Effects of Exercise Interventions on Anxiety in Korean University Students: A Narrative Review

1. Design and search strategy

This study was conducted as a narrative review to synthesize current evidence on exercise and anxiety. Literature searches were performed between June 1 and June 30, 2026, across two domestic databases (RISS, KISS) and three international databases (PubMed, ScienceDirect, Google Scholar). Databases were searched for studies published between January 2000 and June 2026. No language restriction was applied, and searches were conducted in both Korean and English. Google Scholar and hand-searching of reference lists were used as supplementary sources. Google Scholar was searched as a supplementary source using the query: Korean “university students” anxiety exercise. Because Google Scholar ranks results by relevance rather than returning a bounded result set, records were screened in rank order and retrieval was discontinued after the first 41 records, at which point results were no longer topically related to exercise or anxiety in university-student populations. These 41 records were retained for screening. In addition, the reference lists of included articles were hand-searched, yielding 7 further records. Because the domestic databases do not fully support nested Boolean operators, a single combined query could not be executed. Because RISS and KISS are Korean-language database, the original Korean search terms are reported alongside their English translations to ensure the search strategy is reproducible. Instead, separate queries of the form [대학생 AND 불안 AND X] were run for X = 운동 (exercise), 스포츠 (sports), 신체활동 (physical activity), 요가 (yoga), 필라테스 (Pilates), and 무용 (dance), and the results were merged. Preliminary searching confirmed that 유산소운동 (aerobic exercise), 무산소운동 (anaerobic exercise), 복합운동 (combined exercise), 심신운동 (mind-body exercise), and 태극권/타이치 (Tai Chi) were fully captured by the 운동 and 요가 queries, and these terms were therefore not run as separate queries. In RISS, the search was restricted to domestic journal articles, theses and dissertations, and research reports; monographs were excluded as they do not report original research data. In KISS, journal articles were searched. For PubMed, the following single strategy was applied: (“college student*”[tiab] OR “university student*”[tiab] OR “undergraduate*”[tiab]) AND (Korea*[tiab] OR Korea[ad]) AND (exercise[tiab] OR "physical activity"[tiab] OR yoga[tiab] OR pilates[tiab] OR "tai chi"[tiab] OR dance[tiab] OR sport*[tiab]) AND (anxiety[tiab] OR STAI[tiab] OR "state-trait"[tiab]). ScienceDirect limits the number of Boolean terms per query, so the intervention concept was divided across three queries, each combined with (“college student” OR “university student” OR undergraduate) AND (Korea OR Korean) AND anxiety: (a) exercise OR “physical activity”; (b) yoga OR Pilates OR “tai chi”; and (c) dance OR sport OR aerobic. The affiliation field tag [ad] was used as a supplementary proxy to capture studies conducted in Korea; because this field reflects author affiliation rather than participant nationality, all retrieved records were manually screened to confirm that the study sample consisted of Korean university students.

2. Eligibility criteria and study selection

The searches yielded 598 records in total: 387 from RISS, 144 from KISS, 18 from PubMed, 1 from ScienceDirect, 41 from Google Scholar, and 7 from hand-searching reference lists. The large number of domestic records reflects the use of multiple parallel queries in RISS and KISS, which produced substantial overlap. After removal of 177 duplicates arising from overlap between queries within a database and across databases, 421 unique records remained. Titles and abstracts of these 421 records were screened, and 348 were excluded, most commonly because anxiety was not assessed as an outcome, because the anxiety construct examined was situation- or domain-specific (competitive anxiety, performance anxiety, social physique anxiety, career-related anxiety) rather than generalized anxiety, or because no exercise or physical activity variable was present. The full texts of the remaining 73 records were assessed, and 53 were excluded. Twenty studies met eligibility criteria and were included in the review.

Figure 1.

Process of selecting literature for analysis

Included studies were organized into two groups according to how completely they satisfied the eligibility criteria elaborated below. Studies meeting all four criteria are referred to as fully eligible studies, and those satisfying them only partially are referred to as partially eligible studies. These labels describe only how well each study fit our inclusion criteria. They were created for this review to facilitate narrative synthesis of a heterogeneous literature, and are not a formal rating of study quality, risk of bias, or evidence strength. The criteria were not pre-registered: they were developed during full-text review of the retrieved records and finalized before data extraction and synthesis.

Fully eligible studies were defined as those meeting all of the following criteria: (a) a Korean university-student sample; (b) exercise or physical activity as the independent variable (an intervention program); (c) anxiety measured directly with a standardized instrument (STAI, SCL-90 anxiety subscale, Self-rating Anxiety Scale (SAS), State Anxiety Index (SAI)); and (d) anxiety analyzed as a primary, not incidental, outcome.

