Content uploaded by Ronald J Ozminkowski
Author content
All content in this area was uploaded by Ronald J Ozminkowski on Jun 29, 2018
Content may be subject to copyright.
CME AVA I L A B L E F O R THIS ARTI CL E AT ACOEM.ORG
Improving Employee Productivity Through Improved Health
Rebecca J. Mitchell, MPH, Ronald J. Ozminkowski, PhD, and Seth Serxner, PhD
Objective: The objective of this study was to estimate productivity-related
savings associated with employee par ticipation in health promotion programs.
Methods: Propensity score weighting and multiple regression techniques
were used to estimate savings. These techniques were adjusted for demo-
graphic and health status differences between participants who engaged in
one or more telephonic health management programs and nonparticipants
who were eligible for but did not engage in these programs. Results: Em-
ployees who participated in a program and successfully improved their health
care or lifestyle showed significant improvements in lost work time. These
employees saved an average of $353 per person per year. This reflects about
10.3 hours in additional productive time annually, compared with similar, but
nonparticipating employees. Conclusions: Participating in health promotion
programs can help improve productivity levels among employees and save
money for their employers.
Employees who have health problems are often less productive at
work .1,2 Thus, employee absences and on-the-job productivity
losses due to poor health cost employers nearly $230 billion in the
United States each year.3For some companies, productivity-related
losses may even exceed the costs of direct medical care.4In an effort
to control costs and improve health, employers seek solutions that
improve employee health habits and increase the appropriateness or
coordination of their health care—these may also lead to a more
productive workforce.5
When health promotion programs are effective, it has been
shown that they may save an average of $3.27 in medical costs and
$2.73 in absenteeism costs, per dollar spent on these programs.6To
address the increasing interest in productivity as a key element to the
value proposition of health promotion programs, the objective of this
study was to examine the relationship between program engagement
and productivity metrics across a variety of employers. By examining
a large sample of workers with identified gaps in health care or
lifestyle, we estimate the productivity impact associated with better
managing their health risks and their need for health care services.
This study is organized as follows: First, we briefly describe
the wellness, disease management, and other interventions that were
designed to help improve health and productivity among the employ-
ees of several companies. Then, we describe the sample members
whose data were used for our main analyses of program impact.
We also describe the samples used in sensitivity analyses that were
designed to relax the key assumptions of the main analyses. We
then proceed by noting how the productivity loss variables were con-
structed for this study, and how variables that may influence program
engagement and productivity were constructed. After that we explain
From OptumHealth, Golden Valley, Minn.
Conflicts of interest and source of funding: Funding for the study was provided by
and the authors are employed by Optum. The authors’ compensation was not
tied to any revenues gained from or the outcomes associated with the program
services described in this article.
Authors Mitchell, Ozminkowski, and Serner have no relationships/conditions/
circumstances that present potential conflict of interest.
The JOEM editorial board and planners have no financial interest related to this
research.
Address correspondence to Rebecca J. Mitchell, MPH, OptumHealth, 6300 Ol-
son Memorial Highway, Golden Valley, MN 55427 (Rebecca.Mitchell@
Optum.com).
Copyright C2013 by American College of Occupational and Environmental
Medicine
DOI: 10.1097/JOM.0b013e3182a50037
Learning Objectives
rOutline the wellness and disease management interventions
evaluated in the current study, and the methods used to
assess their impact on employee productivity.
rSummarize the reduction in lost work time among pro-
gram participants who successfully improved their health or
lifestyle.
rDiscuss the implications for calculating the economic bene-
fits of employee health promotion programs.
the statistical analyses used to estimate the impact of participation
in health promotion programs on work productivity. Results from
the main analyses and sensitivity analyses are then presented, fol-
lowed by a discussion of the implications of our findings and some
concluding comments.
THE INTERVENTIONS
Health promotion programs typically provide services that ed-
ucate consumers about their health conditions and treatment needs.
These programs also help consumers navigate the health care sys-
tem and stay on track with their health improvement goals. In the
programs that are the focus of this study, care advocates, health
coaches, and nurses delivered wellness, lifestyle coaching, disease
management, and decision support services to employees with one
or more identified health improvement opportunities. These services
were delivered by Optum, a UnitedHealth Group business dedicated
to making the health system work better.
