Background

Globally, university students could be considered a privileged group given the significant variation in percentage of national populations with a university education [1]. However, for those who do attend university usually do so at a developmentally high risk period for the emergence of mental heath problems [2, 3]. Psychological distress, encompassing symptoms ranging from normal fluctuations in mood to the emergence of a serious mental illness, is an increasingly common experience among university students which can have significant consequences for individuals [4, 5]. Recent international evidence suggests 35% of first year students report symptoms indicative of lifetime mental disorder, and 31.4% report symptoms in the previous 12 months [6]. International longitudinal research is more limited. Studies in Norway, the UK and the USA has shown both psychological distress and common mental disorders (CMD) have increased in prevalence among both students and similar aged non-student populations over the last 10 years [7,8,9,10,11]. Suicidal behaviour, while lower in students compared to matched non-student populations, has also increased over a similar timeframe in England and Wales [12]. International estimates among students suggest around 4.3% have attempted suicide in their lifetime [6]. The short- and longer-term consequences of mental health difficulties can be significant including poorer academic performance, relationship breakdown, and exclusion from the labour market [6, 13, 14]. Current students face greater financial and academic pressures compared to 20 years ago, which may be contributing to poorer mental health outcomes [2, 15,16,17]. These findings suggest a significant mental health need among this population. [1].

For students in mental distress, the support available to them is likely to vary signficiantly between and within countries. For example, in many high-income countries (HIC) students may have a range of effective mental health services available to them but these services are often fragmented, uncoordinated and underutilised [6, 19, 20]. For example, US studies suggest around a 1/3 of students received treatment [9], while epidemiological studies suggest this varies widely independent of need based on sex and gender, ethnicity, age, and where they attend university [6, 20,21,22,23]. Barriers such as self-stigma, perceived need, and self-reliance influence when and how they seek help, while student’s also report a lack of awareness of appropriate services, concerns about confidentiality and discrimination, cost, or may perceive services to be ineffective or inappropriate [19, 24, 25]. These barriers may explain why some students only seek help in crisis and others tend to rely on informal sources of support [26, 27]. International studies suggest very few students with need, receive support globally. One recent international cross-sectional study found 19.8% of first year university students, and 36% of those who may meet criteria for CMD report having ever used a mental health service, defined as medication or psychological counselling [6]. Compared to HICs, much less is known about students in Lower and Middle Income Countries (LMIC), although individual studies suggest very small numbers of students report accessing support when in distress [18, 28].

While a limited number of studies have highlighted the scale and nature of the problem outside of the USA, there is a renewed effort to understand and address barriers to treatment that stop some students reaching help in the first place [4, 16, 27]. The World Health Organization’s (WHO) World Mental Health International College Student Initiative (WMH-ICS) aims to provide greater clarity on the unmet need of this group [16]. In the UK, there has been a policy focus on improving access to mental health interventions through greater integration between the National Health Service (NHS) and Universities, and an emphasis on mobilising university resources towards the mental health of students [29, 30]. Previous reviews in the USA have looked at which students are most likely to seek help [20, 31], however this is obviously confounded by the nature of services available to them. There are no systematic reviews conducted on the variety of services available to students internationally, how these integrate with each other and how use varies by types of service that deliver interventions to support mental health and wellbeing. Studies have examined individual services such as university counselling centres, external psychological services, or inpatient settings but have not compared the differential use of these by students with different clinical presentations. Given the developmental period in which many students attend university these settings are important in contributing to improving overall population mental health [3, 32]. By understanding where variation occurs could indicate areas of differential access, highlighting where care pathways could be improved and inform policy initiatives.

This systematic review was conducted to address this gap, by answering two review questions: (1) what proportion of university students use mental health services when experiencing psychological distress? And (2) does utilisation differ across health service type?

Method

This review was reported in accordance with PRISMA guidelines [33] (see Additional file 1: Appendix S1). A protocol for this review was pre-registered on the 22/02/21 on PROSPERO (https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD42021238273).

Deviations from initial protocol

On the 26th of April 2021 we made an amendment to only include studies published in the year 2000 or after over concerns around changes to the student population that would create issues of comparability [4]. On the 27th of July 2021 we amended the focus of the review as the original aims were considered too broad for a coherent synthesis. The amendment removed one review question related to student characteristics associated with service use which could be explored in future analysis.

Eligibility criteria

Studies were included that:

  1. 1)

    Measured the use or utilisation of mental health services (as a primary or secondary outcome).

  2. 2)

    Studies that included adults (aged 18 +) studying at a university.


Studies were excluded:

  1. 1)

    That employed an empirical study design that aimed to test an intervention or approach to address or effect access or use of healthcare services.

