Predictors of Mental Health Service Use by Young Adults: A Systematic Review

Eligible studies were identified through systematic searches of the PubMed, PsycINFO, Scopus, CINAHL, InformIT, Cochrane Library, Web of Science, Embase, and ERIC databases. [A list of key search terms is included in an online supplement to this article.] The initial database search was deliberately kept broad to maximize identification of studies (31). Reference lists of all eligible articles, relevant meta-analyses, and reviews (5,9,16,24,32–35) were manually searched for additional eligible studies.

Eligible studies examined sociodemographic or standardized psychological (that is, affective or cognitive-behavioral) correlates of mental health service use among young adults (ages 18 to 24 years [3,36]). This included, but was not limited to, college students age 18 years and older—a subpopulation primarily consisting of young adults (2,15). Studies targeting specific vulnerable or special-service populations—namely, homeless young adults, young veterans, and chronic illness or disability groups—were excluded due to their different treatment needs (37,38). Consistent with existing research (39,40), mental health services were broadly defined as community mental health services, counseling services, or services provided in hospital departments, with treatment delivered by trained mental health professionals (that is, general practitioners, nurses, psychiatrists, psychologists, psychotherapists, counselors, or social workers), regardless of duration of service use or data collection method (that is, prospectively via self-report or retrospectively via medical records). Eligible studies were published in English in a peer-reviewed journal between January 1990 and June 2015. To be eligible, a study had to provide sufficient data for calculation of odds ratio (OR) effect sizes (for example, Pearson’s r). Finally, to ensure validity and generalizability of these findings, empirical studies had to report variables that were investigated in at least three included studies (41). If studies employed overlapping samples, only the study providing the largest amount of information (that is, with the largest sample or with a similar sample size and more independent variables) was included to ensure data independence (31,42). Similarly, from the single longitudinal study reviewed, only the most recent data set was utilized (that is, data from the third wave of data collection—young adults with a mean age of 21.5 years) (43).

In total, 18 eligible studies with independent data were included in this systematic review. [A flow chart of the study selection process is included in the online supplement.]

Consistent with reporting guidelines for systematic reviews (44), a data extraction sheet was developed. This included sample characteristics (for example, sociodemographic variables) and methodology (for example, study design) per study.

For ease of data interpretation, individual outcome measures were classified according to the variables they represented—specifically, prior use of services, sexual orientation, gender, race-ethnicity, social support, psychological distress, perceived need for help, depression, and anxiety. These were then broadly grouped as predisposing, enabling, or need factors. Examination of age as a moderator was considered but precluded by the narrow age range of the sample. Data collection was completed by the first author.

Included studies were assessed for compliance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement (45). Each STROBE criterion was rated as met, met with some limitations, or not met. This evaluation process was independently conducted by the first and second authors, with good interrater reliability demonstrated (r=.90, p<.001).

The OR was the main effect size metric used in this review. An OR represents the odds that an outcome (that is, use or nonuse of mental health services) will occur given a particular exposure (that is, the examined predisposing, enabling, or need variables) (31,46). An OR provides a common standardized metric when analyses compare otherwise nonidentical outcomes of interest, such as continuous versus categorical variables (47). Unlike effect size r, the possible range of OR values is not affected by the distribution of the examined variables. For this review, an OR value of 1 suggested that there was no relationship between service use and the investigated biopsychosocial variable. A value less than 1 meant that the biopsychosocial variable was associated with decreased odds of service use, whereas a value greater than 1 implied that the variable was associated with greater likelihood of use (31).

