Determinants of Burnout among Teachers: A Systematic Review of Longitudinal Studies

We aimed to review the determinants of burnout onset in teachers. The study was conducted according to the PROSPERO protocol CRD42018105901, with a focus on teachers. We performed a literature search from 1990 to 2021 in three databases: MEDLINE, PsycINFO, and Embase. We included longitudinal studies assessing burnout as a dependent variable, with a sample of at least 50 teachers. We summarized studies by the types of determinant and used the MEVORECH tool for a risk of bias assessment (RBA). The quantitative synthesis focused on emotional exhaustion. We standardized the reported regression coefficients and their standard errors and plotted them using R software to distinguish between detrimental and protective determinants. A qualitative analysis of the included studies (n = 33) identified 61 burnout determinants. The RBA showed that most studies had external and internal validity issues. Most studies implemented two waves (W) of data collection with 6–12 months between W1 and W2. Four types of determinants were summarized quantitatively, namely support, conflict, organizational context, and individual characteristics, based on six studies. This systematic review identified detrimental determinants of teacher exhaustion, including job satisfaction, work climate or pressure, teacher self-efficacy, neuroticism, perceived collective exhaustion, and classroom disruption. We recommend that authors consider using harmonized methods and protocols such as those developed in OMEGA-NET and other research consortia.

hazards (e.g., increased job demands or limited job resources) [21]. Of note, burnout is one of the most frequently studied adverse effects of psychosocial exposures at work. Indeed, specific job demands in the teaching profession, especially increased responsibilities and tight deadlines, identify teaching as one of the most stressful occupations [7,[22][23][24].
It is well known that exposure to chronic workplace stressors can result in development of burnout [25,26]. The job demands-resources (JD/JR) model of burnout assumes that the workplace context is characterized by a variety of physical, psychological, social, or organizational factors (also referred as job demands) that require prolonged physical or psychological efforts in workers. Job demands are not necessarily negative, but they may turn into workplace stressors when the invested personal efforts are high, meaning job demands may be associated with certain physiological or psychological costs [27], leading to overtaxing and emotional exhaustion. Additionally, the lack of job resources may result in withdrawal behavior (depersonalization) and disengagement [26,28,29]. Accordingly, job resources are those aspects of the job that reduce job demands (and the associated costs) and stimulate personal growth and learning and development [27]. In the context of reduced job resources (e.g., inappropriate performance feedback, low salary, job insecurity, inadequate supervisory coaching and teamwork), job demands are particularly detrimental [28][29][30].

Burnout in Teachers: Current State of Knowledge
Schaufeli and Taris [31][32][33] conceptualized burnout as the "combination of the inability and unwillingness to spend the necessary effort at work for proper task completion". In this context, inability and unwillingness are two inseparable components, representing energetic and motivational dimensions, respectively [34].
Three dimensions (exhaustion, cynicism, and lack of professional efficacy) usually define burnout, but evidence indicates that lack of professional efficacy plays a divergent role as compared to exhaustion and cynicism [30,35]. Empirical results confirm the exceptional role of lacking efficacy compared with the other two burnout dimensions, illustrated by: a low correlation of lack of efficacy with exhaustion and cynicism; findings that burnout manifests itself via exhaustion and cynicism in psychotherapeutic clients, but is not manifested by lacking efficacy; and evidence that lack of efficacy shows a different pattern of correlations with job characteristics when compared with exhaustion and cynicism [36].
Exhaustion at both physical and psychological levels constitutes the core dimension of occupational burnout. According to the harmonized definition of occupational burnout elaborated within the framework of the EU COST Action CA16216 (The Network on the Coordination and Harmonization of European Occupational Cohorts-OMEGA-NET; http://omeganetcohorts.eu, accessed on 5 May 2022), it is characterized as a state of physical and emotional exhaustion due to prolonged exposure to work-related problems. In this EU COST Action, occupational burnout was chosen as priority health outcome for harmonization of its definition, measurement, and research protocols for future epidemiological studies. To do so, systematic reviews have been conducted [37][38][39][40], including on the predictors of occupational burnout onset, regardless of the type of occupation activity or job [41].
In this article, a systematic review on determinants of occupational burnout among teachers is presented. We include only longitudinal studies, since cross-sectional studies do not consider temporality.

Objective
The objectives of this systematic review focused on longitudinal studies were to identify the determinants of burnout onset in teachers and to show how much further beyond qualitative analysis we can go with quantitative synthesis.

Methods
The study was conducted according to a study protocol registered in PROSPERO (CRD42018105901), with a focus on teachers. We performed a literature search for the period  1990 (January) to 2018 (August) in three databases: MEDLINE, PsycINFO, and Embase. Because there was a possibility that additional studies were published during or after finalizing this review, we checked databases for new publications up until December 2021, and no additional prospective longitudinal studies on burnout in teachers were identified.

Study Selection and Criteria for Inclusion and Exclusion
We included only prospective longitudinal studies where burnout was a dependent variable (outcome), with a final sample of at least 50 teachers per exposure group (i.e., studies with sufficient power), published between the years 1990 (January) and 2018 (August) in peer-reviewed journals, with no language limitation. The full search strategy can be found in Supplementary Figure S1. In cases where we identified multiple publications describing a single study, we included the study only once and chose the most complete or most recent publication. We excluded studies where exposure was not assessed prior to the outcome assessment. Other reasons for exclusion were: no full text available; and studies where participants were not professionally employed (e.g., students). For one study, written in German [42], we could not assess the risk of bias (RoB) and excluded this study at this stage.
Titles and abstracts were screened independently by two reviewers (D.C. and S.C.M.) against the inclusion criteria. For studies that were not excluded on the basis of the title or abstract, full-text manuscripts were obtained and assessed by two reviewers. Any discrepancies were resolved through discussion, and if required a third reviewer (IGC) was consulted.

