Health risk factors and absenteeism among university employees.

Abstract: To examine relationships between health risk factors and sick leave usage, a retrospective design was used for analysis of a sample of university employees. Binary logistic regression identified variables characteristic of high sick leave usage compared to low. Predictive factors identified were." perceived stress, food choices, body mass index, systolic blood pressure, tobacco use, and cancer for females; and body mass index, physical activity, perceived stress, social support, and heart disease for males. While health risks were found to be related to amount of sick leave used, the analysis suggests that additional factors impact high rates of sick leave usage.

**********

The causes of employee absenteeism are multiple and complex. In efforts to both understand and impact absenteeism, researchers and practitioners have focused attention on modifiable risk factors associated with employee illness expecting that changes in these risk factors would result in corresponding changes in absenteeism. Recent reviews suggest that health risks are associated with absenteeism (Aldana, 2001; Reidel, Lynch, Baase, Hymel, & Peterson, 2001). Aldana (2001) concludes that obesity, stress, and the presence of multiple risk factors are associated with increased rates of absenteeism. Aldana (2001) also indicates that the association between absenteeism and other health risks and behaviors such as seat belt use, cholesterol, physical activity, hypertension or alcohol abuse remains unclear. However, the body of literature upon which these conclusions are based is small. Additionally, few of the studies included in his review assess differences in extremely high and low rates of absenteeism. This is important since the vast majority of direct and indirect costs associated with poor health are attributable to 10-15% of a given population (Yen, Edington, & Witting, 1992).

Individual studies have identified a significant association between rates of absenteeism and obesity (Burton, Chen, Schultz, & Edington, 1998; Narbro et al., 1998; Thompson, Edelsberg, Kinsey, & Oster, 1998; Tucker & Friedman, 1998), tobacco use (Bertera, 1991), diabetes (Burton, Conti, Chen, Schultz, & Edington, 1999), seat belt use (Burton et al., 1999), physical inactivity (Steinhardt, Greenhow, & Stewart, 1991) and stress (Jacobson et al., 1996; Jamal, 1984; Neubauer, 1992; Tang & Hammontree, 1992; Woo, Yap, Oh, & Long, 1999). Additionally, hypertension has been associated with absenteeism (Burton et al., 1999), but this association does not appear to be consistent (Leigh, 1990). Finally, it has been suggested that one of the best predictors of absenteeism is past absenteeism (Yen et al., 1992) suggesting that absenteeism may be either related to chronic illness or simply be habitual.

Reaching a definitive conclusion about the association between absenteeism and risk factors is tenuous due to differences in the way that the independent variables have been employed, in the way that the absenteeism dependent variable has been measured, and in the way that demographic variables have been used to provide clarity to results. For example, previous research on the association between health risks and absenteeism has been limited by the use of a single risk factor, such as hypertension (Leigh, 1990) or cardiovascular fitness (Tucker, Aldana, & Friedman, 1990). In the limited research that has considered multiple risk factors, either a limited number of risk factors were used (Bertera, 1991) or cost-related or productivity data were reported (Burton et al., 1999) rather than rates of absenteeism. To date, only one study has incorporated a large number of individual risk factors in a predictive model to identify which risk factors are characteristic of employees with high versus low rates of absenteeism (Yen et al., 1992). Further research is needed to identify differences in risk profiles of high and low users of sick leave.

Further, different methods have been used to gather data for absenteeism when it was used as dependent variable. In several studies (Cole, Tucker, & Friedman, 1987; Jacobson et al., 1996) absenteeism data was self-reported. Where self-reports were used, the authors stated the case for self-reports by noting that correlational data exist to support self-report compared to an objective measure (Jacobson et al., 1996). In this study we used an objective measure of absenteeism, along with health risk appraisal data, and clinically gathered biometric data. Additionally, while some studies have reported some results by gender (Bertera, 1991 ; Steinhardt et al., 1991), others have assessed only a single gender (Narbro et al., 1998; Neubauer, 1992). The potential for different prevalence of health risks, as well as rates of absenteeism due to differing health risks, indicates the need to look at results by gender within a single population.

