.75 110,570,391.41 109,739,020.73 108,907,650.05 Sum of total expenditures in year
2001 dollars
Potential benefits of risk management, not discounted
Potential benefits of risk management, discounted 3% per year
Program cost (investment), discounted 3% per year
Return on investment ratio
1,416,480,742.49 Not applicable—base case Not applicable—base case
1,279,199,882.86 137,280,859.63 112,939,178.42 112,939,178.42 1.00 * Return on investment is calculated relative to scenario in which demographics and risk shift as according to pre-existing trends. Discounted benefits are found by taking difference between reference case expenditures and break-even expenditures in each year, then dividing by the dis-count rate, then adding the results. Disdis-counted costs are found by multiplying the program in-vestment of $282 per employee by the number of employees per year, then discounting by 3% per year. The discount rate is 1.03 for 2002, 1.03 squared (i.e., 1.0609) for 2003, 1.03 cubed (i.e., 1.0927 for 2003, etc.). All dollars are reported in 2001 equivalents. Values for 2001 are re-ported in actual dollars. Values for subsequent years are forecast as noted in the column head-ers.
Table 6
Impact of Alternative Discount Rates and Program Costs on the Amount
of Risk Reduction Required to Break Even Program Cost ($) Discount Rate 0% 3% 6% 9% 12% 150 200 250 282 0.44 0.63 0.87 1.08 0.46 0.67 0.94 1.15 0.49 0.72 1.01 1.24 0.52 0.77 1.09 1.32 0.56 0.82 1.16 1.42 300 350 400 1.19 1.59 2.12 1.28 1.72 2.27 1.37 1.85 2.44 1.48 2.00 2.62 1.59 2.14 2.79
figures, were used to estimate the an-nual level of risk reduction required to break even on Motorola’s health promotion program. Since the true cost of the programs designed to re-duce risk could not be separated from other programs not designed to do so, the risk-reduction percentages required to break even were report-ed for several alternative investments in programming ranging from $150 to $400. Using a 3% discount rate (which is arguably the one most economists would prefer8), risks for
the overall Motorola population must decline by .45% to 1.15% annually to break even (assuming costs are less than or equal to $282 per employee per year). Higher discount rates or higher program cost rates would
yield higher risk reduction require-ments, as shown in Table 6.
The likelihood that such risk re-ductions would occur would need to be estimated, perhaps by reviewing the relevant literature, talking to ven-dors, and surveying employees. Armed with this information, senior managers would be able to decide how large an investment in health promotion is appropriate for Moto-rola.
Key Assumptions
As noted throughout theMethods section, several assumptions were made to generate the results report-ed. Most importantly, it was assumed that the relationships between demo-graphics, job type, risk, and expendi-tures that were reported in Leutzin-ger et al.5and Goetzel et al.2for
HERO-member organizations could be used for Motorola as well. This as-sumption is thought to be valid be-cause the risk and expenditure pro-jections made for 2001 (the only year for which real data existed) were simi-lar to the projections made for those years, in most cases. Even so, adjust-ment factors were applied to scale HERO projections to the Motorola experience. Nevertheless, the results reported here may still carry some unknown amount of bias and may not be generalizable beyond Motorola.
A second key assumption is that utilization of medical services and re-lated expenditures would decline as a result of the application of risk re-duction programs. The literature on this issue is small, and the effort to make the link between reductions in risk and subsequent reductions in ex-penditures is difficult. This could be because few individuals change their risks over time, or because of the large variability in medical claims data.10
Nonetheless, some researchers have tackled this problem. For exam-ple, Edington11showed that changes
in costs may follow changes in risk. His analysis of health risk and medi-cal claims data over a period of sev-eral years has shown that as health risks declined, so did costs.
A similar analysis (although un-published) was performed by Ozmin-kowski and Goetzel for Citibank.
They found that as the net number of risks declined (defined as the dif-ference between the number of risks that improved minus the number of risks that deteriorated), so did medi-cal costs. In their analysis, employees who were able to achieve a net im-provement of three or more risk fac-tors over a 2-year period saved the company an average of $146 per em-ployee per month.
In sum, there are few studies sug-gesting a direct relationship between improvements in health risks and consequent medical cost savings. Clearly, more research using larger populations and longer time periods is needed. The early findings noted here are suggestive, however, and seem to support the assumption that reductions in expenditures would fol-low reductions in risk.
Limitations
Senior managers are often con-fronted with the need to make deci-sions with imperfect information. The results reported are imperfect for the following reasons.
First, projections were based on work conducted in other studies (Leutzinger et al.5and Goetzel et
al.2); thus, the limitations noted in
those studies will carry over to this study as well.
