Global prevalence and associated factors of social anxiety disorder among medical students: a systematic review and meta-analysis

Abstract

Background

Social anxiety disorder (SAD) is common and can seriously affect daily functioning. Medical students face high academic pressure and frequent social evaluation, but how common SAD is in this group globally remains unclear. This review therefore aimed to estimate the worldwide prevalence of SAD among medical students and summarized what factors are linked to it.

Methods

We followed PRISMA guidelines and searched 13 databases from inception to September–October 2025, with no language restrictions. We included observational studies that reported SAD prevalence among undergraduate medical, nursing or allied health students using validated tools. Two reviewers independently extracted data and assessed study quality with the JBI checklist. A random-effects model was used to pool prevalence. Heterogeneity was examined with I² and prediction interval. We also did subgroup analyses and meta-regression. Associated factors were summarized narratively using a grading framework developed by the authors.

Results

Thirty-four studies from 18 countries were included, covering 25,914 students. The pooled prevalence of SAD was 45.8% (95% CI: 38.0%–53.8%), but heterogeneity was very high (I² = 99.01%). The 95% prediction interval ranged from 4.1% to 84.3%. Prevalence varied significantly by the screening tool used and increased with more recent publication year (β = 0.1341, p = 0.0033). Female proportion and sample size showed no significant association. Other subgroup analyses (Geographical region, study quality, sampling method, participant type) did not show significant differences. Factors consistently associated with SAD included female sex, early academic years, depression/anxiety symptoms, low self-efficacy, neuroticism, body dissatisfaction, family conflict, negative peer experiences, academic stress, low socioeconomic status, and tobacco use.

Conclusions

Globally, SAD is very common among medical students, though estimates vary widely mainly because of differences in how it is measured. The factors identified point to a mix of personal vulnerability and stressful training conditions. Standardized screening and interventions that address both student resilience and the learning environment are urgently needed.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40359-026-04874-y.

Introduction

Social Anxiety Disorder (SAD), also known as social phobia, is defined in the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) as a marked fear or anxiety regarding one or more social situations in which an individual may be scrutinized by others. Common examples include social interactions like conversations or meeting strangers, being observed while eating or drinking, and performing in front of an audience, such as delivering a speech [1]. As a prevalent psychiatric condition, SAD is closely associated with significant functional impairment and reduced quality of life. Epidemiological studies indicate that the lifetime prevalence of SAD in the general population is 13%, with a 12-month prevalence of 7.4%, positioning SAD as one of the most common mental disorders, following only depression and substance use disorders [2–4].

SAD typically emerges in adolescence, with a mean age of onset between 14 and 16 years, and often follows a chronic course. Individuals with SAD may either endure social situations with intense distress or resort to avoidance behaviors, both of which can severely hinder academic, occupational, and social functioning. Furthermore, high rates of comorbid psychiatric conditions further complicate the clinical picture and long-term outcomes of SAD [2, 4].

Within higher education, medical students represent a population at particular risk of SAD owing to the unique demands of medical training. They are confronted with heightened academic pressure, substantial clinical responsibilities, ethical dilemmas, and frequent exposure to human suffering. These stressors are often accompanied by intense social evaluation—from peers, instructors, and patients—which can directly amplify the core features of SAD, namely the fear of negative evaluation and social avoidance [5–7]. Additional stressors, such as competitive learning environments, financial burdens, separation from support networks, and limited leisure time, may further exacerbate their concerns about being scrutinized or judged by others. Over time, these cumulative stressors can not only induce general psychological distress but also specifically precipitate or exacerbate SAD symptoms. In turn, SAD may impair academic performance, diminish empathy, increase cynicism, and reduce overall well-being [8, 9].

In recent years, a growing body of research has investigated the prevalence of SAD among medical students across different regions. However, reported prevalence rates demonstrate substantial variability. For instance, studies conducted in Germany, India, and China have documented rates of 47.0%, 78.0%, and 25.6%, respectively [10–12]. This significant heterogeneity is likely attributed to methodological differences, including the use of diverse diagnostic or screening tools, cultural and regional variations, differences in medical education systems, and varying operational definitions of SAD [13, 14], and substantial variation across studies is therefore expected. Accordingly, rather than aiming to resolve these inconsistencies through a single pooled estimate, this systematic review and meta-analysis seeks to quantify the extent of heterogeneity and explore its sources through prespecified subgroup analyses and meta-regression.

Despite the accumulating body of primary research on SAD among medical students, two notable research gaps persist in current literature. First, no comprehensive study has yet generated a global pooled prevalence estimate focused exclusively on medical students. Second, existing evidence lacks a systematic synthesis of SAD-related influencing factors across diverse geographical regions and educational contexts.To address this gap, the present systematic review and meta-analysis aims to achieve two primary objectives: (1) To estimate the global pooled prevalence of SAD among medical students, and (2) To identify and analyze key associated factors, including demographic characteristics, academic stress, personality traits, and socio-cultural influences. By integrating existing findings, this review seeks to provide a clearer epidemiological overview, inform the design of future research, and offer evidence-based foundation for developing targeted interventions, supportive institutional policies, and preventive strategies. Strengthening mental health support for medical students is crucial not only for their personal well-being but also for cultivating resilient and competent future healthcare professionals.

Methods

Protocol registration

We attempted to prospectively register this systematic review on the International Prospective Register of Systematic Reviews (PROSPERO) but did not complete the submission due to our initial unfamiliarity with the platform’s workflow. To ensure transparency and adherence to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [15], we registered the review on the Open Science Framework (OSF) on 7 April 2026 (DOI: 10.17605/OSF.IO/SGKY7). The OSF record contains the full protocol, including search strategies, eligibility criteria, and analysis plans. A related PROSPERO protocol (CRD420251150961, Mohta et al.; https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD420251150961) exists, but our review differs in population (undergraduate medical students only, including nursing and medical sciences) and search coverage (13 databases, no language restriction), and it was conducted independently.

Search strategy

This systematic review and meta-analysis were conducted in accordance with the PRISMA guidelines [15]. To ensure comprehensive retrieval of relevant literature, a systematic search was performed across multiple electronic databases, including PubMed, Embase, PsycINFO, Cochrane Library, Scopus, ScienceDirect, Web of Science, Google Scholar, ProQuest Dissertations & Theses, and preprint servers (medRxiv and bioRxiv). To minimize language and regional bias, major Chinese databases-China National Knowledge Infrastructure (CNKI), Wanfang Data, and VIP Database were also included, with no language restrictions applied. The reference lists of all included studies and relevant review articles were manually screened to identify additional publications. The search covered records from the inception of each database up to the respective search date (conducted between September and October 2025), with the search strategies and exact search dates for each database provided in Supplementary File 1. For full-text articles not written in English or Chinese, we obtained a Chinese translation of the entire article before data extraction. One German article was identified and included in this way.

