Introduction

Medical students are more likely to suffer from poor sleep quality [1] compared to other college students [2], which may have a negative impact on their academic performance, physical and mental health, and quality of life [3]. Poor sleep quality may be related to emotional problems (e.g., stress, depressive, and anxiety symptoms) [4, 5], clinical placements [6, 7], heavy study workload (e.g., hectic schedule, vast syllabus, various clinical training, and onerous academic load) [8,9,10], and significant economic pressures [11, 12],

The Pittsburgh Sleep Quality Index (PSQI) is the most widely used instrument to evaluate subjective sleep quality in the past month. It covers a broad range of indicators relevant to sleep quality [13], including subjective sleep quality, sleep latency, sleep duration, sleep efficiency, sleep disturbance, use of sleep medication, and daytime dysfunction. The PSQI has been validated in numerous languages with satisfactory psychometric properties [14], and is commonly used across a wide range of clinical and research settings [15]. The PSQI has been also validated in college students [16,17,18,19,20] including medical students [21].

In order to develop effective interventions and lower the risk of the negative outcomes related to poor sleep, such as burnout [22], depression and anxiety [23, 24], and poor academic and work performance [25, 26], it is essential to examine patterns of poor sleep quality. To date, the findings regarding the patterns of poor sleep quality among medical students has been mixed across studies [27, 28]. No meta-analysis or systematic review has yet been conducted to examine the prevalence of poor sleep quality in this population. Thus, we conducted a comprehensive meta-analysis of the prevalence of poor sleep quality worldwide and its associated factors in medical students.

Methods

The study protocol was registered in the international prospective register of systematic reviews (PROSPERO; registration number CRD42019076413).

Data sources and search strategies

The preferred reporting items for systematic reviews and meta-analyses (PRISMA) checklist and PRISMA study flow chart were used. Four investigators (WWR, WL, HQ, and LH) conducted literature search in PubMed, EMBASE, Web of Science, PsycINFO, and Medline Complete from their inception dates until Aug 20, 2018 using the following queries: Pittsburgh Sleep Quality Index, PSQI, medical students, health occupations students, and medical education. Titles and abstracts of relevant publications were independently screened, and then the full texts were reviewed for eligibility by the same four investigators. If the same dataset was used in more than one publication, only the one with the largest sample size was included. Any disagreement was resolved after a discussion with a senior investigator (YTX). Fig. 1 presents the literature search and selection process.

Fig. 1
figure 1

Flowchart of literature selection

Full size image

Study eligibility

Original studies were included if they fulfilled the following inclusion criteria: (1) cross-sectional or cohort studies (only the data at baseline were extracted) on medical students; (2) available data on sleep quality measured by the Pittsburgh Sleep Quality Index (PSQI); (3) those published in English. Review articles were excluded. The reference list of included studies was also reviewed for additional studies.

Data extraction

The following information from included studies was extracted and recorded by four investigators using an Excel data collection spreadsheet, such as mean age, gender, sampling method, sample size, year of publication, study site, response rate, country/region, and PSQI cut-off and total score.

Quality assessment

The methodology quality of the studies was independently assessed by the same four investigators using the quality assessment instrument for epidemiological studies [29,30,31], with the total score ranging from 1 (lowest quality) to 8 (highest quality) points. The eight domains were: (1) target population was clearly defined; (2) probability sampling was used or the entire population was surveyed; (3) response rate was ≥ 80%; (4) non-responders were clearly described; (5) sample was representative of the target population; (6) data collection methods were standardized; (7) validated criteria were used to measure the target diagnosis or symptom; and (8) prevalence estimates were given with confidence intervals and specified by subgroups. Any discrepancies in quality assessment were resolved after a discussion with the senior researcher (YTX). This quality assessment instrument has been widely used in previous studies [32, 33].

Statistical analysis

Data were analyzed by the STATA, Version 12.0 for Windows (Stata Corporation, College Station, Texas, USA) R, version 3.3.0 and R Studio, version 0.99.903. The pooled prevalence of poor sleep quality was calculated as effect size (ES); the estimate pooled prevalence and its 95% confidence intervals (CIs) were calculated by the “metaprop” command in Stata 12.0 using the Freeman-Tukey double arcsine transformation and DerSimonian and Laird random effects model. Heterogeneity was measured by I2 statistics and Q-statistic, with I2 > 50% as high heterogeneity. Subgroup analyses were performed according to the following categorical variables: sampling methods (Cluster/Random/Convenience/Others), cut-off of PSQI (≥ 5/≥ 6/≥ 7/≥ 8), regional classification (Africa/the Americas/Europe/Asia/Oceania continents), publication year (in and after 2016/before 2016 according to the median splitting method) and clinical medical student (Yes/No/Both). For the prevalence of poor sleep quality, meta-regression analyses were performed based on continuous variables, including publication year, sample size, response rate, quality assessment score, mean age, and sex ratio. Begg and Mazumbar’s rank correlation test was used to explore publication bias. A bilateral alpha risk of 0.05 was set.

Results

Study selection

A total of 1,109 relevant articles were identified in literature search, and finally, 57 studies with 25,735 medical students were included for the analyses (Fig. 1). Of these, 50 studies had reported the prevalence of poor sleep quality and 41 had reported the PSQI total scores. One study [34] examined sleep quality in both clinical and nonclinical medical students separately; hence, this study was analyzed as two samples in subgroup analyses. Study characteristics are presented in Table 1.

Table 1 Characteristics of studies included in the meta-analysis
Full size table

Quality assessment and publication bias

The scores of study quality assessment ranged from 3 to 8 with the mean of 6. No publication bias for poor sleep quality was found in funnel plot (Fig. 3) and Begg’s test (z = 0.31, P value = 0.757).