The Optum programs included in this study addressed health
conditions commonly associated with productivity loss, including
asthma, coronary artery disease, diabetes, chronic obstructive pul-
monary disease, cancer, back pain, weight problems, and tobacco use
(see Table 1). Information about the existence of these conditions
was reported on the health risk appraisal (HRA) surveys completed
by each sample member.
Program services included one or more telephone-based dis-
cussions with nurses, health advocates, or lifestyle coaches about
the need for high-quality, evidence-based ways to treat these prob-
lems and how to avoid complications in the future. Each program
participant also received up to four mailings over the course of the
study period. These mailings described the need for preventive ser-
vices, the kinds of symptoms or problems that should be discussed
with physicians, and potentially more affordable care alternatives
to discuss with their doctors as well (eg, pill splitting or generic
pharmaceuticals that can be used instead of brand name drugs). Re-
minders to make appointments with doctors to address these issues
were also sent via mail.
The goals of working with consumers of these program ser-
vices were to give people the information and tools they need to
improve their health, to help reduce their medical costs, and to help
them increase their work productivity. This article focuses on the
productivity-related savings estimated for those who participated in
these programs.
Copyright © 2013 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
1142 JOEM rVolume 55, Number 10, October 2013
JOEM rVolume 55, Number 10, October 2013 Improving Employee Health and Productivity
STUDY SAMPLES FOR THE MAIN ANALYSIS AND
SENSITIVITY ANALYSES
Sample for the Main Analyses
The main study sample included individuals employed by
Optum clients between May 2010 and April 2011. Among these
individuals were program participants and program-eligible nonpar-
ticipants. Participants were those who worked with a nurse, health
coach, or health advocate to address one or more health improvement
opportunities during this program period. Working on an opportu-
nity means that employees completed at least one telephone call with
one of these professionals, in an effort to close gaps in their needs
for health care or to improve their lifestyle choices in ways that can
help make them healthier. Nonparticipants included those who were
eligible for these programs but did not participate in a telephonic
program.
To be included in the study, sample members must have
completed two HRA surveys. These surveys were the source of
information about work productivity and other key variables needed
for the analysis. For program participants, the first HRA must have
been completed up to 12 months before the initial date of program
activity, and the second HRA must have been completed within 6
to 18 months after that program start date. For nonparticipants, the
first HRA must have been completed up to 12 months before May
1, 2010 (the start of the program period), and the second HRA must
have been completed within 6 to 18 months after that date. These
HRA completion date criteria were meant to ensure that any re-
ported productivity gains or losses occurred after the start of the
Optum programs.
TABLE 1. Health and Disease Management Programs
Included in Analyses
Program Type Condition or Health Risk Addressed
Disease management/
decision support
Asthma
Back pain
Cancer
Chronic obstructive pulmonary disease
Diabetes
Heart disease/heart failure
Wellness/lifestyle Diabetes lifestyle
Exercise
Heart health
Nutrition
Tobacco cessation
Weight management
Several exclusion criteria for this study were also applied.
First, we excluded people who called only the Optum nurse line
telephonic triage program to receive information on where best to
seek medical care for an acute problem that may need treatment
immediately or within a very short period. Members who call the
nurse line often have problems that are not typically addressed in
wellness or disease management programs, which have a longer-
term focus. Next, we excluded those who used only the mail-based
program services mentioned previously, because they would not have
received any telephonic services.
We also excluded those who used only programs devoted to
maternity, neonatal, transplant, or chronic kidney disease issues.
These exclusions were made because HRA respondents did not
report productivity losses related to those health conditions on the
HRA.
In addition, we excluded members who did not have insur-
ance coverage from UnitedHealthcare Insurance Company (or Unit-
edHealthcare Insurance Company of New York, for residents there),
because such members may have completed a different type of health
risk assessment and would likely have different features of insurance
coverage that could have influenced results. Finally, we excluded
members younger than 18 years or older than 70 years at the time of
the study, so the focus would be on a working-age population.
All study data were de-identified. After applying the inclusion
and exclusion criteria just mentioned, there were 131,011 individuals
eligible for analyses (3793 participants and 127,218 nonparticipants;
see Fig. 1).
Samples for the Sensitivity Analyses
To determine whether differing levels of participation affected
outcomes, we conducted sensitivity analyses based on the intensity
of program participation. For our first sensitivity analysis, we distin-
guished between (1) members who mentioned closing at least one
gap in their need for health care or lifestyle changes during con-
versations with their program providers and (2) members who had
program activity but who had not yet closed any such gaps during
the study period. Savings were expected to be higher for the former
group and lower for those who had not yet closed any gaps.