Data Extraction and Quality Assessment
We developed a MS Excel data extraction form, which was tested by all reviewers and improved until reaching consensual approval. From each included study, two reviewers (D.M. and D.C.) independently extracted the data as follows: study reference, country, objective, design, hypothesis tested and result (confirmed or not confirmed hypothesis), burnout definition used, tool used for burnout measurement, inclusion and exclusion criteria, number of occupational groups for which separate data are available, number of samples used, initial sample size, final sample size of occupational group, sex ratio (F/M), mean age (min/max), setting (urban/rural), participation rate, participation rates for other waves if several (max 4 waves), time interval between waves, number of burnout domains investigated, definition of predictive (explanatory) variables (including burnout determinants), statistical method used, type of result reported (e.g., slope or relative risk with differences across burnout measures), final model description, burnout domain name, wave number, occupational group, category of predictor and outcome (cut-off values of category boundaries, ordinal level, or continuous variable), result, result variability measure (e.g., SE, CI, p-value), variability value, and author's interpretation. Particular attention was paid to the outcome definition and assessment, along with the precision of the method or tool used and the cut-off values. The same attention was paid on the determinants studied.

Risk of Bias Assessment
The RoB was assessed using the Methodological Evaluation of Observational Research (MEVORECH) [43], which automatically produces a RoB report in MS Access format, taking into account the external validity (sampling of subjects, assessment of sampling bias, response rate, exclusion rate from the study, reported methods used to address sampling bias) and internal validity of the study (methods used to obtain data on dependent and independent variables, reference period, reported validation, and reliability of used methods; treatment of confounding factors; data on the loss of follow-up, appropriateness of the used statistical methods, reporting of the tested hypotheses, precision of the estimates, and sample size justification).

Qualitative and Quantitative Synthesis
Studies were summarized in a narrative synthesis with two summary tables: per study and per type of independent variable (determinants). All determinants were grouped into 4 types according to the previous studies, knowledge, and experience as follows: support, conflict, individual characteristics, and organizational context. Support is defined as "an interpersonal transaction of help from a support source to the help receiver that involves emotions, material assistance, and information and that takes place in a specific family, work, or care-giving context" [44]. Conflict refers to a workplace phenomenon that is not harmful when handled objectively and in a timely manner but that leads to lost communication, affecting people and work performance when not handled at all or handled wrongly [45]. In addition to other factors, individual characteristics, such as perceived self-efficacy, coping strategies, sense of defeat, or demographic characteristics, could be related to burnout. The organizational context of burnout is described by the areas of work life, including workload, reward, fairness, and values [46].
The support category included support from colleagues, support from a supervisor, support from the community, emotional support, and social facilitators [47][48][49]. The conflict determinant group comprised four factors: conflict with colleagues, emotional strain, parent criticism, and obstacles from parents or students. All of these factors were identified in two studies [48,49]. By reading about how each of those factors were defined, we decided to remove emotional strain from our list, as it represented the same construct as emotional exhaustion, which was not taken into account as a predictor but only as an output. Conflict with colleagues and parent criticism were found in the study by Feuerhahn et al. [48], and we selected their beta coefficients and standard errors in the most complete model. In the study by Salanova et al. [49], a single model predicting emotional exhaustion was available, which we took the data from. We ended up with three data points.
The category of individual characteristics comprised seven determinants: emotional dissonance, teacher self-efficacy (in managing student behavior), emotional exhaustion at the first wave of data collection (T1), depersonalization at T1, cynicism at T1, neuroticism (defined as "individuals who score high on neuroticism are more likely than average to be moody, and to experience such feelings as anxiety, worry, fear, anger, frustration, envy, jealousy, guilt, depressed mood, and loneliness") [50], and job satisfaction. Each of those determinants was tested in one of the four following studies, namely the studies by Feuerhahn et al. [48], Malinen et al. [51], Salanova et al. [49], and Goddard et al. [52], except for teacher self-efficacy, which was tested in two different studies.
Furthermore, given the fact that most studies have investigated emotional exhaustion as the main dimension of burnout, this was our primary focus of the study. Additionally, and since not all studies provided the standard error value but only the p-values, the next step was to transform the values using a two-tailed standard normal Z-table. We divided the beta regression coefficient by the z-score and obtained the standard error. We standardized each beta regression coefficient by dividing them by their standard error, and the obtained data points were plotted using R.
For the organizational context category, in the table of determinants we included 10 factors: time pressure, classroom disruption, perceived collective exhaustion, perceived collective cynicism, workload stressors, technical obstacles, effective class management, work climate (pressure), supportive school climate, and collective teacher efficacy. The factors were distributed between five articles [48,49,[51][52][53].
Besides the narrative synthesis, we summarized quantitative results reported for different determinants, considering the effect estimate (a standardized slope, for example, structural equation modeling, multiple linear regression, hierarchical linear regression, or random coefficient modeling) and variability estimate (SE, CI, or at least sample size and exact p-value of the chosen model). The reported estimates were standardized (Supplementary Tables S1-S4) and plotted using R software to distinguish between harmful and protective determinants, following the same procedures as in the general review [41].
With respect to the outcome, in most studies the three burnout dimensions were explored separately, while in others only emotional exhaustion was studied. Therefore, our synthesis of qualitative estimates focused on emotional exhaustion, the core dimension of burnout.

Description of Selected Studies
The PRISMA statement flowchart (Figure 1) describes the literature screening, study selection, and reasons for exclusion. The extracted data from the selected articles were stored in a table that included 240 articles for which the abovementioned variables, if available, were reported. After precise study selection and exclusion of full-text articles that described studies not conducted in teachers or studies not in compliance with inclusion or exclusion criteria, a total of 33 studies (Table 1) Figure 1. Flow chart of study identification and selection process.

Support
-Small negative effect of social support from colleagues at T2 on burnout at T3 for primary and secondary school principals-P -Social support from colleagues at T3 had a significant negative relationship with burnout (via stress) at T4 among the secondary school principals-P -Support from the broader community via the downside of empathy had a positive relationship with burnout at all measurement times (whether in primary or in both primary and secondary schools)-D -Social support from the broader community via stress showed significant negative indirect relationships with burnout at T1 (primary and secondary school principals) and T2 (primary school principals)-P Individual characteristics -Strong positive relationship between stress and burnout for both primary and secondary schools principals across all points of measurement-D

External validity
Major flaws: -Response rate in total sample <40% Minor flaws: -Non-general population -Self-selection of participants -Not addressed sampling bias in analysis Poor reporting: -Not reported exclusion rate

Internal validity
Major flaws: Individual characteristics -Burnout symptoms at T1 were the best predictor of cases of burnout at T2-D -Participants with interpersonal rejection sensitivity presented a 119% increase in the risk of being burned out-D