Second, no information was avail-able to estimate the impact of Moto-rola’s health promotion program ini-tiatives on absenteeism or disability program use. Because of this, the break-even risk reduction percentag-es reported here may be a bit too high. It is unclear whether these bias-es would be offset or exacerbated by other limitations. One example could be the inability to measure the effect of health promotion programming on turnover or morale. There may be other reasons for bias as well, so the results reported herein are im-perfect, as any prospective Return on Investment (ROI) analysis would be.
Third, the projections made here were based on a relatively small num-ber of variables (age, gender, race, job type, and 11 risk factors). Some of these (in particular most of the health risks) were self-reported, and their reliability and validity is un-known. The risk percentages
report-ed for Motorola may reflect some problems with recall. They may also differ from percentages reported for other companies or populations. In addition, there are likely to be fac-tors that influence health care ex-penditures that could not be includ-ed, and their influence is unknown. The expenditure analyses reported here were based on the expenditure regressions conducted by Goetzel et al.,2which explained 19% of the
vari-ability in health care expenditures among HERO-member organizations.
Fourth, a 10-year time horizon was adopted here. Other companies may prefer different time horizons. Short-er horizons would require greatShort-er re-ductions in risk in order to break even, and longer horizons would re-quire lesser reductions in risk.
Finally, the nature of any forecast-ing process involves some random er-ror that cannot be avoided. No one can predict the future with 100% ac-curacy, and neither can the methods proposed here. There will always be some residual error in any forecast-ing process. The value of the meth-ods proposed here lies in whether the projections they produce are bet-ter than those that would result from alternative approaches that are either not data driven or use different types of data to forecast expenditures. Re-searchers are therefore encouraged to develop alternative approaches to compare with the one presented here.
CONCLUSIONS
In a highly competitive and fluid business environment, business man-agers require analytic tools to make judgments about where to invest company resources. These managers are often faced with the decision of whether to offer, or maintain, health promotion and wellness programs for their employees.
One rationale for offering these programs is that they can save the company money by reducing medical expenditures. If risk reduction will yield savings in medical expenditures, there may be a level of risk reduction associated with enough savings to jus-tify the monetary investment in health promotion programming
re-quired to help reduce those risks. This paper illustrated a process that can be used to estimate that break-even level of risk reduction at Moto-rola. The estimates obtained from this process can then be used by se-nior managers to determine how much health promotion and wellness activities to offer. Final investment decisions might be influenced by the perceived likelihood that risks can be reduced to a break-even point. For any particular health promotion or wellness program, the decision to of-fer it may be influenced by its attrac-tiveness relative to other investments that have their own expected return on investment.
SO WHAT? Implications for Health Promotion Practitioners and Researchers
Information from published studies and a company’s own data can
be used to estimate the annual amount of risk reduction required to break even on the company’s investment in health promotion programming. If this assertion holds true, senior managers can use these estimates to help decide how much to invest in health pro-motion. Accordingly, practitioners can use such estimates to help set risk-reduction targets for the pro-grams they design and use to manage health risks. Researchers can either use the regression-based methods noted here to fore-cast break-even risk reduction per-centages for their clients, or they can conduct their own studies rep-licating the regression analyses and then use the methods de-scribed
here to make projections tailored to their own organizational experi-ence.
Acknowledgments
Work on this project was funded by Motorola; their sup-port is greatly appreciated. In addition, the authors would like to thank Kathy Reus and Laura Christian at Medstat for programming assistance; Arlene Guin-don at Medstat for managing work on the project; and Gary Fine at Wellness, Inc, for contributing data and assisting with technical aspects of the research. Finally, the authors would like to thank Ernie Meyer, Motorola’s Director, Global Rewards—Wellness & Education, for his critical review. The opinions expressed herein are the
authors’ and do not necessarily reflect the opinions of their affiliated organizations.
References
1. Centers for Disease Control. Recommenda-tions for prevention and control of hepatitis C virus (HCV) infection and HCV-related chronic disease.MMWR.1998;47:RR-19. 2. Goetzel RZ, Anderson DW, Whitmer RW, et
al. The relationship between modifiable health risks and health care expenditures: an analysis of the multi-employer HERO health risk and cost database.J Occup Environ Med.
1998;4:843–857.
3. Ozminkowski RJ, Mark TL, Goetzel RZ, et al. Relationships between substance use, medical expenditures, and the occurrences of injuries at a large manufacturing firm.Am J Drug Alco-hol Abuse.2003;29:151–167.