The purpose of this study was to examine the relationships between health risk factors and absenteeism for non-academic university employees that participated in a voluntary health screening program. The primary purposes were twofold: to explore associations between risk factors and absenteeism for men and women and to identify health risks that predict high and low rates of absenteeism.

METHODS

DESIGN AND SAMPLE

A retrospective study design was used. The sample consisted of 940 employees of Oklahoma State University. Primary criterion for inclusion in the study was classification as non-faculty, non-administrative, 3/4 time or greater personnel (N=3,469). Only those employees who worked 3/4 time or greater were eligible for employee benefits and subsequently had sick leave tracked. An additional criterion for inclusion was voluntary participation in a health screening program offered by the O.S.U. Wellness Center during the 19951996 fiscal year (N = 1,233). Those participants who were employed for the entire 24-month period from July 1994 to June 1996 (N=940) were included in the study.

These data had been used for programming purposes but had never been formally evaluated and reported. Analysis and reporting of these data is important for three reasons. First, HIPAA, and increased employee concern over privacy, has made it more difficult to access and combine both employee health risk and absenteeism data. Thus, the data used in this study contains information that has, since collection, become more difficult to obtain. Second, very few worksite wellness studies have been able to employ an objective measure of absenteeism. Third, the underlying risk factors that are associated with absenteeism in this study have not changed since 1996. In fact, on a population basis, many of these risk factors have worsened. Thus, if anything, the results reported herein perhaps understate the existing association between health risk factors and proxy measures of productivity in today's workforce. These three reasons were adequately compelling for the data to be analyzed and reported.

The sample represents 27.1% of total potential subjects based on employment classification (the primary inclusion criteria). An informed consent, approved by the University Institutional Review Board, was signed by all participants in the health screening program.

MEASURES

Total sick hours were collected monthly for each employee over the two-year period. Sick leave records for all non-faculty, non-administrative, 3/4 time or greater personnel were matched to health screening records through a confidential social security number derivative. Total sick hours for fiscal year 1995-1996 (also referred to as year two) was selected as the dependent variable. Subjects were grouped into deciles according to sick leave hours. Separate analyses were completed for males and females.

The health screening program included clinical measures and the completion of a health risk appraisal. Clinical measures included height and weight, non-fasting finger stick cholesterol procedure using the Abbot Vision enzymatic test, as recommended by the National Cholesterol Education panel, seated blood pressure after a three-minute rest period, using a Critikron Dynamap Automated Sphygmomanometer, and a 7 site body fat estimate using Lange skin-fold calipers.

Additional health and behavioral data were collected using a self-reported 135 item health risk appraisal (HRA) developed by People Karch International. This HRA, is based on norms produced by the American Cancer Society, the American Heart Association, the Centers for Disease Control, the Framingham Study Group and the Harvard Alumni Study Group. The HRA as a health measurement tool has shown evidence of predictive validity for mortality (Gazmararian, Foxman, Yen, Morgenstern, & Edington, 1991) and morbidity (Smith, McKinlay, & Thorington, 1987). However, in the only known review of the effectives of the HRA in the worksite, caution against using the HRA as a stand-alone intervention tool is given (Anderson & Staufacker, 1996).

Data for this analysis included only responses for the individual participant. Answers regarding illness and disease history for parents or siblings were omitted. From the answers provided, some composite variables were created for purpose of analysis (see Table 1). Because some of the variables were gender specific, self-care and cancer variables were coded separately for men and women, and two additional variables (osteoporosis and age at birth of first child) were included for women only. A description of the variables used in this study is provided in footnotes to Table 1.

Age, race and annual income were initially included in the health risk appraisal data; however, an administrative decision was made to omit these fields for confidentially reasons when the health risk data and sick leave data were combined. Therefore, it was not possible to include these variables in any analyses.

DATA ANALYSIS

Data were analyzed separately for males and females. In order to first check the conclusion of Yen and colleagues (1992) that previous sick leave usage is the best predictor of future usage, sick leave in year one and year two was compared using a simple bivariate relationship. For both males and females, sick leave usage in year two was moderately correlated with year one (324 males; Pearson's r = 0.54; P = 0.01; 615 females; Pearson's r = 0.42; P = 0.01). Since the health risk appraisal was conducted during the second year, subsequent analyses were performed using only sick leave data from year two.

