Study design, setting and participants
The study employed a combination of stratified and cluster sampling to identify the sample population. A total of 85 hospitals across seven provinces (Shandong and Jiangsu from Eastern, Gansu and Yunnan from Western, Guangdong from Southern, Beijing from Northern and Hubei from Central) in China were invited to participate in the study, which ran from July 2014 to April 2015; of these, 77 hospitals (90.59%) agreed to participate. Within each hospital, nurses were recruited from four different surgical specialty units and four internal medicine units (obstetric and pediatric units were excluded). Convenience sampling was used to select nurses from each of the participating units. Participation was voluntary, and all data were confidential. In all, 9093 nurses, from 528 units, were asked to complete survey. A total 6667 (73.32%) nurses completed the survey. After excluding 1085(11.93%) nurses who provided invalid questionnaires and a further 697 (7.67%) nurses who provided questionnaires missing key responses in this study, we were left with a study sample of 4885 nurses (Figure S1 in the Additional file 1).
Data collection
Most study data, including primary outcomes, primary independent variables and the control variables, were collected using a self-administered paper survey questionnaire (Additional file 2). Trained survey interviewers sent copies of the questionnaire to each unit, along with an explanation of the survey purpose and method. After 1 or 2 days, the survey interviewers returned to collect completed questionnaires. Additional data on the number of physicians, nurses, and beds in the unit were obtained from the unit head. The academic status of each hospital was obtained from the official hospital website and in the case of teaching hospitals, verified first with the hospital administration and then, with the respective affiliated universities (since some teaching hospitals in China have nominal teaching activity) (Additional file 3).
Study measures
The primary independent variables consisted of three organizational behaviors (organizational justice, leadership attentiveness, and group interaction) and two patient behaviors (patient trust and unreasonable demands by patients). Existing scales assessing organizational justice, including most notably Colquitt’s Organizational Justice Scale [22], have previously been reported to be laborious and impractical. Therefore, in the study organizational justice was assessed using just two items adapted from the full Colquitt’s scale, namely pay justice (a component of Colquitt’s distributional justice) and task justice (a component of Colquitt’s procedural justice). Similarly, leadership attentiveness was assessed by the items attention to staff interests (reflecting material requirements) and attention to staff opinions (reflecting spiritual requirements). Group interaction was also assessed by two items, the number of dinners with colleagues per month (reflecting social interactions) and the number of clinical case discussions per month (reflecting work-related interactions). Finally, patient behaviors were assessed using the two items patient trust (an intrinsic behavior) and unreasonable demands by patients (an explicit behavior). Each question was categorized on a 5-point Likert scale with possible responses ranging from “not at all” to “extremely”. Once categorized, each item was recoded into quarter variables.
In previous studies [23, 24], well-being has been measured by several items, including quality of life, job satisfaction, life happiness, and burnout. In the present study, the primary outcomes were limited to three measures of nurse well-being: job satisfaction (“Overall, how would you rate your satisfaction with your work?”), turnover intention (“If you had another opportunity to choose your profession, how likely are you to become a nurse?”), and life happiness (“Overall, what do you think your happiness score is?”).
The outcome variable questions including job satisfaction and turnover intention were categorized on a 5-point Likert scale with possible responses ranging from “very low” to “very high”; the response for life happiness was converted to categorical variable (0–19 =very low; 20–39 =lower; 40–59 =average; 60–79 =higher; 80–100 =very high) and then, recoded as binary variables (i.e., satisfied versus dissatisfied, low versus high turnover intention).
Previous research [9] has demonstrated an association between nurse well-being and several nurse- and institutional-related factors. Based on this, our analysis controlled for the following potentially confounding factors: sex, age, marital status, education level, title, economic status, family support, hospital level, hospital type, academic status, nurse specialty, and nurse-to-patient ratio.
Data analysis
To adjust for nonresponses, the data were weighted by respondent age and sex, according to the available hospital personnel demographic information issued by the National General Hospital in 2015. Standard descriptive summary statistics were used to characterize the nurses. For crude comparisons, chi-square tests or Kruskal-Wallis tests were used to analyze categorical variables, with type Ⅰ error rates of 0.05. Binary logistic regression models were used to examine the association of organizational and patient behaviors with nurse well-being.
Sensitivity analyses were conducted to test the robustness of the results, including: (1) modeling without the few cases of male nurses; (2) adjustment for interactions between the measures of well-being, i.e., adjusting for the effects of life happiness in the modeling of job satisfaction and for the effects of job satisfaction in the modeling of turnover intention and life happiness, respectively; (3) all new models adjusted for the previously described potentially confounding nurse- and institutional-related factors. All analyses were done using IBM SPSS Statistics for Windows, version 22.0 (IBM Corp., Armonk, NY).