In a second sensitivity analysis, we limited the sample of
participants and nonparticipants to those who were found to be qual-
ified for Optum programs by a predictive model that used medical
claims data to identify people at risk for serious chronic health condi-
tions. These sample members were more likely to qualify for disease
management programs than for wellness/smoking cessation/weight
reduction programs.
Measures of Work Loss and Other Variables Needed
for Analysis
To estimate the impact of program participation on productiv-
ity, it was necessary to create two measures of work loss, one related
to absenteeism from work and one related to lower productivity while
FIGURE 1. Inclusion criteria for partici-
pants and nonparticipants.
Copyright © 2013 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
C2013 American College of Occupational and Environmental Medicine 1143
Mitchell et al JOEM rVolume 55, Number 10, October 2013
at the workstation (ie, presenteeism). These are described first. Then
we describe other variables that are associated with decisions to par-
ticipate in the Optum programs and variables that influence work
loss. These variables are needed for subsequent statistical analy-
ses that are designed to distinguish between the impact of program
participation and the impact of those other variables on work loss.
Work Loss Measures
We measured work productivity losses from the HRA survey
responses that were provided before and after the start of the Optum
programs. The HRA instrument used in this study was developed,
validated, and administered by the University of Michigan Health
Management Research Center.
Average annual rates of absenteeism and presenteeism were
measured for each sample member. Absenteeism was measured as
the self-reported number of days of work missed in past year due to
illness. Presenteeism was meant to account for productivity losses
while at the workstation due to poor health. The HRA included the
eight-item short form of the Work Limitations Questionnaire (WLQ)
developed by Lerner and colleagues7to measure presenteeism. Using
the WLQ responses and methods to estimate at-work productivity
loss that are described in detail by Lerner and her colleagues, we
calculated each sample member’s percentage of at-work time that
was lost because of health reasons.
Once the measures of absent days and percent of time lost due
to presenteeism were obtained, we transformed those measures into
monetary equivalents. Although monetization of these productivity
losses is not necessary to estimate the impact of program partic-
ipation, it is often desired by employers who would like to better
understand the financial bottom-line impact of these programs.
We monetized work loss attributable to absenteeism in the past
year, using the lost wages method often used by health economists.8,9
That is, we multiplied the total number of workdays missed because
of poor health by an estimate of average daily compensation. Daily
compensation figures were neither reported on the HRA nor available
from Optum clients, so the monetization process was based on the
average dollar value of daily wages plus benefits for civilian workers
in the United States in 2011. This information was obtained from
the Bureau of Labor Statistics.10 Thus, the monetary value of absen-
teeism was first estimated as $240.88 per day of work missed, multi-
plied the number of days that were missed as reported on the HRA.
To complete theabsence cost monetization process, we multi-
plied the absenteeism dollar figure by an average wage “multiplier”
of 1.61. The multiplier was derived by Nicholson et al11 to account
for additional costs associated with absence. These additional costs
reflect the potential inability of an employer to replace the absent
worker when necessary and the potential effect of absence on the
work conducted by team members. Then, given the lack of con-
sensus in the health management community about monetization
methods for productivity loss, we estimated absenteeism losses in
two ways, once with and once without applying the wage multiplier.
Turning now to presenteeism, note that the WLQ questions
used to measure the percentage of time lost at the workstation due
to poor health were asked with reference to the most recent 2-week
period from the time the respondent completed the HRA. Never-
theless, our desire was to annualize presenteeism losses because
most employers evaluate health and productivity costs over a 1-year
timeframe.12 We, therefore, monetized presenteeism by extrapolat-
ing the estimated percentage of productivity loss in the past 2 weeks
to a 1-year time period.
For example, if a respondent’s WLQ data led to an estimated
5% productivity lost during the last two weeks, we assumed that
this 5% figure would apply for the entire year. Total presenteeism-
related losses for this person would, therefore, equal the result of
the following math. First, multiply 5% (ie, the percentage of days of
work lost due to presenteeism over the year) by 240 eligible workdays
per year; this yields the equivalent loss of 12 days of work. Then,
multiply that number of days lost by the average daily compensation
figure of $240.88 noted previously. For this sample member, the
estimated presenteeism loss in dollar terms would be $2890 (ie, the
equivalent of 12 days lost because of lower performance at work ×
$240.88 per day =$2890).