Previous research suggests that participants who voluntarily participate in health screenings are healthier (Yen et al., 1992). However, in a study with a sample similar to the one included in the present study, volunteer wellness participants were relatively less healthy (Haynes, Dunnagan, & Smith 1999). To determine whether the volunteer participants in the present sample were relatively healthier, sick leave rates over both the second 12-month period and the entire 24-month study period were compared between the subjects who had participated (n = 1,233) versus those who had not (n = 2,236). No differences were found in either analysis (12-month: t = .32, p = .75; 24-month: t = .55, p = .58).

In order to assess associations between health risk data and sick leave usage, Pearson correlation coefficients were calculated for continuous predictor variables and Spearman Rho correlations were calculated for categorical predictor variables. To address the primary purpose of identifying health risk factors associated with high and low rates of absenteeism, a binary logistic regression model was constructed. Participants were grouped into deciles according to sick leave usage. Regression analyses reported characteristics predictive of the highest (top decile) compared to the lowest (bottom decile) sick leave usage.

When building the prediction models, previous research on risk factors and absenteeism was considered. Additionally, any variable known to be associated any chronic disease condition was also included. No variable reduction schemes were employed prior to analysis because the objective was to determine which variables in a full model would predict high versus low rates of sick leave usage.

When considering the previous research on blood pressure and various economic outcomes, it was determined that blood pressure medication might have a mitigating effect on blood pressure values. In other words, those on blood pressure medication should have blood pressure measurements closer to normal than abnormal. To account for the potential impact of blood pressure medication on study outcomes, two analyses were performed. First, blood pressure medication was entered as a variable into both prediction models. This would determine whether using blood pressure medication would predict sick leave usage. Second, the mean systolic and diastolic blood pressure was compared between those that did and did not take blood pressure medication. In both men (142/86) and women (136/73) who took blood pressure medication, both systolic and diastolic blood pressure was significantly higher than for those who did not (132/ 75 and 124/71 respectively).

Statistical significance was determined by p = .05. Ali Data were analyzed using Statistical Package for Social Sciences Software (SPSS) Version 11.5.

RESULTS

Data were collected for 325 males and 615 females. Mean sick hours for males (46.47, S.D.=45.66) were considerably lower than females (80.37, S.D.=72.91). For both genders, a small number of subjects utilized a large number of sick hours (See Table 2). Females in the bottom decile used an average of 3.82 sick hours (S.D.=3.15), compared to females in the top decile (Mean=242.69, S.D.=78.30). Males in the bottom decile used no sick hours over the year period, compared to men in the top decile (Mean=147.71, S.D.=45.12).

Some significant correlations (p < .05) were identified between predictor variables and sick leave usage. For females there were negative correlations between sick leave usage and systolic blood pressure, diastolic blood pressure, and history of heart disease; and a positive correlation between sick hours and eating habits (a composite variable). A positive correlation was also identified between sick leave usage and a woman's age when her first child was born. Mean sick hours used increased linearly among women who never had children (Mean=63.45, S.D.= 61.48) to women who had their first child at age 30 or older (Mean=101.90, S.D.=98.30). For males, statistically significant correlations were identified between sick leave usage and diabetes, as well as with higher values for BMI, body fat percentage, and diastolic blood pressure. A negative correlation between strength training and sick leave usage was also statistically significant.

For both males and females, increased sick leave usage correlated negatively with increased physical activity, but the correlation was only statistically significant for the males. Roughly one quarter of the sample (24.2% of females, 24.3% of males) reported meeting the U.S. Surgeon General's recommendation for physical activity; identified by those who exercise one hour per week or more. 16.0% of males (N= 52) and 15.4% of females (N=92) indicated they had a physical problem or limitation that affected their ability to exercise. Out of this subgroup, 62.1% of the females (N=59) and 75.0% of the males (N=39) indicated they were physically active for at least one hour per week.