External validity
Minor flaws: -Non-general population -Self-selection of participants Poor reporting: -Response rate in total sample not reported -Exclusion rate not reported

Internal validity
Minor flaws: -Intensity/dose of the exposure (independent) variable not assessed in the study (only presence/ absence) Poor reporting: -Reliability of independent variable not reported Individual characteristics -The relationship between depersonalization and perceived self-efficacy showed an effect of the former on the latter, while the time frame was longitudinal-P -The relationship between personal accomplishment and perceived self-efficacy showed an effect of the former on the latter, while the time frame was synchronous-P

External validity
Major flaws: -Response rate in total sample <40% Minor flaws: -Non-general population -Self-selection of participants Organizational context -Lack of stimulation showed significant and independent correlations with burnout in all cases-D -Narrow client contacts showed significant and independent correlation with the dependent variables in almost all cases (DP, LPA, total MBI)-D -Conflict and ambiguity-only with EE-D -Type of school was significantly correlated with depersonalization-D Individual characteristics -Unmet expectations-only with DP-D -Individual demographic characteristics-contributed significant levels of explained variance on EE, DP, and total MBI score-D

External validity
Major flaws: -Response rate in total sample <40% Minor flaws: -Non-general population -Self-selection of participants Poor reporting: -Not reported exclusion rate Organizational context -Sources of stress and psychological burnout (positive)-D -Lack of social support and psychological burnout through sources of stress-D -Work setting characteristics and psychological burnout (positive)-D

Support
-Lack of social support and burnout (positive relationship)-D

External validity
Major flaws: -Response rate in total sample <40% Minor flaws: -Non-general population -Self-selection of participants Organizational context -Red tape work is the predictor of burnout for total sample (teachers and administrators within a single board of education), men, administrators, and teachers. The best predictor for the total sample, men, and administrators-D Individual characteristics -Self-doubt is the predictor of burnout for men-D Support -Lack of social integration is the predictor of burnout for teachers-D

Conflict
-Disruptive students is the predictor of burnout for total sample, women, and teachers. The best for women and teachers-D

External validity
Major flaws: -Response rate in total sample <40% Minor flaws: -Non-general population -Self-selection of participants -Not assessed sampling bias -Sampling bias not addressed in analysis Poor reporting: -Not reported exclusion rate

Internal validity
Minor flaws: -Intensity/dose of social integration not assessed in the study (only yes/no) Poor reporting: -Validity and reliability of independent variables not reported Organizational context -Stress intrinsic to the job predicted burnout at the kindergarten level-D Individual characteristics -In the total sample, men, and at the secondary level (low and high), a sense of defeat was the major predictor of a change in burnout-D -Loss of status-the most important predictor of burnout in women and at the kindergarten level-D Conflict -Stress due to societal demands was positively associated with burnout (total sample)-D -Stress due to relationship with students was positively associated with burnout (total sample, men, and high secondary school)-D -Stress due to societal demands predicted burnout at the kindergarten level

External validity
Minor flaws: -Non-general population -Self-selection of participants Individual characteristics -Downward identification had an independent positive relation with burnout-D -There was a significant negative effect of the use of a direct coping style on a change in burnout-P

External validity
Minor flaws: -Non-general population -Self-selection of participants Poor reporting: -Not reported exclusion rate

Internal validity
Minor flaws: -Reference period different from recommended Individual characteristics -Self-determined work motivation was positively associated with personal accomplishment-P Support -High quality of relationships with colleagues was negatively associated with EE, DP, RPA-P IC x Support -Having high-quality relationships with coworkers was negatively associated with EE over time, but only for employees with low self-determined motivation-P -High-quality relationships were beneficial only for employees with low self-determined motivation (concerning DP)-P -High-quality relationships were more positively associated with PA over time for employees with low self-determined motivation-P

External validity
Major flaws: -Response rate in total sample <40% Minor flaws: -Non-general population -Self-selection of participants Individual characteristics -Harmonious passion had a cross-lagged effect on professional efficacy-P -Obsessive passion had a cross-lagged effect on emotional exhaustion-D

External validity
Major flaws: -Response rate in total sample <40% Minor flaws: -Non-general population -Self-selection of participants Poor reporting: -Not reported exclusion rate Organizational context -Significant main (single) effect of cognitive job demand time pressure on T2 emotional exhaustion-D -Significant main (single) effect of cognitive job demand classroom interruptions on T2 emotional exhaustion-D Individual characteristics -Emotional job demand emotional dissonance with emotional support-significant interaction (buffering effect)-P -Emotional job demand emotional dissonance with self-efficacy-significant interaction (buffering effect)-P Conflict -Emotional job demand parents' criticism-significant main (single) effect on T2 emotional exhaustion-D -Emotional job demand conflicts with colleagues with emotional support-significant interaction (buffering effect)-P

External validity
Minor flaws: -Non-general population -Self-selection of participants -Incomplete justifications of the sample size Poor reporting: -Response rate in total sample not reported Individual characteristics -Academics exhibiting a self-critical form of perfectionism were found to report significantly higher exhaustion (T3, not T4)-D -Indirect effects of perfectionism on respite to post respite change in exhaustion, anxiety, and fatigue via worry and rumination during the respite-D -Academics who reported greater work-related worry and rumination during the respite showed elevated emotional exhaustion (T3 not T4)-D

External validity
Minor flaws: -Non-general population -Self-selection of participants -Incomplete justifications of the sample size Poor reporting: -Response rate in total sample not reported Organizational context -Innovation-negative relation with EE and DP, and positive relation with PA-P -Work pressure (climate)-positive relation with EE-D Individual characteristics -Neuroticism-positive relation with EE, and negative relation with PA-D

External validity
Major flaws: -Exclusion rate from the analysis >10% Minor flaws: -Non-general population -Self-selection of participants -Incomplete justifications of the sample size

Internal validity
Minor flaws: -Some subscales of the work climate scale with low reliability coefficients Organizational context -Perceived collective EE and cynicism at T1 positively predicted individual EE and cynicism at T2, respectively-D -Teacher-student ratio negatively predicted cynicism at T2-P Individual characteristics -Individual EE and cynicism at T1 positively predicted EE and cynicism at T2, respectively-D -Individual collective cynicism at T1 positively predicted individual cynicism at T2-D

