The relationship of health-related quality of life to workplace physical violence against nurses by psychiatric patients

Abstract

Purpose

To clarify the relationship of health-related quality of life (HRQoL) to workplace physical violence (WPV) against nurses by psychiatric patients.

Methods

A Web-based WHOQOL-BREF was used to assess HRQoL and an intranet Workplace Violence Report System was used to record WPV incidents in order to design prospective longitudinal repeat measures over a period of 12 months, as well as using a subject-domain approach to detect any significant association between HRQoL and WPV.

Results

A total of 129 physical violence events were reported, and the WHOQOL-BREF was completed a total of 860 times. A lower HRQoL score within 7 days before an event in psychological domain was a predictor of WPV after adjustment for other variables. Other consistent risk factors for violence were being married, shorter employment duration, and quite a bit or extreme worry about violence. The social domain scores worsened within 7 days after an event.

Conclusion

A worsened psychological domain score might predict violence, and improving HRQoL seems to be helpful in preventing WPV against nurses by psychiatric patients.

Appendix

The daily reported WPV event can be generally represented by y _it for the i-th staff on the t-th day. y _it is equal to 1 when subject i filed a WPV report and zero otherwise. x _it is the i-th subject’s score of an HRQoL domain on the t-th day. We may assume that y _it is a realization of a Bernoulli distribution with a rare rate which may be affected by subject i’s personal HRQoL on the few days before day t, and other personal factors. Since WPV is a rare event and HRQoL was measured only a few times in the study period, it is difficult to select a proper time-domain model for testing the association between x _it and the sparse y _it directly. One possible solution is to apply the subject-domain approach proposed by Hwang et al. [19] which aggregates the sparse event data by staff. If x _ij for some days j close to day t has affected subject i such that y _it  = 1 for some day t’s in the study period, the average of these x _ij ’s can be treated as having an effect on the subject’s total WPV reports in the study period, $Y_{i} = \sum\nolimits_{t} {y_{it} }$. Similarly, if x _ij for some days j after the day t have been affected by the subject with y _it  = 1, the average of these x _ij ’s will be different from the subject’s baseline quality of life which can be estimated by the average of those scores measured far from the days on which the violence occurred. In this study, we divided each subject’s available repeated measured domain scores into three groups, those measured within 7 days before a report of violence, within 7 days after a report, and normal days. The three group average scores for the i-th staff are denoted as B _i , A _i , and N _i , respectively. Since N _i is an average of scores measured on normal days, we may assume that it is not affected by a violent event and treat it as the i-th staff’s baseline domain score. To adjust staff-to-staff variation, we define the baseline subtracted variables $B_{i}^{\text{adj}} = B_{i} - N_{i}$ and $A_{i}^{\text{adj}} = A_{i} - N_{i}$ for testing their associations with total number of events Y _i . Under the framework, we can reasonably assume Y _i to approximate a Poisson distribution. Obviously, the mean of Y _i is also affected by other personal characteristics and confounding variables. We consider six variables, which are religious belief, marital status, encouragement to report the violent event, employment duration, level of worry and the ward, respectively. The first three binary variables are denoted by Z _1i, Z _2i , and Z _3i , respectively, for the i-th subject. Variable Z _4i is the number of years the i-th subject has been employed. We denote two binary variables Z _5i and Z _6i as being extreme and moderate worry, respectively. We also use two binary variables Z _7i and Z _8i for working in an acute ward and intensive care unit, respectively. The Poisson model is given with the mean:

$$ E(Y_{i} ) = \exp \left[ {\beta_{0} + \beta_{1} B_{i}^{\text{adj}} + \beta_{2} A_{i}^{\text{adj}} + \alpha_{1} Z_{1i} + \cdots + \alpha_{8} Z_{8i} } \right]. $$

The model was implemented for physical, psychological, social, and environmental domains, separately, using the glm function in the statistical package R, version 2.10.1 (http://www.r-project.org/) for the estimation of the model parameters.