Our systematic review was a priori registered on PROSPERO (CRD42022384066). During the whole research process, we considered the preferred reporting items for systematic reviews and meta-analyses (PRISMA) [30].

The systematic literature search included three core concepts, namely, “depressive symptoms,” “higher education students,” and “Germany.” Within each concept, we used different free text terms with truncations and controlled vocabulary terms with automatic explosion, which were joined with ‘OR’. The concepts were linked using ‘AND’. Both German and English search terms were included (Table 1). We restricted our search to “Title and Abstract” in order to be more precise and to avoid obtaining results where a higher education institution is indicated in the authors’ affiliations. Moreover, filters on English or German language, peer reviewed journal, and publication date within the years 2002–2023 were applied as we decided to focus on the period after the implementation of the Bologna reform in Germany to make the results more comparable.

We searched the databases MEDLINE and Embase through Ovid and PsycINFO through EbscoHOST, respectively for eligible publications on July 15, 2022. We set an alert, which continued until data extraction was completed on June 30, 2023. Additional searches were conducted by one author (EH) via Elicit [31] on July 26, 2023 and by reviewing the references of the identified articles and the citing literature via Google Scholar. The search in Elicit was based on the underlying research question of the review.

We included articles which met the following criteria: 1) the population studied are students from universities or universities of applied sciences, 2) data was assessed at higher education institutions in Germany, and 3) information on depressive symptoms as outcome is provided, including diagnosis of affective disorders (International Statistical Classification of Diseases and Related Health Problems (ICD) codes F30-F39) or prevalence rates indicated by cut-off values of instruments for assessing depressive symptoms. In addition, we considered only 4) original articles published in peer-reviewed journals in 5) German or English language 6) between 2002 and 2023. Our inclusion criteria were modified during the screening process because the data appeared too comprehensive. Therefore, we decided to include only prevalence rates and not to include studies that reported mean values.

Qualitative studies, opinion papers and commentaries, review articles, and studies published only as abstracts were excluded. In addition, studies that examined depressiveness as symptoms of other conditions than affective disorders (e.g., postnatal depression) or in other population groups (e.g., high-school students) were also excluded.

Rayyan [32], an online tool for conducting collaborative reviews, was used to maintain the search results. Duplicates were automatically removed by the search portals and can therefore not be reported. Two reviewers (EH and KH) independently screened all results of the literature search and excluded irrelevant articles based on titles and abstracts. Cohen’s kappa was calculated for this selection step to account for interrater agreement. Full text copies were obtained and independently assessed by two reviewers (EH and KH). Publications were excluded only if the two reviewers agreed, and any disagreements were resolved by discussion until consensus was reached. Reasons for exclusion were documented during full-text screening.

One reviewer (EH) extracted data from all included publications using a standardized excel spreadsheet. The following information was extracted and synthesized: bibliographical data (first author’s name, year of publication), study design, period of data assessment, details on study population (sample size, higher education institution, age and gender distribution, study field(s)), instruments used to assess depressive symptoms, prevalence of depressive symptoms in the whole sample and additional prevalence rates (e.g., for subgroups)—if applicable.

The appraisal of study quality was conducted by two authors, EH and KH, ensuring a thorough and objective assessment process during which each person checked the other person’s appraisal. By employing the Joanne Briggs Institute (JBI) Critical Appraisal Tool for Studies Reporting Prevalence Data [33], we ensured a rigorous and standardized assessment of the quality of prevalence studies included in our review. This tool allowed us to systematically assess various aspects of the studies, enhancing the credibility and reliability of our findings. The JBI Critical Appraisal Tool is commonly utilized for evaluating the quality of prevalence studies [34, 35], and its use has been recently recommended following a systematic review of quality assessment tools [36]. This checklist comprises 9 items, which assess various aspects of the study, including the representativeness of the sample, appropriateness of recruitment methods, adequacy of sample size, detailed description of subjects and setting, valid ascertainment and measurement of the condition, thorough reporting of statistical analysis, use of standard measurements for all participants, and adequacy of the response rate. Although the tool does not provide a numerical score, we assigned proportions to rate the quality of the included articles. However, poor-quality articles with a high risk of bias were not excluded from data synthesis as this could lead to selection bias [37] and their contributions will be thoroughly discussed within the context of the overall discussion.

For descriptive data synthesis, a comprehensive table was generated, presenting the sociodemographic characteristics of each study along with their respective prevalence rates of depressive symptoms among higher education students. This tabular format provides an overview over the individual study findings and their respective sample compositions.

Additionally, meta-analyses were undertaken using SPSS Version 29.0 to synthesize the data. Given the variability in prevalence and variance within studies and as recommended by Borges Migliavaca et al. [38], a random effects model was employed. The range of effects was evaluated through visual examination of the forest plot depicting estimates and 95% confidence intervals (CIs), along with consideration of the weight assigned to each point estimate. For studies in which more than one subgroup prevalence rate was reported for the entire sample, a weighted mean was calculated to avoid distorting the p-value of the meta-analysis by including multiple data points per study.