External validity
Minor flaws: -Non-general population -Self-selection of participants Poor reporting: -Not reported exclusion rate

Internal validity
Minor flaws: Organizational context -T1 workload has a negative longitudinal relationship with T2 EE due to negative suppression (generally, positive)-D/P Individual characteristics -T1 negative affectivity has a negative longitudinal relationship with T2 EE due to negative suppression (generally, positive)-D/P

External validity
Major flaws: -Response rate in total sample <40% -Exclusion rate from the analysis >10% Minor flaws: -Non-general population -Self-selection of participants Support -Work-to-family facilitation at T1 caused low levels of exhaustion and disengagement at T2-P -High level of family-to-work facilitation at T1 predicted low levels of DE at T2-P Conflict -Work-to-family and family-to-work conflict produced lagged positive effects on EE and DE-D

External validity
Major flaws: -Response rate in total sample <40% Minor flaws: -Non-general population -Sampling bias not assessed -Sampling bias not addressed -Self-selection of participants Poor reporting: -Exclusion rate from the analysis not reported

Internal validity
Major flaws: -Major confounding factors not assessed Organizational context -Positive direct and indirect (perceived stress mediates this) effect of workload on EE and RPA-D -Positive indirect effect of inequity on EE (perceived stress mediates this)-D -Perceived stress has a positive direct effect on EE-D -Perceived stress has a negative effect on DP-P -Significant positive indirect effect is observed between perceived stress and DP (coping centred on traditional teaching methods mediates this)-D Individual characteristics -Coping centred on the problem-direct negative effect on all burnout dimensions-P -Adopting a traditional style of teaching-direct positive effect on all burnout dimensions-D -Avoidance coping-direct positive effect on DP and RPA-D -Self-efficacy-direct negative effect on all burnout dimensions-P Support -Social support (satisfaction)-significant indirect negative effect on the RPA (coping centred on the problem mediates this)-P

External validity
Major flaws: -Response rate in total sample <40% -Exclusion rate from the analysis >10% Minor flaws: -Non-general population -Self-selection of participants

Internal validity
Minor flaws:

External validity
Minor flaws: -Non-general population -Self-selection of participants Poor reporting: -Exclusion rate from the analysis not reported

Internal validity
Minor flaws: -Reference period different from recommended Organizational context -General school climate had a negative indirect effect on burnout-P Individual characteristics -Job satisfaction-direct negative effect on burnout-P -Teacher self-efficacy-direct negative effect on burnout-P

External validity
Minor flaws: -Non-general population -Self-selection of participants -Sampling bias not assessed, but justified the missing data Poor reporting: -Exclusion rate from the analysis not reported Organizational context -Temporary workers (type of contract) reported more EE at each wave. However, the Group × Time interaction effect was not fully consistent-D Support -In the absence of work-family enrichment, temporary employees reported the highest level of EE. The level of WFE did not affect the level of EE in permanent employees over time-D

External validity
Major flaws: -Exclusion rate from the analysis >10% Minor flaws: -Non-general population -Self-selection of participants Individual characteristics -Mastery was a strong positive predictor of problem-focused coping at both time waves and a significant negative predictor of emotion-focused coping-P -Failure avoidance was a strong predictor of emotion-focused coping and negative predictor of problem-focused coping-D -Problem-focused coping was a negative predictor of burnout-P -Emotion-focused coping was a strong positive predictor of teacher burnout-D

External validity
Minor flaws: -Non-general population -Self-selection of participants Poor reporting: -Exclusion rate from the analysis not reported

Internal validity
Poor reporting: -Reference period not reported Individual characteristics -Deep acting had a negative effect on EE over the period of a year-P

External validity
Major flaws: -Response rate in total sample <40% -Exclusion rate from the analysis >10% Minor flaws: -Non-general population -Self-selection of participants

Internal validity
Major flaws: -Major confounding factors not assessed Minor flaws: -Reference period different from recommended Individual characteristics -Work-avoidance orientation is positively related to burnout-D -Work-avoidance goal orientation emerged as the only significant predictor of burnout-D

External validity
Major flaws: -Response rate in total sample <40% Minor flaws: -Non-general population -Self-selection of participants -Incomplete justifications of the sample size Poor reporting: -Exclusion rate from the analysis not reported Individual characteristics -Teacher self-efficacy appears to be a protective resource against job stress, whereas job stress translates directly into burnout (EE and DP)-P

External validity
Minor flaws: -Non-general population -Self-selection of participants -Sampling bias not addressed in the analyses -Sampling bias not assessed Poor reporting: -Exclusion rate from the analysis not reported

Internal validity
Minor flaws: -Major confounding factors partially assessed Organizational context -For the stressors of heterogeneous classes, the T1 value of this stressor significantly positively predicted T2 burnout-D

External validity
Major flaws: -Exclusion rate from the analysis >10% Minor flaws: -Non-general population -Self-selection of participants

Internal validity
Minor flaws: -Major confounding factors partially assessed -Reference period different from recommended -Low reliability of the Heterogeneous classes and Physical conditions scales Individual characteristics -Burnout at T1 significantly related to burnout at T2-D -Self-efficacy negatively related to burnout at T1-P -Proactive attitude negatively related to burnout at T1-P

External validity
Major flaws: -Exclusion rate from the analysis >10% Minor flaws: -Non-general population -Self-selection of participants -Incomplete justifications of the sample size Conflict -Stress experienced in relationship with the students, colleagues, and organization-positive with EE, DP, and negative with PA (only students and organization)-D -Perceived inequity in relationships with students, colleagues, and organization-positive with EE, DP, and negative with PA (only students and organization)-D

External validity
Minor flaws: -Non-general population -Self-selection of participants Poor reporting: -Response rate not reported Conflict -Perceived inequity in relationships with students, colleagues, and organization-small magnitude of association and inconsistent results

External validity
Minor flaws: -Non-general population -Self-selection of participants Organizational context -Job demands (overload and role conflict positively predicted burnout-D -Job resources (autonomy and climate) negatively predicted burnout-P Individual characteristics -Self-efficacy at T1 was negatively related to burnout at T2-P -Also, self-efficacy showed mediation effect between JDs and burnout-P