Partially eligible studies were defined as those in which one of these criteria was only partially satisfied, specifically: (a) anxiety was assessed only as a secondary or subscale outcome, or through a physiological proxy rather than a validated anxiety questionnaire; (b) the design was correlational, mediational, or observational (assessing participation-level physical activity rather than an intervention program), precluding causal inference; or (c) a study met the formal STAI/SAI direct-measurement criterion but the direction of the reported effect could not be reconciled with the authors' own operational definition of anxiety, and this inconsistency could not be resolved from the published text.

Studies were excluded when anxiety was not measured at all, when a Korean-language publication drew its sample from a non-Korean student population, or when participants were not university students. Following screening, eleven studies were retained as fully eligible and nine as partially eligible (N = 20 total).

1. Overview of included studies

The eleven fully eligible studies spanned aerobic exercise (k = 3) (Han, 2003; Kang, 2016; Park, 2007), resistance/strength‑oriented exercise (k = 1) (Kang, 2016) combined aerobic-resistance exercise (k = 2) (Kang, 2016; Lee & Yu, 2018), mind-body exercise (k = 4) (Park & Kim, 2016; Kwak, 2023; Kim et al., 2023; Yook et al., 2024), and other movement-based activities (k = 3) (Kim, 2016; Kim & Park, 2000; Jung, 2015). Study designs varied, including a single-blind randomized controlled trial (k = 1) (Yook et al., 2024), multi-arm controlled trials or nonequivalent-control pre-post designs (k = 4) (Kang, 2016; Park, 2007; Lee & Yu, 2018; Park & Kim, 2016), two-group or quasi-experimental pre-post designs (k = 3) (Kwak, 2023; Kim, 2016; Kim & Park, 2000), and single-group pre-post designs with no comparison group (k = 3) (Han, 2003; Kim et al., 2023; Jung, 2015). Sample sizes were generally small, ranging from 16 to 109 (median < 30), with several fully eligible studies enrolling single-sex cohorts, which limits generalizability by sex.

Table 1.

Summary of Fully Eligible Studies on Exercise and Anxiety in Korean University Students

2. Partially eligible studies evidence

Nine additional studies were classified as partially eligible, providing directionally informative but methodologically weaker or ambiguous evidence. To avoid conflating evidence of different types, these studies are divided into two subgroups and summarized separately in Table 2a and 2b.

Table 2.

Summary of Partially Eligible Studies

Table 2a summarizes five studies that used an exercise intervention but did not fully meet the eligibility criteria above (Jang & Kim, 2013; Kim, 2015; Kim et al., 2004; Sim, 2023; Song & Shin, 2022), most often because the reported anxiety effect was non-significant, indirectly measured, or showed an unresolved scoring inconsistency. Among these, Kim et al. (2004) met the formal direct-measurement criterion for anxiety, but the study was reclassified from fully eligible to partially eligible because the reported score change (an increase) was logically inconsistent with the original authors' own case definition of anxiety, which most likely reflects an unresolved item-coding issue rather than a genuine anxiolytic effect.

Table 2b summarizes four studies that examined the association between physical activity and anxiety using an observational or correlational design rather than an intervention (Jung & Yook, 2022; Kim & Kim, 2007; Kim et al., 2025; Kim, 2009). Because these studies did not administer an exercise intervention, they are interpreted as descriptive, hypothesis-generating evidence on the exercise-anxiety association rather than as evidence of intervention effect, and are not used to support causal claims elsewhere in this review.

3. Level of evidence

Because fully eligible status alone does not equate to methodological rigor, the eleven fully eligible studies were further stratified by design strength (Table 3). Only one randomized controlled trial was identified (Yook et al., 2024). The remaining fully eligible studies relied on nonequivalent control groups, two-group or quasi-experimental comparisons, or single-group pre-post designs without a comparison group, which cannot rule out regression to the mean, maturation, or practice effects as alternative explanations for observed anxiety changes.

Table 3.

Level of Evidence of Fully Eligible Studies

4. Exercise-prescription parameters

Table 4 summarizes explicit frequency, intensity, time, and type (FITT) parameters for the interventions in the eleven fully eligible studies as these variables are crucial for physical therapists’ exercise prescription. Reporting was inconsistent across studies: intensity was unreported or vaguely described (“individualized”) in several studies (Park, 2007; Kwak, 2023, Kim et al., 2023, Yook et al., 2024, Kim, 2016, Kim & Park, 2000), and frequency or session length was missing entirely in the two lowest-quality single-group studies (Kim & Park, 2000; Jung, 2015). Total program duration ranged widely, from 4 weeks (Yook et al., 2024) to 15 weeks (Kim et al., 2023), with no discernible dose-response pattern. A significant within-group reduction in state anxiety was observed in the 7-week Tai Chi program (Park & Kim, 2016), whereas the 15-week Pilates program showed a significant increase (Kim et al., 2023). Duration alone does not appear to predict the direction of the effect in this literature.