  2. 2)

    Where it was not possible to extract sociodemographic and utilisation data for student participants.

  3. 3)

    Where participants under 18 were recruited.

  4. 4)

    Where participants weren’t all university students.

Studies needed to be published in English due to the languages spoken by the primary reviewer (TO).

Search strategy

The following electronic databases were searched on the 9th of March 2021, 3rd of November 2021 and the 23rd of August 2022: MEDLINE (Ovid); EMBASE (Ovid); PsycINFO (Ovid); ERIC (ESBCO); and CINAHL plus (ESBCO). The search strategy using a Context, Condition, Population (CoCoPop) framework with the concepts of “students”, “mental health/illness”, “access” and “mental health services” [34]. Key words and MeSH terms were developed in Medline between 2nd of December 2020 and 9th of March 2021, and adapted for each database (see Additional file 1: Appendix S2). On the 16th and 17th of June 2021, the 14th of December 2021 and the 16th of November 2022 forward and backward citation searching was conducted. The publicly available reference list of studies published by the WHO’s WMH-ICS was searched on the 23rd of April 2021, the 14th of December 2021 and the 16th of November 2022. The authors of the originally included studies were contacted on the 18th of June 2021, where possible, to help identify any unpublished or ongoing research.

Data extraction

Records retrieved from electronic database searches were exported to Endnote X9, where duplicates were removed. Abstracts and full texts of potentially relevant articles were screened against the inclusion and exclusion criteria on Rayyan software. A random sample of approximately 10% of titles and abstracts identified in the initial searches were screened independently by a second reviewer (SL) using a purpose designed screening tool (see Additional file 1: Appendix S3). Data from the included studies were extracted independently by two reviewers (TO and SL) using a pre-defined data extraction framework (see Additional file 1: Appendix S4). Data were extracted into Excel. After data were extracted for two studies, the data extraction framework was checked for interpretation by both TO and SL. Study authors were contacted where additional data or clarification was required. The main items of interest were:

i Condition: use or utilisation

We defined use as the occurrence or number of uses of a mental health service over a defined time-period [35]. Indicators could include attendances, usage, inpatient days, admissions, contacts, episodes, or costs due to the receipt of treatment or attendance [35]. These indicators may be measured through self-report, clinical records, and/ or other routinely collected data. As observational or more naturalistic study designs were included in this review, outcomes are likely to be reported as prevalence or incidence and therefore as a proportion of the total study sample. Therefore, the effect measures were proportions with a 95% confidence interval as the main outcome [34].

ii Context: mental health service

An amended version of the WHO’s definition of a mental health service was used, this being ‘the means by which effective interventions are delivered for the dominant or subdominant intention to improve wellbeing or mental health’ [36]. This included outpatient services, day treatment, inpatient wards, community mental health teams, General Practice, mental health hospitals, and university counselling services [36]. To facilitate comparison of proportions by service type an adapted version of the Description and Evaluation of Services for Disabilities in Europe (DESDE) instrument was used (see Appendix S5) [37]. This is a hierarchical classification system, with six initial categories: (1) Information for care, (2) Accessibility to care, (3) Self-help and volunteer care, (4) Outpatient Care, (5) Day care, and (6) Residential care. A random 10% sample were double coded by two reviews (TO and SL). No service descriptions could be classified beyond the first level of the DESDE hierarchy. Therefore, to further specify, we used the National Institute for Health and Care Excellence (NICE) treatment stepped care categories, referred to as ‘treatment type’ [38], and the service location—being either on campus, off campus, or potentially either.

iii Other items

We also collected sociodemographic characteristics, study design, duration of study, data collection methods, data analysis methods, setting and date of study, raw data for the outcome, indicator(s) used, and time point(s) outcomes where reported, source of funding and conflicts of interest.

Quality assessment

We assessed risk of bias using the Joanna Briggs Institute (JBI) appraisal checklist for systematic review reporting prevalence data [34]. The checklist prompts the reviewer to answer nine questions with four possible response options: “yes”/ “no”/ “unclear”/ “not applicable”. Each study was assigned low, moderate, or high quality based on the number of yes answers it scored to indicate study quality. Studies with 1–3 ‘yes’ were low, 3–6 indicating moderate, and 7–9 as high quality. Quality appraisal was conducted independently on all studies meeting the inclusion criteria by two reviewers (TO and SL). Where there were disagreements, these were discussed until agreement was reached. No studies were excluded based on the study quality to enable sensitivity analyses to be conducted by removing studies rated as low quality.