Binary logistic regression was utilized in order to further classify variables that were characteristic of the highest compared lowest amount of sick leave usage. This allowed incorporation of all predictor variables--continuous and categorical--into regression models predicting highest and lowest deciles of sick leave usage for each gender. Binary logistic regression models were performed separately for males and females. All predictor variables were entered for each group: twenty-eight variables for females, and twenty-six variables for males. Stepwise entry of the variables was used to identify the most significant predictors of highest (top decile of sick leave hours) compared to lowest (bottom decile) sick leave usage.

For females, the overall model included six variables that predicted with 75.0% percent accuracy whether the participant used a high or low amount of sick hours by predicting top and bottom deciles of sick leave usage (See Table 3). The level of stress/strain in life by itself predicted with 98.1% accuracy which women were in the top decile of sick leave usage (high usage), but was overall only 60.7% accurate in predicting top or bottom decile when additional variables were taken into consideration. Four variables were 81.7% accurate in predicting the top decile alone (stress, food choices, body mass index, and systolic blood pressure). Two additional variables (tobacco use and cancer) increased accuracy for the entire model, but did not add any additional information for the prediction of the top decile. The model for males included five variables that were 90.3% accurate in predicting the top and bottom deciles of sick leave usage (See Table 4). The combination of only two variables (body mass index and physical activity) was 91.9% accurate in predicting the bottom decile (no sick leave usage) and 60.0% accurate in predicting the top decile of sick leave usage for men. Additional variables (stress, quality of social support, and heart disease) increased accuracy of the model as a whole, contributing predictive value to the highest users of sick leave.

DISCUSSION

This analysis identifies different levels of sick leave usage, as well as different sets of predictive factors, for males and females. As identified in previous research (Aldana, 2001) stress, excessive body weight, and multiple risk factors figured prominently in this analysis. The large number of predictive variables utilized in this analysis does not minimize potential impact of other health promotion issues (such as self-care) but indicates the importance of a few variables that impact sick leave usage for this sample. Both body mass index and stress were primary predictive variables for high compared to low sick leave usage for both males and females. Body mass index and physical activity together were highly accurate in identifying males who used no sick leave hours in an entire year.

Correlational data for the entire sample indicated some significant associations between sick leave usage and predictive variables. For males, correlations between some of the clinical measures were statistically significant, indicating possibilities for multiple risk factors. Additionally for males, perceived stress was significantly negatively correlated with both number and quality of social relationships.

For females, variables that were significantly correlated with sick leave were not necessarily predictive of amount of sick leave usage. Stress alone was highly accurate in predicting females in the top decile, but did not significantly correlate with amount of sick leave usage. Quality of social support was a primary predictive factor for males, yet not females. Only 37.2% of the males rated the quality of their social relationships as very helpful, supporting, and encouraging, compared to 56.3% of the females. 60.9% of the males, and 53.0% of the females indicated that they had three or more people with whom they could speak honestly with about their problems and concerns in life.

Certainly personal illness is not the only reason that employees are absent from work. Other factors associated with absenteeism include personal conflicts, poor employee morale, unsatisfactory compensation and benefit programs, unrealistic job expectations, stressful working conditions, smoking, caregiving and family responsibilities, and personal business. (Jacobson, et al, 1996). Unfortunately, because sick leave data utilized in the analysis was reported by the employer, it is not possible to determine whether absence was due to illness or other reasons. A small number of subjects in this analysis accounted for a large amount of sick hours. Although cost analysis was not available with this data, the nature of employment for study participants would suggest that high rates of sick leave usage were costly in terms of lost work hours and productivity.

In summary, these findings provide support for health education efforts for at least three reasons. First, excess body weight and high stress levels appear to be strongly related with persons who used the greatest amount of sick leave. This supports existing health education efforts to encourage weight loss and reduction of stress levels and suggests, along with previous research, that these changes may improve employee productivity as indicated by sick leave usage. However, it should be noted that these data are correlational, thus causality cannot be assumed.

Second, these findings support the use of targeted interventions since much of the total sick leave usage burden was accounted for by a small group of employees. Finally, the findings support previous research that suggests an association between specific health risks and absenteeism (Edington, 2001).