Effect sizes were calculated in several stages with the assistance of meta-analysis with Interactive eXplanations (MIX) 2.0 software (48). First, data related to any association between service use and a sociodemographic or psychological variable were extracted from each study. Second, Pearson’s r (26,49–51) and 2 × 2 frequency tables (52,53) were converted to ORs by using Wilson’s effect size calculator (54). Most studies directly provided OR estimates, necessitating few conversions. To ensure statistical independence, if a total score and subscores for a standardized psychosocial measure were reported, only the total score was used. If a study provided more than one OR for a single sociodemographic variable (for example, for each racial-ethnic minority group), an average OR was calculated for that study (31,55). Third, ORs that were operationalized in the same manner across studies that used the same design were pooled to produce an average OR. Prior to pooling, each OR was converted into a log OR and then weighted by its inverse variance. The weighted mean log ORs were then transformed back to ORs for interpretation (31,46,55). Finally, 95% confidence intervals (CIs) were calculated to examine the precision of each OR (46). The conservative random-effects model was selected for these calculations (56).

Eighteen independent studies met all the inclusion criteria, resulting in a total of 96,297 participants (Table 1). The studies included 16 cross-sectional studies, one longitudinal study (seven years of data), and one single-birth cohort study of young adults (born in 1987 and followed to 2008). More studies recruited among college students (11 studies) than from the general young adult population (seven studies); however, because of larger samples (N=22,194 versus N=74,103, respectively), data from young adults in the general population predominate in this review. The pooled mean age was 21±2.3 (six studies), although this mean was based on limited data. Although all studies reported an eligible age range, not all specified the mean age of the sample. Similarly, data were not always reported for gender ratios (14 studies; male-to-female ratio of 1 to 1.21) or for key sociocontextual variables, such as employment status (four studies).

Only 16% of participants (16 studies) had used any form of mental health services; nine studies recruited young adults from a nonclinical population, and nine used clinical samples (that is, individuals with a diagnosed substance use, anxiety, or affective disorder). There were no significant differences between these two sample groups in terms of sample size (18 studies; U=37.0, Z=–.31, p=.76, r=–.08), age (six studies; U=2.0, Z=–.29, p=.77, r=–.12), or gender (14 studies; U=22.0, Z=–.26, p=.80, r=–.07).

Most studies used prospective data from a self-report measure (15 studies). Only three studies extracted retrospective data from medical records. Thirteen studies defined mental health services in terms of a multidisciplinary team of health professionals, whereas five investigated a single specific service (for example, general practitioner). The specific type of care provided (that is, primary, secondary, or tertiary) was not routinely indicated.

Reporting of results across the included studies revealed an attrition bias; no studies reported sensitivity analyses, and only one reported on the management of missing data (26). Selection biases were also inherent, with studies routinely failing to report key sociocontextual variables (for example, employment status) and potentially confounding variables (for example, psychotropic medication use). However, studies generally addressed potential detection and reporting biases, in accordance with STROBE, by providing a clear description of theoretical and empirical backgrounds, study objectives, and design and by using valid and reliable psychological instruments (57).

Both clinical and nonclinical samples displayed similar sociodemographic patterns of mental health service use. Young women were twice as likely as young men to access services, and members of racial-ethnic minority groups reported lower use (Table 2). The gender difference was confirmed by the longitudinal study (43) and the cohort study (58). Prior contact with services was associated with increased future use in clinical samples: individuals with a history of service use were almost four times more likely than those without such a history to use services. With regard to social vulnerability, two studies found that homosexual or bisexual individuals were more likely than heterosexuals to seek help (59,60).

Only one enabling variable, social support, was identified, and it was measured inconsistently (Table 2). Of the four self-report measures, only the Multidimensional Scale of Perceived Social Support and a single item measuring the existence of “warm and trusting relationships” produced significant associations, with young adults who reported a higher quality of social support being less likely to access services (50,61). Limited data were obtained for other potentially enabling variables, including health insurance (two studies), residential area (two studies), and financial status (three studies). These contextual variables were also defined inconsistently. For example, one study defined residential area in terms of province (62), whereas another study categorized residence as rural, semiurban, or urban (58).