Subgroup analyses were carried out to explore heterogeneity among studies. The subgroups were defined based on sex (male/female), field of study (medicine and dentistry), first-year students only, and time-point of data collection (before/during the COVID-19 pandemic). Subgroup analyses were also conducted for the two most frequently used screening tools, the Patient Health Questionnaire (PHQ) and the Beck Depression Inventory (BDI). Differences between subgroups were assessed based on 95% CIs. Moreover, the I² statistic was utilized to assess the extent of study heterogeneity. Values above 25%, 50%, and 75% were considered to represent low, moderate, and high levels of heterogeneity, respectively [39]. A funnel plot was additionally created to visualize the potential heterogeneity and publication bias.

In case of sample overlap across multiple articles, for example, by utilizing the same dataset, only the one with the largest sample size was incorporated into the meta-analysis. This was done to mitigate potential bias. If studies used multiple assessment instruments, we included results from the instrument validated and established in student mental health research and if more than one prevalence rate was reported (e.g., for sex-specific subgroups only), we calculated a weighted mean for consideration in the meta-analysis.

In case of longitudinal studies with multiple time points, our meta-analysis focused on using the baseline prevalence data, as it was done in previous student mental health research, e.g., by Ahmed et al. [40]. This decision was primarily based on two considerations. Firstly, the baseline data frequently represents the largest sample size among all data collection time points, ensuring a more robust estimate of prevalence. Secondly, using baseline data helps to minimize potential confounding factors, such as demand characteristics or selection bias, which might arise at subsequent data collection time-points due to biased study participation and external factors occurring over time [40]. In addition, the inclusion of data from interventional studies for mental health would also result in selection bias.

Figure 1 shows the literature selection process. The initial search resulted in 911 records, and the ongoing alert yielded a further 81 records. We retrieved another 16 records through cross-checking the references and citing literature as well as through an additional search via Elicit (5 articles). After removing duplicates, 622 articles remained for title and abstract screening, of which 38 were inconsistently judged by the two reviewers (EH and KH). The resulting Cohen’s kappa was κ = 0.9, indicating a level of almost perfect agreement (93.9%). Of the 622 articles, 404 were excluded because they did not meet the eligibility criteria, leaving 218 articles for full text analysis. After full text screening, 60 articles remained for data extraction, and 56 could be used for meta-analysis. A total of four articles were excluded from the overall meta-analysis due to sample overlap as mentioned above. These articles comprised Berger et al. [41], Burger et al. [42], Mikolajczyk et al. [43], and Weber et al. [7].

Table 2 provides a comprehensive overview of the study characteristics found within the included articles. Most studies employed a cross-sectional design (n = 51). The sample sizes varied, ranging from n = 50 [45] to n = 7,203 participants [46], with a total of 75,107 participants and a mean sample size of 1,252 persons per study. The articles were published between 2008 and 2023, and data were gathered from 2004 to 2022.

Among the studies, 29 focused on medical and/or dental students, while 15 publications reported data on sex-specific prevalence rates. Notably, 15 studies were conducted during the pandemic, and 13 studies specifically provided information regarding first-year students. A diverse range of 14 different assessment instruments were utilized to measure depressive symptoms. Most of the studies employed versions of the PHQ (PHQ-2/PHQ-8/PHQ-9) in 30 instances, followed by the BDI (BDI-I/BDI-II/M-BDI) in 16 cases and the Hospital Anxiety and Depression Scale - Depression subscale (HADS-D) in 4 instances. Different cut-off levels were applied across these assessment instruments. Studies utilizing the PHQ primarily reported at least moderate levels of depressive symptoms, while those employing the BDI mostly indicated at least mild depressive symptoms.

The prevalence rates of depressive symptoms exhibited considerable variability across the studies. In general, the rates ranged from 2.2% (weighted mean prevalence rate for the overall sample [47]) to 59.1% [22]. For female students, the rates ranged from 5.9% [28] to 59.8% [23], while for male students, the rates ranged from 6.2% [28] to 49.2% [23].

All 60 articles were assessed based on the nine items of JBI Critical Appraisal Tool for Studies Reporting Prevalence Data [34]. The score of each included publication can be found in Table 2. The assessment of these publications resulted in cumulative scores that ranged from 3 to 9, with a mean score of 6.3, out of a possible 9 points. The majority of studies employed suitable sampling methods (e.g., data assessment in compulsory lectures) (65.0%), provided sufficient details regarding the characteristics of their study subjects and settings (e.g., information on age and sex of the participants) (66.7%), and maintained adequate sample sizes that ranged from 306 [9] to 7,203 participants [46] (68.3%). Of 60 studies, 96.7% used an appropriate sample frame to address the target population, which substantially mitigates the risk of overgeneralizing the findings: a minimum sample size of 305 was determined based on the calculation for prevalence studies [94, 95]. This calculation considered an assumed prevalence rate of 20%, drawing upon findings from prior systematic reviews on depressive symptoms in comparable population groups [4–6].

