The aim of this prospective study was to investigate the risk factors for junior doctors’ intention to leave clinical practice after specialty training. Our study contributes particularly to research through investigating the joint impact of on-the-job training and working in direct patient care. In the simultaneous model, training and work conditions show different effects on junior doctors’ intention to leave. In regard to working conditions, improvements in autonomy over time may reduce a junior doctor’s intention to leave clinical practice, whereas for the training conditions, a higher training standard in the first years of training may prepares the ground for a lower intention to leave clinical practice.
A potential explanation for the different effects of training and working conditions on intention to leave can be found by looking at the organization of residency training. Residency training in Germany, as in other countries , is organized as an apprenticeship with rotating assignments through different clinical departments and specialties. During this period, junior doctors experience various working conditions depending on the work organization and workload in the respective clinical department. Job autonomy covers the aspect of job control and describes junior doctors’ freedom in the arrangement of their own working process [11, 36]. In clinical rotation, job autonomy depends on the working processes and decision latitudes in the actual department. Furthermore, the job autonomy of junior doctors may be limited at the beginning of residency but increases once they gain more experience. Therefore, an increase in job autonomy during residency training will be expected by senior residents with a certain level of medical knowledge and skills. In line with our study, Heponiemi et al.  indicated that high job control, as measured by decision authority, was associated with a lower level of intention to leave the profession in a random sample of Finnish physicians (2,650 physicians). Furthermore, job autonomy in physician workplaces has been found to have positive effects on general psychological (e.g. psychiatric distress , depressive symptoms ) and job-related well-being (e.g. job dissatisfaction , emotional exhaustion ).
In regard to training conditions, it is an overall high standard of the training conditions as opposed to advancement in the conditions that appeared to be associated with lowered intention to resign from clinical practice. In our study sample, junior doctors in hospitals with higher standards of specialty training reported less ITL. In the context of clinical rotation structured residency training and dealing with lack of knowledge are influenced less by workload or work organization; these training conditions are rather an underlying prerequisite of successful on-the-job training. Structured residency training encompasses the existence of and compliance with a training schedule that guides clinical practice and is adhered to by supervisors. Poorly structured residency training is often criticized by junior doctors and can lead to an extension of the residency duration [9, 12]. Dealing with lack of knowledge is important for postgraduate junior doctors within training, as working and learning at the same time requires a learning environment in which it is feasible to ask questions openly and admit lack of knowledge or skills [40, 41]. In addition, this training condition is to a large extent independent of the working organization and workload of a department, as it describes a supportive learning environment in which it is possible to talk openly to supervisors and medical colleagues.
In terms of our underlying theoretical model, the significant effect of working conditions (in the final model 3) refers to the control dimension of the JDCS model. The fact that we did not find consistent evidence of additive and interactive effects of the JDCS model is in line with previous longitudinal research. De Lange et al.  found in their review that of 16 high-quality longitudinal studies examining the JDCS model, only 3 provided support for joint effects of all DCS dimensions.
As junior doctors have different training and working conditions in different countries the generalizability of our findings needs to be considered for each country. We assume larger potential for generalization for countries where training and working conditions are comparable to Germany (e.g., Austria, Switzerland). In other countries generalizability might be partly limited. For example, in the US similar training routines for residents occur like clinical rotation . However, working demands are different since residents in the US frequently work more than 70 hours per week  whereas German residents work less than 50 hours per week. Due to the European Working Time Directive working conditions will thus be in general less diverse within European countries.
Strengths and limitations
The longitudinal research design with three data collection waves is a particular strength of the study. This design provides empirically based insights how changes in the training and working conditions affect intention to leave. Our simultaneous analysis of junior doctors’ work and training conditions takes specific account of the different and joint demands of specialty training. The study therefore reflects the actual working situation of junior doctors in direct patient care and training. A further strength refers to the study population coming from all medical specialties involved in hospital care and private practice.
Our dataset provides no information about non-response after t2. Thus, we cannot infer potential reasons why doctors dropped out of further assessments. Because maximum likelihood estimates were used, this missing information should not introduce inconsistent estimation results like over- or underestimation, as long as the doctors’ dropouts depend on covariates observed in t1 or t2 [30, 44]. Therefore, the risk that missing information introduces inconsistent estimation is strongly reduced. In fact, the maximum likelihood estimation is a particular strength of the study as more dataset information could be used in the prospective analysis.
Although a validated standard questionnaire was used to obtain information on training and working conditions and intention to leave all data were based on self-reports. Therefore, common method bias may be a problem for estimating the true associations of the predictor and outcome variables .
A potential limitation may be the use of a single-item measure for the outcome variable. Many studies investigating intention to leave use single-item measures, such as thinking about giving up clinical practice  or intention to change profession . To our knowledge, no validation study comparing single-item measures and multiple-item measures of intention to leave has yet been conducted. As the construct used to measure intention to leave clinical practice in this study is homogeneous and leaves very little scope for interpretation, we believe that a single-item measure is sufficient in this context .
Personal characteristics are not modelled within the JDCS model. Interactions of job stress and personality (e.g. locus of control ) have been studied and may be also relevant in our study context. As personal characteristics can´t be influenced by supervisors or managers, we relied on the training and working conditions were a potential improvement may convince junior doctors to stay in the direct patient care environment after residency training. Further research is needed to clarify the role of dispositional characteristics (e.g. self-efficacy, resilience) of junior doctors in the examined relationship.
Another limitation worth discussing is that intention to leave clinical practice may differ from the actual behaviour of leaving. An recent investigation of 1,174 UK family physicians aged 50 or below confirmed that intention to leave direct patient care is a significant predictor of actual leaving . This study used data on doctors’ intention to leave direct patient care within 5 years and followed-up the physicians over the next 5 years to assess the actual leaving behaviour. A former investigation found that intention to leave clinical practice is not an accurate predictor of actual leaving behaviour . However, the study design was criticized because time lags in updating practicing status in the data source (the Physician Masterfile) took much longer than the time span investigated . Drawing on the sound result of the family physician study we think that intention to leave is a valuable predictor of actual leaving.
Our study indicates that both improving job autonomy and establishing a high standard of specialty training conditions may prevent junior doctors’ intention to leave clinical practice after residency training. The improvement of job autonomy during residency training could be established via a continuous process in order to reflect the clinical experience that junior doctors have obtained. This may be achieved by permitting experienced junior doctors to use their decision-making scope in the organization of work processes and the application of work techniques. Structured schedules in specialty training seem to be essential for junior doctors in order to achieve training objectives without hurdles and within an appropriate time frame. This may be supported by the implementation of a training plan, the use of training log books and/or the organization of rotation and assignment schedules . Openly dealing with lack of knowledge in a department reflects an underlying learning climate that may facilitates junior doctors learning on the job and would be part of the organizational culture of a hospital. Moreover, a climate of dealing with lack of knowledge openly appeared to give junior doctors security in clinical practice and may also have a positive effect on patient safety .