The I mpact of the Citibank, NA, Health
Management Program on Changes in
Employee Health Risks Over Time
Ozminkowski, Ronald J. PhD; Goetzel, Ron Z. PhD; Smith,
Mark W. PhD; Cantor, Richard I. MD; Shaughnessy, Anita
MEd; Harrison, Mary JD
Address correspondence to: Ronald J. Ozminkowski, The MEDSTAT
Group, Inc, 777 East Eisenhower Parkway, Office 915R, Ann Arbor, MI
48108; e-mail: email@example.com.
The MEDSTAT Group, Inc, Ann Arbor, Mich. (Dr Ozminkowski),
Washington, D.C. (Dr Goetzel, Dr Smith), and Stamford, Conn. (Dr
Harrison); and Citibank, New York (Dr Cantor, Ms Shaughnessy).
This study estimated the impact of the Citibank Health Management Program on changes in
health risks among Citibank employees. McNemar chi-squared tests compared the probability of
being at high risk for poor health when the first and last health-risk appraisal surveys were taken.
Logistic regression controlled for baseline differences in subsequent analyses when those who
participated in more intensive program features were compared with those who participated in
less intensive features. Declines in risk were noted for 8 of 10 risk categories. Most changes were
small, except those related to exercise habits, seatbelt use, and stress levels. For nine health risk
categories, those who participated in more intensive program services were significantly more
likely than others to reduce their health risks. Thus, the Citibank Health Management Program was
associated with significant reductions in health risk.
The ultimate test of the effectiveness of a worksite health management program is its ability
to influence behavior change and effect risk reduction in the target population. A science-based,
well-designed, and well-implemented program should produce measurable improvements in
population health habits that, in turn, result in lower health care service use and, ultimately,
reduced medical expenditures.
This logic supports most current investments in worksite health management programs,
including those introduced at Citibank, NA. Prior to its introduction in 1994, the Citibank medical
department developed a business case for a Health Management Program (HMP) that promised a
positive return on investment (ROI). The business case was founded on the assumptions that
employees at risk for disease and those with chronic disease conditions would benefit from a
program that would educate them about ways to reduce their health risks, better manage their
chronic disease states, and motivate them to lead healthier lifestyles. These changes were
expected to lead to lower benefit utilization, fewer health care expenditures, reduced absenteeism,
© 2000 Lippincott Williams & Wilkins, Inc.Volume 42(5), May 2000, pp 502-511
Page 1 of 16Ovid: Ozminkowski: J Occup Environ Med, Volume 42(5).May 2000.502-511
and improved worker productivity.
As reported elsewhere, 1 an objective external econometric analysis was commissioned by
Citibank in 1997 to determine whether the HMP was cost beneficial. Results from the analysis
were impressive and consistent with other applications of the Healthtrac program 1 (the vendor
chosen for the Citibank HMP). A rigorous financial analysis documented an ROI ranging from $4.56
to $4.73 in medical expenditure savings for every dollar invested in the program, depending on
the discount rate used in the analysis. A separate, unpublished internal analysis uncovered
additional savings resulting from reduced employee absenteeism attributable to the program.
The earlier analysis of the Citibank Health Management program focused on its financial
aspects, and several questions remained once it was completed. For example, in addition to cost
savings, could health improvement and risk reduction be documented for all employees who
participated in any aspect of the HMP? Additionally, did employees who participated in more
intensive high-risk interventions improve their behavior and lower their risks more so than those
who received less intensive programming? This article reports the results of these follow-up
studies. Specifically, the analyses reported here focus on short-term improvements in behavior
change and risk reduction for employees who participated in the HMP and for whom cost/benefit
analyses were performed. These studies expand on the existing literature that has investigated
the health and productivity outcomes resulting from multicomponent health promotion programs.
