Pooled prevalence of depressive symptoms among medical students: an individual participant data meta-analysis

Depression is a common mental disorder affecting more than 264 million people globally [1]. Common symptoms of a depressive episode include depressed mood, loss of interest or pleasure in activities, changes in appetite, sleeping problems, loss of energy, feeling worthless or guilty, difficulty in thinking, and thoughts of death [2]. People suffering from depression may function poorly at work, in school, and within their family [1]. Additionally, depression is an important risk factor for suicide [3].

Studying at university is a transitory period in life from adolescence to adulthood, during which students have to make important life decisions. Students often experience pressure from economic stress, academic demands, and interpersonal relationships [4]. Among undergraduate programs, medicine is one of the most stressful, given the heavy workloads and high expectations from society [5]. Medical students have been reported to exhibit heightened psychological morbidity, daytime somnolence, impaired social interactions, and compromised general wellbeing, leading to mental health problems such as anxiety, stress, depression, suicidal ideation, and burnout [6], and these problems may appear as early as the first year of medical school education [7, 8].

The prevalence of depressive symptoms among medical students has been estimated from primary studies to vary from 1.4% to 80.6% [9, 10]. The pooled prevalence of such symptoms has been reported in systematic reviews and meta-analyses [5, 11, 12], some of which have explored factors including age, sex, geographical region, and sample size [5, 11]. Despite their enlightening insights from previous studies, the methodological choice of aggregated estimates from primary studies to compute the pooled estimate may be notable for some common drawbacks. Firstly, variations in the cut-off values of depression, e.g. using > 4 (mild depression) or > 9 (moderate depression) as cut-off for Patient Health Questionnaire-9 (PHQ-9) [13], might lead to inaccurate reporting of prevalence, i.e. a lower cut-off score would lead to a higher prevalence, and that may misestimate the disease burden of depression. Secondly, the use of aggregated data for analyzing the associated factors (through subgroup analysis or meta-regression) might lower its statistical power.

To address the research gap, our meta-analysis aimed to refine the precision of previous findings on the prevalence of depressive symptoms among medical students, through gathering individual participant data (IPD) as identified from our previous reviews [5, 14]. In addition, we can verify the associated factors of depressive symptoms among medical students with a larger sample size of individual data, which is another added value of the present study.

An IPD meta-analysis has been registered in the PROSPERO register of systematic reviews (CRD42018091917). This paper was written in accordance with the PRISMA statement for IPD [15].

Of the 249 studies [5], 197 provided email addresses, based on which contact was attempted and 75 responses were received. Nine authors declined to provide the data or lost contact after the initial reply, 14 authors implied the unavailability of the datasets, and one redirected us to his collaborators, but no response was received from them. Among those providing positive responses, 14 were unable to provide the data because they lost the labelling or coding of the data or because of other reasons. Finally, the datasets of 34 studies (from 37 publications) were included in this meta-analysis (Supplementary Fig. 1).

An IPD meta-analysis was conducted to examine the prevalence and associated factors of depressive symptoms among medical students. The pooled prevalence using IPD among medical students was found from our meta-analysis to be 18.1% (95% CI: 14.1%, 22.1%). Based on the two-step method (using standard cut-off values for depression), significant differences were observed for year of study, study region, and measurement tool, but not for sex. However, based on the one-step method (converting continuous assessment scores of depressive symptoms into z-scores), multiple linear regression suggested that age, sex, and year of study were significantly associated with the depression z-score, but the study region and measurement tool were not. It is remarkable that since the cut-off values of depression varied across studies, the estimation of prevalence by simply using aggregate data might not be accurate. The strength of an IPD meta-analysis is illustrated by the standard cut-off value to classify depression in this paper.

The estimated prevalence (18.1%) using the one-step method is lower than not only the pooled prevalence (27.6%) from our included studies using the 2-step method, but also the pooled prevalence from the previous meta-analyses, i.e., 27.0% [5], 27.2% [11], 28.0% [12], and 30.6% [28]. Although our analysis was conducted based on data from only around 15% (37/249) of the total number of papers were acquired, the pooled prevalence based on 2-step method did not differ much with the previously published meta-analysis. Therefore, the representativeness of studies might not be the major cause of the difference between the estimates from the 1-step and 2-step method, one potential explanation of the difference is that the former considered each subject had the same weighting but the latter still assigned the weighting based on the inverse of the variance from the estimate in each study.