Synthesis methods

i Narrative synthesis

Initially, a non-statistical narrative synthesis was conducted to describe the included studies relevant to the review questions [34]. Study participants and the measures of psychological symptoms were not universally well described. Therefore, the samples were qualitatively summarised and then categorised based on whether this was a general student sample, subgroup sample or a sample of students with more severe current psychological distress, referred to as ‘at risk’.

ii Meta-analysis

Most studies provided data for multiple service types, therefore three-level mixed effects models were used to account for clustering. Where the study provided a single estimate or an overall estimate of service use they were included in one of three conventional random effects meta-analytic models: (1) overall service use (any service), (2) overall outpatient service use, (3) overall residential service use reflecting the service types commonly observed in the data. Following this, to specifically test differences between these service types all estimates were then included into a three-level mixed effects model, where sub-group analysis and meta-regression were also conducted [39]. Further analyses were conducted for studies providing multiple estimates within the same study using two three-level mixed effects models to account for clustering: (1) outpatient service use; (2) service use where the service could be classed within multiple DESDE service categories.

For all pooled proportions, a priori subgroup analysis and meta-regression were conducted based on population group. Post-hoc analyses were conducted based on service location, treatment type, reporting timeframes, publication year, study design, and country, due to the substantial estimated heterogeneity. To conduct meta-regression for recall time-period a continuous variable was created based on the number of months participants were asked to recall service use (e.g., 12 months). If the reporting time-period did not use months (e.g., the student’s lifetime), it was estimated using the average age of the participants.

Heterogeneity was further explored by identifying outliers above or below the 95% confidence interval of the pooled proportion; by conducting influencer analysis; drafting a Baujat plot and conducting Graphic Display of Heterogeneity (GOSH) plots [39].

Sensitivity analyses were conducted for pooled estimates where low quality studies, estimates of lifetime service use and outliers and influential cases were excluded then all described analyses were repeated. Publication bias was not assessed due to the substantial between study heterogeneity [39].

Results

Search results

A total of 7739 unique titles / abstracts were identified through database searches, and a further 52 through other search strategies (see Fig. 1 and Additional file 1: Appendix S6). Inter-rater agreement for data screening was Cohen’s Kappa (K) = 0.85 indicating strong agreement [40].

Fig. 1
figure 1

PRISMA flow diagram

Full size image

As a result of these search strategies, 44 studies were deemed eligible for inclusion. Within these studies there were 123 estimates of service use. Seven of these studies were smaller analyses of larger surveys conducted in the USA [23, 41,42,43,44,45,46]. These seven studies were excluded from meta-analysis as their estimates would double count participants. 29 studies and 42 estimates were included in conventional two-level meta-analyses pooling estimates of overall service use, and then a three-level meta-analysis to test differences by service type. 25 studies and 60 estimates were included in further analyses using three-level meta-analysis. Inter-rater agreement for data extraction was K = 0.82 indicating strong agreement [40].

Study characteristics

i Study origin

Studies were conducted in a range of mostly high-income countries. The majority were from the United States, where 34 of the 44 studies were based [9, 23, 41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72]. The remainder from Australia [73, 74], Brazil [75, 76], China [77], Canada [78], Ethiopia [79], Bangladesh [28], and Italy [80]. A total of nineteen studies were samples of students from separate individual universities [43, 46, 48,49,50,51,52,53,54,55, 67, 68, 70, 73, 75,76,77, 79, 80]. Whereas the remaining twenty-four were samples across multiple universities [9, 20, 23, 28, 41, 44, 45, 47, 56,57,58,59, 61,62,63,64,65,66, 69, 71, 72, 74, 78].

ii Study design and methods

Most studies (n = 36) were either primary or secondary analyses of cross-sectional surveys [9, 20, 23, 41, 43,44,45, 47, 49,50,51, 53,54,55,56, 58, 61,62,63,64,65,66,67,68,69, 73,74,75, 78, 79] (see Table 1). Outcomes were assessed using standardised questionnaires and open questions. Of the remaining seven studies, one was a longitudinal study [46], one was a cohort study using a mix of a baseline survey and linked electronic medical records from the university counselling centre [77], two were secondary data analyses of electronic medical records from university counselling or health centres [52, 59, 60], and two were mixed method studies [48, 80].