The impact of multicomponent worksite health management programs on worker health, risk
status, absenteeism, and attitudes has been closely studied in the past 25 years. A comprehensive
review of the literature on this subject conducted by Heaney and Goetzel 2 found that worksite
health management programs varied widely in terms of their comprehensiveness, intensity, and
duration. Consequently, the measurable impact of these programs was shown to be uneven
because different intervention and evaluation methods were employed. As the rigor of the
evaluations improved, however, findings became more equivocal. Nonetheless, the authors
concluded that there was “indicative to acceptable” evidence supporting the effectiveness of
multicomponent worksite health management programs in achieving long-term behavior change
and risk reduction in the target population. According to the authors, the most effective programs
offered individualized risk-reduction counseling to the highest risk employees within the context of
a comprehensive awareness building program. In short, they found that changing behavior
patterns and reducing health risks were achievable in a worksite health management program,
assuming favorable conditions including proper program design and execution.
Recently, Pelletier 3 conducted the fourth in a series of comprehensive literature reviews that
focused on the outcomes of health promotion programs. In this most recent review, Pelletier
added 11 new studies to the 77 previously analyzed. After reviewing the current and prior
literature on this topic, Pelletier concluded, “None of the multifactorial comprehensive intervention
programs reviewed… reduced all indicators of risk. However, the majority of programs of sufficient
intensity, breadth, and duration did result in a decrease in an adequate number of risks to result
in an overall [population] risk reduction.” Because multicomponent—or as Pelletier calls them,
multifactorial—programs focus on several risk factors and behavior change initiatives
Page 2 of 16Ovid: Ozminkowski: J Occup Environ Med, Volume 42(5).May 2000.502-511
simultaneously, it is often challenging for the program evaluator to measure multiple program
impacts and to distill which interventions affected which outcome.
The research reported in this article builds on prior efforts at documenting the health impact
of worksite-based health management programs. In addition, we report on regression-based
evaluation methods employed in the current study that we hope improve on traditional
approaches applied in an employer setting. Our evaluation uses both pretest/posttest
(preexperimental) and intervention versus comparison group (quasi-experimental) research
designs. In the pretest/posttest design, we describe a method of stratifying subjects based on the
length of follow-up. In the intervention versus comparison group study, we report on the use of
multivariate statistics to control for known differences between study groups. A detailed
description of the methods, findings, discussion, and remaining evaluation program limitations
The Citibank HMP was originally offered to all 47,838 active Citibank employees eligible for
medical benefits during January 1994. The program was repeated again in 1996. Of the 47,838
eligible employees, 25,931 (54.3% ) participated in the HMP.
The health improvement, risk reduction analysis reported here included HMP participants who
• a unique identification number (formed by scrambling the Social Security number) that enabled
us to track participants and merge data from different sources;
• at least two health risk appraisal (HRA) surveys to track changes in health risks over time;
• at least 180 days between the first and last HRAs to allow sufficient time for risk improvements
• non-missing data at the first and last HRAs for the health risk category of interest.
These criteria yielded as many as 9234 employees with at least two observations, depending
on the number of missing data for each risk category (see Table 1). The sample sizes were
substantially lower for risks related to diastolic blood pressure and cholesterol levels because
many program participants did not participate in recent biometric screenings or could not recall
the values reported to them at screenings.
Page 3 of 16Ovid: Ozminkowski: J Occup Environ Med, Volume 42(5).May 2000.502-511
Table 1. Descriptive Statistics for Analytic Sample (n = 9234)
Intervention: The Citibank Health Management Program
The Citibank HMP was administered by HealthtracSM, Inc (Menlo Park, CA). The program began
in 1994, and its evaluation continued through 1997. The program had three goals:
• to help Citibank employees improve their health practices and behaviors, thereby reducing the
prevalence of preventable disease;
• to help employees better manage their chronic medical conditions; and
• to reduce the demand for unnecessary and inappropriate health services.
It was expected that if each of these goals was successfully accomplished, the program would
result in significant cost savings and a positive ROI.
The health risk appraisal.
The Citibank HMP made substantial use of a comprehensive HRA survey developed by
Healthtrac. This HRA elicited self-reported information from employees on their demographics, risk
factors, use of medications, chronic health problems, perceptions of their health status, and
perceptions of their most important health problems. Our evaluation focused on the program’s
Page 4 of 16Ovid: Ozminkowski: J Occup Environ Med, Volume 42(5).May 2000.502-511
success in modifying these ten risk factors: dietary fiber intake, fat consumption, salt intake,
diastolic blood pressure values, cholesterol levels, exercise habits, obesity as measured by body
mass index, stress levels, seatbelt use, and smoking habits. Changes for all 10 risk categories
were assessed in our statistical analyses of the HMP, using information reported on the HRA forms
at different points in time.