Our results have shown that female medical students exhibited significantly higher depression z-scores than their male counterparts. This finding has been echoed in the literature [14], alongside insights into sex disparities in medical education [29]. Dahlin et al. reported that female medical students were subjected to sex discrimination [30], for which they would therefore seek psychological help more often than their male counterparts [31]. Besides, a survey conducted in four medical schools United States has found that being female was the most significant risk factor for experiencing sexual harassment among medical students [32], which might in turn associate with higher psychological distress [33]. In a qualitative study [34], female students felt that they struggled to define their roles in the wards and experienced different workplace relationships compared with male students; thus, experiencing higher level of depression. From a biological perspective, sex difference has been found depression-related gene expression, neuroplasticity, and immune signatures and are in opposite directions for some of the parameters (e.g. synapse-related genes increased in women with depression and were decreased in men), which may also account for the sex-specific prevalence of depression [35].

In terms of age, older medical students have been found to exhibit significantly higher depression z-scores after adjusting the effect from sex and year of study. This finding lends corroborative evidence to the literature. Rotenstein et al. have reported a 0.2% increase in the prevalence of depression for every year of increase in age [11]. Dyrbye et al. have found that fourth-year medical students aged above 24 would exhibit significantly higher odds (OR = 1.33) of being in a depressive status than their juniors [36]. Chen et al. posited that higher prevalence originated from more stressful events confronted by older students, such as employment, finances, graduation, and marriage-related pressures [37].

Our results have also revealed that medical students in earlier years of study would exhibit higher depression z-scores. Similar findings have been reported by Quince et al., who found higher mean depression scores among preclinical (i.e. first- to third-year) medical students than their clinical (i.e. fourth- to sixth-year) counterparts [38]. Saravanan and Wilks also posited that the transition from the pre-university to the university might underlie the higher level of depression among first-year medical students [39, 40]. Aktekin and Akdemir also found that the mental health of the students was adversely affected in the first year of medical school education [7, 8]. While our meta-analysis has not found differences between regions, previous studies have suggested that Asian medical students deal with greater stress, possibly because of the cultural emphasis on diligence and filial piety as inculcated by Confucian values and concomitant high expectations of their parents, teachers, and themselves [41]. Besides, we have noted the controversy that both "older medical students" but "medical students in earlier years of study" exhibit higher depression. We suspect that the controversy is due to the different system of medical schools. For example, students can enroll to medical school directly after A-level examination in United Kingdom system, so probably one can enroll at the age of 18–19. However, students in United States are required to have a first degree before they can enroll to medical school, therefore, their age could be 21–22 at least. Taken together, first year medical students may have experienced more depressive symptoms if they entered medical school at an older age. This may explain the finding from multiple linear regression, i.e. age and year of study were both significant predictors of the depression z-score.

The strengths of our IPD meta-analysis are threefold: its large sample size with data from more than 18,000 medical students; the adoption of commonly used cut-off values for the scales characterizing depressive cases; and the use of a consistent method for analyzing data from different studies. However, some limitations are noteworthy. Firstly, the generalizability of data was limited by only acquiring individual level data from 15% (37/249) of the total number of papers and only involving 19 countries worldwide, it might introduce bias due to the non-responses. Secondly, different scales were used in the included studies, across which possible variances in definitions might still be present and they might use different cut-off due to cultural difference, thereby using a standard cut-off might increase the false negative rates. Thirdly, as self-reported instruments were used, recall bias and reporting bias were inevitable. Fourthly, the analysis included only common factors (i.e., age, sex, year of study, and region), thus potentially meaning the omission of some factors. Lastly, some studies lacked data on specific groups: studies from Africa and North America had no data on sixth-year students.

To conclude, the pooled prevalence of depressive symptoms from the IPD meta-analysis was 18.1% (95% CI: 14.1%, 22.1%), which was lower than the previous meta-analyses using aggregated data. Age, sex, and year of study were significantly associated with the depression z-score. Despite the additional efforts in acquiring and integrating individual data from smaller proportion of included studies, IPD meta-analysis may provide a more accurate estimation of disease burden, and allow verification of associated factors.