External validity
Minor flaws: -Non-general population -Self-selection of participants Poor reporting: -Exclusion rate from the analysis not reported

Internal validity
Major flaws: -Major confounding factors not assessed Minor flaws: -Reference period different from recommended Organizational context -Effective class management-negative and significantly associated with EE and cynicism but these associations are mediated by EE and cynicism in T1-P Individual characteristics -Burnout dimensions at T1-mediating in all cases the relationship among obstacles/facilitators and burnout in T2-D -Gender-woman more exhausted than men-D

Conflict
-Social obstacles regarding to students and parents-positive and significantly associated to cynicism and DP but these associations are mediated by DP in T1-D

External validity
Minor flaws: -Non-general population -Self-selection of participants -Sampling bias not addressed in the analyses -Sampling bias not assessed Poor reporting: -Response rate in total sample not reported -Exclusion rate from the analysis not reported -Number of screened and eligible not reported

Internal validity
Minor flaws: -Reference period different from recommended Poor reporting: -Validity of independent variables not reported Individual characteristics -Problem-focused coping proved to be functional by attenuating burnout at T2 (reduces EE and DP)-P -Avoidance coping had a deleterious effect by worsening subsequent burnout (increases DP and RPA)-D -Conservative (traditional) pedagogical style had a deleterious effect by worsening subsequent burnout (increases DP and RPA)-D

External validity
Minor flaws: -Non-general population -Self-selection of participants -Sampling bias not addressed in the analyses -Sampling bias not assessed Poor reporting: -Response rate in total sample not reported -Exclusion rate from the analysis not reported -Not reported sampling method -Number of screened and eligible not reported

Internal validity
Minor flaws: -Major confounding factors partially assessed -Reference period different from recommended -Did not obtain methods to reduce bias Individual characteristics -Reduced self-efficacy increases burnout-D

Conflict
-Exposure to obstacles enhance a lack of perceived efficacy that in turn enhances burnout (EE and cynicism)-D

External validity
Minor flaws: -Non-general population -Self-selection of participants Four studies reported response rates above 60%, 8 studies between 40% and 60%, and 12 studies reported response rates below 40%. Nine studies did not provide response rates. The initial sample size varied from 78 to 5575 participants and the final simple size from 56 to 2235 persons.
Of the 33 included studies, 13 studies used a single composite measure of burnout as a dependent variable and the other 20 studies focused on one (seven studies using emotional exhaustion (EE) as a dependent variable), two (two studies), or three (ten studies) burnout dimensions. One study [77], besides EE and personal accomplishment (PA), additionally used two dimensions of depersonalization, one regarding the students and the other one regarding their colleagues.

Findings of the Risk of Bias Assessment
The RoB assessment showed that most studies assessed the major confounding factors (e.g., gender, age, marital status, work experience, level of education, income), but had external and internal validity issues due to limitations in sampling, inadequate reporting of response rates and exclusion rates, or use of self-reported instruments with uncertain validity and reliability.
In more detail, the analysis of the external validity of the included studies (N = 33) within the RoB assessment demonstrated that each analyzed study was designed to detect determinants of burnout in teachers as an occupational group. Hence, all involved studies were conducted in a non-general population [2,[47][48][49]. Similarly, self-selection of participants was demonstrated in each of the analyzed studies during a RoB assessment. Concerning the response rates across the whole sample, these were frequently either not reported [48,49,55,63,64,67,76,77,79] or were lower than 40% [47,54,[56][57][58]61,62,65,66,68,72,73].
Regarding the validity and reliability of measures used for assessing the independent variables, most were self-reported measures. Often authors did not report on the validity or reliability of these measures [49,[53][54][55]79], while others used subscales with low reliability coefficients [51,52,75]. Rarely did we find that the intensity or dose of the exposure (independent) variable was not assessed in the study [54,55]. The authors in these studies only used binary variables, such as "presence/absence" or "yes/no".

Qualitative Synthesis of Burnout Determinants in Teachers
The qualitative analysis identified 61 burnout determinants studied among teachers ( Table 2). Table 2 refers to the list of burnout determinants that were detected within the qualitative analysis.

Study (1st Author, Year of Publication)
Conflicts and interpersonal problems -Refers to conflicts or strained relations with parents; lack of respect, arrogance or violence on the part of some students; being blamed by some parents for their child's scholastic difficulties; pressure from the school inspector to improve their work or to work differently; fear of committing a professional error [58] D, JD (Emotional) Laugaa

Study (1st Author, Year of Publication)
Coping-Centred on the problem -Attempting to objectively analyze the situation and controlling one's emotions, thinking about the positive aspects of teaching, taking stock of the situation and attempting to rationalize it, giving the students positive encouragement, attempting to always remain coherent and honest in the relation with the students [57,58]; -Refers to persistence (keep trying at difficult things in work), planning (usually stick to a work timetable or work plan), and self-management (usually tries to find a place where one can prepare well) [   Inequity -Refers to lack of consideration for the job of teaching, little perspective of career advancement and promotions, an inadequate salary in light of the responsibilities and the work put in, the lack of recognition for the work and the efforts put in, the inflexibility of working hours [58] D, JD (Emotional) Laugaa et al., 2008 [68] Innovation -The perception of work climate as rich with workplace innovation; how innovative the school environment was [42] P, JR Goddard et al., 2006 [52] Lack of stimulation -Refers to challenging and stimulating nature of the job [46] D, JD (Cognitive) Burke et al., 1995 [57]   Time pressure/Work pressure (climate)/Workload -Refers to frequency of being pressed for time at work [38]; -The perception of work climate where workers experience significantly high work pressures [42]; -Quantitative and qualitative demanding aspects in the work situation, such as working under time pressure, working hard, and strenuous work [55]; -Refers to difficulty making progress with children who are failing academically, lack of time to monitor the progress of students individually, feeling responsible for their students' results, having too many things to do and not enough time to do everything, heavy workload [58]; -Refers to quantitative overload and too much work to do [59,69] D, JD (Physical) Feuerhahn  Work setting characteristics -Refers to inadequate orientation, workload, lack of stimulation and autonomy, unclear goals, poor leadership, and social isolation [47] D, JD (Physical) Burke et al., 1995 [58] * D-Detrimental effects; P-protective effects; JD-job demands; JR-job resources.
Support determinants were analyzed less frequently than the other determinants (e.g., lack of social integration and work-to-family or family-to-work facilitation by Burke et al. [54] and Innstrand et al. [66], respectively). Social support from colleagues, social support from the broader community, and lack of social support or lack of work-family enrichment were examined in two or three studies.
The qualitative analysis revealed specific determinants related to conflict relationships. Hence, stress due to societal demands [59] or stress due to relationships with colleagues or the organization [76] was detected as a significant determinant in a single study, while stress due to relationship with students was detected in four studies [49,59,76,79] and stress due to parental criticism was detected in three studies [48,49,79].
Finally, within the category of organizational context determinants, a lack of stimulation and narrow client contacts (e.g., "I spend most of my time in my job in direct contact with other people") were detected as significant determinants of burnout only in the study by Burke et al. [57]. On the contrary, ambiguity and conflict were detected as significant determinants in three studies [57,69,78], while time pressure, work pressure, and workload were detected in six studies [48,52,65,68,69,78].