The search strategy combined Medical Subject Headings (MeSH) terms and free-text keywords related to three core concepts: (1) SAD, (2) medical students, and (3) prevalence and associated factors. Boolean operators (“AND” and “OR”) were used to combine these terms. The search strategy was initially developed for PubMed and subsequently adapted to match the syntax and controlled vocabulary of each of the other databases. The PubMed search string was as follows:

((“phobia, social“[MeSH Terms] OR “social phobia*“[Title/Abstract] OR “Sociophobia“[Title/Abstract] OR “Social Anxiety Disorder“[Title/Abstract] OR “SAD“[Title/Abstract] OR “social anxi*“[Title/Abstract] OR “Social Fear“[Title/Abstract] OR “social avoidance“[Title/Abstract]) AND (“students, medical“[MeSH Terms] OR “medical student*“[Title/Abstract] OR “medical undergraduate“[Title/Abstract] OR “Undergraduate Medical Students“[Title/Abstract] OR “Medical undergraduate Students“[Title/Abstract] OR “schools, medical“[MeSH Terms] OR “medical school*“[Title/Abstract] OR “Medical university“[Title/Abstract] OR “Medical college“[Title/Abstract] OR “medical school student*“[Title/Abstract]) AND (“prevalence“[MeSH Terms] OR “prevalence“[Title/Abstract] OR “epidemiology“[MeSH Terms] OR “epidemiology“[Title/Abstract] OR “incidence“[MeSH Terms] OR “incidence“[Title/Abstract] OR “burden“[Title/Abstract] OR “frequency“[Title/Abstract] OR “cross-sectional studies“[MeSH Terms] OR “cross sectional study“[Title/Abstract] OR “cohort studies“[MeSH Terms] OR “cohort stud*“[Title/Abstract] OR “survey“[Title/Abstract] OR “risk Factors“[MeSH Terms] OR “risk Factors“[Title/Abstract] OR “associated factor*“[Title/Abstract] OR “predictor*“[Title/Abstract] OR “determinant*“[Title/Abstract] OR “influence factor*“[Title/Abstract] OR “impact*“[Title/Abstract] OR “correlate*“[Title/Abstract])) AND (humans[Filter])

Eligibility criteria

Studies were included if they met the following criteria: (1) Observational studies (cross-sectional, cohort, or case-control) reporting original empirical data. (2) Participants were undergraduate students enrolled in medicine, nursing, or allied health programs. (3) Studies reported the prevalence of SAD or provided sufficient raw data (number of cases and total sample) to calculate prevalence. (4) SAD was assessed using a validated diagnostic instrument (e.g., Mini International Neuropsychiatric Interview [MINI]) [16] or a standardized screening tool with established cutoff scores (e.g., Social Phobia Inventory [SPIN]) [17], Liebowitz Social Anxiety Scale [LSAS] [18].

Studies were excluded if any of the following applied: (1) Qualitative studies, case reports, case series, interventional studies (e.g., RCTs), systematic reviews, meta-analyses, commentaries, or editorials. (2) Insufficient data to extract or calculate SAD prevalence, even after contacting corresponding authors. (3) Studies focusing solely on non-medical students (except where such students belonged to a broader medical sciences faculty as defined above). (4) Publications available only as abstracts or conference proceedings, as these typically lack sufficient methodological detail and raw data (e.g., sample size, number of cases) to calculate a reliable prevalence estimate.(5) Use of a non-specific anxiety measure that did not assess SAD specifically.

Data extraction

A standardized, pre-piloted data extraction form was adopted for this study (Supplementary 2). Two reviewers (Wang YX and Hou CY) independently extracted the following information from each included study: (1) Study identification and characteristics: first author, year of publication, country, study design, and sampling method; (2) Participant characteristics: academic year of the students, sample size, proportion of female participants, and mean age or age range; (3) Outcome measurement: assessment tool for SAD and its specific cut-off value, number of SAD cases, and prevalence estimate.All included studies used self-report screening instruments rather than structured clinical diagnostic interviews (e.g., MINI); (4) Associated factors: where reported, data on factors potentially associated with SAD were also extracted, including the specific factor (e.g., sex, academic year, depressive symptoms, family conflict), the measure of association (e.g., odds ratio, correlation coefficient), and its statistical significance.

Any discrepancies between the two reviewers were resolved through discussion, or by a third reviewer if necessary. The extracted data on associated factors were synthesized narratively, which was used to inform descriptive analyses and the interpretation of heterogeneity.

Quality assessment

The methodological quality of each included cross-sectional study was independently evaluated by two reviewers (Huang ZX and Yan MX) using a standardized 9-item JBI checklist [19]. This checklist assessed the following key domains: (1) representativeness of the sampling frame; (2) appropriateness of the sampling method; (3) adequacy of the sample size; (4) description of subjects and study setting; (5) coverage of the sample in data analysis; (6) validity of the method used to identify SAD; (7) standardization and reliability of the measurement; (8) appropriateness of statistical analysis; and (9) adequacy and handling of response rates.

For each item, judgments were recorded as “Yes,” “No,” “Unclear,” or “Not Applicable.” A score of “1” was assigned for “Yes” responses, while “No,” “Unclear,” and “Not Applicable” received a score of “0.” Based on the total score, studies were categorized as low quality (scores ≤ 3), moderate quality (scores 4–6), or high quality (scores ≥ 7).

Discrepancies between reviewers were resolved through discussion, with arbitration by a third reviewer when necessary. The final quality rating for each study was integrated into the data extraction table (Table 1). These ratings were taken into account during evidence synthesis and informed a pre-specified sensitivity analysis to test the robustness of the pooled prevalence estimates.

Table 1.

Characteristics of the 34 studies included in the systematic review and meta-analysis