ADDITIONAL VARIABLES NEEDED FOR THE STUDY
In addition to the productivity loss variables mentioned pre-
viously, other variables were created for the subsequent statistical
analyses that were used to estimate the impact of the Optum pro-
grams. First, to help refine the extrapolation of presenteeism re-
ported in past 2 weeks to a 1-year period, we created a seasonal
variable to indicate the time of year in which the HRA was com-
pleted either during winter or spring (January–June) or summer or
fall (July–December). We examined the distribution of productivity
losses across these two periods and found that health-related pre-
senteeism rates differed slightly but significantly by season, ranging
from 3.25% in winter/spring to 2.87% in summer/fall (P≤0.0001)
in the baseline period, and 2.95% to 3.02% in the follow-up period
(P=0.04). Therefore, in the subsequent statistical analyses, we in-
cluded the seasonal indicator to adjust for any seasonal impacts on
productivity levels.
A number of other covariates were also measured because
these might be associated with the decision to participate in the
Optum programs and/or workplace productivity. These included the
sample member’s age and sex; the existence of chronic health con-
ditions referenced on the HRA; his or her productivity loss in the
baseline (preintervention) period; and self-reported lifestyle risk fac-
tors related to stress levels, tobacco use, job satisfaction, and alcohol
use. Table 2 reports the criteria used to determine whether a partic-
ipant was at high risk based on any of the lifestyle factors. Table 3
lists the chronic conditions that were measured for this study. Binary
(yes or no) indicators for each person were then generated for being
at high risk due to these factors, and these indicators were included
in subsequent statistical models.
Finally, we included an overall estimate of the sample mem-
ber’s relative health status. This was done by calculating Symmetry
Episode Risk Group (ERG) scores based upon information in mem-
bers’ medical and pharmacy claims data from the 12-month inter-
vention period. Episode Risk Group scores use diagnosis codes, de-
mographic, and other information on medical and pharmacy claims
to create a score for each sample member that reflects the health
and pharmacy costs one would expect for the member in the coming
year, relative to the sample average. The ERG score has been shown
to be a rough proxy for health status.13 Higher values of the ERG
score roughly reflect more problematic health conditions.
STATISTICAL ANALYSES
The objective of the study was to estimate the health-related
productivity savings to employers associated with participation in
telephonic health promotion programs. Specifically, our intent was
to learn whether participation in wellness, disease management, and
decision support programs would reduce absenteeism and presen-
teeism, as measured via HRA survey data.
TABLE 2. Definitions of Health Risk Categories
Risk Factor High Risk Criteria
Stress level S-scale score >18
Tobacco use Still smoke
Job satisfaction Partly or not satisfied
Alcohol use >14 drinks/week for males; >7 drinks/week for females
Copyright © 2013 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
1144 C2013 American College of Occupational and Environmental Medicine
JOEM rVolume 55, Number 10, October 2013 Improving Employee Health and Productivity
TABLE 3. Participant Demographics Before and After Propensity Score Weighting
Before Weighting After Weighting
Demographic
Participant
(N=3793)
Nonparticipant
(N=127,218) Std Diff
Participant
(N=3793)
Nonparticipant
(N=127,218) Std Diff
Age 46 42 0.34 42 42 0.00
% Female 59 58 0.03*
ERG risk score 2.62 1.57 0.34 2.0 1.58 0.16†
Monetized baseline productivity loss $2,679 $2,493 0.04*
Season of HRA (% completed in summer/fall) at
follow-up
55 39 0.33 40 39 0.02
Lifestyle risk factor (% at risk)
Tobacco use 6 9 0.12 8 9 0.02
Stress risk 24 21 0.07*
Job risk 12 10 0.06*
Alcohol risk 2 3 0.05*
Chronic health condition (% reported “have currently”)
Allergies 34 28 0.13 28 28 0.01
Arthritis 19 11 0.23 11 11 0.00
Asthma 8 5 0.12 5 5 0.01
Back pain 23 14 0.24 15 14 0.02
Cancer 2 0.3 0.16 0.3 0.3 0.00
Depression 11 7 0.13 7 7 0.01
Diabetes 13 5 0.30 5 5 0.02
Heart disease 4 2 0.13 2 2 0.00
Migraine 7 7 0.00*
Pain 12 6 0.21 6 6 0.01
*Metric not statistically different by participation status at baseline so removed from propensity score model.