The Healthtrac HRA has as its foundation the Health Assessment Questionnaire developed by
the Centers for Disease Control in 1980. The reliability and validity of these HRA questions have
been studied in several applications; a review of these studies was recently published by
Eddington et al 4 They noted that “there is general agreement that the HRA has a high degree of
face validity” and that “[r]eliability issues are not a major problem with HRA calculations, since the
results are minimally affected by minor changes in answers to most survey questions.”
Citibank Health Management Program implementation.
The Citibank HMP was implemented as follows. The standard HRA questionnaire was offered to
all 47,838 Citibank employees in 1994 and again in 1996. As an incentive to complete the HRA
and participate in the program, subjects were given a $10 credit that was applied to their portion
of the cost of the company’s employee medical benefits program.
All HRA participants received a confidential letter and report identifying their individual health
and lifestyle risks, based on the information submitted on their HRA forms. Low-risk HRA
participants received the initial HRA and a personalized follow-up report along with general health
education and self-care materials. All participants, regardless of risk, were provided access to
Healthtrac’s toll-free customer service telephone line to ask questions about Healthtrac program
On the basis of the initial HRA results, those found to be at overall high risk for poor health
outcomes or high medical expenditures were invited to participate in a high-risk intervention
program called Accent. Methods for assessing overall risk status are described in more detail
elsewhere. 1 Those who self-reported the existence of specific conditions, including asthma,
arthritis, back pain, high blood pressure, chronic lung disease, diabetes, heart disease, tobacco
use, and high body weight, were also invited to participate in Accent programs tailored to those
conditions. Accent programs were also offered to those who had multiple risk factors (eg, stress,
overweight, poor nutrition, and sedentary lifestyle), even when any one of those risk factors, by
itself, was not considered serious enough to lead to high-risk status. For these individuals, the
combination of risk led them to the Accent High Risk Program.
Participants in the Accent program received three additional HRA questionnaires, offered at
roughly 3-month intervals after the initial HRA. These additional HRAs were tailored to the risks
that led to Accent program assignment, with some individuals receiving a combination of
intervention modules depending on their risk classification.
After their completion of each Accent module questionnaire, participants received a
personalized letter and report, identifying their most current health risks and recommending
Page 5 of 16Ovid: Ozminkowski: J Occup Environ Med, Volume 42(5).May 2000.502-511
actions to reduce risks and improve health. In addition, after each questionnaire, participants
received health education materials, books, and videos specific to their chronic disease conditions
or lifestyle risks. Accent program participants identified in 1994 also received an Accent follow-up
HRA in 1995 and two follow-up HRAs in 1996, along with personalized reports. A random sample
of Accent participants from 1996 also received one outbound telephone counseling call by a health
educator to check on their progress in reducing risk and to support risk reduction behaviors.
In 1996, Citibank also introduced an inbound telephone counseling service through a
Healthtrac subcontractor. This service was offered to high-risk participants who wished to obtain
advice or additional health education from a nurse regarding their Accent modules. The inbound
telephone service also featured telephone access to an audio library of health tapes.
Finally, all HRA participants were mailed their choice of one of three popular self-care books
that provided advice and instruction on personal health care, child health care, informed use of
health care services, and healthy lifestyles.
Three databases were integrated for the Citibank evaluation. Healthtrac provided data on
participation in the Citibank HMP and its high-risk Accent modules. Participant risk data were
recorded from the multiple HRA forms administered. In addition, data on health plan enrollment
were provided by Citibank health plan administrators. These data were independently processed
and merged for analysis by the authors.
Assigning levels of risk for each HRA category.
