Description of the DRECT
The modified DRECT includes 35 questions, divided into 9 sub-scales: “Educational atmosphere” (5 questions), “Teamwork” (3 questions), “Role of the speciality tutor” (6 questions), “Coaching and assessment” (6 questions), “Formal education” (4 questions), “Resident peer collaboration” (3 questions), “Work is adapted to residents’ competence” (3 questions), “Accessibility of supervisors” (3 questions) and “Patient sign-out” (2 questions). The question can be answered on a five point Likert-scale (1 = totally disagree, 2 = disagree, 3 = neutral, 4 = agree, 5 = totally agree)[27]. The total DRECT score is the mean of all 35 questions. The score of each subscale is the mean of the questions within the subscale.
Translation and adaptation
The original questionnaire [14] was translated from English into French independently by two bilingual doctors (Forward translation). This translation was consolidated during a meeting of the research team. Then the French version was translated back to English by a professional translator (Backward translation). The translated questionnaires were evaluated during a meeting to reconcile the French and the English versions. The research team modified two questions to be more adapted with the local residency system (Supplementary material 1)[26, 28].
Pre-test
The initial version was submitted to 10 residents[29], during face-to-face interviews with a member of the research team. For each question, residents were asked if the question was clear and understood, and to propose changes to improve the questionnaire[30]. All remarks were noted and evaluated by the research team and eventually incorporated to the final version of the questionnaire in French (Supplementary material 1).
Distribution of the questionnaire
In addition to the French version of the D-RECT, the final questionnaire included demographic and professional questions also in French. An electronic questionnaire was created using the Google Form platform (https://www.google.com/forms/about/) and was submitted to Moroccan residents between July 1 and September 30, 2018. In the absence of residents’ email databases in medical schools, it was not possible to directly target residents in a consistent manner. To overcome this difficulty, referent doctors were designated in each university hospital and they were responsible for distributing the form to residents. Participation in the study was voluntary and data was collected anonymously.
Statistical methods
Quantitative variables are expressed as mean and standard deviations, or medians and quartiles as appropriate. For DRECT answers, normality of data distribution was assessed by the asymmetry test and the Kurtosis test. Absolute asymmetry values less than 3 and Kurtosis values less than 10 are considered acceptable for confirmatory factor analysis[31].
Quantitative variables are expressed in number and percentage. Missing values were replaced using the expectation-maximization (EM) technique.
For the DRECT scores, the means and standard deviations of each item and of the nine subscales were calculated. In addition, for each item, discrimination (rit), or item-total-correlation for each of the 9 subscales were calculated. Item-total correlations above 0.4 were considered good[18].
Evaluation of the construct validity
Confirmatory factor analysis was used to evaluate the validity of the construct [29, 32, 33]. The fit of the model was evaluated by the following indices[32]: SRMR (standardized root mean square residual), RMSEA (root mean square error approximation), CFI (Comparative Fit Index) and TLI (Tucker- Lewis Index). Threshold values for these indices were predetermined according to Brown's recommendations [32, 33] (SRMR <0.08 for a good fit and <0.12 for an acceptable fit, RMSEA <0.06 for a good fit and <0.10 for an acceptable fit CFI and TLI > 0.95 for a good fit and > 0.90 for an acceptable fit).
Convergent validity was assessed using factor loadings (or regression coefficients) and average variance extracted (AVE). Factor loadings above 0.55 and AVE above 0.5 were considered satisfactory[34, 35]. Inter-scale correlation was used for discriminant validity. Correlations between subscales of 0.85 or above indicate poor discriminant validity [35].
Fidelity analysis
Internal consistency
Internal consistency was assessed using Cronbach Alpha test [36]. A result greater than 0.7 was considered satisfactory[37]. The internal consistency was measured for the entire DRECT questionnaire and for each of the nine subscales. In addition, Corrected item-total correlations were calculated to examine the homogeneity of each subscale. Corrected item-total correlation above 0.40 was considered satisfactory[38].
Test – Retest
We invited 15 residents to respond to the questionnaire a second time to assess the stability of responses over time. A minimum of two weeks was necessary between the first and second measurements. Test-retest intraclass correlation coefficient greater than 0.6 was considered satisfactory[29].
IBM SPSS statistics 21 application was used for descriptive statistics, analysis of internal consistency, and test-retest reliability. The SPSS Amos Application Version 21 was used for confirmatory factor analysis.