Predictive factors of quality of life among medical students: results from a multicentric study – BMC Psychology

This is a cross-sectional study to evaluate factors contributing toward Quality of Life among medical students. This study is described in accordance with the STROBE (Strengthening the Reporting of Observational studies in Epidemiology) guidelines [41].

The Institutional Review Board of the School of Medicine of the University of São Paulo (CEP-FMUSP number 181/11), as well as the institutional review boards of all other participating medical schools approved the study, with all participants providing informed consent before the implementation of any study protocol.

We conducted this investigation as part of a multicenter study involving 22 Brazilian medical schools (the VERAS study, translated to English as “Students’ and Residents’ life in health professions”). Detailed description of this study was previously published [2, 36,37,38,39,40, 42]. Schools participating in the study were from all regions of Brazil, and with a diverse legal status and location (13 public and 9 private); 13 in state capital cities and 9 in other cities). All medical schools included (Universidade Federal do Rio de Janeiro, Universidade Federal de Ciências da Saúde de Porto Alegre, Universidade Estadual do Piauí, Faculdade de Medicina de Petrópolis, Faculdade de Ciências Médicas da Paraíba, Pontifícia Universidade Católica de São Paulo, Universidade Federal do Ceará, Universidade Federal de Goiás, Universidade Federal de Mato Grosso do Sul, Escola Baiana de Medicina e Saúde Pública, Faculdade de Medicina de Marília, Faculdade de Medicina de São José do Rio Preto, Faculdade de Ciências Médicas da Paraíba, Faculdade Evangélica do Paraná, Faculdade de Medicina do ABC, Fundação Universidade Federal de Rondônia, Pontifícia Universidade Católica do Rio Grande do Sul, Universidade Federal do Tocantins, Universidade Federal de Uberlândia, Universidade Estadual Paulista Júlio de Mesquita Filho, Centro Universitário Serra dos Órgãos, Universidade de Fortaleza and Universidade de Passo Fundo[43]) approved the study.

Students were stratified into clusters by gender and program year, using a computer-generated list of random numbers. Medical students included in the study received a link from both institutional and personal emails to access an electronic survey platform designed specifically for this project. The first page of the survey was the written informed consent, the student could only continue if he/she read and agreed to participate in the study. Apart from data collected through formal, validated instruments (described below), we also collected socio-demographic data: age, sex, location and type of medical school (public or private), year of training education, information on financial support and housing, heigh and weight. Students were asked to complete the survey within 10 days. After completing the survey, students received automatic feedback on their scores for each domain of each questionnaire along with information about its interpretation. There were psychologists in the research group, so students could contact any of the coordinating researchers for guidance or emotional support throughout the study. Data were collected between August 2011 and August 2012.

Participants

A total of 153 Brazilian medical schools were considered for inclusion in our study. We only included medical schools with at least one class in the process of graduation, bringing the total target population to approximately 86,000 medical students across all six years of training. We defined our sample size (n = 1152) to achieve an effect size of 0.165 comparing two groups with the same size, also assuming a statistical power of 80% and a 0.05 significance level. We later increased the sample to 1650 students to account for a 30% loss of participants. We then randomly selected at least 60 students from each of 22 medical schools. Next, we stratified all participants into clusters by gender and program year, i.e., five males and five females per each of the six program years, using a computer-generated list of random numbers. We excluded students who have not answer 100% of the questionnaires. To avoid sampling bias due to a low percentage of responses, we defined a minimum response rate of 60% in each medical school to include the responses in the final analysis, according to the recommendation of Huston [44].

Outcomes

Our outcomes of interest included: (a) personal quality of life (QoLp) measured using a self-reported analog scale with a score ranging from 0 (worst) to 10 (best), and (b) quality of life (QoLmc) related to medical course activities measured using a score ranging from 0 (worst) to 10 (best).

Predictors

We selected the following scales and instruments to investigate predictive factors:

World Health Organization Quality of Life Assessment abbreviated version (WHOQOL-BREF) is a brief version of the WHOQOL 100 and consists of 26 items, clustered into the following four domains: physical health, psychological health, social relationships and environment. Higher scores represent better QoL. This questionnaire was translated and validated to Brazilian Portuguese [45, 46]. The Cronbach´s alpha for each domain was 0.66 for social relationships, 0.73 for environment, 0.75 for physical health and 0.79 for psychological health.

VERAS-Q is a questionnaire created to evaluate the QoL of students in health professions (Additional file 1). It consists of 45 items with a 5-point Likert scale divided into four domains (time management, psychological, physical health, and learning environment) and a global score. The score ranges from 0 (worst quality of life) to 100 (best quality of life) [47]. The Cronbach´s alpha for each domain was 0.77 learning environment, 0.79 for physical health, 0.82 for both psychological and time management and 0.91 for global score.

Epworth Sleepiness Scale (ESS) is a self-reported questionnaire evaluating the likelihood of falling asleep in eight situations involving daily activities, on a scale of 0–3 (where 0 = would never doze and 3 = high chances of dozing). The overall score ranges from 0 to 24 with higher scores representing a person’s daytime sleepiness. A version has been translated and validated for Brazilian Portuguese [48, 49]. The Cronbach´s alpha for this questionnaire was 0.74.