CONCLUSION

The results suggest that behavioral and clinical risk factors are predictive of high compared to low rates of absenteeism, and that different risk factors were predictive of absenteeism for men and women. Perceived stress had a prominent role for both males and females. Stress is not a single conditioning but a response to physical and psychological stimuli (Thomas, 1993). Many other studies have addressed aspects of stress in relation to absenteeism, including related-costs and productivity (Jacobson et al., 1996; Jamal, 1984; Neubauer, 1992; Tang & Hammontree, 1992; Woo et al., 1999). In additional to the identification of stress as a predictor of sick leave usage, several of the predictive variables are related to stress (high blood pressure, food choices, tobacco and alcohol use, physical inactivity). This indicates the potential for general health promotion strategies to address stress as well as the individual risk factors.

However, it is apparent that more research is needed to understand what part risk factors and health issues play in sick leave usage. As previously stated, issues surrounding absenteeism are complex. Certainly health promotion programs, such as screenings, behavior change, and utilization of healthcare resources can impact health status, and therefore have a positive effect on worker productivity and absenteeism (Reidel et al., 2001). The role these programs play in self-care and patient education is valuable; especially screenings and early detection of potentially health-threatening conditions. While research has focused on individual factors associated with absenteeism, it would be worthwhile to assess the relationship between absenteeism and structural variables such as morale, availability of flex-time, dissatisfaction with workplace conditions. This type of study calls for a large-scale study across a representative sample of the working population so that industry or occupation-specific confounding variables could be contained.

An additional area that warrants further examination is why employees are absent from work. Sick leave usage for personal reasons, such as care giving responsibilities and personal business is a difficult area to capture and understand. Women in this analysis who never had children used significantly less sick leave than women who had children. (Due to the phrasing and nature of the question, data were not collected for men.) The number of women, compared to men, who are primary caregivers to children, aging parents and grandparents--or both will impact why women are absent from work compared to their male colleagues. In theory, identification of a problem such as care giving issues, could promote understanding and solutions. Here too there is much potential for negative impact, such as discrimination in hiring, employee evaluation and promotion practices.

On a positive note, information gleaned from this study suggests potential for use of these methods in future analyses; especially in different industries and among different occupations. Good correlational data exist across many studies in this field. But the use of binary logistic regression offered the opportunity to include a comprehensive list of predictive variables that were expressed in different ways (continuous and categorical). The identification of different predictive health risk variables, by population and by gender, indicates the need for specificity in health promotion research.

REFERENCES

Aldana, S. G. (2001). Financial impact of health promotion programs: a comprehensive review of the literature. American Journal of Health Promotion, 15(5), 296-320.

Anderson, D. R., Staufacker, M. J. (1996). The impact of worksite-based health risk appraisal on health-related outcomes: A review of the literature. American Journal of Health Promotion, 10(6), 499-508

Bertera, R. L. (1991). The effects of behavioral risks on absenteeism and health-care costs in the workplace. J Occup Med, 33(11), 1119-1124.

Burton, W. N., Chen, C. Y., Schultz, A. B., & Edington, D. W. (1998). The economic costs associated with body mass index in a workplace. Journal of Occupational and Environmental Medicine, 40(9), 786-792.

Burton, W. N., Conti, D. J., Chen, C. Y., Schultz, A. B., & Edington, D.W. (1999). The role of health risk factors and disease on worker productivity. Journal of Occupational and Environmental Medicine, 41(10), 863-877.

Cole, G. E., Tucker, L. A., & Friedman, G. M. (1987). Absenteeism data as a measure of cost effectiveness of stress management programs. American Journal of Health Promotion, 1 (4), 12-15.

Gazmararian, J. A., Foxman, B., Yen, L.T., Morgenstern, H., & Edington, D. W. (1991). Comparing the predictive accuracy of health risk appraisal: The Centers for Disease Control versus Carter Center program. American Journal of Public Health, 81(10): 1296-301.

Haynes G., Dunnagan, T., & Smith V. (1999). Do employees participating in voluntary health promotion programs incur lower health care costs? Health Promotion International, 14(1): 43-51.

