Study design:
According to the aim of our study, a prospective cross-sectional survey was conducted.
Inclusion criteria and study settings:
The study concerned consenting medical practitioners, regardless of age or gender. These included medical doctors and doctoral medical students working part-time in clinical practice in Gabon at the moment of the study. Gabon is a sub-Saharan African middle-income country. It is populated by about 1.8 million inhabitants with about 3 to 4 physicians per 10,000 inhabitants (6), or approximately a total between 540 to 720 physicians. But these physicians are unequally distributed on the national territory (7).
Sample size:
We determined the minimum sample size required for our survey based on an acceptable margin error of 5%, a confidence level of 95% and an expected prevalence of severe burnout of approximately 7%, average prevalence reported in the literature (8). Thus, by applying the formula n = z² x p (1 - p) / m² (9) (z: for 95% = 1.96, p: assumed prevalence = 7% and m: acceptable margin error= 5%), a minimum of 100 practitioners were required.
Sampling and data collection:
The source population was physicians and medical doctoral students working in clinical practice in Gabon and registered on the professional Facebook forum of physicians in Gabon, and the WhatsApp forum of doctoral medical students in Gabon. The Facebook forum of physicians is a professional online forum (named “physicians of Gabon”) where are registered 457 physicians on a total of 540 to 720 physicians in Gabon (6). They represent around 63.5% to 84.6% of physicians in Gabon according to medical density and population size. All doctoral medical students are registered in a WhatsApp forum. In fact, in Gabon, there is only one medical school, and doctoral medical students were 153 in Gabon at the moment of study. For academic needs, all doctoral students are registered in a WhatsApp forum to facilitate communication in medical school of Gabon. These online forums bring together the majority of medical practitioners and all medical doctoral students in Gabon. Thus, we estimated that these forums had representative population of physicians and medical doctoral students for the study. We made a list on Excel sheet 2016 using Facebook addresses of the 457 physicians and WhatsApp addresses of the 153 doctoral medical students. The Facebook addresses are easily accessible to each physician registered in the professional Facebook forum, and the WhatsApp addresses were available from the student delegate. Probability sampling was performed using the ALEA function in Excel 2016. According to sample size expected, the first 100 (+ 30%) addresses were selected and the link of anonymous questionnaire (with a cover text that succinctly explained the rationale of the study) has been sent to these online addresses. From time to time, reminders were necessary to mobilize the participants. The practitioners who answered the anonymous questionnaire according to their consent, were our study population (figure). Each anonymous questionnaire completed and submitted online was systematically collected on the database via google form. Data collection took place from November 1, 2018 to April 1, 2019. Data exported in Excel sheet to statistical analysis software.
Variables operationalization:
- Practitioners feelings about burnout were collected by using the validated French version of Maslach Burnout Inventory - Human Services Survey for Medical Personnel (MBI-HSS (MP)), widely used (10). Therefore, the following outcomes criteria have been defined:
- severe burnout defined by high level score in all 3 dimensions of the Maslach Burnout Inventory Scale: emotional exhaustion (EE) ≥27 and depersonalization (DP) ≥10 and personal achievement (PA) ≤33. Severe burnout was coded as a binary categorical variable (yes = 1, no = 0).
- burnout (or burnout symptom) defined by the existence of at least one high level score in one of the 3 dimensions of the Maslach Burnout Inventory Scale: EE score ≥27 or DP score ≥10 or AP score ≤33. The burnout (the burnout symptom) was coded in binary categorical variable (yes = 1, no = 0).
- the psychometric data related to the perception of the practitioners were collected by a self-administered Likert-scale questionnaire at 7 levels (proportionally to assertion worded) and treated as a numeric variable. These were the following parameters: a) the easy access to the hospital: in Gabon, public transport is poor and this suggests that the difficulty of access to the workplace could affect motivation, and therefore indirectly interacts to burnout. Thus, we have identified it as a potential factor to explore. b) the weekly frequency of the professional medical activity outside the public hospital: organization and equipment in private hospitals would make the working environment more satisfactory than in public hospitals. Or, working in the two hospitals could be overworked. Thus, we have identified it as a potential factor to explore. c) the frequency of activity in non-professional associations: taking part in extra-professional associative activities could be an environment that promotes humanization, or empathy skill, which is inversely correlated to burnout. Or, working simultaneously in non-professional associations could be overworked. Thus, we have selected it as a potential factor to explore. d) opinion favorable to traditional medicine: traditional medicine has a preponderant psychological orientation, supposing an understanding listening skill, which could interact with depersonalization feeling. Also, traditional medicine could be an alternative medicine where practitioners suffering of burnout symptoms would refer patients. Thus, we selected “to be favorable to traditional medicine” as potential factor to explore.
- demographic and socio-professional data: gender, marital status, hospital status (doctor versus doctoral medical student), hospital facility attended (university hospital center (UHC) versus other), extra-professional associative activity (yes versus no), means of transport (taxi versus personal car), place of residence relative to hospital (to live in same borough versus to live in different borough where the hospital is located) were coded as dichotomous categorical variables. Age, number of dependent children (living with the practitioner), number of elderly dependents (living with the practitioner), estimated average number of patients taken care of per day were coded as numerical variables.
Statistical analysis
Descriptive study:
Categorical variables were summarized as a percentage expressed with a 95% confidence interval (95% CI). The none normality of distributions of majority of numeric variables has been checked by One-Sample Kolmogorov-Smirnov test. Thus, numeric variables have been summarized by median with interquartile range (IQR).
Analytical study:
- univariate analysis to evaluate the link between each variable and burnout symptoms was performed by Fisher's Chi-square test for categorical variables and the Mann-Whitney U-test for numeric variables. Variables associated to burnout symptoms with statistical significance less than 25% (p <0.25) were selected for the multivariate analysis. In view of the exploratory nature of our study which tested original factors, we chose 25% to reduce the risk of excluding possible factors masked by confounding factors (11). Age, sex and place of residence relative to hospital (same or different borough from that of the hospital) were considered as forced variables for multivariate analysis.
- multivariate analysis was performed by binary multiple logistic regression using backward stepwise method (with Wald’s statistic). The adjusted odd ratio (aOR) was the estimator of the association between the burnout symptoms event and potential factor. Backward stepwise method generated a final model of probable predictive factors of prevalent burnout symptoms. The thresholds for the probability of entering and removing factor in the model were set at 5% and 10% respectively. The quality of the final model of potential predictor factors was evaluated by the Nagelkerke’s R-squared and the Hosmer-Lemeshow test. The French version of the SPSS 21 software (Statistical Package for the Social Sciences) was used.