4. Reynolds C. The health and productivity management movement (HPM)—three key insights, one radical conclusion.Employer Health Management eNews.June 2001. 5. Leutzinger JA, Ozminkowski RJ, Dunn RL, et
al. Projecting future medical care costs using four scenarios of lifestyle risk rates.Am J Health Promot.2000;15:35–44.
6. Krahn M, Gafni A. Discounting in the eco-nomic evaluation of health care interven-tions.Med Care.1993;31:403–418. 7. Nas T.Cost-Benefit Analysis.Thousand Oaks,
CA: Sage; 1996.
8. Gold MR, Siegel JE, Russell LB, Weinstein MC.Cost-effectiveness in Health and Medicine.
New York: Oxford University Press; 1996. 9. Gramlich EM.A Guide to Benefit-cost Analysis.
2nd ed. Englewood Cliffs, NJ: Prentice-Hall; 1990.
10. Goetzel RZ.The Role of Business in Improving the Health of Workers and the Community. Wash-ington, DC: National Academy of Sciences (NAS); 2001.
11. Edington DW. Emerging research: a view from one research center.Am J Health Promot.
2001;15:341–349.
12. Duan N. Smearing estimate: a nonparametric retransformation method.J Am Stat Assoc.
1983;78:605–611.
Appendix
Equations One Must Solve to Estimate the Amount of Risk Reduction Required to Break Even in Health Promotion Programming at Motorola or Other Firms
Estimating Risk
The following equation was used to es-timate the value of risk factoriin yeart for Motorola. This equation refers to re-gression coefficients obtained from the Leutzinger et al.5study, which are abbrevi-ated by the letterL,followed by the vari-able of interest as a subscript. For exam-ple,LAgerefers to the Leutzinger et al.5 regression coefficient for the age variable used in their analysis.
Riskit
5 Intercept from Leutzinger et al. for risk factori
1 (LAge3Mean age value for Motoro-la for yeart)
1 (LFemale gender3% of female workers for Motorola for yeart)
1 (LAfrican-American race3% of African-American employees for Motorola for yeart)
1 (LHispanic3% of Hispanic employ-ees for Motorola for yeart) 1 (Lother non-White race3% of other
non-White employees for Motorola for yeart)
1 (LSales job type3% of Motorola em-ployees with sales jobs in yeart) 1 (LProfessional job type3% of Motorola
employees with professional jobs for yeart).
Since there were 11 risk factors and 11 years in the study period, there were 121 total calculations like the one noted in the equation above. The results from each calculation produced a log odds val-ue for each particular risk factor and year of interest. These log odds values were then transformed into the predicted prob-abilities of being at high risk using the following mathematics equation to find the predicted probability of being at high risk for a risk factoriin yeart:
Predicted probabilityit
5 e(log odds)/(11e(log odds)). In other words, the log odds value for any particular risk factor was transformed into the predicted probability of being at high risk by exponentiating the log odds value, then dividing the result by (1.01 the exponentiated value). This process was used to predict the probabilities that Motorola employees would be at high risk for each risk factor for each year in the study period.
Estimating Medical Expenditures
The following equations describe how medical expenditures were estimated for each year of interest. Some of the infor-mation for these equations was obtained from Goetzel et al.2The same naming process for regression coefficients that was described above for Leutzinger et al.5was used to refer to the regression coefficients obtained from Goetzel et al.2(i.e., such variables are abbreviated by the letterG, followed by the variable of interest noted as a subscript).
The first equation predicted the proba-bility that the typical employee would have positive dollars in medical expendi-tures in the year of interest based on the demographic and risk projections for that year. The second equation predicted the magnitude of those expenditures for those expected to have any. The third equation combined the results from the
previous two to find the overall medical expenditure estimate per employee.
The following is the equation to pre-dict the likelihood of having any medical expenditures per employee for yeart:
Likelihood of having any (non-zero) ex-penditures
5 Intercept from the Goetzel et al. logis-tic regression equation
1 (GAge3Mean age value for Motoro-la for yeart)
1 (GFemale gender3% of female workers for Motorola for yeart)
1 (GAfrican-American race3% of African-American employees for Motorola in yeart)
1 (GHispanic3% of Hispanic employ-ees for Motorola for yeart) 1 (Gother non-White race3% of other
non-White employees for Motorola in yeart)
1 (GSales job type3% of Motorola em-ployees with sales jobs in yeart) 1 (GProfessional job type3% of Motorola
employees with professional jobs in yeart)
1 (GExercise risk3% of Motorola em-ployees expected to be at high exer-cise risk in yeart)
1 (GEating habits risk3% of Motorola em-ployees expected to be at high risk for poor eating habits in yeart) 1 (GDeviation from ideal body weight3% of
Motorola workers expected to devi-ate substantially from ideal body weight in yeart)
1 (GCurrent smoking3% of employees ex-pected to smoke in yeart)
1 (GFormer smoker3% of employees ex-pected to be former smokers in yeart)
1 (GHigh cholesterol3% of employees ex-pected to be at high risk due to cholesterol values in yeart) 1 (GHigh blood glucose3% of Motorola
employees expected to have high blood glucose in yeart)
1 (GHigh blood pressure3% of Motorola employees expected to have high blood pressure in yeart) 1 (GHigh stress3% of employees
ex-pected to have high stress in yeart) 1 (GDepression3% of employees
ex-pected to have depression in yeart) 1 (GHeavy alcohol use3% of workers
ex-pected to drink heavily in yeart).