†Standardized difference in means is higher for these variables than others so included in final weighted reg ression models.
Std diff, difference in means or proportions divided by standard error.
Before making inferences about the impact of the Optum
programs, it was necessary to ensure that program participants and
nonparticipants were similar to each other, at least in terms of mea-
surable variables that influenced program participation decisions and
productivity levels. Propensity score analyses were used to do this.
Developed in the early 1980s,14 there is a long history of using
propensity score analyses in a number of ways in program evalua-
tion work. Useful discussions of alternative methods (including the
propensity weighting method we used) can be found in Faries et al15
and in Curtis et al.16
The notion behind propensity score adjustment is that each
person has an underlying nonzero probability of participating in
an intervention, regardless of whether he or she actually did
participate. If we can estimate that underlying probability for each
sample member based upon measurable factors such as demograph-
ics and other variables, we can then create a case weight that is based
upon that probability.17 Applying the case weight in subsequent sta-
tistical analyses helps make program participants look more similar
to nonparticipants, and make the nonparticipants look more simi-
lar to participants, based upon variables we can measure for each
group. Doing this helps equalize the two groups on average, making
the subsequent comparisons of their productivity more valid.
To apply the propensity-weighting technique, we first created
a propensity score for each program participant and nonparticipant.
This was done by using a logistic regression model to estimate the
probability that each person participated in an Optum program, based
on his or her age, sex, productivity losses before the Optum programs
began, their reported health conditions, lifestyle risk factors, and
their ERG health risk scores. We then used the predicted probability
of program participation for each sample member to create the case
weight for him or her.
More specifically, as suggested in the literature, the case
weight was constructed as the multiplicative inverse of the predicted
probability of being in the participant or nonparticipant group each
sample member was actually in.18 This propensity score-weighting
approach allowed us to use the data from all of the program par-
ticipants and nonparticipants in our analyses, enhancing the gener-
alizability of the results to all who met the program inclusion and
exclusion criteria.
To assess whether the propensity score weighting process
worked to make program participants and nonparticipants look more
like each other, we measured differences in their demographics,
health risks, chronic conditions, and ERG scores before versus af-
ter propensity weighting was applied (weighting should drastically
reduce these differences). To do this, we calculated standardized dif-
ferences in variable means or percentages before versus after weight-
ing. This technique showed whether weighting reduced demographic
and other differences between participants and nonparticipants in a
way that controlled for the large differences in sample sizes between
these two groups.
Finally, to estimate the impact of program participation on the
absenteeism and presenteeism metrics, propensity-weighted gener-
alized linear regression models were used. The dependent variables
for these analyses were the sum of the annual health-related ab-
senteeism and presenteeism metrics mentioned previously, in their
monetized forms. The independent variables for the regression anal-
yses included a binary indicator to denote participant status and any
covariates with standardized differences after weighting that were
Copyright © 2013 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
C2013 American College of Occupational and Environmental Medicine 1145
Mitchell et al JOEM rVolume 55, Number 10, October 2013
greater than 0.10 (as suggested in the literature to denote meaningful
imbalance between groups).19 Using the regression results, we then
calculated net lost productivity by participation status as the differ-
ence in the least squares means of monetized lost productivity for
program participants (cases) and nonparticipants. All analyses were
performed using SAS software version SAS 9.1.3 (SAS Institute,
Inc, Cary, NC).
RESULTS
Sample Characteristics
Of the employees who participated in health promotion
programs during the 12-month intervention period, 9320 (7%)
completed two HRA surveys and 3793 (3%) met further study inclu-
sion criteria with regard to applicable dates of HRA survey comple-
tion before and after program activity dates (Fig. 1). Of the 322,853
employees who completed two HRA surveys but did not participate
in a program during the study period, more than 127,000 (39%) met
study inclusion criteria.
Some demographic variables and lifestyle risk factors, the sea-
son in which HRA surveys were completed, the presence of health
conditions, and the average ERG score differed significantly between
participants and nonparticipants before propensity weighting. Par-
ticipants were older, had higher ERG scores, and were more likely
to report having select health conditions than nonparticipants. Sex,
baseline productivity loss, and stress, job, and alcohol risk status did
not differ between participants and nonparticipants before weighting.