Healthtrac assigned a “high-risk” indicator flag to participants meeting specific criteria for
inclusion in the Accent program. However, Healthtrac did not assign individuals to high or lower
risk status for each individual risk category. Because such assignments at an individual level were
required to perform the change-in-risk analyses described below, we adopted rules for risk
assignment from other published research. Where possible, we adopted criteria for high-risk
assignment from previously published work conducted for the Health Enhancement Research
Organization (the HERO study). 5 We also conferred with Healthtrac staff to develop “markers” for
high-risk categories not addressed in the HERO study. As a result, we set the following criteria to
identify persons at high risk for poor health outcomes for each risk category.
Respondents were classified as being at high risk for problems related to dietary fiber intake if
they consumed fewer than 34 servings of fiber per week. Respondents were asked separate
questions about consumption of whole grain breads, whole grain cereals, fruits, and vegetables.
The total number of servings of dietary fiber per week were summed and compared with the 34-
serving minimum criterion to identify those at high risk.
High risk for problems related to salt intake was denoted for respondents who consumed 22 or
more servings of salt per week. The same criterion was used for dietary fat intake. Respondents
were asked separately about their consumption of red meat, cheese, eggs, butter, ice cream, and
Page 6 of 16 Ovid: Ozminkowski: J Occup Environ Med, Volume 42(5).May 2000.502-511
whole milk. These responses were summed and compared with the 22-serving maximum
criterion to denote high risk status.
High risk owing to diastolic blood pressure was denoted for those whose values exceeded 100
mmHg. High risk for problems related to cholesterol intake was denoted for those whose total
serum cholesterol levels were reported to be above 239 mg/dL. High risk for problems related to
lack of exercise was denoted for those who reported that they exercised less than 45 minutes per
week. High risk for problems related to body mass was denoted for those whose body mass index
was greater than 27. Body mass index is defined as weight in kilograms divided by height in
High risk for problems related to seatbelt use was denoted for those who estimated that they
used seatbelts less than 90% of the time when riding in a moving vehicle. High risk for stress was
denoted for those who reported extremely high levels of stress compared with others they know.
High risk owing to smoking habits was denoted for those who smoked any cigarettes on a typical
day. Data on cigar and pipe smoking habits were not available.
Two sets of statistical analyses were performed. The first set employed a pretest/posttest
cohort group research design. The study used data from all HMP participants who completed at
least two HRA surveys. For each risk category examined, McNemar chi-squared tests were used to
test whether the proportion of individuals at high risk differed over time. Program effectiveness
was determined if the proportion of participants at high risk was significantly lower at the second
HRA administration when compared with baseline.
The time period between the first and last HRAs administration varied considerably across the
study group, with a mean of 725 days (approximately 2 years), a standard deviation of 185 days
(approximately one-half year), and a range of 180 to 1111 days (approximately 3 years). Because
of the variability in time periods between the first and last HRAs administrations, we adjusted the
analysis accordingly. First, we estimated changes in risk regardless of the time interval between
first and last HRAs, using the entire sample. We then conducted the same analyses, this time
stratifying the total sample according to the length of the time interval between HRAs. One
analysis included only those subjects whose time between first and last HRAs was 180 to 365
days. A second analysis included those whose time between first and last HRAs was 366 to 547
days. Finally, a third analysis focused on those whose time interval exceeded 547 days.
The second set of analyses employed a quasi-experimental research design in which Accent
program participants were compared with non-participants in terms of risk reduction over time.
Because the Accent program was, by design, more intensive than the general program offered to
all Citibank HMP participants, the analysis sought to determine whether participants in that
program improved their risk profile to a greater extent when compared with non-Accent program
Because this was a non-randomized study, logistic regression analyses were used to test the
Page 7 of 16 Ovid: Ozminkowski: J Occup Environ Med, Volume 42(5).May 2000.502-511
hypothesis that changes in risk would be greater for Accent program participants than non-
participants. The logistic regressions controlled for measurable baseline differences between the
two groups of subjects. These differences included age group, gender, business unit within
Citibank, type of health care plan (managed care versus indemnity), type of insurance coverage
(single versus family), hourly versus salary status, full versus part-time employment status,
annual wage rate, and the length of time between the first and last HRAs. These factors were
previously shown to exert a significant influence on health care expenditures for Citibank
employees, 1 and they were expected to influence health care risks as well.