The output from this equation was the log odds of having any medical expendi-tures in yeart.This was transformed into a predicted probability using the same process described above for health risks.
Equation to Predict Magnitude of Medical Expenditures
A similar equation was used to predict the amount of medical expenditures em-ployees with non-zero expenditures were expected to have. This equation had the same format as the one noted immediate-ly above for the log-odds equation, except that the coefficients were taken from an ordinary least squares regression (not a logistic regression equation) published by Goetzel et al.2Since the expenditure equation estimated by Goetzel et al.2was
cast in terms of log dollars, the predicted values obtained from the use of that re-gression were also cast in terms of log dollars. These were then exponentiated and multiplied by a smearing estimate to transform them into actual dollars, as sug-gested by Duan.12
Putting the Two Equations Together to Obtain Total Expenditure Estimates
For each year, the predicted probabili-ty of having any expenditures (obtained
from the first expenditure equation) was multiplied by the predicted magnitude of those expenditures (which was obtained from the second expenditure equation) to estimate total medical expenditures per employee for Motorola:
Total $t
5 (Probability of having any $t31.32)
3(Expected magnitude of those $t
A fusion of the best of science and the best of practice —
together, to produce the greatest impact.
Stay on top of the science and art of health promotion with
your own subscription to the American Journal of
Health Promotion.
“ The American Journal of Health Promotionprovides a forum for that rare commodity — practical and intellectual exchange between researchers and practitioners. ”
Kenneth E. Warner, PhD
Avedis Donabedian Distinguished University Professor of Public Health School of Public Health, University of Michigan
“ The contents of the American Journal of Health Promotionare timely, relevant,and most important, written and reviewed by the most respected resesarchers in our field. ”
David R. Anderson, PhD
Vice Programs and Technology, StayWell Health Management
Definition of Health Promotion
“Health Promotion is the science and art of helping people
change their lifestyle to move toward a state of optimal
health. Optimal health is defined as a balance of physical,
emotional, social, spiritual and intellectual health. Lifestyle
change can be facilitated through a combination of efforts
to enhance awareness, change behavior and create
environments that support good health practices. Of the
three, supportive environments will probably have the
greatest impact in producing lasting change.”
(O’Donnell, American Journal of Health Promotion, 1989, 3(3):5.)
DIMENSIONS OF
OPTIMAL HEALTH
Subscribe today...
Editor in Chief
Michael P. O’Donnell, PhD, MBA, MPH
Associate Editors in Chief
Bradley J. Cardinal, PhD Diane H. Morris, PhD, RD Judy D. Sheeska, PhD, RD EDITORS Interventions Fitness Barry A. Franklin, PhD Medical Self-Care Donald M. Vickery, MD Nutrition Karen Glanz, PhD, MPH Smoking Control Michael P. Eriksen, ScD Weight Control Kelly D. Brownell, PhD Stress Management Paul J. Rosch, MD Mind-Body Health Kenneth R. Pelletier, PhD, MD (hc) Social Health Kenneth R. McLeroy, PhD Spiritual Health Larry S. Chapman, MPH Strategies Behavior Change James F. Prochaska, PhD Culture Change Daniel Stokols, PhD Health Policy Kenneth E. Warner, PhD Applications Underserved Populations Ronald L. Braithwaite, PhD
Health Promoting Community Design
Jo Anne L. Earp, ScD Research Data Base David R. Anderson, PhD Financial Analysis Ron Z. Goetzel, PhD
Method, Issues, and Results in Evaluation and Research
Lawrence W. Green, DrPH The Art of Health Promotion
Larry S. Chapman, MPH
ANNUAL SUBSCRIPTION RATES:
Individual
Institution
U.S.
$99.95
$137.95
Canada and Mexico
$108.95
$146.95
Other Countries