Author/Year	Country	Research design/Sampling method	Subject	Medical Year of Study	Sample Size	Female%	Average Years or Range	Scale/Cutoff	Cases/Prevalence rate	JBI score
Regis et al. 2018 [21]	Brazil	Cross-sectional study /Convenience sampling	Medical students	1st to 6th year	479	58.7%	22.5 ± 2.6	SPIN/≥19	174/36.3%	7
Chu et al. 2023 [22]	China	Cross-sectional study /Convenience sampling	Nursing students	1st to 4th year	204	94.6%	21.09 ± 1.99	IAS/≥50	74/36.3%	6
He & Xie 2011 [23]	China	Cross-sectional study/Random sampling	Medical students	1st year	522	53.6%	Not reported	SADS/≥13	115/22.0%	4
Wang et al. 2011 [24]	China	Cross-sectional study /Convenience sampling	Nursing students	1st to 3rd year	1032	91.86%	21.44 ± 1.85	IAS/≥49	174/16.86%	5
Liu et al. 2019 [25]	China	Cross-sectional study /Stratified cluster sampling	Medical students	4th to 5th year	664	62.8%	Not reported	BSPS/≥10	652/98.2%	6
Ba.209 [26]	China	Cross-sectional study /Stratified cluster sampling	Medical students	1st to 2nd year	1446	82.4%	20.5 ± 1.7	IAS/≥49	370/25.6%	8
Wang et al.2011 [12]	China	Cross-sectional study/Stratified cluster sampling	Medical students	1st to 2nd year	1448	63.8%	20.4 ± 1.8	IAS/≥49	370/25.6%	6
Reta et al.2020 [27]	Ethiopia	Cross-sectional study/Stratified random sampling	Medical and health science students	1st to 5th year	293	41.3%	22.13 ± 2.18	SPIN/≥20	96/32.8%	6
Baldauf et al. 2014 [10]	Germany	Cross-sectional study/Convenience sampling	Medical students	1st, 2nd, and part of 4th year	525	73.5%	21.95 ± 3.13 Range: 17–39	LSAS/≥30	251/47.%	4
Solanki et al. 2025 [28]	India	Cross-sectional study/Systematic random sampling	Medical students	1st to 3rd year (Parts 1 & 2)	480	54.8%	20.28 ± 1.23	SPIN/≥21	451/94.0%	6
Singh et al. 2024 [29]	India	Retrospective study/ Convenience sampling	Medical students	1st to 4th year	600	63.7%	19.4	SPIN/≥21	468/78.0%	5
Ratnani et al. 2017 [30]	India	Cross-sectional study/Convenience sampling	Medical students	1st to 3rd year	290	53.1%	19.0 ± 1.1	SPIN/≥24	33/11.37%	5
Honnekeri et al. 2025 [31]	India	Cross-sectional study/Convenience sampling	Medical students	Not specified	176	65.9%	Not reported	SIAS/≥24	26/14.8%	5
Preeti et al. 2019 [32]	India	Cross-sectional study/Convenience sampling	Medical students	1st to 4th year	404	26.7%	20.7 ± 1.5	SIAS/>34	51/12.62%	6
Sundaram et al. 2023 [33]	India	Cross-sectional study /Simple random sampling	Medical students	1st to 5th year & Interns	538	63%	21.27 ± 1.73	SPIN/≥21	221/41.1%	7
Dsouza et al. 2019 [34]	India	Cross-sectional study /Convenience sampling	Medical students	2nd, 3rd, and Final year	272	53.3%	20.8 ± 1.3	SPIN/≥21	83/30.5%	7
Kommisetty et al.2025 [35]	India	Cross-sectional study /Simple random sampling	medical students	1st to 3rd year	400	56%	Range: 18–25	SPIN/≥21	292/73.0%	5
Jogdande et al.2017 [36]	India	Cross-sectional study/Convenience sampling	Medical students	1st to 3rd year	150	57.3%	< 21 yrs: 86; >21 yrs: 64	SPIN/≥21	69/46.0%	4
Utkarsh et al.2025 [37]	India	Cross-sectional study / Convenience sampling	Medical students	1st to Final year	250	63.2%	Not reported	SPIN/≥21	109/43.6%	4
Charishma et al.2020 [11]	India	Cross-sectional study /Convenience sampling	Medical students	1st to 4th year	600	63.7%	19.4 years	SPIN/≥31	468/78.0%	4
Mascarenhas et al. 2019 [38]	India	Cross-sectional study/stratified sampling by gender	Medical and paramedical students	1st year	100	50%	18–25 years	SPIN/≥21	28/28%	4
Hamzei et al.2025 [39]	Iran	Cross-sectional study /Convenience sampling	Medical sciences students	not specified	200	65%	22.87 ± 2.2	LSAS/≥55	75/37.5%	6
Kadhim et al.2024 [40]	Oman	Cross-sectional study/Convenience sampling	Medical students	1st to 7th year	558	62.4%	20.05 ± 1.87	LSAS/≥30	355/64%	6
Ruzhenkova et al.2020 [41]	Russia	Cross-sectional study/Convenience sampling	Medical students	1st to 6th year	724	74.2%	20.5 ± 1.8 Range: 17–26	SPIN/≥21	244/33.7%	5
Alkhalifah et al.2017 [42]	Saudi Arabia	Cross-sectional study/Convenience sampling	Medical students	1st to 7th year	1447	61.7%	Not reported	SPIN/≥21	1333/92.1%	5
Mawgod et al. 2025 [43]	Saudi Arabia	Cross-sectional study/Convenience sampling	Medical students	2nd to 6th year	150	46%	22.5 ± 1.6 range(19–26)	SPIN/≥21	84/56.0%	5
Al-Johani et al.2022 [44]	Saudi Arabia	Cross-sectional study/Convenience sampling	Medical students	2nd year to internship	5896	55.12%	22.43 ± 1.68	SPIN/≥21	3004/50.95%	8
Elhadad et al.2017 [45]	Saudi Arabia	Cross-sectional study/Convenience sampling	Medical students	2nd to 6th year	368	42.9%	Not reported	SPS/>66	219/59.5%	4
Quronfulah et al.2024 [46]	Saudi Arabia	Cross-sectional study/Convenience sampling	Health profession students	1st to 7th year	461	87.4%	Not reported	SPIN/≥24	259/56.2%	7
Alhazmi.2024 [47]	Saudi Arabia	Cross-sectional study/Convenience sampling	Nursing students	3rd and 4th year	141	100%	20–22 years	SPIN/≥21	67/47.5%	7
Suleiman et al.2021 [48]	Sudan	Cross-sectional study/Convenience sampling	Medical students	1st to 5th year	375	67.2%	21.4 ± 1.46 (18–30)	SPIN/≥23	230/61.3%	7
Mohsen et al.2022 [49]	Syria	Cross-sectional study/Convenience sampling	Medical students	1st to 6th year	492	48%	Not reported	LSAS/≥55	244/ 49.6%	5
Latifeh et al. 2018 [50]	Syria	Cross-sectional study/Random sampling	Medical students	2nd to 6th year	285	40.7%	Not reported	LSAS/≥55	62/21.8%	4
Laidlaw.2009 [51]	United Kingdom	Cross-sectional study/Convenience sampling	Medical sciences students	1st to 3rd year	247	~ 60%	Not reported	SIAS/≥34	19/7.69%	7

SPIN Social Phobia Inventory, IAS Interaction Anxiousness Scale, SADS Social Avoidance and Distress Scale, BSPS Brief Social Phobia Scale, SIAS Social Interaction Anxiety Scale, LSAS Liebowitz Social Anxiety Scale, SPS Social Phobia Scale

Throughout the review, all procedures followed the predefined protocol. No changes were made to the research question or eligibility criteria, and no pilot testing altered study inclusion rules. All screening, data extraction and quality assessment were completed as planned to maintain methodological rigor.

Data analysis

Statistical analyses were conducted using Comprehensive Meta-Analysis 3.0. The pooled prevalence of SAD was calculated with 95% confidence intervals (CI) under a random-effects model to account for between-study heterogeneity. Heterogeneity was assessed via the I² statistic and Cochran’s Q test, whereby an I² value > 50% indicated substantial heterogeneity. Because proportions are bounded between 0 and 1, the analysis was conducted on the logit‑transformed scale. The resulting pooled effect size was back‑transformed to the original proportion scale. The Z‑test tests the null hypothesis that the true proportion equals 0.5 (i.e., 50%).

To further characterize the extent of between-study heterogeneity, we calculated the 95% prediction interval (PI) for the overall pooled prevalence using the method of Riley et al. (2011) based on a random-effects model with a logit transformation [20]. Prediction intervals were not calculated for subgroup analyses due to the limited number of studies within each subgroup.