Two logistic regressions were estimated for each HRA risk category, except the one addressing
cholesterol levels, which was omitted from this analysis because no Accent members at high risk
for cholesterol problems could be found. The first logistic regression assessed differences between
Accent participants and non-participants and the probability of their being at high risk at the
administration of the first HRA. The second assessed those differences for the last HRA. The
impact of the Accent program was estimated from these regressions as the mean difference over
time in the predicted probability of being at high risk for Accent participants, minus the mean
difference over time for non-participants, controlling for baseline characteristics. A t test of the
differences in these two mean values was used to determine statistical significance.
Table 1 provides unadjusted descriptive statistics for HMP participants. Most of the participants
(67% ) were under age 45 and male (58% ). The average base salary of participants (in 1996
dollars) was $58,762. The sample included a high percentage of exempt staff (67% ), a low
percentage of hourly employees (4% ), and a few part-time employees (7% ). Approximately 52%
of the sample members were enrolled in the Citibank indemnity health plan. Most worked in the
credit card or banking business units, and a little over one third (37% ) were Accent program
Changes Over Time in Health Risk for All HMP Participants
Table 2 shows risk changes over time for the entire sample, regardless of the time interval
between the first and last HRAs. As shown, changes in health risk over time were statistically
significant and in the expected direction (with risks declining over time) for 8 of the 10 risk
categories examined. Significant risk improvement was found for the study sample in the following
categories, organized from greatest to least change over time: seatbelt use, exercise habits, fiber
intake, stress levels, fat intake, salt intake, cigarette use, and diastolic blood pressure. One risk
factor, obesity as measured by body mass index, worsened significantly over time, and another,
total cholesterol, showed no significant change.
Page 8 of 16 Ovid: Ozminkowski: J Occup Environ Med, Volume 42(5).May 2000.502-511
Table 2. Proportion of Citibank Health Management Program Participants at High Risk Over Time,
by Category (Sorted by Difference Over Time)*P < = 0.05, McNemar chi-squared test.† NO,
number of observations.
Table 3 shows the difference over time in the proportions of employees at high risk by risk
factor, stratified according to the length of time between the first and last HRAs. As shown,
changes in risk were statistically significant in at least two of the three analyses for the following
categories: exercise, salt intake, fat consumption, seatbelt use, stress, and dietary fiber. Changes
in risk were not significant consistently when stratified according to time intervals between first
and last HRAs for the following categories: cholesterol and smoking. Risks for high blood pressure
and obesity showed statistically significant decrements in only one of the three time strata
Table 3. Change in Percentage of Participants at High Risk, by Health Category and Number of
Days Between First and Last HRA* Significant at the P < 0.05 level (McNemar chi-squared test).**
Page 9 of 16 Ovid: Ozminkowski: J Occup Environ Med, Volume 42(5).May 2000.502-511
Significant at the P < 0.01 level (McNemar chi-squared test).
Although not always true, declines in risk tended to be greater when the time period between
HRAs was shorter (ie, less than 1 year). With three exceptions (smoking patterns, blood pressure,
and body mass index), the direction of the change in risk over time was consistent for each
category, regardless of the time interval between the first and last HRA. 3
Changes Over Time for Accent Program Participants Versus Non-Participants
Table 4 shows the impact of Accent program membership by noting the adjusted probabilities
of being at high risk at the first and last HRA administrations, separately for Accent participants
and non-participants. The probabilities of being at high risk were computed from a logistic
regression model and account for baseline differences between the two comparison groups. The
Accent program impact (shown in row 3) was derived by computing the change over time in the
adjusted probability of being at high risk for Accent members (in row 1), minus the change over
time in the adjusted probability of being at high risk for non-members (in row 2).
Table 4. Difference in Probability of Being at High Risk Over Time* Comparing Accent Program
Participants to Non-Participants, by Category, Adjusted for Covariates†* Difference = last HRA -
first HRA.† Negative values imply a greater reduction in risk over time for Accent program
participants.‡ Statistically significant at the P < 0.05 level.