Table 3 shows mean values and standardized differences in
means before and after propensity score weighting. After weighting
the data to adjust for differences, only the ERG risk score had a
standardized difference over 0.10. It was, therefore, included as a
covariate in the subsequent regression analyses.
ESTIMATING PROGRAM IMPACT—RESULTS FROM
WEIGHTED REGRESSION
Table 4 presents results from the weighted multiple regression
analyses. These analyses demonstrate the impact of participating in
a health management program on changes in self-reported produc-
TABLE 4. Productivity Gained From Program
Participation: Results From Propensity-Weighted Regression
Analysis
βSE TP 95% CL
Monetized absenteeism and presenteeism
Intercept $1,621 $68 23.8 <0.0001 $1,488–$1,755
Participant $2,054 $67 $1,922–$2,186
Nonparticipant $2,407 $12 $2,384–$2,429
Difference $353 $68 5.2 <0.0001 $219–$487
Risk score $282 $5.2 54.6 <0.0001 $272–$293
Days absent
Intercept 1.04 0.05 22.1 <0.0001 0.94–1.13
Participant 1.57 0.05 1.48–1.66
Nonparticipant 1.78 0.01 1.76–1.79
Difference 0.21 0.05 4.4 <0.0001 0.12–0.30
Risk score 0.33 0.00 97.8 <0.0001 0.33–0.34
Time unproductive, %
Intercept 2.12 0.11 19.3 <0.0001 1.9–2.3
Participant 2.50 0.11 2.3–2.7
Nonparticipant 2.95 0.02 2.9–3.0
Difference 0.45 0.11 4.1 <0.0001 0.24–0.67
Risk score 0.24 0.01 29.5 <0.0001 0.23–0.26
tivity loss over time. Program participants reported higher levels
of productivity than nonparticipants, yielding average annual sav-
ings of $353 per employee. Removing the wage multiplier from
monetization calculations reduced the average savings to $325 for
participants, and applying the wage multiplier to both absenteeism
and presenteeism increased savings to $386. When compared with
nonparticipants, differences in productivity levels were statistically
significant.
Sensitivity Analysis
Sensitivity analyses revealed that findings varied according
to participation levels and according to whether the claims-based
predictive model was the source of information about program qual-
ification status (Table 5). When the study sample was limited to
those flagged as qualified by the predictive model, potential annual
productivity savings for participants averaged $357. As expected, all
participants who closed one or more health care or lifestyle gaps in
care had higher levels of productivity, with annual average savings
of $401 per employee. This increased to $485 when limited to those
who closed at least one gap and who qualified for the programs via
the claims-based predictive model. In contrast, participants who did
not close at least one health care or lifestyle gap in care did not re-
port significantly different levels of productivity loss compared with
similar nonparticipants, as one might expect.
DISCUSSION
This study examined whether participation in telephonic
health promotion programs was associated with increased produc-
tivity among employees of several firms. A sample of employees
with data over a 3-year observation period was available to test this
hypothesis. After we made propensity weighting and regression-
based adjustments, annual productivity levels were higher among
employees who participated in these programs, compared with sim-
ilar employees who did not participate in the programs.
The productivity metrics used for this study were based upon
information reported in HRAs. One might wonder whether self-
reported data are valid and reliable, but at least one study found that
self-reported recall of health-related absence from work was reliable
and valid, particularly when recall periods are short.20 Moreover,
without detailed absenteeism records from each employer (which
are often difficult to get for all employees, even when studies are
limited to just one firm) and with no information from employers
about detailed work processes, self-reported data are often the only
data available for a study like this.
One may also wonder whether using national data from the
Bureau of Labor Statistics is sufficient to monetize absenteeism and
presenteeism losses. Actual costs of absences and unproductive time
are likely to differ by industry and employer and were not available
from each employer in this study. If we avoid the monetization ex-
ercise and just focus on time lost instead of the associated dollars,
we see that participants gained an average of about 10.3 hours in
productive time annually compared with similar, nonparticipating
employees. For a typical person who may work about 240 days per
year, this amounts to about a 0.5% productivity loss. Nevertheless,
for a large firm even a low percentage of productivity loss can be
equivalent to many total days of lost productivity. For example, for
a large employer with 10,000 employees and a 10% program partic-
ipation rate, these findings may translate to the equivalent of about
five full-time employees’ worth of work that would be saved per year
by program participation.