Table 1 Study characteristics
Full size table

iii Study participants

Sample sizes varied substantially ranging from 15 to 730,785 participants. Most studies included general samples of student attending a university with fifteen studies studying specific subgroups of students [41, 44, 51, 52, 58, 59, 61, 63, 65, 69,70,71, 73,74,75,76]. Thirteen studies included samples of students ‘at risk’ [23, 48,49,50, 56, 57, 62, 64, 66, 68, 72, 79, 80]. Two studies sampled university faculty members, in addition to university students, although these participants were not asked about mental health service use [41, 47]. One study included students at community college and 4-year institutions in the USA [23].

iv Mental health services

Overall, most estimates were associated with services classified into the outpatient service category of the DESDE instrument (see Table 2). Seventy-four estimates associated with thirty-seven studies were outpatient services [9, 20, 28, 41, 43,44,45,46,47,48,49,50,51,52, 54, 55, 57, 59, 61,62,63,64,65,66,67, 70,71,72,73, 75,76,77,78,79,80]. Thirty-seven estimates associated with twenty-two studies could be classed as multiple service categories [9, 20, 23, 41, 47, 50, 53, 56, 57, 61,62,63,64,65,66, 68,69,70,71, 74, 78]. Residential service category was appropriate for seven estimates associated with five studies [9, 57, 61, 66, 70]. Inter-rater agreement for service coding was Κ = 0.89, indicating strong agreement [40].

Table 2 Service categories and mental health service use
Full size table

Across the service categories, 38 estimates related to services providing a range of treatments, 1 providing advice and support, 25 providing low intensity treatment, 35 related to high intensity treatment and 17 related to specialist treatment. Of these estimates thirteen related to services located off campus; 29 were on campus, whereas the remaining 79 estimates could have been located on or off a university campus.

v Defining and measuring use of health services

While all studies implicitly conceptualised mental health service use as an event or occurrence by a person in a time-period, the operational assessment was heterogeneous. In the cross-sectional and longitudinal studies, measurement varied by recall period and by item wording [9, 20, 23, 28, 41, 43,44,45, 47, 49,50,51, 53,54,55,56, 58, 61,62,63,64,65,66,67,68,69,70,71,72,73,74,75, 78, 79]. Only one study used a validated instrument assessing use over the previous two weeks [79], one asked student about their use over the previous two months [49], sixteen over the last 12 months [9, 23, 28, 42,43,44,45,46, 50, 56,57,58, 67, 70, 72, 74], four while students were at university [41, 47, 68, 71], and ten asked participants to report about previous use in their lifetime or ever [55, 61,62,63,64,65,66, 69, 78]. One cross-sectional study asked student participants to both recall use of university counselling centre while at university, and the students use of other mental health service over their lifetime [66]. Nearly all cross-sectional studies gave participants a binary response option—either yes or no. Only one study used an ordered categorical response option where participants were asked to state whether they had used a particular service using a Likert scale ranging from 1–5 (never-often) [50]. Of the two mixed methods studies one reported current use [48], and the other reported on lifetime use [80]. Secondary analyses of electronic medical records examined number of unique visits per student over the study period [52, 59, 60].

Quality appraisal

Overall, the quality of the studies included in the review were moderate with around a quarter of the total samples rated as either high [43,44,45,46, 56, 67, 72, 79], or low quality [49, 52, 54, 61, 65, 69, 76]. The main area of weakness came from questions related to the validity and reliability of the assessment of mental health service use, with only six studies being rated as “yes” in both questions [45, 46, 56, 67, 74, 79]. A further area of significant weakness was found in question eight which related to whether appropriate statistical analyses had been conducted with four studies rated as “yes” [49, 53, 59, 63] (see Table 1 and Additional file 1: Appendix S7). Inter-rater agreement for quality appraisal was Κ = 0.88 indicating strong agreement [40].

What proportion of university students use mental health services when experiencing psychological distress?

i. Overall use of any mental health service

Narrative summary (n = 10; k = 11)

Ten studies reporting on students’ use of any mental health service use with estimates ranging between 13.7 and 68.6% of the study population reporting use [9, 41, 47, 50, 53, 57, 61, 64, 70, 71, 74, 78]. Estimates ranged from 13.7 to 68.6% of the study population reporting using a service. It was difficult conclude the source of this variation. The highest estimate, at 68.6%, was the only for an on-campus service. Treatment offered by the service did not appear to be associated with variation across estimates. Broader operational service definitions tended to have higher estimates [53, 74]. For example, in one study 49% of Chinese international students reported using “any form of help”, whereas all other estimates within the same study relating to specific services were low.

There was some evidence to suggest more severe current psychological distress was associated with higher previous mental health service use. For example, in studies with at risk samples reported estimates between 25.7 and 49% [50, 57, 74]. Whereas estimates in general populations of students had a lower range between 19.7 and 45% [9, 47, 53, 78]. Variation also appeared to be related to the reporting period, where studies reporting on lifetime mental health service use tended to have higher estimates [61, 78] (see Tables 1 and 2).

Meta-analysis (n = 9; k = 9)

The overall pooled proportion effect size using a random effects model was estimated to be 0.35 (95%CI: 0.22;0.50) (see Fig. 2). The between study heterogeneity was estimated at τ2 = 0.69, and Ι 2 = 99.9%. The prediction interval ranged from 0.06 to 0.81. This indicated a wide range of future possible estimates. Overall, these results indicate substantial heterogeneity across the included estimates of mental health service use.