Subgroup analyses were performed to explore potential sources of heterogeneity across the following categorical factors:

Geographical region (Asia, Europe, Africa, South America): Pooled prevalence was synthesized for continents with two or more eligible studies (Asia, Europe and Africa), while the single study from South America was presented separately without pooling.

Assessment instrument: All tools for SAD measurement were summarized. Pooled estimates were calculated for scales used in no fewer than two studies, including the Social Interaction Anxiety Scale (SIAS), Liebowitz Social Anxiety Scale (LSAS), Social Phobia Inventory (SPIN) and Interaction Anxiousness Scale (IAS). For single-use instruments (Social Phobia Scale, SPS; Social Avoidance and Distress Scale, SADS; Brief Social Phobia Scale, BSPS), only individual prevalence and 95% CI were reported.

Participant type: Categorized as medical students, nursing students, and medical and health science students.

Sampling method: Divided into convenience sampling and random sampling.

Methodological quality (JBI score): Stratified into high, moderate and low quality according to established cut-off values.

Meta-regression was further applied to explore continuous moderators, including publication year, female proportion and sample size. This method avoids artificial grouping, retains raw information and maintains statistical power. Both subgroup analyses and meta-regression were undertaken using a random-effects framework.

Subgroup analyses were exploratory; no adjustment for multiple comparisons (e.g., Bonferroni) was applied.

Synthesis of associated factors

A narrative synthesis was conducted to summarize evidence on factors associated with SAD. Due to substantial heterogeneity in study designs, measurement tools, and analytical methods across the included studies, a quantitative meta-analysis was not deemed appropriate. Extracted data on correlates were categorized into broad domains (e.g., demographic, psychological, social). The strength and consistency of evidence for each specific factor were evaluated based on three primary criteria: (1) the number of studies reporting the association; (2) the direction, magnitude, and statistical significance of reported effect measures (e.g., odds ratios, beta coefficients); and (3) the consistency of findings across studies.

Based on this evaluation, evidence was qualitatively classified into the following levels

Strong evidence: Supported by ≥ 5 studies with highly consistent (≥ 75% reporting statistically significant associations) and robust effect estimates, often from analyses controlling for potential confounders.

Moderate evidence: Supported by 3–4 studies with consistent significant associations, but with a more limited total evidence base.

Limited evidence: Supported by 1–2 studies, indicating a preliminary association requiring further replication.

Inconsistent evidence: Multiple studies available but reporting conflicting directions of association or mixed statistical significance.

The evidence classification scheme described above was developed by the authors specifically for this review, as no universally accepted standard exists for grading evidence in narrative syntheses of observational studies on associated factors.

The results of this synthesis are presented in the Results section, both in a summary table (Table 3) and descriptive text, detailing the identified factors and the robustness of their associations with SAD among medical students.

Table 3.

Summary of factors associated with SAD among medical students (with illustrative examples)

Broad category	Specific factor	Association direction & evidence strength	Representative research findings
Demographic factors	Female sex	Strong evidence, highly consistent risk factor (27 studies; majority report significant positive association)	Utkarsh et al.[37],2025: AOR = 1.98 (1.25–3.12) Kadhim et al.[40],2024: Prevalence 70% vs. 52% Sundaram et al.[33],2023: AOR = 2.00 (1.36–2.95)
	Earlier academic year	Strong evidence, consistent risk factor (13 studies; majority indicate higher risk in earlier years)	Latifeh et al.[50],2018: Prevalence decreased with year (35.5% in Year 2 vs. 9.7% in Year 6) Singh et al.[29],2024: Social anxiety score decreased by 0.765 per year
	Rural/urban residence	Inconsistent evidence (13 studies; conflicting results)	Rural as risk: Sundaram et al.[33],2023: AOR = 1.75 (1.18–2.58) Urban as risk: Singh et al. [29], 2024: Urban students scored higher (Coef. =0.553)
Psychological & Health Factors	Depressive/anxiety/stress symptoms	Strong evidence, consistent risk factor (9 studies; all report significant positive associations)	Ruzhenkova et al.[41],2020: Comorbid depression OR = 5.09 (3.6–7.3) Regis et al.[21], 2018: Depressive symptoms OR = 2.78 (1.37–5.66)
	Personality traits (high neuroticism /low extraversion)	Moderate evidence, consistent risk factor (3 studies; all report significant associations)	Wang et al.[24],2011: Neuroticism (r = 0.31), Extraversion (r=–0.42); combined contribution > 65% Ba.[26],2009:Neuroticism (r = 0.520)
	Self-efficacy	Strong evidence, (5 studies; all report significant negative associations)	Liu et al.[25],2019: β=-1.101, OR = 0.333(0.225–0.491) Wang et al.[24],2011: r=–0.34
	Body image dissatisfaction	Moderate evidence, consistent risk factor (4 studies; all report significant positive associations)	Regis et al.[21],2018: OR = 1.56 (1.03–2.36) Preeti & Das.[32],2019: Facial dissatisfaction OR = 2.98 (1.21–7.38)
Social & Behavioral Factors	Family functioning & conflict	Moderate evidence, consistent risk factor (4 studies show significant effect of conflict; some support for family support/function)	Utkarsh et al.[37],2025: Family conflict AOR = 2.60 (1.30–5.10) Chu et al.[22],2023: Family function negatively correlated with social anxiety (r=–0.232)
	Negative peer experiences	Moderate evidence, consistent risk factor(3 studies; all report significant associations)	Sundaram et al.[33],2023: Humiliation experience OR = 2.3 (1.5–3.6) Regis et al.[21],2018: Difficulty making friends OR = 2.08 (1.34–3.22)
	Academic pressure & performance	Strong evidence, mixed associations (6 studies; subjective stress and poor performance are risks; high performance is protective)	Jogdande & Gupta.[36],2017: Poor academic performance AOR = 7.94 Liu et al.[25], 2019: Subjective stress OR = 1.542
	Low socioeconomic status	Moderate evidence, consistent risk factor (6 studies; majority report significant associations)	Sundaram et al.[33], 2023: Maternal illiteracy AOR = 3.64 (1.43–9.27) Utkarsh et al.[37], 2025: Low maternal education AOR = 3.25 (1.28–8.22)
Substance Use & Lifestyle	Tobacco use	Moderate evidence, consistent risk factor (4 studies; majority report significant positive associations)	Solanki et al.[28], 2025: Tobacco use positively correlated with social anxiety severity (p < 0.0001) Latifeh et al.[50], 2018: Higher prevalence in heavy smokers (19.4%)
	Lack of physical activity	Limited evidence, potential risk factor (1 study)	Solanki et al.[28], 2025: Physical activity as a protective factor (p < 0.0001)

Sensitivity analysis and assessment of publication Bias

A leave-one-out sensitivity analysis was performed by iteratively removing each individual study to examine its impact on the overall pooled prevalence estimate. Additionally, a post hoc sensitivity analysis was conducted by excluding outlier studies that reported prevalence of ≥ 90% (deviating substantially from the overall distribution). This analysis aimed to explore the contribution of extreme values to the high between‑study heterogeneity and to validate the robustness of the pooled findings after removal of influential outliers.