Prevalence of poor sleep quality, subgroup analyses, and meta regression

The pooled prevalence of poor sleep quality across 50 studies with 24,884 medical students was 52.7% (95% CI: 45.3%–60.1%; I2 = 99.22; P < 0.001; Fig. 2). Subgroup analyses found that compared to other cutoff values (≥ 6, ≥ 7, and ≥ 8), studies using the PSQI cutoff value of ≥ 5 was associated with higher prevalence of poor sleep quality (P = 0.0003). Across the continents, the prevalence of poor sleep quality was highest in the studies conducted in Europe (65.13%), followed by in the Americas (59.92%), Africa (54.54%), Asia (47.44%), and Oceania (30.51%). Meta regression analyses revealed that smaller sample size (slope = − 0.0001, P = 0.009) was associated with higher prevalence of poor sleep quality.

Fig. 2
figure 2

Forest plot of the prevalence of poor sleep quality in medical students. The horizontal axis refers to effect size. Note: ES=Effect Size; CI=Confidence Interval

Full size image
Fig. 3
figure 3

Funnel plot of publication bias for studies of sleep quality (n=50)

Full size image

PSQI total score and subscale scores

The pooled PSQI total score from 41 studies with 16,748 medical students was 6.058 (95% CI: 5.614–6.538; I2 = 71.8; P < 0.001). The pooled mean score of the 7 PSQI subscales were as follows: subjective sleep quality: 1.22 (95% CI = 1.04–1.41), sleep latency: 0.99 (95% CI = 0.88–1.11), sleep duration: 1.05 (95% CI = 0.92–1.18), sleep efficiency: 0.27 (95% CI = 0.19–0.34), sleep disturbance: 1.17 (95% CI = 1.01–1.33), use of sleep medications: 0.33 (95% CI = 0.23–0.43), and daytime function: 1.32 (95% CI = 1.11–1.53) (Table S1).

Sleep habits

The data of sleep duration and sleep habits are shown in Table S2. The proportion of medical students who slept less than 7 h/day was 58.7% (95% CI = 45.3%–72.0%), while the proportion of more than and equal to 7 h/day was 41.3% (95% CI = 28.0%–54.7%).

The pooled bedtime across 6 studies with 1,332 medical students was 0:23 am (95% CI: 11:13 pm–1:33 am). The pooled mean sleep latency across 13 studies with 2,930 medical students was 21.53 min (95% CI: 18.65–24.41). The mean sleep duration across 22 studies with 4,851 medical students was 6.45 h (95% CI: 6.03–6.87) and time to get up across 5 studies with 1,393 medical students was 7:13 am (95% CI: 5:46 am–8:41 am).

Discussion

To our best knowledge, this was the first comprehensive meta-analysis of studies worldwide on the pooled prevalence of poor sleep quality in medical students using the PSQI. The main finding was that the majority of medical students had self-reported poor sleep quality (52.7%, 95% CI: 45.3%–60.1%).

In this meta-analysis, the prevalence of poor sleep quality in medical students (52.7%, 95% CI: 45.3%–60.1%) is significantly higher than the corresponding figures (23.9%, 95% CI: 20.8%–27.4%; by the PSQI) in university students [35] and in older population (38.3%; 95%CI = 32.4%–44.2%; by the PSQI) [36]. This is probably related to the high academic pressure in medical schools [4] and short sleep duration among medical students necessary to meet such academic demands [37]. Additionally, certain psychological factors, such as anxiety and depressive symptoms, and even suicidality, are relatively common in medical students [38], which is associated with higher risk of sleep problems [39,40,41]. On the other hand, there is an assumption that medical students may have more medical knowledge than the general population [42], and therefore may be prone to over-reporting symptoms in surveys, which could increase the prevalence of self-reported poor sleep quality.

This study found that smaller sample size was associated with higher prevalence of poor sleep quality. Due to limited statistical power, small sample size may bias the findings to an uncertain extent [43]. As expected, lower PSQI cutoff values were associated with higher prevalence of poor sleep quality, which is consistent with previous findings [35]. In addition, the use of different PSQI cutoff values may be a major source of heterogeneity between studies. Study region was significantly associated with the prevalence of poor sleep quality in medical students. The prevalence of poor sleep quality was highest in Europe (65.13%), followed by the Americas (59.92%), Africa (54.54%), Asia (47.44%), and Oceania (30.51%). Most of the high-ranking medical schools globally are located in Europe and the Americas [44]; therefore, medical students in these regions are more likely to have rigorous academic requirements and pressure compared to those in other areas, which is associated with higher risk of poor sleep quality [45]. Moreover, medical students in Western countries may also have high self-expectation to perform [46], and may therefore have a higher likelihood of poor sleep quality.

The strengths of this meta-analysis includ the large number of studies and the large pooled sample size. However, several limitations need to be addressed. First, similar to other meta-analysis [47, 48], substantial heterogeneity was inevitable in meta-analysis of epidemiological studies, although subgroup analyses alleviated this limitation to some extent. Second, some factors related to sleep quality, such as academic achievement and pressure and family support, were not recorded in most studies. Third, only studies using the PSQI were included. However, the PSQI is considered the most widely used tool to measure poor sleep quality, and in order to minimize bias caused by different measures, other instruments on sleep, such as the Epworth Sleepiness Scale (ESS) or Insomnia Severity Index (ISI), were not included.

In conclusion, poor sleep quality is common in medical students globally, particularly in Europe and the Americas. To reduce the negative health outcomes of poor sleep quality, education on the impact of poor sleep, regular monitoring of sleep and practicing sleep hygiene should be promoted in medical students. Finally, longitudinal research on the association between poor sleep quality and other demographic and clinical variables in medical students should be conducted in the future.

Table 2 Meta-regression and subgroup analyses of prevalence of poor sleep quality
Full size table