In Table 4, a negative score in the third row reflects a greater reduction in the probability of
being at high risk over time for Accent participants compared with non-participants. For all nine
risk categories, the computed scores favor Accent participants, whose risks were reduced to a
significantly greater extent than for non-participants. The impact was greatest for risks related to
exercise patterns, seatbelt use, stress levels, and body mass index. For each of these risk
categories, the net impact of the Accent program was at least a 4.3% reduction in high-risk status
(ie, 8.7% for exercise habits, 6.5% for seatbelt use, 6.1% for stress levels, and 4.3% for body
mass index). Impact estimates for the other risk categories were 1.2% (for smoking habits) or
Page 10 of 16Ovid: Ozminkowski: J Occup Environ Med, Volume 42(5).May 2000.502-511
less (Table 4, row 3).
Detailed Regression Results
Finally, the adjusted probabilities noted in Table 4 were based on the logistic regressions
described above, and some readers may wish to view the detailed regression results. These are
noted in the Appendix and in Tables 5 and 6.
Table 5. Effect of Independent Variables on the Probability and Odds of Being at High Risk, at the
First HRA* Square root of (annual wage/1000).
Page 11 of 16Ovid: Ozminkowski: J Occup Environ Med, Volume 42(5).May 2000.502-511
Table 6. Effect of Independent Variables on the Probability and Odds of Being at High Risk, at the
Last HRA* Square root of (annual wage/1000).
A prior analysis of the Citibank HMP established that the program was cost beneficial,
returning between $4.56 and $4.73 in medical expenditure savings for every dollar invested in the
program. With these ROI results as background, Citibank program managers sought to determine
whether program participants also achieved health improvements that would coincide with the
cost savings. Consequently, follow-up studies were initiated, focusing on the same group of
participants and the same time period previously analyzed in the cost/benefit study. This article
reports the results of two follow-up investigations.
The first investigation used a pretest/posttest preexperimental research design. This analysis
focused on the entire cohort of employees who participated in at least two HRA screenings: the
first when the HMP was introduced and the second at an average of 2 years after the baseline
screening. As many as 9234 subjects were followed over a minimum of 6 months and a maximum
Page 12 of 16 Ovid: Ozminkowski: J Occup Environ Med, Volume 42(5).May 2000.502-511
of almost 3 years.
Analyses of the data revealed statistically significant improvements in 8 of the 10 risk
categories. These included risks related to seatbelt use, exercise habits, stress levels, fiber intake,
fat consumption, salt intake, cigarette use, and diastolic blood pressure. Health status seemed to
decline over time for one category (being overweight, measured as body mass index). No
significant change in risk was found for one final category (total cholesterol level).
When risk factors were examined by using stratification methods to control for variability in
the length of follow-up, improvements by risk category noted above for the total sample were
sustained for the most part, despite reductions in the number of subjects within individual data
cells. Exceptions to this general pattern included diastolic blood pressure (for which a tiny minority
of participants, about 1% , reported values that placed them in a high-risk category at either the
first or last HRA) and risks related to smoking patterns and total cholesterol levels. In the
stratified analyses, changes in risk over time for smoking and cholesterol were not significant
regardless of the time interval examined.
When the entire sample was analyzed regardless of the length of time between first and last
HRAs, we found that risks related to high body mass index increased over time. Risks related to
cholesterol levels increased as well, but not significantly. These findings were not surprising
because the study tracked a cohort group of subjects longitudinally, and comparison to a
reference or control group was not feasible. Similar results indicating deterioration in health status
for biometric measures such as body weight, cholesterol, and blood pressure for a cohort group
have been reported elsewhere 6 and may be attributable to an internal validity limitation common
to these types of studies called maturation. This refers to an increase in risk resulting from the
The second set of analyses we conducted used a quasi-experimental research design to
examine health improvements for participants in the high-risk intervention program called Accent,
as compared with those of non-participants. These analyses sought to determine whether a more
intensive intervention program directed at the highest risk population achieved better results than
a more generic, less intensive program.
Results from these analyses supported prior research findings that more intensive,
personalized, frequent, and targeted health management programming achieves better health
outcomes than those that are less intensive and directed toward awareness building. 2 In all
comparisons between Accent participants and non-participants across risk factors, high-risk
program participants achieved better outcomes than non-participants. This was true even when
controlling for known differences between Accent participants’ and non-participants’
demographics, insurance coverage, organizational factors, and length of exposure to the program.