Diverse measures were used to assess mental health status. Psychological distress (as measured by Adult Self-Report and the DSM-IV) was the only significant predictor for the clinical group (63,64), whereas depression (as measured by Center for Epidemiological Studies Depression Scale [CES-D] and Patient Health Questionnaire [PHQ-9]), anxiety (as measured by the Beck Anxiety Inventory and PHQ-9), and psychological distress (as measured by the dysphoria domain of the Trauma Symptom Inventory) were each significant predictors of service use among the general (nonclinical) young adult population (26,43,51,60,65). Notably, probable diagnoses of depression (based on CES-D screening) significantly increased the odds of service use in the longer term among the nonclinical group (Table 2). Regarding perceived need, participants in clinical samples who reported a need for mental health support were almost five times more likely to access services than those who perceived no need for professional help (61,63). This significant relationship was also found in a nonclinical sample (65).

Our review had several limitations. First, although the narrow 95% CIs demonstrated the strength of the findings, the small number of included studies for some variables (for example, prior use of services) limits the generalizability of those results. Second, methodological moderators (for example, study location) were not formally analyzed because most of the studies reviewed were conducted in the United States. Furthermore, important contextual variables, such as employment status, could not be examined due to the limited available data (74). Future empirical studies should consistently report living arrangements, health insurance benefits, and employment status to facilitate examination of their influence on young adults’ help seeking.

Third, we adopted a broad definition of mental health services because our focus was on variables associated with use of any type of service (use versus nonuse). Research suggests, however, that correlates of service use may vary by service type and amount (72,75,76). For instance, gender differences in types of services used have long been acknowledged, with females preferring general medical settings (69,76). Future empirical studies and subsequent reviews should separately investigate the correlates of using different types of service. The broad definition and primary focus on service correlates also make it difficult to comprehensively evaluate and draw conclusions about characteristics of young adults using different forms of services (for example, community clinics or student counseling services). Future studies are therefore recommended to examine the characteristics of young adults who are receiving treatments or who successfully complete treatments.

Fourth, this study examined only independent relationships between the identified variables, whereas researchers have outlined dynamic and bidirectional interactions between predisposing, enabling, and need factors associated with service use (14,15,69,77). This includes an interactive effect between sociodemographic variables and psychological distress in the prediction of service use (69). Future longitudinal research on help seeking may elucidate the interactions of the identified variables.

Consistent with Andersen’s theoretical framework, personal and attitudinal factors (that is, predisposing and need variables) were found to significantly influence help seeking among young adults. Similar findings have also been obtained with older adults; need factors were the most important predictors of service use in this age cohort (6,78). It follows that psychoeducational campaigns that inform perceptions of mental health problems and services, which have been recommended to increase service use among older adults (6), can also enhance young adults’ utilization of services. This education, which could be widely disseminated via telecommunication technology and social media, might target knowledge of key physical, cognitive, and behavioral symptoms of mental disorders (for example, depression) and treatment efficacy in order to improve attitudes toward mental health conditions and treatment (13,20,21). Broader public education campaigns should also be considered, particularly because young adults’ perceptions of urgency can be influenced by parents and peers, who may initiate help seeking on their behalf (13,21,68).

The identified gender differences may partly reflect masculine norms, including reliance on one’s own strengths and resources (79). Thus, by challenging stereotypes and reducing perceived, use of social media to normalize utilization of professional services may help narrow the gender gap in service use and encourage young men to seek professional help (80).

The findings support the importance of enhancing young adults’ satisfaction with services by promoting positive experiences in order to facilitate future use. Mental health professionals should consider the specific demands, preferences, and needs of young adults, especially those from potentially vulnerable subgroups, when providing care. This includes their preference for personalized and age-appropriate services and flexible interpersonal communication styles (80,81). Clinicians must be specifically trained in cultural sensitivity and competence to increase intake, retention, and successful treatment of individuals from racial-ethnic minority groups and other vulnerable young clients (24,82).