Jacobson, B. H., Aldana, S. G., Goetzel, R. Z., Vardell, K. D., Adams, T. B., & Pietras, R. J. (1996). The relationship between perceived stress and self-reported illness-related absenteeism. American Journal of Health Promotion, 11(1), 54-61.

Jamal, M. (1984). Job stress and job performance controversy: an empirical assessment. Organizational Behavior and Human Performance, 33, 1-21.

Leigh, J. P. (1990). Hypertension, gender, job hazards and absenteeism in a 1973 national sample of U.S. workers. Health Policy, 16, 221-232.

Narbro, K., Johsson, E., Larsson, B., Waaler, H., Wedel, H., & Sjostron, L. (1998). Economic consequences of sick leave and early retirement in obese Swedish women. International Journal of Obesity. 20, 895-903.

Neubauer, P. J. (1992). The impact of stress, hardiness, home and work environment on job satisfaction, illness, and absenteeism in critical care nurses. Medical Psychotherapy, 5, 109-122.

Reidel, J. E., Lynch, W., Baase, C., Hymel, P., & Peterson, K. W. (2001). The effect of disease prevention and health promotion on workplace productivity: a literature review. American Journal of Health Promotion, 15(3), 167-191.

Smith, K., McKinlay, S. & Thorington, B. (1987). The validity of health risk appraisal instruments for assessing coronary heart disease risk. American Journal of Public Health, 77(4), 419-424.

Steinhardt, M., Greenhow, L., & Stewart, J. (1991). The relationship of physical activity and cardiovascular fitness to absenteeism and medical care claims among law enforcement officers. American Journal of Health Promotion, d5(6), 455-460.

Tang, T. L., & Hammontree, M. L. (1992). The effects of hardiness, police stress, and life stress on police officers' illness and absenteeism. Public Personnel Management, 21(4), 493-510.

Thomas, C. L. (Ed.). (1993). Taber's Cyclopedic Medical Dictionary (17 ed.). Philadelphia, PA: EA. Davis Company.

Thompson, D., Edelsberg, J., Kinsey, J. L., & Oster, G. (1998). Estimated economic costs of obesity to U.S. business. American Journal of Health Promotion, 13(2), 120-127.

Tucker, L. A., Aldana, S. G., & Friedman, G. M. (1990). Cardiovascular fitness and absenteeism in 8,301 employed adults. American Journal of Health Promotion, 5(2), 140-145.

Tucker, L. A., & Friedman, G. M. (1998). Obesity and absenteeism: an epidemiologic study of 10,825 employed adults. American Journal of Health Promotion, 12(3), 202-207.

Woo, M., Yap, A. K., Oh, T. G., & Long, F. Y. (1999). The relationship between stress and absenteeism. Singapore Med J, 40(9), 590-595.

Yen, L. T., Edington, D. W., & Witting, P.. (1992). Prediction of prospective medical claims and absenteeism costs for 1284 hourly workers from a manufacturing company. J Occup Med, 34(4), 428-435.

HEALTH EDUCATION RESPONSIBILITY AND COMPETENCY ADDRESSED

Responsibility I: Assessing Individual and Community Needs for Health Education

Competency B: Distinguish between behaviors that foster and those that hinder well-being

Sub-competency 2: Identify behaviors that tend to promote or compromise health

Troy B. Adams, Ph.D.

Virginia S. Cowen, M.A.

Troy B. Adams, Ph.D. and Virginia S. Cowen, MA., are affiliated with the Department of Exercise and Wellness at Arizona State University. Address all correspondence to Troy Adams, Ph.D., Arizona State University, Department of Exercise and Wellness, 7350 E. Unity Drive, Mesa, AZ 85212, PHONE: [phone], FAX: [phone], E-MAIL: [e-mail].