Publication bias was assessed through visual inspection of a funnel plot for asymmetry and statistically evaluated using Egger’s linear regression test. However, we acknowledge that funnel plot asymmetry may also reflect genuine between-study heterogeneity or other study characteristics beyond publication bias. Therefore, the results of Egger’s test were interpreted cautiously and in conjunction with the leave-one-out sensitivity analysis.

Results

Search results

The literature search and screening process is summarized in the PRISMA flow diagram (Fig. 1). A total of 898 records were initially identified from database searches and other sources. Specifically, 886 records were retrieved from electronic databases, including PubMed (n = 63), Embase (n = 138), Web of Science (n = 67), Scopus (n = 103), PsycINFO (n = 71), ScienceDirect (n = 6), Cochrane Library (n = 0), Google Scholar (n = 22), ProQuest Dissertations & Theses (n = 12), CNKI (n = 46), Wanfang Data (n = 46), Weipu Data (n = 7), and the preprint servers medRxiv and bioRxiv (n = 305). An additional 12 records were identified through manual searches of reference lists.

Fig. 1.

Flowchart of the study selection process

After removing 213 duplicates, 685 unique records remained for title and abstract screening. During this phase, 588 records were excluded as irrelevant. The remaining 97 full-text articles were assessed for eligibility. Of these, 63 were excluded for the following reasons: inability to extract usable prevalence data (n = 31); study population not focused on medical students (n = 3); publication only as an abstract or conference proceeding (n = 5); outcome not pertaining to SAD prevalence (n = 4); duplicate or overlapping study population (n = 1); use of a non-specific instrument for assessing SAD (n = 1); prevalence not estimated using a validated cutoff score (n = 3); interventional studies without prevalence data (only mean scores) (n = 3); and classification as a review article (n = 2).

Consequently, 34 studies met all predefined eligibility criteria and were included in the present systematic review and meta-analysis.

Characteristics and methodological quality of included studies

This systematic review and meta-analysis included 34 studies [10–12, 21–51]; their main characteristics and methodological quality are summarized in Table 1.

1. Temporal and geographical distribution: The included studies were published from 2009 to 2025, demonstrating sustained research interest over the past 15 years. Geographically, the studies covered 18 countries, representing a global scope. Most studies were conducted in Asia (26 studies, 76.5%), with the largest contributions from India [11, 28–38], China [12, 22–26], and Saudi Arabia [42–47]. The remaining studies were distributed across other regions: the Middle East (Iran [39], Oman [40], , Syria [49, 50]), Europe (Germany [10], Russia [41], United Kingdom [51]), South America (Brazil [21]), and Africa (Ethiopia [27], Sudan [48]), contributing to the rich cultural and educational diversity of the synthesized evidence.

2. Study design and sampling: All included studies adopted an observational design. Among them, 33 were cross-sectional studies [10–12, 21–28, 30–51], and one was a retrospective study [29]. With respect to sampling strategies, convenience sampling was the most commonly used method (23 studies) [10, 11, 21, 22, 24, 29–32, 34, 36, 37, 39–49, 51]. Probability-based methods, including simple random sampling, stratified random sampling, and cluster sampling, were employed in 11 studies [12, 23, 25–28, 33, 35, 38, 50].

3. Participant demographics: Across the 34 included studies, the pooled total sample comprised 25,914 undergraduate students enrolled in medicine, nursing, and allied health programs. Sample sizes varied substantially across studies, ranging from 100 [38] to 5,896 [44] participants, with a median of 432. The proportion of female participants ranged from 26.7% [32] to 100% [47], with a median of approximately 60%. Of the 26 studies that reported age [10–12, 21, 22, 24, 26–30, 32–36, 38–41, 43, 44, 47, 51], mean ages corresponded to early adulthood, typically between 19 and 23 years.

4. Assessment of SAD: A key source of methodological heterogeneity across studies was the assessment instrument used for measuring SAD. Social Phobia Inventory (SPIN) was the most frequently utilized tool, adopted in 18 studies (52.9%) [21, 27–30, 33–38, 41–44, 46–50], The Liebowitz Social Anxiety Scale (LSAS) was used in five studies [10, 39, 40, 49, 50], the Interaction Anxiousness Scale (IAS) in four [12, 22, 24, 26], and the Social Interaction Anxiety Scale (SIAS) in three [31, 32, 51]. Of note, the cut off scores for the same instrument varied substantially across studies (e.g., SPIN: 19–31; LSAS: 30–55; SIAS: 24–34), further reflecting heterogeneous operational definitions of SAD caseness. Other instruments-including the Brief Social Phobia Scale (BSPS) [25], the Social Phobia Scale (SPS) [45], and the Social Avoidance and Distress Scale (SADS) [23]-were each used in a single study. This variability in assessment tools and diagnostic thresholds directly contributed to the substantial heterogeneity in reported SAD prevalence estimates, which ranged from 7.69% [51] to 98.2% [25].

5. Methodological quality: Methodological quality was appraised using the JBI critical appraisal checklist for prevalence studies. Total scores ranged from 4 to 8 out of a maximum possible score of 9. The score distribution was as follows: 4 points [10, 11, 23, 36–38, 45, 51]; 5 points [24, 29–31, 35, 41–43, 49]; 6 points [12, 22, 25, 27, 28, 32, 39, 40]; 7 points [21, 33, 34, 46–48, 51]; and 8 points [26, 44]. These results indicated an overall moderate methodological quality across the included studies. All quality appraisal scores are presented in Table 1 and were included in sensitivity analyses.

The prevalence of SAD among medical students

The pooled prevalence of SAD among medical students was synthesized from the 34 included studies using a random-effects model, which accounts for the substantial observed heterogeneity. The overall estimated prevalence was 45.8% (95% CI: 38.0%–53.8%). The Z-test for the null hypothesis that the true proportion equals 50% was not significant (Z = − 1.036, p = 0.300), which is consistent with the 95% confidence interval containing 50%. This indicates that the average prevalence does not differ significantly from 50%, although the wide confidence interval reflects the substantial heterogeneity among studies.

Heterogeneity among the studies was exceptionally high, as quantified by Cochran’s Q test (Q = 3347.715, p < 0.001) and the I² statistic (I²=99.01%). This I² value signifies that approximately 99% of the total variability in the observed prevalence estimates is attributable to genuine differences between studies (heterogeneity) rather than sampling error. The forest plot visualizing the individual study estimates and the pooled result is presented in Fig. 2.

Fig. 2.

Forest plot of the prevalence of SAD among medical students

The 95% prediction interval for the overall prevalence ranged from 4.1% to 84.3%, indicating that the true prevalence of SAD in a future study could plausibly fall within this wide range, reflecting the substantial heterogeneity among the included studies.