Fig. 2
figure 2

Forest plot for overall mental health service use by population group

Full size image
Subgroups and meta-regressions for overall use

No variables were associated with an overall reduction in between study heterogeneity using meta-regressions. Subgroup analyses found differences by service location (Q = 40.41, df:2, p < 0.001), and reporting period (Q = 5.92, df:2, p = 0.05), However, meta-regressions found lower proportions were associated with off-campus service (β = − 1.35, 95%CI:− 2.52; − 0.18, p = 0.03), and higher proportions associated with longer reporting periods (β = 0.0043, 95%CI:− 0.001; 0.0075, p = 0.02) (see Additional file 1: Appendix S8).

ii Overall outpatient use

Narrative summary (n = 25; k = 27)

Twenty-five studies reported estimates of students overall outpatient service use with between 2.6 and 75% of the study populations reporting service use [9, 28, 41, 43,44,45,46,47,48,49,50,51,52, 54, 57, 59, 61,62,63, 66, 67, 69,70,71,72,73, 75,76,77, 80]. Use of on-campus services were lower ranging between 2.6 and 33.5% [9, 41, 47, 50,51,52, 58,59,60, 66, 69, 73, 77]. There was only one estimate of off-campus service use at 13.7% [49], whereas the remaining estimates were for services that could be either on or off campus between 7 and 75%. These differences could also be partly explained by differences in population group and treatment offered by the service. The lowest two estimates overall were in subgroups of students namely international students (2.6%) [52], and students in China (5.1%) [77], and among students Bangladeshi universities (7.1%) [28]. Whereas the highest estimates overall and in the category of either on campus or off campus services were in a study of medical students with more severe current psychological distress using services offering potentially any treatment (75%) [73]; previously homeless students or who had been in care where a broad service model had been developed for them (68%) [48], and veterinary students (62.5%) [61]. For this estimate participants reported against the use of “counselling”—which could have a broad interpretation in the USA. A further study also using a broad outpatient service definition was associated with a high estimate of 68% [49]. Overall, studies asking students to recall service use over their lifetime reported a higher range of estimates [61,62,63, 69, 80], compared to studies with shorter recall periods (see Tables 1 and 2).

Meta-analysis for overall outpatient use (n = 24; k = 26)

The overall pooled proportion effect size using a random effects model was estimated to be 0.21 (95%CI = 0.15;0.30) (see Fig. 3). The between study heterogeneity was estimated at τ2 = 1.12 and Ι 2 = 99.9%. The prediction interval ranged from 0.03 to 0.72. This indicated a wide range of future possible estimates. Overall, these results indicate substantial heterogeneity across the included estimates of residential mental health service use.

Fig. 3
figure 3

Forest Plot for outpatient overall service use by population group

Full size image
Sub-group analyses and meta-regressions for overall outpatient use

No meta-regression model resulted in a significant reduction in overall between-study heterogeneity. Subgroup analyses found overall differences by service location (Q = 9.03, df:1, p = 0.002), population group (Q = 35.40, df:2, p < 0.001), study design (Q = 94.68, df:3, p < 0.001) (see Additional file 1: Appendix S9). Meta-regressions were conducted finding lower proportions of service utilisation were associated with service providing low intensity treatment (β = − 0.91; 95%CI = − 1.78;− 0.04; p = 0.04), and on campus services compared than those either on or off campus (β = − 1.10, 95%CI: − 1.85; − 0.36, p = 0.005). Higher proportions of use were associated in ‘at risk’ to general populations of students (β = 1.62, 95%CI:0.88; 2.37, p < 0.001), and mixed methods studies (β = 2.41, 95%CI:0.08; 4.73, p = 0.04).

iii Overall residential service use

Narrative summary (n = 5; k = 7)

Four studies reported six estimates of residential service use [9, 57, 61, 66, 70], ranging from 1 to 5.4%. Population group appeared to be associated with this variation, with the study reporting on general populations of students having a lower estimate than other groups (see Tables 1 and 2, and Additional file 1: Appendix S10 for a detailed narrative summary).

Meta-analysis for overall residential service use (n = 5; k = 7)

The overall pooled proportion effect size using a random effects model was estimated to be 0.03 (95%CI:0.02;0.05) (see Fig. 4). The between study heterogeneity was estimated at τ2 = 0.30, and Ι 2 = 99.4%. There was a prediction interval which ranged from a proportion of 0.007 to 0.12. This indicated a wide range of future possible estimates. Overall, these results indicate substantial heterogeneity across the included estimates of residential mental health service use.