Table 1: Correlation of Predictor Variables with Sick Hours

                                           FEMALES

Continuous                           Pearson        Sig.
Variables                          Correlation    2-tailed

Body mass index                       0.059        0.147
Total cholesterol                     0.038        0.351
Body fat percentage                   0.012        0.774
Systolic blood pressure              -0.138        0.001 *
Diastolic blood pressure             -0.098        0.015 *
# of drinks per week                 -0.016        0.698
# of times had 5+ drinks in
    past month                        0.042        0.304
Composite: back safety (1)           -0.014        0.724
Composite: eating habits (2)          0.081        0.047 *
Composite: food choices (3)           0.011        0.786
Composite: tobacco use (4)           -0.058        0.147
Composite: coping style (5)           0.037        0.362
Composite: self care (6)              0.021        0.608

Categorical                        Spearman Rho     Sig.
Variables                          Correlation    2-tailed

Take blood pressure medication       -0.050        0.214-
Diabetes (7)                         -0.041        0.313
Heart Disease (7)                    -0.079        0.050 *
Osteoporosis (7)                     -0.035        0.392
Physical Activity (8)                -0.070        0.084
Physical limitations affect
    ability to exercise (9)          -0.055        0.173
Stretch (10)                          0.032        0.432
Strength train (10)                   0.003        0.945
Wear seatbelt (10                    -0.010        0.795
Age at birth of first child           0.103        0.010 *

Composite: cancer (11)               -0.064        0.114
Level of life stress/strain (12)      0.062        0.125
Satisfaction with life (13)           0.017        0.674
Quantity of social support (14)       0.037        0.361
Quality of social support (15)       -0.035        0.392

                                              MALES

Continuous                           Pearson        Sig.
Variables                          Correlation    2-tailed

Body mass index                       0.115        0.039 *
Total cholesterol                     0.045        0.416
Body fat percentage                   0.131        0.018 *
Systolic blood pressure               0.048        0.391
Diastolic blood pressure              0.109        0.050 *
# of drinks per week                 -0.003        0.962
# of times had 5+ drinks in
    past month                        0.068        0.230
Composite: back safety (1)           -0.118        0.033
Composite: eating habits (2)         -0.078        0.163
Composite: food choices (3)          -0.105        0.059
Composite: tobacco use (4)            0.089        0.108
Composite: coping style (5)          -0.094        0.091
Composite: self care (6)              0.047        0.401

Categorical                        Spearman Rho     Sig
Variables                          Correlation    2-tailed

Take blood pressure medication        0.051        0.358
Diabetes (7)                         -0.142        0.011 *
Heart Disease (7)                    -0.028        0.619
Osteoporosis (7)
Physical Activity (8)                -0.133        0.017 *
Physical limitations affect
    ability to exercise (9)          -0.130        0.019 *
Stretch (10)                         -0.067        0.227
Strength train (10)                  -0.145        0.009 *
Wear seatbelt (10                     0.055        0.319
Age at birth of first child

Composite: cancer (11)                0.003        0.961
Level of life stress/strain (12)     -0.020        0.715
Satisfaction with life (13)           0.012        0.824
Quantity of social support (14)      -0.019        0.730
Quality of social support (15)       -0.048        0.391

* Indicates statistical significance at p = 0.05 level.

(1) Composed of 7 variables that assessed performance of protective
measures such as stretching and exposure to risk factors for back
problems

(2) Composed of 2 variables that assessed regularity of eating and
habitual lower fat food preparation

(3) Composed of 9 variables that assessed frequency of healthy food
consumption

(4) Composed of 9 variables that assessed frequency of tobacco use
including cigarettes, pipes, and chewing tobacco

(5) Composed of 6 variables that assessed use of healthy and unhealthy
coping techniques

(6) Composed of 10 variables for women and 8 for men that assessed
frequency of preventive medical screenings

(7) Single item that assessed lifetime occurrence of this condition

(8) Single item with 8 response options that assessed habitual
physical activity

(9) Single dichotomous item that assessed presence of physically
limiting conditions

(10) Single item that assessed habitual practice of this activity

(11) Composed of 5 items for women and 4 items for men that assessed
lifetime prevalence of various forms of cancer

(12) Single item with 5 response options that assessed the level of
stress/strain in life

(13) Single item with 3 response options that assessed the level of
satisfaction in life

(14) Single item with 5 response options that assessed the quantity
of social connections

(15) Single item with 6 response options that assessed the collective
quality of social connections

Table 2. Sick Leave Usage deciles by year

                        FEMALES

                   Sick Hours: Year One

  Deciles     Mean Hours    N    Std. Dev.