Subgroup analyses

To explore potential sources of between-study heterogeneity, subgroup analyses were conducted according to predefined categorical variables, including geographical region (continent level), assessment instrument, study quality, participant type, and sampling method. All subgroup results are summarized in Table 2.

Table 2.

Subgroup analyses of the prevalence of SAD among medical students based on random-effects model

Subgroup	No.of reports	Prevalence(%)	95%CI(%)	I2(%)	P value for heterogeneity	P value between groups
Geographical region						0.257
Asia	28	48.4	39.4-57.6	99.1	< 0.001
Africa	2	46.8	18.2-77.7	99.1	< 0.001
Europe	3	25.7	10.1-51.7	97.9	< 0.001
the Americas†	1	36.3	32.1-40.7	-	-
Assessment instrument						0.000
IAS①	4	25.4	13.8-42.0	93.8	< 0.001
LSAS②	5	43.5	28.1-60.3	96.9	< 0.001
SIAS③	3	11.4	5.0-23.9	65.7	0.059
SPIN④	19	54.1	45.4-62.6	98.6	< 0.001
BSPS#⑤	1	98.2	96.8-99.0	-	-
SPS#⑥	1	59.5	54.4-64.4	-	-
SADS#⑦	1	22.0	18.7-25.8	-	-
Study quality						0.246
High	9	37.6	23.6-54.0	98.3	< 0.001
Moderate	25	48.9	39.0-58.8	99.2	< 0.001
Participant type						0.081
Medical students	26	50.8	41.8-59.7	99.1	< 0.001
Nursing students	3	31.9	13.8-57.7	97.6	< 0.001
Medical and health science students	5	29.6	15.5-49.2	97.1	< 0.001
Sampling method						0.344
Convenience sampling	24	43.4	34.4-52.7	98.9	< 0.001
Random sampling	10	51.7	37.4-65.8	99.1	< 0.001

^†Only one study from the Americas; the reported prevalence and 95% CI are from that single study (not pooled)

^#Only one study used this tool; the reported prevalence and 95% CI are from that single study (not pooled)

^①IAS cut-off: three studies used ≥49, one study used ≥50

^②LSAS cut-off：two studies used≥30, three studies used ≥55

^③SIAS cut-off：two studies used≥34, one study used ≥24

^④SPIN cut-off：one study used≥19, one study used≥20, 13 studies used≥21, one study used≥23, two studies used≥24, one study used≥31

^⑤BSPS cut-off：≥10

^⑥SPS cut-off：>66

^⑦ SADS cut-off：≥13

Geographical region: The pooled prevalence of SAD was 48.4% (95% CI: 39.4–57.6) for Asia (n = 28), 46.8% (95% CI: 18.2–77.7) for Africa (n = 2), and 25.7% (95% CI: 10.1–51.7) for Europe (n = 3). A single study from the Americas reported a prevalence of 36.3% (95% CI: 32.1–40.7) and was presented individually without pooled estimation. No significant intercontinental difference was detected (p = 0.257).

Assessment instrument: Pooled SAD prevalence differed significantly across assessment tools (p < 0.001). The Social Phobia Inventory (SPIN) yielded the highest pooled estimate at 54.1% (95% CI: 45.4–62.6), followed by the Liebowitz Social Anxiety Scale (LSAS) at 43.5% (95% CI: 28.1–60.3) and the Interaction Anxiousness Scale (IAS) at 25.4% (95% CI: 13.8–42.0). The Social Interaction Anxiety Scale (SIAS) showed the lowest pooled prevalence at 11.4% (95% CI: 5.0–23.9). For scales used in only one primary study (i.e., the Brief Social Phobia Scale [BSPS], Social Phobia Scale [SPS], and Social Avoidance and Distress Scale [SADS]), individual prevalence estimates with 95% CI are listed in Table 2 without meta-analytic pooling.

Study quality: No significant difference in SAD prevalence was found between high-quality studies (37.6%, 95% CI: 23.6–54.0) and moderate-quality studies (48.9%, 95% CI: 39.0–58.8; p = 0.246).

Participant type: The pooled SAD prevalence was 50.8% (95% CI: 41.8–59.7) among medical students (n = 26), 31.9% (95% CI: 13.8–57.7) among nursing students (n = 3), and 29.6% (95% CI: 15.5–49.2) among medical and health science students (n = 5). No significant difference was observed across the three participant subgroups (p = 0.081).

Sampling method: Studies adopting convenience sampling reported a pooled SAD prevalence of 43.4% (95% CI: 34.4–52.7; n = 24), while those using random sampling had a pooled prevalence of 51.7% (95% CI: 37.4–65.8; n = 10). The between-group difference was not statistically significant (p = 0.344).

Prediction intervals were not calculated for these subgroup analyses due to the limited number of studies within each subgroup.

Meta-regression results

To avoid arbitrary categorization and retain complete information from the 34 included studies, publication year, proportion of female participants, and sample size were examined as continuous moderators using meta-regression. A random-effects model with the Knapp–Hartung adjustment was applied, consistent with the analytical framework described in the Methods section. Degrees of freedom were set at 32 (34 studies minus 2 model parameters), in line with standard meta-analytic practice.

Publication year: Meta-regression revealed a significant positive association between publication year and SAD prevalence (β = 0.1341, 95% CI: 0.0480–0.2202, p = 0.0033). This result indicates a gradual increase in SAD prevalence over time among the study population.

Proportion of female participants: The proportion of female participants exhibited no significant moderating effect on SAD prevalence (β = 0.0036, 95% CI: −0.0272 to 0.0345, p = 0.8118).

Sample size: Sample size was not a significant moderator of pooled SAD prevalence (β = 0.0001, 95% CI: −0.0004 to 0.0006, p = 0.6041).

Bubble plots illustrating the associations between these moderators and SAD prevalence are presented in Figs. 3 and 4, and 5.

Fig. 3.

Meta-regression bubble plots based on year of publication

Fig. 4.

Meta-regression bubble plots based on female%

Fig. 5.

Meta-regression bubble plots based on sample size

Sensitivity analysis and assessment of publication bias

The leave-one-out sensitivity analysis showed that no single study significantly skewed the overall pooled estimate. All pooled prevalence estimates obtained after sequentially removing each individual study fell within the 95% confidence interval of the primary analysis (38.0–53.8). After excluding outlier studies with a prevalence of ≥ 90%, the pooled prevalence decreased from 45.8% (95% CI: 38.0–53.8) to 38.5% (95% CI: 32.2–45.2), yet the finding of a high prevalence remained credible. Collectively, these results demonstrate the robustness of our main conclusions.

Potential publication bias was assessed through visual inspection of a funnel plot (Fig. 6) and quantified using Egger’s linear regression test. The funnel plot, which plots the logit event rate against the standard error for each study, showed a broadly symmetrical distribution. This visual impression was confirmed by Egger’s test, which yielded a non-significant result (intercept=-0.636, t = 0.190, p = 0.850). Nevertheless, given the extremely high heterogeneity in this meta-analysis (I² = 99.01%), the credibility of publication bias evaluation is restricted. In meta-analyses with substantial between-study variability, conventional publication bias tests such as Egger’s regression tend to have reduced statistical power and may produce biased results. Therefore, the current findings cannot fully rule out the existence of unpublished small-sample or negative studies.

