Design and participants
Data source and study sample
In this cross-sectional study, we adopted the nationwide hospital survey Needs Assessment Survey on Physical and Mental Health and Occupational Safety for Full-time Staff in Healthcare Workplace (Additional file 1) which was conducted by Bureau of Health Promotion in 2011. The structured questionnaire survey was developed to assess the health, health-related behaviors, and work conditions of hospital staffs in Taiwan. Work hours and turnover intention are two main dimensions assessed in this survey. Previous research has shown that the questionnaire has exhibited acceptable validity and introduced its sampling method clearly (Additional file 1). Among the 127 selected hospitals, 100 (78.7%) agreed to participate in this survey, and all full-time staff members at these hospitals were requested to participate. The questionnaire was anonymous, and a return envelope was provided with each questionnaire. Hospital staffs were requested to return the completed questionnaires in a sealed envelope to collecting sites at the hospitals. We distributed 98,817 questionnaires from May 2011 to July 2011, and 70,622 (71.5%) validated questionnaires were returned. Among those who returned their questionnaires, 4538 respondents reported that they were physicians. After we excluded responses with incomplete information (678 people) and those below 35 years old (1437 people), the final sample in our study comprised responses from 2423 attending physicians. For those below 35 years old, they are very likely still under trainee. After accomplishing trainee, it is reasonable to change to other hospitals to pursue their new medical lives. To avoid overestimating the meaning of intention to leave current hospital as the willing of turnover, we only analyzed the physician group above 35 years old. Furthermore, following previous studies, we chose to focus on physicians’ intention to withdraw, instead of actual turnover, as the dependent variable for several considerations [11, 19]. First, the cross-sectional nature of the survey prevented the measurement of physicians’ actual turnover from practice. Second, previous studies have shown turnover intention is a reliable predictor of actual turnover [20, 21]. The study protocol was approved by the institutional review board at the Bureau of Health Promotion prior to distributing the survey (BHP investigation number 0990800708).
Measurements
Dependent variable
Turnover intention was assessed using one question (“What is the likelihood that you will leave your current hospital?”. The responses were measured using a 5-point Likert scale ranging from 1 (none) to 5 (very strong). We classified turnover intention as mild (i.e., physicians who answered none or mild), moderate (i.e., physicians who answered moderate), or strong (physicians who answered strong or very strong).
Independent variable and moderating variable
The main independent variable was the number of work hours, which was measured using the survey item “Please recall how many hours you worked in the last week”. According to Taiwan Labor Standards Act, the normal weekly work hours shall not exceed 48 h, and the overtime weekly work hours shall not exceed 60 h [22]. Physicians in Taiwan are not considered laborers; thus, the Labor Standards Act does not apply to this profession. No legislation exists to regulate hospital physicians’ work hours, so the proposed reference of an 88-h maximum for residents’ work hours per week by the American ACGME (The Accreditation Council for Graduate Medical Education) was also adopted as one cut point [23]. Based on those cut points, we categorized the sample into five groups (< 49, 49–59, 60–88, and > 88 h) to further evaluate the impact of work hours on turnover intention.
The moderating variable considered in this study was pay satisfaction, which was assessed by one question (“Do you think it is reasonable for your current work pay?”. The responses were measured using a 5-point Likert scale ranging from 1 (very unreasonable) to 5 (very reasonable). We classified pay satisfaction as bad (i.e., physicians who answered very unreasonable or little unreasonable), moderate (i.e., physicians who answered moderate), or good (physicians who answered not bad or very reasonable).
Control variables
In the analyses, we also included sociodemographic variables (age, gender, and marital status) and work characteristics variables (seniority at current hospital, clinical setting, supervisor position, accredited hospital level, hospital ownership, and health promoting hospital (HPH) status). Furthermore, we subsequently included health status and job satisfaction as control variables to adjust their impact on the relationship between work hours and turnover intention. Self-rated health status was separately categorized into three groups (good, moderate, and bad), and job satisfaction was divided into two groups (good and bad).
Statistical analyses
We employed ordinal logistic regression models to analyze the association between the number of work hours and turnover intention. This method is suitable for dependent variables with multiple ordered response categories, and we verified the appropriateness of using this model, which means the relationship between any two pairs of outcome groups is statistically the same. To consider the cluster effect of hospitals, we used the “gllamm” command in the statistical software package Stata Version 12.1(StataCorp, 4905 Lakeway Drive College Station, Texas 77845-4512 USA), which can estimate generalized linear latent and mixed models. First, multivariable regression was applied to determine the relationship between work hours and turnover intention by adjusting the sociodemographic and work characteristics variables (Model 1). Next, we analyzed the association between work hours and turnover intention by incorporating health status, pay satisfaction, and job satisfaction into the models (Model 2). Finally, to evaluate whether pay satisfaction moderated the relationship between work hours and turnover intention, we conducted stratification analyses based on the level of pay satisfaction.