Table 4.

Exercise Prescription Parameters (FITT Framework) of Interventions in the Fully Eligible Studies

5. Findings by exercise category

1. Implications for physical therapy practice

From a physical therapy standpoint, three observations stand out.

First, none of the eleven fully eligible studies were designed or delivered as a formal therapeutic-exercise prescription in the sense used in physical therapy practice, that is, with individualized progression criteria, objective load monitoring, and symptom-based dose adjustment. The programs were instead delivered by nursing, sport-science, psychology, or general-education instructors, and intensity and progression were frequently undocumented (Table 4). This represents a concrete opportunity for physical therapists, who are trained in exercise prescription and dosing, to design and test anxiety-focused exercise programs with the same rigor already applied in musculoskeletal or cardiopulmonary rehabilitation.

Second, one of the two most methodologically rigorous studies in this review, the single-blind randomized trial of an online mindful somatic movement program, was co-authored by a physical therapy researcher (Yook et al., 2024). This suggests that physical therapy expertise is already contributing meaningfully to this evidence base and supports the relevance of this topic to physical therapy science. We specifically looked for additional studies in which a physical therapist was directly involved in intervention delivery or exercise prescription, but none could be identified among the twenty studies screened. This absence is itself informative: it suggests that physical therapy-led exercise trials for anxiety in this population remain largely unexplored, and it tempers any claim that physical therapy involvement has already been shown to improve outcomes in this literature.

Third, the pattern of unexpected results identified in this review carries a clinically important point for physical therapists who prescribe exercise for psychological outcomes. Among the fully eligible studies, a 15-week Pilates program showed a significant increase in anxiety (Kim et al., 2023), and a related but methodologically ambiguous study showed a similar numerical pattern that could not be reconciled with the authors' own scoring criteria (Kim et al., 2004). Among the observational studies, higher self-reported sport participation was associated with higher, not lower, state anxiety (Kim & Kim, 2007), and unsupervised participation-level exposure showed inconsistent or null associations with anxiety in two further studies (Kim et al., 2025; Kim, 2009). None of these counter-directional or null findings arose from a program that was individually dosed, progressed, or monitored by a trained professional; each involved either an unsupervised participation measure or a fixed, non-individualized group protocol. This pattern is consistent with the broader principle in physical therapy practice that a therapeutic exercise program requires ongoing, symptom-based adjustment to remain both safe and effective, rather than a uniform prescription applied without monitoring. Taken together, these findings suggest that not every exercise modality or delivery format produces an anxiolytic effect in every subgroup, and that unsupervised, self-reported participation, as opposed to a dosed, monitored intervention, appears particularly prone to null or paradoxical results. This underscores a concrete role for physical therapists, whose training in individualized assessment, progression, and symptom monitoring positions them to supervise exercise programs in which anxiety is tracked as an explicit outcome and adjusted for in real time, rather than relying on general, unguided activity participation delivered by non-specialist instructors.

2. Methodological considerations & confounders

Several methodological considerations are worth noting when interpreting the findings in this review.

First, a clear divergence was observed between state and trait anxiety outcomes. Trait anxiety reached statistical significance only in the intervention with higher session frequency and total duration (three sessions per week over 36 sessions) (Kang, 2016), whereas the lower-dose online program reduced state anxiety alone, without a significant change in trait anxiety (Kwak, 2023). While consistent with the theoretical distinction between transient state anxiety and stable dispositional traits, this observation remains hypothesis-generating as only three fully eligible studies evaluated both subscales.

Second, because exercise modality, session frequency, total duration, and sample demographics covary almost completely across the literature, it is impossible to isolate the true effect of exercise type from these confounded design features. Any apparent superiority of specific modalities must be interpreted with caution.

3. Limitations

This review has several limitations. As a narrative rather than a systematic review, the search was limited to five databases without a structured PRISMA flow, dual-reviewer screening, or a formal risk-of-bias assessment. In addition, none of the twenty included studies reported an a priori sample size or power calculation. Because most studies enrolled fewer than 30 participants per arm, it remains unclear whether they were adequately powered to detect anxiety reductions; this raises the possibility that some non-significant findings reflect insufficient power rather than a true absence of effect, and conversely that some significant findings may be inflated in a small-sample literature. Additionally, Google Scholar does not permit reproducible, bounded searching. Retrieval was therefore terminated at the point where results ceased to be topically relevant, and this cut-off reflects a judgment by the reviewers rather than a predefined stopping rule. Records ranked beyond this point may have included eligible studies. Furthermore, reliance on Korean-language sources and a small number of international sources may have introduced publication or language bias. Finally, extracted statistics (particularly for the multi-arm trial, mediation analyses, and scoring directions in the Pilates and meridian-exercise studies) warrant careful re-verification against original manuscripts.