For this study, we implemented stringent inclusion require-
ments to create similar participant/nonparticipant groups based on
chronological dates of participation and HRA survey completion.
These temporal relationships resulted in a smaller participation rate
(3%) than is typical for health promotion programs in total (for
Optum’s telephonic programs the overall participation rate is about
Copyright © 2013 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
1146 C2013 American College of Occupational and Environmental Medicine
JOEM rVolume 55, Number 10, October 2013 Improving Employee Health and Productivity
TABLE 5. Sensitivity Analyses of Savings Gained From Program Participation: Results
From Propensity-Weighted Regression Analysis
Monetized Productivity Gains by Predictive
Model Qualification Status
Participant Status
Qualified Via All
Available Sources of
Information
Subset Flagged by
Predictive Model as
Qualified for a Program
All participants vs nonparticipants $353*,†$357†
Participants who closed one or more gaps in
health care or who reduced one or more health
risks vs nonparticipants
$401†$485†
Participants who have not yet closed one or
more gaps or reduced one or more risks vs
nonparticipants
$100‡$151‡
*Main study sample.
†Difference between participants and nonparticipants statistically significant at P<0.05 level.
‡Difference between participants and nonparticipants was not statistically significant.
11%). One might, therefore, question to whom our results can be
generalized. Technically, results can be generalized to the sample
used here, and to other samples from similar firms using similar
telephonic programs, provided that two or more HRAs like the ones
we used are completed. We estimate this to be about 27% of all Op-
tum program participants and nonparticipants (ie, 3%/11% =27%).
We do not know the associated percentages for those who are not
served by Optum.
Next, the sensitivity analyses we conducted revealed that pro-
ductivity levels differed according to the level of participation. Sav-
ings were higher for those who successfully closed at least one gap
in the need for health care or in the need for lifestyle changes that
would reduce their risks for poor health in the future. Nevertheless,
one might not expect to see big savings from those who are still
working with their program providers but who have not yet made
definitive progress toward their goals. In short, it takes time and
commitment for program participation to yield success. Further re-
search could help determine the likelihood that a typical participant
will eventually be successful at closing at least one gap in care or
reduce at least one lifestyle risk.
There are a few remaining limitations to this study that are
worth noting. Two of these may have resulted in overestimating pro-
gram impact, and two may have resulted in underestimating program
impact. First, we tried to limit selection bias that might overestimate
program impact by using weighted propensity score analyses to make
the participant and nonparticipant groups more comparable, but we
could not measure all of the potential ways that these groups may
differ. Factors that may have contributed to productivity loss that we
were unable to measure in this study include the levels of techno-
logical support or resources available to employees at home, on the
Internet, or at their worksite that might influence their productivity.
We also could not measure each employee’s workload directly, nor
could we measure their motivation to make health improvements.
Thus, our results may still be influenced by some selection bias,
and the extent of such bias is unknown. This is typical in program
evaluation work.21
Next, as mentioned earlier, the intent was to estimate produc-
tivity loss on an annual basis so we converted presenteeism estimates
to annual rates. Extrapolating values from shorter reporting periods
to values that extend to 1 year could overestimate time losses.6We
attempted to alleviate overestimates of presenteeism by controlling
for season in which study members reported productivity loss in the
regression models.
The two reasons why productivity gains may be underesti-
mated in this study are as follows. First, any productivity increases
among those who used Optum interventions delivered via health por-
tals or on-site at workplaces could not be examined, as we could not
measure to what extent such resources were used. Second, complet-
ing the HRA might motivate positive changes in health or lifestyle,
even in the absence of program participation. This is possible be-
cause the HRA process involves feedback to the employee about their
health status and recommendations for making healthy lifestyle im-
provements. Therefore, even nonparticipants in the Optum programs
received this information. This may have affected productivity levels
reported in the follow-up period and could have worked to reduce
differences in productivity gains that are estimated when comparing
participants with nonparticipants.
Despite these limitations, this study applied a robust evalua-
tion design using data from various employer types to determine the
implications of better managing employee health risks and health
conditions. We enhanced the credibility of findings by establishing
a temporal sequence of events (ie, by investigating productivity re-
ported before and after the intervention), by controlling for several
measurable predictors of productivity loss, and by examining differ-
ing levels of participation among employees who were engaged in
programs.