Fig. 4
figure 4

Forest Plot for overall residential service use

Full size image

Subgroup analyses and meta-regressions for overall residential service use

Meta-regressions only a found a reduction in between study heterogeneity association with population group (τ2 = 0.19, Ι 2 = 86.6%). High estimates were associated with ‘at risk’ students (β = 1.29, 95%CI: 0.84; 1.73, p = 0.001), and subgroup of students (β = 1.50, 95%CI: 0.80; 2.21, p = 0.0041) when compared to general populations of students (see Additional file 1: Appendix S10).

Does service use differ across health service type?

i Differences in use by service type

Subgroup analysis conducted using a three-level meta-analysis suggested differences between service types (F = 63.25, df:2,39, p < 0.001). A meta-regression was conducted where compared to overall service use, both overall outpatient service and overall residential service use was associated with lower proportion of university students reporting using these services (outpatient: β = − 0.77, 95%CI: − 1.26; − 0.29; p = 0.01; residential: β = − 3.05, 95%CI: − 3.63; − 2.47, p < 0.001).

Sensitivity analyses found mixed results (see Table 3). For example, excluding estimates of lifetime service use had an attenuating effect on all pooled proportions, whereas removing low quality studies resulted in a lower pooled proportion only in overall service use. When outliers and influential estimates were removed the pooled proportion for overall service use was higher. A reduction in between study heterogeneity was only observed when outliers and influential cases were removed (see Table 3). Sensitivity analyses continued to suggest differences by service location and treatment type for overall outpatient service use, by service location for overall service use, except when excluding estimates of lifetime use (see Additional file 1: Appendix S11, 12 and 13).

Table 3 Sensitivity analyses
Full size table

Further analyses using three-level meta-analysis

i Estimates meeting multiple service categories

Narrative summary (n = 12; k = 23)

Twelve studies reported on twenty-one estimates associated with services that could be classified as any DESDE classifications [9, 47, 53, 55, 56, 62,63,64,65, 70, 74, 78]. These estimates ranged from 5 to 68%. Lower estimates were reported in services offering specialist or high intensity treatment compared to a range of treatments, whereas higher estimates tended be in campus services. In general, studies asking students report service use over their lifetime were associated with higher estimates [55, 62,63,64,65, 78] (see Tables 1 and 2).

Meta-analysis (n = 12; k = 23)

The pooled proportion based on the three-level meta-analytic model was 0.20 (95%CI:0.13; 0.31, p < 0.001). Ι 2level 3 = 82.9% of the total variation can be attributed to between-cluster, and Ι 2level 2 = 13.76% to within-cluster heterogeneity. We found that the three-level model provided a significantly better fit compared to a two-level model with level 3 heterogeneity constrained to zero (χ21 = 8.10, p 0.004).

Subgroup analyses and meta-regressions

Subgroup analyses found differences by service location (F = 11.201, df:2,18, p < 0.001). Meta regressions found on campus, and off campus location was associated with a high proportion when compared service potentially located in both locations (On campus:β = 1.83, 95%CI:0.83, 2.83, p = 0.001; off campus:β = 0.91, 95%CI:0.003, 1.81, p = 0.05) (see Additional file 1: Appendix S14, and Appendix S16 for sensitivity analyses).

ii Specific outpatient services

Narrative summary (n = 13; k = 37)

Between 6.98% and 62.5% of students reporting outpatient service use out of the ten studies and twenty-seven estimates [49, 55, 61, 64,65,66,67,68, 70, 71, 76, 79]. These estimates were between 6.98% and 62.5% of the study populations reporting outpatient service use. It was difficult to determine what this variation was associated with. The definitions used to measure service use may explain some variation. For example, the highest estimate of 62.5% related to individual counselling, and lowest estimate of 6.98% related to group counselling within the same study, and both classed as low intensity treatments [61]. The country a service was located appeared to potentially be associated with some variation. Estimates in a study of students at risk in Ethiopia were both low compared to most other estimates in the USA [79]. In general, higher estimates tended to be in studies asking students to report whether they had ever used a mental health service [49, 55, 61, 64, 65, 68, 78].

Meta-analysis (n = 13; k = 37)

The pooled proportion based on the three-level meta-analytic model was 0.19 (95%CI:0.13; 0.28, p < 0.001). Ι 2level 3 = 31.3% of the total variation can be attributed to between-cluster, and Ι 2level 2 = 64.3% to within-cluster heterogeneity. We did not find that the three-level model provided a significantly better fit compared to a two-level model with level 3 heterogeneity constrained to zero (χ21 = 1.99, p = 0.16).

