Participants and settings
Using cluster sampling, we conducted a pilot survey to assess the robustness and validity of our variables and the framing of our questions, before undertaking face-to-face and online surveys of medical professionals in Tianjin, China from July 1, 2019 to September 8, 2019. Respondents could scan the QR code on their mobile phones to complete the survey or fill out paper questionnaires. In order to improve the accuracy of the survey responses, the survey purpose, method and the conceptual basis of each variable was explained to participants. The inclusion criteria of participants captured physicians, nurses, medical technicians, pharmacists, administrators and other medical personnel with more than 1 year of work experience engaged in medical and health work in Tianjin public medical and health institutions. All participants were volunteers, who gave informed consent to participate and were capable of completing the questionnaire independently. The responses by participants were anonymous.
The minimum sample size was calculated as 270 completed surveys [43-45]. Of the 350 distributed surveys, 15 uncompleted and inaccurate questionnaires were excluded, providing 335 valid responses, with a 95.71% effective response rate. Of the 335 valid questionnaires, 121 were completed online and 214 were completed using paper questionnaires. There were no significant differences in the responses between the online and paper questionnaires.
Measures
The questionnaire composed five parts: job burnout, empathy, job satisfaction, job commitment and the respondent’s social-demographic characteristics, comprising age, sex, educational background, marital status, salary, workplace, professional titles and average daily working hours. The content of each scale is presented in Supplementary Table 1. Tianjin hospitals were categorized into three types: primary hospitals, equipped with less than 100 beds, only providing basic health services for a mainly local community; secondary hospitals, usually equipped with more than 100, but less than 500, beds, providing comprehensive medical services to several communities, with some teaching and scientific research activities; and tertiary hospitals, usually with more than 500 beds, considered a regional hospital providing high-level, specialty medical services to several districts, as well as performing higher level medical education and significant scientific research. Hospitals were also divided into two types, either specialist or general hospitals. Data on employment type (authorized and unauthorized) and job position were also collected.
Job burnout
To measure burnout, we employed the most commonly used burnout scale, the Maslach Burnout Inventory (MBI) [46]. Comprising a total of 22 items, MBI contained three key dimensions, ‘Emotional Exhaustion’, ‘Depersonalization’ and ‘Low-Personal Accomplishment’, measured on a 7-point Likert scale (1-never to 7-every day). The emotional exhaustion and depersonalization items were scored positively, with higher scores indicating a higher degree of job burnout, and the low personal achievement items used the reverse scoring method, with the lower the score, the higher the degree of job burnout. In order to improve the effectiveness of the questionnaire, we did confirmatory factor analysis (CFA), which allowed us to deleted the items whose coefficients were less than 0.6, yielding four ‘Emotional Exhaustion’ variables and four ‘Depersonalization’ variables and six ‘Low-Personal Accomplishment’ variables. We defined low burnout as a value less than 3 on the MBI scale; medium level burnout between 3 and 5 on the MBI scale; and high level burnout as greater than 5 on the MBI scale.
Empathy
Empathy of medical staff was measured by the Chinese version of the Jefferson Scale Empathy – Health Professions (JSE-HP) [47]. The 20 item scale was divided into three dimensions: perspective-taking(PT), compassionate care (CC) and standing in patients’ shoes (SIPS). The items in perspective-taking were understanding the emotions of patients and their families, empathy, attention to body language; items in compassionate care were the ability to understand and empathize with patients and help patients in timely manner, such as treating patients on the basis of understanding their emotions, paying attention to emotional changes when asking about their condition and considering patient's personal experience; and standing in patients’ shoes referred to the ability of medical staff to think from the patient's perspective, including treating the problem from the patient's perspective and thinking from the patient's perspective. The scale included response categories from 1-completely disagree to 7-completely agree, with the emotional care scale and considerate ability scale reverse scored.
Job satisfaction
Adopted from Zhang et al [48,49], the job description index (JDI) scale measured the medical staff’s job satisfaction, based on five dimensions: the satisfaction with the job itself, promotion, salary, manager and partner. In the interviews with participants in the pilot survey, one common feedback was that the doctor-patient relationship could influence them significantly, so we added “job environment” into the original scale. Therefore, the augmented JDI scale included a 6-dimensions scale, measured by a 5-point Likert scale (1-very dissatisfied to 5-very satisfied), with a higher score indicating the more job satisfied the respondent.
Job commitment
To measure job commitment, we used Meyer's three-dimensional scale, comprising ‘Emotional Commitment’, ‘Continuous Commitment’ and ‘Normative Commitment’ items, and measured by a 5-point Likert scale (1—very dissatisfied to 5—very satisfied).
Statistical analysis
Structure Equation Modelling
Structure equation modelling (SEM) refers to equations using parameters in the analysis of the observable, such as income or working hours, and latent variables, such as satisfaction and emotion. This study constructs a structural equation model of empathy and job burnout in medical staff and uses path analysis and diagrams to explain the internal relationships among variables to verify our hypotheses.
Mediating effect analysis
Mediating variables could explain the process of “how” and “why” between two variables. The influence of independent variables on dependent variables has both direct effects and indirect effects through mediating variables. Our research analyses and verifies the mediating effect of medical staff’s job satisfaction and job commitment between empathy and job burnout.
Multi-group analysis
This method was used to test whether the model we built is suitable for different population groups, which is whether the model is invariant among different sample groups. Additionally, it could verify whether a certain variable will have a moderating effect on the model. To test multi-group invariance, we divided our sample into three different hospital levels (primary hospital, secondary hospital and tertiary hospital); two hospital types (general hospital and specialty hospital); three job positions (physician, nurse and other medical staff); and two employment types (authorized employees, who entered the hospital or institutions by passing the formal examination and had registered in the China Organization Department and Ministry of Human Resources, and unauthorized employees).
Statistical methods
Data entry and conversion was completed with EpiData 3.1, and SPSS 24.0 (IBM Corp, Armonk, NY, USA) and AMOS 23.0 (IBM Corp, Armonk, NY, USA) was used to analyze the data. Descriptive analyses explored the key social demographic factors and the degree of job burnout of medical staff. Next, a structural equation model (SEM) was specified to analyze empathy and job burnout, mediated by job satisfaction and job commitment. Based on the SEM, a multi-group equivalence analysis was conducted to determine the impact of different types and levels of hospital, job positions and employment types on the model. In the process of model fitting, 5000 Bootstrap tests to the mediating variables were conducted; observed variables with path coefficients less than 0.6 were deleted; and the modification index (MI) was used to optimize the model. Cronbach's alpha values of each dimension and the overall scale were greater than 0.7, indicating good factor reliability. Most of the average variance extracted (AVE) values were greater than 0.5, indicating good aggregation validity of the model [126].