Fig. 6.

Funnel plot for the assessment of publication bias

Factors associated with SAD among medical students

A narrative synthesis of factors associated with SAD among medical students was performed, with key findings summarized in Table 3. Identified factors were categorized into four domains: demographic, psychological and health-related, social and behavioral, and substance use and lifestyle.

Among demographic factors, consistent and strong evidence identified female sex and earlier academic year as key associated factors for SAD. Evidence regarding rural versus urban residence was inconsistent, with studies reporting conflicting associations.

Within psychological and health-related factors, strong evidence supported low self-efficacy and comorbid depressive, anxiety symptoms as key associated factors. Specific personality traits (high neuroticism and low extraversion) and body image dissatisfaction emerged as consistent associated factors with moderate supporting evidence.

Regarding social and behavioral factors, moderate-to-strong evidence indicated that family conflict or poor family functioning, negative peer experiences (e.g., humiliation, difficulty forming friendships), and low socioeconomic status (frequently indexed by parental educational level) were associated with elevated SAD. The association with academic pressure and performance was complex: subjective academic stress and poor academic performance were associated factors, whereas high academic achievement appeared to be protective.

In the domain of substance use and lifestyle, tobacco use was moderately associated with greater social anxiety, whereas evidence for physical inactivity was limited to a single study suggesting it as a potential associated factor.

Collectively, these findings indicate that the identified factors do not act independently but interact within a multifactorial framework of vulnerability and precipitants specific to the medical student population.

Discussion

To our knowledge, this systematic review and meta-analysis provides the first comprehensive estimate of the global prevalence of SAD among medical students and has systematically examined its associated factors.

The pooled prevalence of 45.8% reveals that SAD represents a significant mental health concern in this population, with rates markedly higher than those observed in the general young adult population: 36% in a seven-country study [52] and 17% in a systematic review of children, adolescents, and youth [53]. This elevated prevalence highlights the distinct psychosocial vulnerability of medical students, which may reflect the intense academic pressures, frequent evaluative situations, and socially demanding environments inherent to medical training-factors consistently implicated in the development of psychological distress [8, 54–56].

The most striking observation was the exceptionally high heterogeneity (I²=99.01%), indicating that the pooled estimate should be interpreted with caution and emphasizing the influence of contextual and methodological factors. Our subgroup analyses and narrative synthesis identified several major sources of this variability. Foremost among these was methodological diversity in assessment. The use of at least seven distinct screening instruments, combined with inconsistent cutoff values even for the same tool, directly contributed to widely divergent prevalence estimates. For example, the pooled prevalence among studies using the SIAS was 11.4%, compared with 54.1% among studies using the SPIN. Even when a common instrument such as the SPIN was employed, different clinical cutoffs substantially affected case identification and, consequently, pooled prevalence estimates [17, 57]. This pattern of measurement-driven heterogeneity is not unique to SAD. A recent systematic review and meta-analysis on impostor phenomenon among medical students similarly reported an overall prevalence of 49% with extreme heterogeneity (I²=95.6%) and found that different assessment tools yielded substantially different prevalence estimates, highlighting the critical role of measurement instruments in shaping prevalence findings in this population as well [58].

This highlights a critical lack of standardized, diagnostically validated screening for SAD in student populations.

The interpretation of pooled findings should account for cultural disparities and measurement heterogeneity. First, SAD presentation and self-reporting vary across cultural backgrounds. Distinct social norms and stigma may influence the reporting of social anxiety, and the overrepresentation of Asian studies limits the cross-regional generalizability of pooled prevalence. Future research is warranted to explore cross-cultural differences with culturally validated tools.

Second, diverse self-report scales and inconsistent cut-off values were adopted across eligible studies. These instruments lack cross-cultural measurement equivalence, potentially causing discrepant severity interpretation and further increasing overall heterogeneity. Standardized assessments or clear cultural validation information are recommended for future studies to improve result comparability.

Beyond measurement issues, geographical and cultural contexts appear influential. Although our subgroup analysis revealed no difference in SAD prevalence across continents, cultural factors affecting symptom expression, stigma toward help-seeking, and variations in medical education systems likely play a role [59]. However, the predominance of studies from Asia (26 out of 34) limits the generalizability of our pooled estimate and reduces statistical power to detect genuine regional differences.

Our meta-regression showed that SAD prevalence significantly increased with more recent publication year (β = 0.1341, p = 0.0033), suggesting a gradual rise over time among medical students. This trend may reflect a genuine rise in SAD, potentially exacerbated by contemporary academic pressures and the COVID-19 pandemic [60].

To examine potential bias from student population and sampling method, subgroup analyses were conducted. No significant between-group differences were observed (participant type: p = 0.081; sampling method: p = 0.344). These null findings should be interpreted with caution, given small subgroup sizes, high heterogeneity, and potential masking of true effects. Social anxiety may be similarly high across health professional students irrespective of discipline or sampling strategy. These results are exploratory rather than confirmatory. Larger, more homogeneous samples and probability sampling are needed in future research.

Our narrative synthesis identified a broad constellation of factors associated with SAD (Table 3). Beyond demographic characteristics such as female sex and early academic year, which help identify highly associated subgroups, the findings reveal a complex interplay between psychological vulnerabilities and contextual stressors central to understanding SAD in this population.

Notably, self-efficacy emerged as a prominent and modifiable psychological factor. Within the high-stakes, evaluative environment of medical training, diminished confidence in one’s ability to manage social and academic demands can directly promote anticipatory anxiety and avoidance behaviors, consistent with social cognitive theory [61]. Empirical evidence specific to medical students supports this view, demonstrating that self-efficacy buffers against general psychological distress [62] and mediates the relationship between perceived stress and social anxiety symptoms [63, 64]. This suggests that self-efficacy is not merely a correlate but may represent a key psychological pathway through which stressors contribute to clinical anxiety. However, given the cross-sectional nature of the included studies, this mediating role requires confirmation through longitudinal designs.

Concurrently, consistent associations with severe academic pressure, negative peer experiences (e.g., humiliation, inadequate social support), and dysfunctional family dynamics underline the powerful impact of the socio-environmental context. These factors are not passive background conditions but active contributors that erode essential social buffers and foster a pervasive sense of evaluative threat and isolation [8, 65]. The competitive culture of medical education may amplify these effects, creating a dynamic in which environmental stressors and individual vulnerabilities (e.g., low self-efficacy, high neuroticism) interact reciprocally [66].

Thus, the etiology of SAD among medical students is best understood within a multifactorial interaction model. Pre-existing or acquired individual vulnerabilities (e.g., low self-efficacy, high neuroticism) converge with the unique and intense pressures of the medical training environment [65, 66]. This interaction likely depletes psychological resources and triggers maladaptive coping strategies [67], creating a pathogenic pathway toward SAD. The extreme heterogeneity (I² = 99.01%) does not undermine but rather reinforces the clinical importance of the identified associated factors, as they consistently emerged across diverse settings despite varying absolute prevalence estimates.This framework strongly supports interventions that simultaneously target individual resilience and modify the broader learning environment.