CONCLUSION
Results of this study add to the growing body of evidence that
investing in a healthy workforce can help increase productivity levels
of employees. Findings suggest that when an employee engages with
a health coach, health advocate, or nurse and makes improvements
in health or reductions in health risks, he or she is likely to become
more productive.
REFERENCES
1. Burton WN, Chen CY, Conti DJ, Schultz AB, Edington DW. The association
between health risk change and presenteeism change. J Occup Environ Med.
2006;48:252–263.
2. Serxner SA, Gold DB, Bultman KK. The impact of behavioral health risks on
worker absenteeism. J Occup Environ Med. 2011;43:347–354.
3. Integrated Benefits Institute. Poor Health Costs U.S. Economy $576 Billion
(Press Release). September 12, 2012. IBIweb.org.
4. Loeppke R, Taitel M, Haufle V, Parry T, Kessler RC, Jinnett K. Health and
productivity as a business strategy: a multiemployer study. J Occup Environ
Med. 2009;51:411–428.
Copyright © 2013 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
C2013 American College of Occupational and Environmental Medicine 1147
Mitchell et al JOEM rVolume 55, Number 10, October 2013
5. Parry T, Sherman B. A pragmatic approach for employers to improvement
measurement in workforce health and productivity. Pop Health Manage.
2012;15:61–64.
6. Baicker K, Cutler D, Song Z. Workplace wellness programs can generate
savings. Health Aff. 2010;29:304–311.
7. Lerner D, Rogers WH, Chang H. Scoring the Work Limitations Questionnaire
(WLQ) Scales and the WLQ Index for Estimating Work Productivity Loss.
Boston, MA: The Health Institute of the Tufts-New England Medical Center;
2005.
8. Mattke S, Balakrishnan A, Bergamo G, Newberry SJ. A review of methods
to measure health-related productivity loss. Am J Manag Care. 2007;13:
211–217.
9. Berger ML, Murray JF, Xu J, Pauly M. Alternative valuations of work loss
and productivity. J Occup Environ Med. 2001;43:18–24.
10. US Department of Labor. Employer Costs for Employee Compensation—
September 2011. Washington, DC: US Bureau of Labor Statistics; 2011.
11. Nicholson S, Pauly MV, Polsky D, Sharda C, Szrek H, Berger ML. Measuring
the effects of work loss on productivity with team production. Health Econ.
2006;15:111–123.
12. Goetzel RZ, Long SR, Ozminkowski RJ, Hawkins K, Wang S, Lynch W.
Health, absence, disability,and presenteeism cost estimates of cer tain physical
and mental health conditions affecting U.S. employers. J Occup Environ Med.
2004;46:398–412.
13. OptumInsight (formerly Ingenix). Symmetry Episode Risk Groups: A Suc-
cessful Approach to Health Risk Assessment. White paper. Ingenix Inc: Eden
Prairie, MN; 2008.
14. Rosenbaum PR, Rubin DB. The central role of the propensity score in obser-
vational studies for causal effects. Biometrika. 1983;70:41–55.
15. Faries DE, Leon AC, Haro JM, Obenchain RL. Analysis of Observational
Health Care Data Using SAS. Cary, NC: SAS Institute; 2010.
16. Curtis LH, Gammill BG, Eisenstein EL, Kramer JM, Anstrom KJ. Using
inverse propensity-weighted estimators in comparative effectiveness analyses
with observational databases. Med Care. 2007;45(suppl 2):S103–S107.
17. Austin PC. An introduction to propensity score methods for reducing
the effects of confounding in observational studies. Multivar Behav Res.
2011;46:399–424.
18. Harder VS, Stuart EA, Anthony JC. Propensity score techniques and the as-
sessment of measure covariate balance to test causal associations in psycho-
logical research. Psychol Methods. 2010;15:234-249. doi: 10.1037/a0019623.
19. Austin PC. Balance diagnostics for comparing the distribution of baseline
covariates between treatment groups in propensity-score matched samples.
Stat Med. 2009;28:3083–3107.
21. Short ME, Goetzel RZ, Pei X, et al. How accurate are self reports? Analysis
of self-reported health care utilization and absence when compared with
administrative data. J Occup Environ Med. 2009;51:786–796.
22. Heckman JJ. Econometric causality. Int Stat Rev. 2008;76:1–27.
Copyright © 2013 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
1148 C2013 American College of Occupational and Environmental Medicine