Pittsburgh Sleep Quality Index (PSQI) measures self-reported sleep quality and disturbance over the past one month. The PSQI is composed of 19 items, which are combined into seven constructs that are summarized into a global score ranging from 0 to 21, where the highest score indicates worst sleep quality. This questionnaire was translated and validated to Brazilian Portuguese [50, 51]. The Cronbach´s alpha for this questionnaire was 0.65.

Beck Depression Inventory (BDI), is a 21-item, self-reported measure that measures characteristic attitudes and symptoms of depression. Responses are scored on a 4-point scale, ranging from 0 to 3, and total scores can range from 0 to 63. The higher the score, the greater the symptom intensity. This questionnaire was translated and validated to Brazilian Portuguese [52, 53]. The Cronbach´s alpha for this questionnaire was 0.87.

Resilience Scale (RS-14) measures the level of individual resilience, a person’s ability to cope with problems and withstand pressure in adverse situations without suffering physical, psychological, or social harm. It involves adaptive skills, flexibility, interaction, overcoming and resignification of lived experiences. The scale consists of 14-items and was translated and validated to Brazilian Portuguese [54, 55]. The Cronbach´s alpha for this questionnaire was 0.87.

State-Trait Anxiety Inventory (STAI) is a self-reported questionnaire consisting of 40 items used to measure the presence and severity of anxiety symptoms. This scale is composed of two distinct subscales: the state anxiety scale evaluates current behavior while the trait anxiety scale evaluates personality. This questionnaire was translated and validated to Brazilian Portuguese [56,57,58]. The Cronbach´s alpha for this questionnaire was 0.92 and 0.93, respectively for Trait and State anxiety.

Interpersonal Reactivity Multidimensional Scale (IRMS) evaluates empathy and its associated factors. It consists of 21 statements describing personal characteristics and following three main domains including empathic, emotional, and behavioral components. Medical students’ empathy was measured using three subscales of the validated version of the EMRI for Brazilian-Portuguese [59, 60]. The Cronbach´s alpha for each domain was 0.61 for emotional, 0.73 for empathic and 0.77 for behavioral.

Maslach Burnout Inventory (MBI) is a burnout assessment instrument consisting of 22 items divided into three domains: emotional exhaustion, depersonalization, and personal accomplishment. Each domain is scored on a seven-point Likert scale, with values from 0 (never) to 6 (every day). Burnout intensity is determined by the summing the scores of questions pertaining to each domain. Higher scores correspond to a higher degree of burnout. This questionnaire was translated and validated to Brazilian Portuguese [61, 62]. The Cronbach´s alpha for each domain was 0.68 for depersonalization, 0.81 for personal accomplishment and 0.85 for emotional exhaustion.

Dundee Ready Education Environment Measure (DREEM) contains 50 statements relating to the educational environment, scored on a five-point Likert scale (0 = strongly disagree and 4 = strongly agrees), some of which are inverted. Higher scores indicate a positive evaluation. The DREEM has a maximum score of 200, which represents an ideal educational environment. The subscales are as follows: Students’ perceptions of Learning (SPL); Students’ perceptions of Teachers (SPT; Students’ Academic Self Perception (SASP); Students’ perceptions of Atmosphere (SPA); Students’ social self-perceptions (SSSP). This questionnaire was translated and validated to Brazilian Portuguese [63, 64]. The Cronbach´s alpha for each domain was 0.65 for SSSP, 0.71 for SASP, 0.82 for SPL, 0.83 for SPA, 0.84 for SPT and 0.94 for global score.

Potential confounders

We selected potential confounders using a combination of clinical judgment and evidence from the literature, as these joint criteria have been demonstrated to perform better than the isolated selection of isolated clinical or evidence-based criteria [65]. We selected age, gender, body mass index, and school levels as potential confounders [66].

Statistical analysis

We started with a descriptive and visual exploratory analysis of all variables to evaluate their frequency, percentage and near-zero variance for categorical variables, distribution for numeric variables, and their corresponding missing value patterns [67]. Comparisons for the exploratory analysis were conducted through analysis of variance (t-tests being a category of analysis of variance) and Chi-square tests (Fisher exact test when any cell presented a frequency below 5).

Our strategy to evaluate predictors and their relationship with study outcomes (personal and related to medical course QoL) made use of a series of generalized linear models with a Gaussian family, i.e., Multiple Linear Regression Models (MLRM). The regression models were adjusted for age, gender, body mass index, and school levels. We used p for trend to measure the impact of individual numeric predictors on each of the QoL outcomes. In order to calculate predicted mean values for the outcome rather than simply obtaining less clinically interpretable measures of correlation, we categorized predictors at their median value (for example, median of Resilience Scale scores of 81, global DREEM score of 120). Results are reported as predicted means with 95% confidence intervals, with results being interpreted as significant when the confidence intervals do not overlap between different estimates along with p for trend < 0.001 [68].

We also used Regression Trees Model (RTM) with the same set of previously described outcomes and predictors. Regression trees complement the use of machine learning models as they represent the best cut-points for predictor values in the context of a given outcome after previous predictors have been taken into account. In order to avoid overfitting, we used a cost-complexity pruning strategy using the weakest link pruning strategy by successively collapsing the internal node that produces the smallest per-node increase in the cost complexity criterion [69]. When overfitting is detected, those nodes were removed. Otherwise, they were left intact. We have also provided a graphical representation of each model.

All analyses were performed using the R language with packages ggplot2 and rmarkdown; using the SPSS software version 22;