Quantitative Synthesis of Burnout Determinants in Teachers
The quantitative synthesis added a certain value to the qualitative findings. We finally obtained a list of 6 articles [47][48][49][51][52][53] that satisfied the criteria for the quantitative synthesis (Table 3). Table 3 refers to the list of burnout determinants that were analyzed via quantitative synthesis.
The three studies [47][48][49] on the support group (determinants analyzed: support from colleagues, supervisor, and community; emotional support; social facilitators) are represented individually in Figure 2. Each point represents an association between the input (in this case, support) and output (emotional exhaustion). When the coefficient is positive, a positive correlation is shown, which indicates that support increases emotional exhaustion, causing a detrimental relationship. Conversely, for a negative coefficient, a negative correlation is shown, which means that support decreases emotional exhaustion and we have a protective relationship. The three studies [47][48][49] on the support group (determinants analyzed: support from colleagues, supervisor, and community; emotional support; social facilitators) are represented individually in Figure 2. Each point represents an association between the input (in this case, support) and output (emotional exhaustion). When the coefficient is positive, a positive correlation is shown, which indicates that support increases emotional exhaustion, causing a detrimental relationship. Conversely, for a negative coefficient, a negative correlation is shown, which means that support decreases emotional exhaustion and we have a protective relationship. In Figure 2, we first observed that there were statistically significant points only in the study by Beausaert et al. [47], and none in the two other studies. Therefore, emotional support (Feuerhahn et al. [48]) and social facilitators (Salanova et al. [49]) did not have an impact on emotional exhaustion. Concerning the study by Beausaert et al. [47], support from a supervisor was also not significant. Two support types were partially significant: In Figure 2, we first observed that there were statistically significant points only in the study by Beausaert et al. [47], and none in the two other studies. Therefore, emotional support (Feuerhahn et al. [48]) and social facilitators (Salanova et al. [49]) did not have an impact on emotional exhaustion. Concerning the study by Beausaert et al. [47], support from a supervisor was also not significant. Two support types were partially significant: support from colleagues and support from the community. The partially significant effects of both types of support meant that they were not constant over time. For colleagues' support in both schools and community support in secondary schools, the effects were significant only in the second wave but not in the two other waves. Community support in primary school was the only one with consistent results. Finally, we observed that the effect of the community support was opposite to the colleagues' support. Increasing the level of community support actually increases emotional exhaustion while increasing colleagues' support diminishes emotional exhaustion. Social facilitators Salanova et al., 2005 [49] effect of the community support was opposite to the colleagues' support. Increasing the level of community support actually increases emotional exhaustion while increasing colleagues' support diminishes emotional exhaustion. Figure 3 presents the results from two studies [48,49] reporting the estimates for the conflict group (conflict with colleagues, emotional strain, parental criticism, and obstacles from parents or students). We observed that even though parental criticism was quite far from zero, none of the three factors tested was statistically significant. The results for individual characteristics (emotional dissonance; teacher self-efficacy; exhaustion, depersonalization, or cynicism at T1; neuroticism; job satisfaction in studies by Feuerhahn et al. [48], Malinen et al. [51], Salanova et al. [49], and Goddard et al. [52]) are shown in Figure 4. We observed that all points except emotional dissonance had statistically significant detrimental effects on the onset of emotional exhaustion. The results for individual characteristics (emotional dissonance; teacher self-efficacy; exhaustion, depersonalization, or cynicism at T1; neuroticism; job satisfaction in studies by Feuerhahn et al. [48], Malinen et al. [51], Salanova et al. [49], and Goddard et al. [52]) are shown in Figure 4. We observed that all points except emotional dissonance had statistically significant detrimental effects on the onset of emotional exhaustion.   Figure 5 displays the results regarding organizational context determinants (time pressure, classroom disruption, perceived collective exhaustion, workload stressors, technical obstacles, effective class management, and work climate or pressure) in the studies by Feuerhahn et al. [48], Gonzalez-Morales et al. [53], Salanova et al. [49], and Goddard et al. [52]. Classroom disruption, perceived collective exhaustion, and work climate (pressure) exhibited statistically significant detrimental effects on emotional exhaustion. Four studies representing individual characteristics as burnout determinants Malinen et al., 2016;Salanova et al., 2005;Goddard et al., 2006) [48,49,51,52]. Figure 5 displays the results regarding organizational context determinants (time pressure, classroom disruption, perceived collective exhaustion, workload stressors, technical obstacles, effective class management, and work climate or pressure) in the studies by Feuerhahn et al. [48], Gonzalez-Morales et al. [53], Salanova et al. [49], and Goddard et al. [52]. Classroom disruption, perceived collective exhaustion, and work climate (pressure) exhibited statistically significant detrimental effects on emotional exhaustion. Figure 5 displays the results regarding organizational context determinants (time pressure, classroom disruption, perceived collective exhaustion, workload stressors, technical obstacles, effective class management, and work climate or pressure) in the studies by Feuerhahn et al. [48], Gonzalez-Morales et al. [53], Salanova et al. [49], and Goddard et al. [52]. Classroom disruption, perceived collective exhaustion, and work climate (pressure) exhibited statistically significant detrimental effects on emotional exhaustion.