It is useful to compare the analysis performed at Citibank with prior efforts at documenting
health improvement and risk reduction programs in the workplace. For example, Goetzel et al 6
analyzed health profile data from 1540 Duke University employees over an average of 3.3 years in
a pretest/posttest cohort group study. Using methods similar to those reported here, the
Page 13 of 16 Ovid: Ozminkowski: J Occup Environ Med, Volume 42(5).May 2000.502-511
researchers found that health risks improved significantly in 8 of the 11 risk categories
examined for the study sample as a whole. Similar to our findings, Duke employees’ blood
pressure and obesity risks increased over time, most likely as a result of population aging.
In an analysis of over 9000 IBM employees, Goetzel et al 7 used a preexperimental research
design to record significant risk reductions in four health categories when comparing participants
in an intensive health improvement program with non-participants. These related to body weight,
smoking patterns, non-LDL cholesterol levels, and blood pressure. Other significant corporate
health improvement studies using a variety of research designs have been conducted at Johnson
& Johnson, 8,9 Dupont, 10,11 Bank of America, 12 and the California Public Employment Retiree
The Citibank health improvement study described here used several advanced methods that
can be applied in worksite settings generally described as naturalistic, as opposed to clinical. For
one, we introduced subject stratification methods to control for length of exposure to the
intervention. We found that stratification by time produced results that were largely comparable
with the overall study results, with some variability for smoking and blood pressure. We also
introduced multiple regression methods to control for observed differences between participants
and non-participants in the high-risk intervention program, which no doubt improved our
estimates of program impact. Nonetheless, we should not lose sight of a selection bias issue that
is relevant to almost all research conducted in a corporate setting where randomization of subjects
to experimental and control groups is not feasible.
In addition to some advances in the methods used in this study, we note several limitations to
the research. First, we focused only on risks included in the Healthtrac HRAs. Other risks would be
useful to measure over time to gain a more comprehensive picture of health status and health
improvements in the target population. Also, it should be noted that most of the risks assessed
(except for the biometric measures) were based on self-reports, which may be unreliable in some
Next, we did not have access to health risk data for employees who did not participate in the
HMP. Consequently, the generalizability of our results to the overall Citibank population is
In many cases, changes in risk seemed small. This underscores the difficulty of affecting
health habits and risk factors in a population. On the other hand, this study, coupled with a prior
financial analysis, 1 shows that even small changes in health risks may produce large changes in
costs. Additionally, the program may have positively affected employees’ attitudes toward their
health or toward Citibank as an employer. These in turn could have resulted in a reduction in
unnecessary utilization of health services. Attitudes were not directly measured, however, so it is
difficult to draw inferences about the program’s ability to influence these “softer” outcome
Page 14 of 16 Ovid: Ozminkowski: J Occup Environ Med, Volume 42(5).May 2000.502-511
The Citibank health improvement study further supports the growing literature that worksite
health management programs can produce a positive effect on the risk profile of a target
population. We were fortunate in that we had access to a relatively large study population, over
several years, and fairly complete health data. This study supports a health management model
focusing on population health. The Citibank model advocates the application of a large proportion
of intervention resources on a “Pareto group” with the greatest need while providing general
education and behavior support services to the population as a whole. The study of changes in risk
over time in the HMP reported here, along with the financial analysis of program impact
referenced earlier, suggests that even small changes in risk in an identified population can
positively affect the corporate bottom line. This should provide program managers with sufficient
documentation to support program renewal and expansion, and it offers food for thought for
decision makers in other corporate settings who are considering the costs and benefits of health
Work on this project was funded by Citibank. The opinions expressed in this article are those
of the authors and do not necessarily represent corporate positions taken by their employers.
Logistic Regression Results
Table 5 reports the results from the logistic regression analyses of the likelihood of being at
high risk at the first HRA. The table shows the impact of each independent variable on the odds of
being at high risk and on the probability of being at high risk. (The probabilities are simply
mathematical transformations of the odds ratios.)
As expected, participation in the Accent program was associated with significantly higher risk
for each category, as the Accent programs were designed for those who tend to be at higher risk.