Subgroup analyses and meta-regressions

Subgroup analyses found differences by treatment type (F = 34.83, df:3,33, p < 0.001) and service location (F = 35.58, df:2,34, p < 0.001). Meta regressions found low intensity (β = − 0.94, 95%CI: − 1.17, − 0.71, p < 0.001), specialist treatment (β = − 2.06, 95%CI: − 2.81, − 1.32, p < 0.001) and on campus locations were associated with lower proportions (β = − 0.93, 95%CI: − 1.15, − 0.71, p < 0.001) (see Additional file 1: Appendix S15, and Appendix S17 for sensitivity analyses).

Discussion

Main findings

This is the first systematic review and meta-analysis to synthesize evidence relating to the proportion of university students using mental health services, and how this varies by service type. In summary, we found there are wide variety of services available taking varying proportions of students, although overwhelmingly these were from HICs, in particular the USA. Across studies when estimates were grouped and pooled in service categories, we found around a 1/3 of students use services overall while attending university, with around 1/5 of students using outpatient services, and between 1 and 3% have used services that could be classed as residential. Our findings suggest where there is greater availability of support there is greater use, as indicated by higher use being associated with services offering a range of treatments. There was limited evidence to suggest services on campus were used more than those off campus, and students with more severe current psychological distress were associated with greater past service use. However, there are significant limitations with the current literature, including few international studies, particularly from LMICs, little clarity on how services link together, no studies of patient flow and limited consistent description of services.

Findings in the context of existing evidence

The finding of the proportion of students using mental health services is broadly consistent with average proportions of students reporting problems in previous literature from the USA and North America. In 2012 around 18% of students reported receiving any form of mental health treatment, and 36% among students with a likely mental health problem [20]. Annual cross-sectional surveys confirm that service use is aligned with prevalence in the USA and Canada with increases in service utilisation between 2007 and 2017 to around one third of university students using services [8, 9]. Comparisons with estimates in non-student populations are difficult to interpret because of heterogeneous measures used to estimate need, limited international longitudinal analyses, and few studies assessing the effect of university on mental health trajectories [4]. A systematic review of service use among non-student young adults found only 16% reported using any mental health service, lower than our findings [81]. This is unlikely to be due to differences in need as individual studies suggest mental disorder has increased in both groups, at a similar rate [10, 11]. US studies featured predominantly in both this previous review and ours, therefore differences in reported service use may reflect differences in the availability of services and insurance coverage between groups in the USA. Studies in non-students included relatively young populations with an average age of 21 [81]. In the USA context, the transition to university could prompt the earlier emergence of mental health difficulties as students may face significant new pressures, a new social context and new financial challenges prompting earlier help seeking [4, 9, 20, 25, 27, 82].

Our review predominantly reports on studies of US university students in four-year institutions, and therefore our findings likely confounded by what is available there. Higher proportions of students using campus services maybe due to student’s awareness of, and ability to reach and pay for these services in comparison to other services [83]. Four-year US institutions receive comparably higher levels of funding than US community colleges, influencing their ability to provide students with comprehensive mental health services [23, 47, 84]. Studies using both national and regional US samples found four-year university students report higher use of services on campus compared to community college students, despite higher prevalence of mental health problems in community colleges [23, 47]. Cost was cited as the most common barrier to seeking help among community college students [23]. International studies included in this review reported different patterns of service use, which may reflect different patterns of service provision, demand among students, and barriers to help seeking [73,74,75, 78,79,80]. For example, countries such as Australia where there may be fewer barriers to support outside of university, students sought help from a broad range of providers, most frequent being General Practitioners [73]. The limited number of studies outside the USA may reflect the relatively recent increases in the number and diversity of students attending university in other HIC countries, such as the UK [4]. Only recent research has highlighted the very limited research focus on LMIC [85], perhaps the reflecting the potentially smaller proportion of their national populations attending university compared to most HICs [1]. However, recent efforts through the WHO WMH-ICS indicates some change in this field [6, 16]. This in the context of the growing emphasis on the importance of global mental health and the role higher education might play in contributing to improvements in population health [1, 3].

The level of heterogeneity observed was striking when compared to the published literature potentially illustrating the wide range of services, likely with a range of entry requirements, and populations of students. This could also reflect inequalities in population coverage and use of mental health services relative to need across the student populations, as noted in other literature [18, 21, 22]. A review in non-student populations found being female, Caucasian, homosexual, or bisexual meant you were more likely to use services, which is similar to findings in students [81]. However, in our review, some studies of international students had comparably lower use of services, one study reporting only 2.6% used a service [52]. Other studies examining use in other populations in our review reported much higher proportions, as high as 75% [73]. It may be that variation among students is even greater than non-students due to the wide variety of needs among students. Despite students in the USA and other HICs potentially having more available services, such as those on campus, these may be particularly underutilised by some groups who experience more significant barriers to help-seeking both inside and outside university [18, 21, 22]. If some groups of students are consistently underrepresented in services, it is unlikely activities and interventions these services provide will be appropriate for their needs, and will continue to be underutilised by these students [86].

