In this observational study, we aimed to evaluate the association between the introduction of a national shadowing programme and in-hospital mortality during the first Wednesday in August. During the study period, the number of emergency admissions was increasing over time, though the mortality rates were declining. We found no significantly higher odds of death in mortality for patients admitted on the first Wednesday in August compared with patients admitted on the last Wednesday in July. Furthermore, we did not observe a significant change in either the medical, surgical or neoplasm diagnosis groups.
Comparison with other studies
This is the first study that has assessed the impact of a shadowing programme on patient outcomes in healthcare though the evaluation of satisfaction of trainees was performed . The programme is designed for final year medical students to observe an existing doctor undertaking the usual activities required of their role for 4 days before starting work. Only a few previous studies investigated the impact of the various induction programmes for junior doctors; however, all these studies reported only perceived benefits rather than changes in patient outcomes .
A significant amount of research investigated the association between the beginning of newly qualified doctors training and adverse effects on patient care . The majority of these studies focused on 30-day or overall mortality for various subgroups of the patient [3, 18,19,20] or patients admitted to the intensive care unit [3, 21]. Only around 20% of these studies found an increase in mortality during the changeover, with an increase in relative risk between 4.3 and 41% or an adjusted odds ratio of 1.08 to 1.34.
Only two studies in the UK investigated the relationship between changeover and in-hospital mortality. Shuhaiber and colleagues  investigated the effect of cardiothoracic resident turnover on cardiac surgical outcomes and found a 30% higher odds of in-hospital mortality after a complex cardiac operation, but not for CABG alone. Jen et al.  found a 6% (OR = 1.06, 95% CI 1.00 to 1.13) higher mortality for patients admitted on the first Wednesday in August compared with patients on the last Wednesday in July . Furthermore, they found a significant increase in mortality for medical patients (OR = 1.08, 95% CI 1.01 to 1.16, p = 0.03) but not for surgical or cancer patients. In comparison with Jen et al’s study, we found similar results for the pre-intervention period for all patients (1.03 vs. 1.06). However, in this study we looked at different time periods (Jen looked at 2000–8, and we looked at 2003–11). Interestingly, in our study, we found no significant increase in mortality for medical patients. In both studies, CCS diagnostic groups were used to divide admissions into three clinical categories; however, the proportions of medical and surgical diagnostic groups in our study (61.6 and 36.0%) were different from the one in Jen et al. work (85.1 and 12.1%). The reason why these proportions are different is not clear, but we were unable to replicate the groups exactly, which may explain different results.
Strengths and limitations of the study
The main strength of our study is the use of a large and rich national administrative dataset that contains all hospital admissions and information relating to patient characteristics and some key outcomes such as in-hospital death. However, given that we focused on only 2 days of admissions a year and included only 7-day outcomes (to eliminate any potential overlap of care), we lacked power to detect small differences.
Another strength is the use of the generalised estimating equation method. The advantage of GEE models is that it provides unbiased estimation of population-level estimates despite the possible misspecification of the correlation structure. Furthermore, the use of GEE models with a control group enabled us to examine the effect of the shadowing programme despite the overall reduction in mortality in England (Fig. 2). The latter could make simple pre-post analysis potentially misleading.
Limitations mainly concern data availability. It is likely that a number of factors are responsible for the fall in in-hospital mortality over time, from changes in team mix and ways of working to discharge policies. For instance, there has been much work on failure to rescue following surgical complications , and teamwork is known to be important. Patient mix can change over time and can affect intervention effects if there is unmeasured confounding. Although the national hospital dataset includes information related to patient characteristics and the episode of care, ICD-10 is poor at capturing the severity of the disease. This could differ between the two patient groups, but its impact is likely to be modest. Furthermore, the dataset does not include any information related to intra-hospital transfers, which have been linked to various adverse outcomes,  or interpersonal and organisational dynamics within hospitals . The programme may also have had an impact on other patient outcomes and have had educational benefits for staff that we could not assess.
Another potential factor that we were not able to account for was the emotional preparation of junior doctors. It has been shown that the start of a job in a new hospital can be a stressful time for junior doctors, which may affect the number of errors they make . Moreover, every new medical member of staff has to attend organisational induction and mandatory training, which requires many doctors to be absent from patient care for a significant period .