The risk of bias for outcome measurement was consistently low due to the overall use of assessment instruments with clearly defined cut-offs for depressive symptoms (98.3%). A total of 68.3% of the studies utilized a consistent measurement approach for all participants. However, we were unable to confirm this for 19 studies due to missing information, such as whether assessments were conducted with pen and paper and whether a trained instructor was present.

Nearly half of the studies (n = 27) reported the prevalence rate in terms of percentages without specifying proportions, necessitating the classification of such studies as “unclear” in this regard. The response rate for 25 studies was noted to be below 10% or not appropriately managed, or in some instances, simply not reported.

In our analysis, we found a pooled prevalence rate of 21.1% (95% CI: 17.6%–24.6%) for depressive symptoms among the higher education student population (Figure 2). The funnel plot showed no publication bias but reflected the large heterogeneity of results (Figure 3), and so did the I² statistic (99%). When analyzing studies using the same assessment tool only, the pooled prevalence rates for moderate depressive symptoms based on the PHQ scale and at least mild depressive symptoms using the BDI scale were 24.8% (95% CI: 19.5%–30.1%) and 22.4% (95% CI: 17.4%–27.3%), respectively.

When examining subgroups (Supplementary File S1), we observed that the pooled prevalence rate for female students was 29.0% (95% CI: 21.4%–36.5%), while it was lower for male students at 23.1% (95% CI: 16.5%–29.6%). Furthermore, when considering the subgroups of higher education students, medicine or dental health education students exhibited a pooled prevalence rate of 13.2% (95% CI: 9.9%–16.4%). Among first-year students, the pooled prevalence rate was 11.0% (95% CI: 7.7%–14.3%), indicating a relatively lower prevalence in this subgroup. Our analysis revealed variations in the prevalence of depressive symptoms before and during the COVID-19 pandemic. Prior to the pandemic, the pooled prevalence rate was 18.0% (95% CI: 14.7%–21.2%), but during the pandemic, it was considerably higher than before (30.6%, 95% CI: 22.1%–39.1%).

Stratified by time-point of survey (Supplementary File S2, S3), around one-quarter of female higher education students and a fifth of their male counterparts showed depressive symptoms before the pandemic (25.7%, 95% CI: 18.8%–32.5% and 20.1%, 95% CI: 14.2%–26.1%, respectively). When measured with different versions of the PHQ, the pooled prevalence rate before the pandemic was 18.6% (95% CI: 13.5%–23.8%) and during the pandemic it was 33.7% (95% CI: 25.1%–42.2%). For the subgroup of medical and dental students, the time-point of the survey made no difference regarding the extent of depressive symptoms before the pandemic: 13.2%, 95% CI: 9.4%–17.0%, during the pandemic: 13.3% (95% CI: 7.6%–19.0%). During the pandemic, the BDI was not employed for data collection, only two studies examined sex-specific prevalence rates [43, 103], and only one specifically focused on first-year students [79], thus no stratification was conducted.

This systematic review and meta-analysis provide a valuable estimation of the prevalence of depressive symptoms among higher education students in Germany. The study exhibits several strengths, including the comprehensive approach of searching three databases, the use of alerts for relevant publications until data extraction was completed, and a well-structured and exhaustive search strategy. Furthermore, the study’s adherence to PRISMA guidelines for systematic reviews, coupled with the quality appraisal conducted by two researchers, significantly bolsters the robustness of its findings.

However, it is essential to acknowledge the inherent limitations associated with this research. Utilizing self-report instruments is a valid method for assessing depressive symptoms. These instruments, despite targeting the same latent variable, exhibit variations in their conceptualization, including the application of distinct cut-off levels. This discrepancy served as the rationale for our subgroup analysis on the two out of 14 most frequently employed instruments. Interestingly, we found similar prevalence rates for the PHQ and BDI, even though one assesses moderate and the other mild depressive symptoms. However, our study has notable limitations due to strong over-representation of research with focus on students from medical subjects, the heterogeneity observed among the results, and the variability in measurement tools among the included studies. While subgroup analyses were conducted to explore potential sources of heterogeneity, a more in-depth assessment through meta-regression was not performed, leaving the underlying causes of this heterogeneity not fully elucidated. Accordingly, it is not possible, for example, to determine why there were differences in prevalence rates before and during the COVID-19 pandemic. In addition, these differences need to be interpreted with caution, because they were based on different studies with different sample characteristics and procedures.

Regarding the quality assessment of the articles, we employed a standardized tool to reduce potential biases. Most of the studies featured adequate sample sizes, sampling methods, and measurement of assessing depressive symptoms. Studies with a poor quality (<5 out of 9 possible points) accounted for a relatively small proportion (4,495/75,107 participants = 5.98%) of the overall sample, so that an overall low risk of bias can be assumed. However, there was variability in the quality of reporting in prevalence studies. In this regard, we can fully support the recommendation for the incorporation of checklists like STROBE [107] to ensure comprehensive and consistent reporting practices also in the context of future epidemiological research in Germany.