Strengths and limitations

This is the first systematic review to summarise and pool evidence quantitatively about the management of student mental health. This allowed us to explore and then quantify variation in the way mental health services are used by university students. However, there are limitations to the current review. Firstly, generalising the findings of this review outside of the USA should be cautioned given the limited number of international studies. Secondly, there were specific challenges to classifying services studies described or listed. For example, it was not always clear whether the services were interpreted in the same way by all participants or services with similar names were comparable to each other between studies. While we double coded a random sample of these services, this could have introduced classification bias when grouping the services in this review. We found some outlying estimates that may have been explained by the broad definitions used. For example, ‘counselling’ could provide help for a range of needs or be interpreted differently by students answering a survey. While other reviews have commented that there is variation by treatment received, service location, and by specific populations of students [20, 31]. There was not always detailed and consistent data across our included studies to thoroughly evaluate these relationships quantitatively. However, we used a range of synthesis methods to understand the literature.

The methods to examine use of mental health services in the included studies were heterogeneous. While most included binary response options, the reporting periods varied. This meant there were challenges determining whether students used a service at university or before they were students and whether students continued to use services from before university or were new presentations. This may have led to an overestimation of the proportion of students using mental health services. However, we did conduct sensitivity analyses where we excluded these estimates and used meta-regressions to control for reporting period in all analyses. Most of the studies were in the USA. We would therefore caution generalising the findings of this review beyond the USA given the specificities of the healthcare system and infrastructure available to students there, in contrast even to other Western countries.

Implications for practice, policy, and research

The findings from this review emphasise the importance of a range of service provision being available to students who are experiencing psychological distress, and supports current policy efforts to develop well integrated services to help span levels of need. However, reviews in countries with a significant policy emphasis on integration, such as the UK, highlight the challenges defining this process, and the traditionally top-down approach has led to mixed success [87]. The authors argue this may relate to the highly contextual nature of the problems integration aims to address, therefore it should focus on what needs to be done rather than simply the goal of integration [87]. The findings of our review, particularly the variety of services, groups of students and numbers using mental health services, support this point. This emphasises the need for detailed local needs assessments, the co-production of the process of integration with relevant stakeholders, and adaptations to meet the needs of the local student population [32, 87].

Given the important developmental period students often attend university and the potential important role university’s could play in improving population mental health, the findings of the review suggest a series of important avenues for future research. (1) There is a urgent need to conduct robust international studies to understand student mental health need; (2) international research describing service models available to, acceptable to, and used by, students and similar aged young people; (3) given the few students using formal mental health services across all studies identified in this review, international research should continue to understand alternative models and interventions which might be acceptable and accessible students, such as task shifting, the use of technology, and capacity building within social networks [3, 32]; (4) there are no studies of patient flow and how services are linked together which should be a priority of research particularly given the policy emphasis on integration; (5) there is a limited number of studies examining the adequacy of treatment students receive which could help understand how well services are meeting the needs of students who reach services [42]. (6) To understand how best to adapt current care pathways the experiences of students, healthcare professionals and other stakeholders need to be explored. In some HICs qualitative studies have spoken to students, and staff in counselling services [19, 24, 25, 82], however given the variation of services we found in this review our findings emphasize the need to speak to healthcare professionals, students and other young in a range of settings; (7) The observed differences between the findings of this review and a review in non-student populations [81], it is crucial to understand whether university attendance adds additional risk to mental health trajectories. Our findings suggest significant inequalities in access to mental health services among students and settings, the literature should be systematically reviewed to examine this further.

Globally, future research should pay close attention to health and social inequalities between those with and without a university degree. In many countries, particularly those with a small proportions of people ultimately attaining a university degree, there is the potential to exacerbate inequalities by improving the health of a potentially privileged group of people [1, 88]. Any initiatives aiming to address student mental health should be considered in the relation to wider population as part of a broader strategy to improve population mental health [3].

Conclusion

This review is the first effort to systematically describe mental health services available to students and quantify students’ use of them. Most studies were in HICs, in particularly the USA, where we found around a third of students had used a mental health service, similar to the proportion of students with symptoms indicative of mental disorder. However, we found significant variation in the utilisation of mental health services across populations of students, settings, and countries. There were some services, such as those on-campus, used more than others potentially reflecting supply and demand patterns in the included study settings. The empirical literature to date is very limited in terms of the relatively small number of international studies, and few studies examining how services link together, and how students move between them which limits our understanding of the problems students face. Our findings support the current renewed effort to study student mental health internationally and emphasises the importance of well-integrated services to support students’ needs.