Most of the other variables influenced the likelihood of being at high risk as well, although the
signs and magnitudes of these variables sometimes differed by HRA risk category. The only
variable that did not influence high-risk status for any category was the one denoting working in
the Citibank division.
Table 6 shows the odds of being at high risk and the associated probabilities for the last HRA.
Again, many variables significantly influenced the odds and probabilities of being at high risk, but
the magnitude of these variables often differed from what we observed at the first HRA. For
example, the odds and probabilities of being at high risk were greater for Accent participants
compared with non-participants. However, for most risk categories, the magnitudes of the odds
ratios and probabilities were lower at the last HRA for Accent participants compared with the first
HRA. This is what lead to the greater risk reductions over time shown for Accent participants
noted in Table 4.
1. Ozminkowski R, Dunn R, Goetzel R, Cantor R, Murnane J, Harrison M. A return on investment evaluation
Page 15 of 16 Ovid: Ozminkowski: J Occup Environ Med, Volume 42(5).May 2000.502-511
Copyright (c) 2000-2005 Ovid Technologies, Inc. Download full-text
Version: rel10.2.0, SourceID 1.11354.1.65
of the Citibank, NA Health Management Program. Am J Health Promotion. 1999; 14:31–43. [Context Link]
2. Heaney C, Goetzel R. A review of health-related outcomes of multi-component worksite health
promotion programs. Am J Health Promotion. 1998; 11:290–307. [Context Link]
3. Pelletier K. A review and analysis of the clinical and cost-effectiveness studies of comprehensive health
promotion and disease management programs at the worksite: 1995–1998 update (IV). Am J Health
Promotion. 1999; 13:333–345. [Context Link]
4. Eddington D, Yen L, Braunstein A. The reliability and validity of HRAs. In: Hyner G, Peterson K, Travis J,
Dewey J, Foerster J, Framer E, eds. SPM Handbook of Health Assessment Tools. Pittsburgh, PA: The
Society of Prospective Medicine & The Institute for Health and Productivity Management; 1999. [Context
5. Goetzel R, Anderson D, Whitmer R, et al. The relationship between modifiable health risks and health
care expenditures. J Occup Environ Med. 1998; 40:843–854. Ovid Full Text [Context Link]
6. Goetzel R, Kahr T, Aldana S, et al. An evaluation of Duke University’s Live for Life health promotion
program and its impact on employee health. Am J Health Promotion. 1996; 10:340–342. [Context Link]
7. Goetzel R, Sepulveda M, Knight K, et al. Association of IBM’s “A Plan for Life” health promotion program
with changes in employees’ health risk status. J Occup Med. 1994; 36:1005–1009. Bibliographic Links
8. Breslow L, Fielding J, Herman A, et al. Worksite health promotion: its evolution and the Johnson and
Johnson experience. Prev Med. 1994; 19:13–21. Bibliographic Links [Context Link]
9. Bly J, Jones R, Richardson J. Impact of worksite health promotion on health care costs and utilization:
evaluation of the Johnson and Johnson Live for Life program. J Am Med Assoc. 1986; 256:3236–3240.
10. Bertera R. Planning and implementing health promotion in the workplace: a case study of the Dupont
company experience. Health Ed Q. 1990; 17:307–327. [Context Link]
11. Bertera R. Behavioral risk factor and illness day changes with workplace health promotion: two-year
results. Am J Health Promotion. 1993; 7:365–373. [Context Link]
12. Fries J, Bloch D, Harrington H, Richardson N, Beck R. Two-year results of a randomized controlled trial
of a health promotion program in a retiree population: the Bank of America Study. Am J Med. 1993;
94:455–462. [Context Link]
13. Fries J, Harrington H, Edwards R, Kent L, Richardson N. Randomized controlled trial of cost-reductions
from a health education program: the California Public Employees Retirement System (PERS) study. Am J
Health Promotion. 1994; 8:216–223. [Context Link] [Context Link]
Accession Number: 00043764-200005000-00008
Page 16 of 16Ovid: Ozminkowski: J Occup Environ Med, Volume 42(5).May 2000.502-511