Discussion
This systematic review of 33 longitudinal studies included 78 to 5575 teachers (in the first wave) and 56 to 2235 teachers (in the final sample). The studies were conducted in 11 countries. Several detrimental determinants of exhaustion were identified and classified according to their relative importance (i.e., effect size): job satisfaction as the most

Discussion
This systematic review of 33 longitudinal studies included 78 to 5575 teachers (in the first wave) and 56 to 2235 teachers (in the final sample). The studies were conducted in 11 countries. Several detrimental determinants of exhaustion were identified and classified according to their relative importance (i.e., effect size): job satisfaction as the most predictive determinant; work climate (pressure); teacher self-efficacy; neuroticism; perceived collective exhaustion; and classroom disruption as the least predictive determinant.

Interpretation of Findings
When interpreting the findings, one should take into consideration the risk of bias in the included studies. In this review, the risk of bias assessment resulted in a variety of methodological issues. All analyzed studies showed external or internal validity issues, mainly due to limitations in sampling, inadequate reporting of response rates and exclusion rates, or use of self-reported instruments with uncertain validity and reliability. Therefore, the sources of this bias should be carefully considered in future studies on burnout in teachers.
This systematic review showed that some burnout determinants were studied more frequently than the others (e.g., lack of stimulation and narrow client contacts in only one study; ambiguity or conflict in three studies; time pressure, work pressure, workload, or perceived self-efficacy in six or more studies). Role ambiguity [81][82][83][84] and workload or time pressure [85][86][87][88] were detected as significant burnout determinants in different settings. Perceived self-efficacy is frequently studied as a burnout determinant, mainly in teachers, but also in other occupations [89][90][91]. Surprisingly, support determinants were not so frequently analyzed (e.g., lack of social integration, work-to-family or family-to-work facilitation, social support from colleagues, social support from the broader community, lack of social support, lack of work-family enrichment), although they were frequently detected as determinants in burnout research [92][93][94][95]. Several determinants related to conflict relationships (e.g., stress due to societal demands or stress due to relationships with colleagues) were detected as significant determinants in one study, while stress due to relationships with students or parental criticism was detected in several of the analyzed studies. The interpersonal relationships and their associations with burnout were also studied in health care workers [96][97][98].
The identified determinants mostly showed detrimental effects (e.g., work setting characteristics as a composite score of inadequate orientation, workload, lack of stimulation, scope of client contacts, unclear institutional goals, lack of autonomy, poor leadership, and social isolation; sources of stress; heterogeneous classes; downward identification; lack of social integration; disruptive students or classroom interruptions; or perceived inequity in relationships with students, colleagues, and the organization). In line with the job demandsresources model of burnout [28,99], a lack of stimulation, narrow client contacts, downward identification, and ambiguity conflict are typical job demands, while a supportive school climate, autonomy, effective class management, and social support from colleagues and the broader community are classified as job resources.
The quantitative synthesis refined the importance of burnout determinants in teachers and was conducted on six selected studies. According to our observations, most of the data did not sustain the assumption that support is beneficial in reducing emotional exhaustion, the core dimension of burnout, although colleagues' support seemed to be in line with the assumptions but without strong evidence. In fact, the strongest evidence contradicted the assumptions, since the community support increased the emotional exhaustion. The authors [47] noted that this is a contradictory and unexpected finding that could be explained by 'the downside of empathy' (i.e., principals and teachers who feel supported by the community, are more connected to the community, and also more vulnerable to the community stress). This effect was stronger in principals and teachers working in primary schools because they were part of a smaller community than those working in secondary schools, and might be more connected to the community and more affected by its stressors.
Nevertheless, we need to explain our findings. All notable effects of support were present in the study by Beausaert et al. [47], who took into account several time intervals over one year and focused on burnout over a longer period (4 years in total vs. 2 years for Feuerhahn et al. [48] and 6 months for Salanova et al. [49]). Focusing on only the first wave of this study (as was the case in the other studies), we would only have a single measurement point during follow-up. Therefore, it is necessary to question how the development goes after burnout is first detected, namely whether it improves, deteriorates, or remains stable. This will probably depend on whether any actions are taken.
Lastly, putting these results in perspective, we can consider the few statistically significant effects (positive relationship with community support and negative relationship with colleagues' support) as suggestive rather than solid evidence. In the study by Kim et al. [100], fourteen empirical studies on burnout in students were reviewed and support was found to be negatively associated with all three burnout dimensions. Similarly, a systematic review of the work environment and burnout, not confined to any specific occupational group, showed moderately strong evidence of a relationship between low workplace support and increased emotional exhaustion [101].
Within the conflict group, even though parental criticism was quite far from zero (beta = 0.24, SE = 0.16, beta/SE = 0.24/0.16 = 1.5), none of the factors tested were statistically significant. As there was no significant factor in this group, we interpreted that parental criticism, conflict with colleagues, and obstacles from parents or students did not influence in any way the occurrence of emotional exhaustion. However, qualitative syntheses in the actual systematic review as well as in other studies [102,103] showed the positive relationships of obstacles with burnout. It has to be taken into consideration that the non-significant findings regarding conflict determinants in this review could be also due to the bias in these studies.
We observed that all analyzed individual characteristics, except emotional dissonance, were statistically significant. They all showed a positive relationship, i.e., increasing the determinant's value increases the values of emotional exhaustion. However, the results were not exactly as expected. For the determinant "neuroticism", it seemed logical that this individual characteristic favored the development of emotional exhaustion, and similar findings were also shown in other studies [104][105][106]. The syntheses highlighted the "emotional exhaustion at T1" as a determinant of emotional exhaustion with subsequent waves of data collection. Therefore, future studies must take into account the levels of burnout dimensions at T1 as confounding factors that are necessary to control for. We also observed positive relationships between teacher self-efficacy and job satisfaction with emotional exhaustion. These individual characteristics showed an inverse association of what we expected. Indeed, the literature has shown negative relationships between self-efficacy and job satisfaction with burnout [74,107,108]. A possible explanation for this result could be that the measure of one of those factors hides another factor. For example, a high level of job satisfaction may be reached only by working hard, implying one or more "hidden" burnout determinants (e.g., high workload). Teacher self-efficacy may also depend on how much discipline the teacher has. If the teacher does not encounter any difficulty with students, then there would be no reason for teacher self-efficacy to be especially high, as there was no "testing" occasion. Additionally, extensive indiscipline management may lead the teacher to think that they are a bad teacher and may reduce teacher self-efficacy.
The results for teacher self-efficacy were significant and consistent, showing positive relationships. This strengthened the evidence that teacher self-efficacy contributes to the emergence of emotional exhaustion. However, we cannot directly compare the two studies. Firstly, this was because teacher self-efficacy was not measured with the same tool. Feuerhahn et al. [48] used a scale developed by Schmitz and Schwarzer in 2000, while Malinen et al. [51] used the Finnish version of the Teacher Self-Efficacy for Inclusive Practices (TEIP) scale. Secondly, emotional exhaustion as an output also was not measured in the same way in both studies (Educator MBI vs. Finnish version of the Bergen Burnout Indicator 15 scale). However, the results of the two studies might still be comparable, despite using different measurement tools.
The analysis of the determinants belonging to the category of organizational context highlighted three significant points (perceived collective exhaustion, work climate, and classroom disruption). Perceived collective exhaustion and work climate (measured by "work pressure" or the degree to which the pressure of work and time urgency dominate the job environment) both had detrimental effects on emotional exhaustion. This seemed reasonable, as they probably were connected. Perceiving exhaustion in colleagues is part of a bad work climate. Moreover, classroom disruption could also additionally worsen the work climate. Actually, collective exhaustion and classroom disruption could be seen as "subfactors" of work climate.
Another element that caught our attention was the non-significance of the determinant "time pressure", contrary to our initial beliefs based on the literature that it would have a detrimental effect on emotional exhaustion by increasing stress [109][110][111][112]. A possible explanation for this result could be the lower level of time pressure within the teaching sector. As the teaching occupation's environment is not very dynamic (i.e., the activities are planned in advance), teachers are probably less exposed to time pressure than health care workers for example.