Strengths and Limitations

The main strengths of this review include its comprehensive search strategy across multiple international and Chinese databases without language restrictions, and its strict adherence to PRISMA guidelines [15]. To our knowledge, this is the first study to provide a global pooled prevalence estimate specifically for SAD among medical students and to systematically categorize its associated factors.

Several limitations should be acknowledged. First, the geographical distribution of included studies is skewed, limiting generalizability to under-represented regions. This reflects a persistent systemic gap in global mental health research capacity and evidence production [68]. Second, the high heterogeneity, driven largely by inconsistent measurement (e.g., diverse screening tools and cutoffs), implies that the pooled prevalence should be interpreted as a broad estimate, consistent with well-recognized methodological challenges in psychiatric epidemiology [69]. However, despite extensive subgroup and meta-regression analyses, the heterogeneity remained very high (I² > 95% in most subgroups), suggesting that unmeasured study-level and participant-level factors (e.g., cultural norms, academic pressure, curriculum structure, diagnostic thresholds) may still contribute. This limits the precision of the global pooled estimate. Therefore, our findings should be interpreted as a range of possible prevalence estimates across different contexts rather than a single fixed value. Future primary studies should adopt standardized diagnostic tools and report more contextual moderators to enable more informative syntheses. Third, the cross-sectional design of all included studies prevents causal inference, an inherent limitation of the synthesized evidence base [70]. Accordingly, we have used the term “associated factors” instead of “risk factors” throughout the manuscript to clarify the purely correlational relationships and avoid misleading causal implications. Fourth, a fundamental issue of construct validity arises because we pooled studies using different screening instruments (e.g., SPIN, LSAS, IAS, SIAS) with heterogeneous cut-off scores. This implies that we are not capturing a single homogeneous clinical entity but rather divergent operational approximations of SAD. Moreover, all included studies relied on self-report screening scales rather than structured diagnostic interviews (e.g., MINI). Screening tools prioritize sensitivity over specificity, likely systematically overestimating the true prevalence of clinically diagnosed SAD. Thus, our pooled estimates should be interpreted as the prevalence of screen-positive cases, not clinically diagnosed SAD. To help readers appreciate this heterogeneity, Table 2 details each study’s instrument, cut-off, and reported prevalence. Future research should adopt standardized diagnostic interviews or at least two-stage designs (screening plus clinical confirmation) to provide more accurate estimates. Fifth, although subgroup analyses by participant type and sampling method showed no significant differences, null findings may reflect limited statistical power and high heterogeneity masking true effects. Residual confounding cannot be ruled out. Future studies with larger, more homogeneous samples and probability sampling are needed to confirm our results. Sixth, due to limited resources and the original study scope, no external expert or stakeholder consultation was conducted. Such input would strengthen result interpretation and practical implications. We recommend involving relevant stakeholders in future reviews to supplement quantitative evidence. Finally, although Egger’s test was non-significant, publication bias in such a highly heterogeneous field cannot be completely excluded [71].The assessment of publication bias using Egger‘s test is limited by the extremely high between-study heterogeneity (I² = 99.01%). Under such conditions, Egger’s test may be unreliable, as heteroscedasticity can inflate type I error rates or reduce statistical power. Accordingly, we primarily rely on visual inspection of the funnel plot, which did not suggest obvious asymmetry.

Implications for practice

The high prevalence of SAD and its identified associated factors demand a multi-level response from medical education institutions. First, the urgent implementation of systematic and standardized screening protocols is essential. Given the methodological heterogeneity that inflates prevalence estimates, adopting consistent, validated instruments (e.g., SPIN, LSAS) is critical for reliable early identification [17, 18]. Such screening should be embedded within comprehensive well-being programs that ensure accessible, confidential counseling and actively reduce stigma toward help-seeking. Second, interventions must be both targeted and systemic. Tailored support-including cognitive-behavioral therapy groups and social skills training-should be prioritized for highly associated subgroups (e.g., early-year and female students). Equally important are efforts to address environmental contributors by actively cultivating a more supportive and less evaluative learning climate. This includes curricular reforms to reduce excessive academic stress, promotion of collaborative learning, development of peer support networks, and training for faculty to better identify and respond to student distress [72]. Enhancing student self-efficacy through structured communication training, simulation exercises, and mentorship should be a core component of this strategy [61, 62]. Third, cultural and contextual adaptation of interventions is essential for effectiveness. The observed regional variations in prevalence and the concentration of studies in certain regions demonstrate that a one-size-fits-all model is insufficient. Mental health initiatives must be tailored to align with local cultural norms, stigma related to mental illness, and patterns of help-seeking to ensure acceptability and impact [59].

Conclusion

This study provides robust evidence that SAD is highly prevalent among medical students globally, with a pooled prevalence estimate of 45.8%. By combining quantitative meta-analysis (for prevalence) with narrative synthesis (for associated factors), we were able to address the high methodological complexity.The extreme heterogeneity in reported prevalence is largely attributable to methodological inconsistencies, particularly the use of diverse assessment tools and cutoff values. Rather than undermining the findings, this heterogeneity reinforces the clinical importance of the identified associated factors—female sex, early academic stage, co-occurring psychological distress, specific personality traits, low self-efficacy, and adverse social and academic environments—as they consistently emerged across diverse settings despite varying absolute prevalence estimates.

These findings highlight the urgent need for standardized, validated screening protocols for SAD in medical student populations. Interventions should be multi-dimensional, targeting both individual resilience (e.g., improving self-efficacy and coping skills) and the educational environment (e.g., reducing evaluative stress and enhancing social support). Future research should employ longitudinal designs to clarify causal pathways and explore culturally adapted interventions to reduce this substantial burden, ultimately protecting the mental health of future healthcare professionals.

Abbreviations

SAD

Social Anxiety Disorder

DSM

Diagnostic and Statistical Manual of Mental Disorders

RCT

Randomized Controlled Trial

PRISMA

the Preferred Reporting Items for Systematic Reviews and Meta-Analyses

PROSPERO

International Prospective Register of Systematic Reviews

OSF

Open Science Framework

CNKI

China National Knowledge Infrastructure

MeSH

Medical Subject Headings

CI

Confidence Interval

PI

Prediction Interval

SPIN

Social Phobia Inventory

JBI

the Joanna Briggs Institute

IAS

Interaction Anxiousness Scale

SADS

Social Avoidance and Distress Scale

BSPS

Brief Social Phobia Scale

SIAS

Social Interaction Anxiety Scale

LSAS

Liebowitz Social Anxiety Scale

SPS

Social Phobia Scale

Authors’ contributions

Wang YX and Yan DF conceptualized and designed the study. Wang XY and Hou CY were responsible for the literature search, screening, data extraction, and formal analysis. Huang ZX and Yan MX carried out the risk of bias assessment. Wang YX and Yan DF interpreted the results and were major contributors in writing the original draft of the manuscript. All authors read, provided critical feedback on, and approved the final manuscript for submission.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.