     1            3.27      61      3.15
     2           15.32      61      3.57
     3           28.38      62      4.10
     4           41.60      62      4.07
     5           55.37      61      3.98
     6           69.73      62      5.05
     7           87.98      62      5.50
     8          114.06      61      8.53
     9          145.95      62     11.41
    10          235.43      61     84.10
All Females      79.67     615     71.93

                          MALES

                   Sick Hours: Year One

  Deciles     Mean Hours   N     Std. Dev.

     1             .00      45       .00
     2            3.78      20      2.07
     3           10.06      30      1.98
     4           18.01      35      2.00
     5           26.19      31      2.43
     6           37.23      34      3.50
     7           49.37      33      3.82
     8           65.74      32      5.43
     9           94.75      33     12.40
    10          164.44      32     55.67
 All Males       46.79     325     51.37

                        FEMALES

                   Sick Hours: Year Two

  Deciles     Mean Hours    N    Std.Dev.

     1            3.82      62      3.15
     2           16.00      61      4.18
     3           29.37      61      3.53
     4           40.02      62      3.34
     5           53.62      61      4.57
     6           70.86      62      4.93
     7           89.28      62      7.15
     8          112.42      61      7.22
     9          146.51      62     14.66
    10          242.69      61     78.30
All Females      80.37     615     72.91

                          MALES

                   Sick Hours: Year Two

  Deciles     Mean Hours   N     Std.Dev.

     1             .00      39       .00
     2            4.55      18      2.20
     3           10.18      39      2.51
     4           19.11      33      2.59
     5           29.60      34      3.22
     6           40.40      32      3.40
     7           54.97      33      4.35
     8           69.95      32      4.14
     9           88.28      33      7.04
    10          147.71      32     45.12
 All Males       46.46     325     45.66

Table 3. Binary Logistic Regression Results for Females.

                                BOTTOM DECILE      TOP DECILE
                                 Lowest Sick      Highest Sick

                                 Leave Usage      Leave Usage
                                    (N=60)           (N=53)

                                 Mean    S.D.    Mean     S.D.
Sick Leave Usage (hours)         3.8     3.15    242.7    78.3

Variables in Predictive Model    % Correct *      % Correct *

1) Level of Stress/Strain           17.3%            98.3%
2) Food choices                     51.9             78.3
3) Body Mass Index                  53.8             76.7
4) Systolic Blood Pressure          69.2             81.7
5) Tobacco Use                      65.4             81.7
6) Cancer                           67.3             81.7

                                 ALL FEMALES
                                   (N=615)

                                 Mean    S.D.
Sick Leave Usage (hours)        80.4    72.91

Variables in Predictive Model    % Correct *

1) Level of Stress/Strain           60.7%
2) Food choices                     66.1
3) Body Mass Index                  66.1
4) Systolic Blood Pressure          75.9
5) Tobacco Use                      74.1
6) Cancer                           75.0

* Percent accuracy of model in predicting participants in top
decile or bottom decile.

Table 4. Binary Logistic Regression Results for Males.

                                BOTTOM DECILE    TOP DECILE

                                 Lowest Sick    Highest Sick

                                 Leave Usage     Leave Usage
                                   (N=39)          (N=32)

                                Mean    S.D.    Mean    S.D.
Sick Leave Usage (hours)        0.0     0.0     147.7   45.12

Variables in Predictive Model    % Correct *     % Correct *

1) Body Mass Index                  83.8%           36.0%
2) Physical Activity                91.9            60.0
3) Level of Stress/Strain           91.9            64.0
4) Quality of Social Support        91.9            80.0
5) Heart Disease                    91.9            80.0

                                  ALL MALES
                                   (N=325)

                                Mean    S.D.
Sick Leave Usage (hours)        46.5    45.66

Variables in Predictive Model    % Correct *

1) Body Mass Index                  64.5%
2) Physical Activity                79.0
3) Level of Stress/Strain           80.6
4) Quality of Social Support        87.1
5) Heart Disease                    87.1

* Percent accuracy of model in predicting participants in top
decile or bottom decile.