Methodological Considerations
This systematic review demonstrated extensive variability in the field of research of burnout determinants in teachers, with different time periods between the waves in different studies. This variability was illustrated by several points.
Firstly, the qualitative synthesis stage detected a wide range of determinants of burnout among teachers. This systematic review identified 61 factors that were highly correlated with burnout in teachers, which were studied in longitudinal settings. A wide spectrum of burnout determinants has been previously studied in teachers, not only in longitudinal studies, which we included in our systematic review, but also in cross-sectional studies [113,114]. However, quantitative synthesis revealed only six studies that had highlighted certain factors demonstrating clear significant protective or detrimental effects.
Additionally, we found that the different studies analyzed in this review used different instruments for burnout measurements. The authors mostly used different versions of the Maslach Burnout Inventory (MBI) [14], while several studies based their analyses on the Oldenburg Burnout Inventory (OLBI) [120], Copenhagen Psychosocial Questionnaire [121], Bergen Burnout Indicator [122], or Shirom-Melamed Burnout Measure [123]. Furthermore, we detected differences between studies in the measures that were used as dependent variables (e.g., composite measures of burnout; one, two, or three burnout dimensions; or two different dimensions of depersonalization).
This review identified a lack of consensus on the use of the burnout construct in measuring exposure and responses to occupational stress, with similar findings to those found in a study on physicians [124]. Different burnout definitions and different measurement tools used in research resulted in very low comparability of findings and results obtained through longitudinal studies on teacher burnout. Similar methodological heterogeneity among the studies in terms of shifting definitions of burnout and questions around the measurement tools of the burnout construct was also detected in healthcare workers [124][125][126][127][128][129] and other occupational groups [101,130,131].
Another issue that was revealed in this systematic review was the methodological inconsistency in the longitudinal studies, namely the different numbers of waves of data collection and the time periods between the waves. Similar differences were found through literature searches in other occupational groups as well [66,102,[132][133][134][135][136][137][138][139][140]. It is obvious that there is a need to harmonize the coordination and development of longitudinal studies in burnout research [141]. Taking into account the latency of occupational burnout, as we have previously recommended [41], the longitudinal studies with multiple waves [142] should involve at least 12 months follow-up of exposed workers.
Within the 33 included studies, 12 (36.4%) did not control at all or only partially controlled for confounding factors. Additionally, the reference period was shorter than the one year follow-up in 17 (51.5%) studies.
It is noteworthy that the literature search did not include the gray literature due to the lack of consensus on a standardized method for searching in the gray literature, the lack of full-text studies, and the non-publishing of the gray literature in peer-reviewed journals as a quality indicator [41].
Taking into account the large number of references screened and reviewed and the multiple methodological approaches implemented in this review, there was a possibility that additional studies were published during or after finalizing this review. We checked databases for new publications up until December 2021 and no additional prospective longitudinal studies on burnout in teachers were identified.
The strengths of this systematic review study are as follows. Only longitudinal studies were included, with different durations of follow-up, since cross-sectional studies do not consider temporality [143,144]. Additionally, a focus was placed on emotional exhaustion as the main component of occupational burnout. This is a consensually accepted dimension of occupational burnout that is measured by almost all available tools, including the most valid ones [41]. As in our previously published systematic review [41], we recommend that future research considers a longitudinal design with multiple waves [142], with at least one year follow-up of exposed workers. Finally, we performed a comprehensive risk of bias assessment according to the most validated and appropriate tools (MEVORECH).
Beyond the given recommendations for future research in the field of burnout in teachers, the practical implications of this review can be seen through the identified detrimental determinants of teacher exhaustion. These factors should be targeted as a priority within the development of prevention programs on burnout in teachers. Tackling the detected determinants of teacher exhaustion could reduce the prevalence of burnout among teachers.

Conclusions
This systematic review, particularly the quantitative synthesis phase, identified several detrimental determinants (job satisfaction, work climate (pressure), teacher self-efficacy, neuroticism, perceived collective exhaustion, classroom disruption) of teacher exhaustion. These findings are especially important due to the longitudinal design of the included studies published for this occupational group over a period of 30 years. The results of the protective determinants were inconsistent between studies, varying from wave to wave, while due to the high variability in measures between studies we were not able to clearly state which determinants truly influence emotional exhaustion.
In general, the difficulty in conducting this kind of systematic review has always been the lack of harmonization in the outcome measures and study protocols in general. We recommend that authors in the future research consider using standardized methods and harmonized protocols, such as those assessed and developed in OMEGA-NET [37][38][39]145] and other research consortia.