Resilient interpersonal pathways from loneliness to suicidal ideation among Chinese transgender women during global health crises Authors: Lulu Xu, Jin Xu, Fengyi Wang, Yawen Zheng, Chen Xu, Shangbin Liu, Jiachen Lu, Ruijie Chang, Shan Qiao, Huachun Zou, Ying Wang, Yong Cai Published: 2026 CONFLICT OF INTEREST STATEMENT The authors declare no conflicts of interest. INTRODUCTION Global health crises such as the coronavirus disease 2019 (COVID‐19) pandemic destabilise social structures and strain healthcare systems with disproportionate consequences for marginalised populations. Transgender women—individuals whose gender identity is female but who were assigned male at birth—are especially vulnerable under such conditions. In China, transgender women continue to face harassment, violence and structural barriers to care; during crises, enforced isolation and service interruptions magnify loneliness, psychological distress and barriers to help‐seeking. 1 Although elevated rates of suicidal ideation (SI) and attempts are well documented internationally, mechanism‐focused evidence from China—particularly during COVID‐19—remains scarce, underscoring the need to identify actionable interpersonal targets. 2 At the interpersonal level, loneliness—conceived as the emotions associated with the quality of social connections 3 , 4 —emerges as a salient driver of SI during periods of social restriction. 5 The interpersonal theory of suicide identifies perceived burdensomeness (PB) and thwarted belongingness (TB) as proximal determinants of suicidal desire, constructs that may be especially consequential in cultural contexts emphasising family cohesion and interdependence. 6 Yet it is unclear whether, amid public health upheaval, loneliness influences SI among Chinese transgender women via PB and TB. An additional question is whether these pathways remain stable over time, which is essential for them to serve as actionable targets. 7 DATA FROM CHINA DURING GLOBAL HEALTH CRISES Against this backdrop, we conducted serial cross‐sectional surveys (2017–2022) to test these interpersonal mechanisms and to inform crisis‐ready suicide prevention. Specifically, three waves were implemented in two Chinese metropolitan areas—Shenyang (Northeast) and Kunming (Southwest)—during April–July 2017, April–September 2018 and February–December 2022, spanning the pre‐pandemic and COVID‐19 pandemic periods. Sites were chosen to capture geographic and sociocultural diversity and to leverage established partnerships with transgender‐serving community organisations. This design allowed us to replicate a theory‐driven mediation framework across distinct local contexts and under conditions of substantial health system disruption. Eligible participants were aged 18 years or older, assigned male at birth and self‐identified as women. Given the population's marginalisation and limited sampling frames, we used community‐based snowball recruitment through longstanding collaborations with local Centres for Disease Control and Prevention and transgender‐focused community organisations. In each wave, socially well‐connected ‘seed’ participants initiated chain referrals to extend reach across networks; different seeds were used across waves to minimise overlap. Surveys were administered at community venues under a standardised protocol. Participation was voluntary and anonymous; respondents could use pseudonyms. To facilitate engagement and safeguard welfare, the protocol integrated a brief, self‐administered questionnaire with routine on‐site services. A rapid referral pathway to local mental health services was activated for any participant disclosing acute suicide risk. The study was approved by the Ethics Committee of the School of Public Health, Shanghai Jiao Tong University (2016022) and conducted in accordance with the Declaration of Helsinki. All participants provided written informed consent. Given the sensitivity of gender identity and suicidality in this context, we implemented enhanced confidentiality protections: private data collection spaces, restricted access to contact information and staff training in culturally competent, trauma‐informed practices and crisis management. Validated Chinese instruments were used for all measures. Loneliness was assessed with the 8‐item UCLA Loneliness Scale (ULS‐8; total 8–32; higher scores indicate greater loneliness) with documented reliability in Chinese samples. 8 Internal consistency (Cronbach's α) in our waves was 0.82 (2017), 0.86 (2018) and 0.80 (2022). Interpersonal needs (IN) were measured using the Interpersonal Needs Questionnaire‐15 (INQ‐15) capturing PB (items 1–6) and TB (items 7–15; six reverse‐scored), with higher subscale totals reflecting greater PB or TB; psychometric properties of the Chinese INQ‐15 have been established. 9 Internal consistency for the INQ‐15 in our waves was 0.85 (2017), 0.87 (2018) and 0.84 (2022). SI was assessed using instruments aligned with prior work and crisis‐period feasibility. 10 In 2017–2018, SI was captured with a single item (‘Have you ever considered suicide?’ yes/no), an approach commonly used in epidemiological studies of suicidality. In 2022, we used the 5‐item Beck Scale for Suicide Ideation (BSI), 11 defining SI presence as nonzero responses on Items 4 and 5 and computing a continuous severity index (sum of Items 1–5; 0–10) for sensitivity analysis; internal consistency was high (α = 0.89). Demographic covariates included age, sexual orientation, education, marital status, monthly income and residential stability (local resident, > 5 years, 1–5 years, ≤ 1 year). Quality assurance procedures emphasised consistency across waves. Community staff received standardised training in recruitment, consent and data handling; supervisors conducted on‐site checks for questionnaire completeness and eligibility verification. Continuous engagement with partner organisations supported systematic recruitment in urban sites and timely activation of referral pathways. Modest, locally appropriate compensation was provided to offset the burden of participation. Analyses were performed in SPSS V.26.0 (IBM) using the PROCESS macro. We first described sample characteristics and examined bivariate associations (χ 2 for categorical variables; Mann–Whitney U or t tests, as appropriate, for continuous variables; two‐sided p < 0.05). We then specified parallel mediation models to test whether PB and TB mediated the association between loneliness and SI, estimating total, direct and indirect effects with 5000 bootstrap resamples to derive bias‐corrected 95% confidence intervals (CIs) and reporting standardised coefficients (β). Mediation was inferred when CIs excluded zero. Covariate adjustment was prespecified per wave based on bivariate associations with SI (2017: marital status; 2018: sexual orientation; 2022: monthly income and residential stability). Because SI measures differed across waves (single item vs. BSI), cross‐wave interpretation emphasised pathway direction and relative contribution rather than absolute prevalence. Across three waves, 709 transgender women were enrolled (2017: n = 198; 2018: n = 247; 2022: n = 264). The sample was predominantly young (≥ 78.79% aged < 40 years per wave) and mostly single (≥ 76.89%); most had secondary education or below (≥ 68.83%). Nonlocal residency was common but declined over time (2017: 64.65%; 2018: 62.35%; 2022: 53.41%). Reported SI was 36.87% in 2017, 26.32% in 2018 and 68.56% in 2022. Supporting Information S1: table S1 presents core sociodemographic characteristics by wave; extended distributions are shown in Supporting Information S1: table S2. Bivariate comparisons showed consistent, graded differences in loneliness and IN by SI status across all waves. In 2022, mean loneliness, PB and TB were higher among participants with SI compared with those without SI (ULS‐8: 19.58 vs. 15.25; PB: 23.26 vs. 15.28; TB: 34.97 vs. 28.41; all p values < 0.001). Similar patterns were evident in 2017 and 2018 (2017: all p values < 0.001; 2018: loneliness p < 0.001, PB p = 0.002, TB p < 0.001). These findings support a robust interpersonal gradient accompanying SI in this population (table 1). TABLE 1Distribution of loneliness, perceived burdensomeness and thwarted belongingness by suicidal ideation status, 2017–2022YearSI status n Loneliness, mean (SD) a PB, mean (SD) b TB, mean (SD) b t value p value for group difference2017No12513.41 (3.98)10.14 (5.76)22.80 (9.05)Yes7317.88 (5.88)15.23 (7.48)31.03 (10.67)Total/test198−5.77/−5.02/−5.77< 0.001/< 0.001/< 0.0012018No18213.91 (5.05)11.95 (9.38)25.45 (11.40)Yes6518.35 (5.23)16.26 (10.25)31.92 (9.87)Total/test247−6.03/−3.11/−4.36< 0.001/0.002/< 0.0012022No8315.25 (5.29)15.28 (10.17)28.41 (11.45)Yes18119.58 (3.64)23.26 (7.54)34.97 (5.57)Total/test264−6.76/−6.39/−4.96< 0.001/< 0.001/< 0.001 Note: Continuous variables are shown as mean (SD).Abbreviations: PB, perceived burdensomeness; SD, standard deviation; SI, suicidal ideation; TB, thwarted belongingness. a Loneliness was assessed with the 8‐item UCLA Loneliness Scale. b PB and TB were assessed with the Interpersonal Needs Questionnaire‐15. Distribution of loneliness, perceived burdensomeness and thwarted belongingness by suicidal ideation status, 2017–2022 Note: Continuous variables are shown as mean (SD). Abbreviations: PB, perceived burdensomeness; SD, standard deviation; SI, suicidal ideation; TB, thwarted belongingness. Loneliness was assessed with the 8‐item UCLA Loneliness Scale. PB and TB were assessed with the Interpersonal Needs Questionnaire‐15. Primary mediation analyses, prioritising 2022 to reflect contemporary crisis conditions, supported significant parallel pathways from loneliness to SI via PB and TB, with a residual direct effect of loneliness (figure 1). Loneliness was strongly associated with PB (β = 0.58, 95% CI 0.48–0.68) and TB (β = 0.49, 95% CI 0.38–0.59). In turn, higher PB (β = 0.57, 95% CI 0.21–0.93) and TB (β = 0.46, 95% CI 0.12–0.80) were associated with greater odds of SI. The indirect effects via PB (β = 0.33, 95% CI 0.09–0.61) and via TB (β = 0.22, 95% CI 0.05–0.45) were both significant. Parallel mediation models for the 2022 wave. (A) Composite IN model: loneliness → IN → suicidal ideation. (B) PB as mediator. (C) TB as mediator. Path coefficients are standardised regression coefficients (β) with bias‐corrected 95% confidence intervals. Monthly income and residential stability were controlled for as covariates. IN, interpersonal needs; Lon, loneliness; PB, perceived burdensomeness; SI, suicidal ideation; TB, thwarted belongingness. When PB and TB were combined as a composite IN construct, the indirect effect from loneliness to SI remained substantial and significant (β = 0.50, 95% CI 0.23–0.87), accounting for approximately 48% of the total effect. The direct path from loneliness to SI persisted after including IN, indicating partial mediation. Directionally similar mediation patterns were observed in 2017–2018, although the relative contributions differed: TB appeared relatively more salient pre‐pandemic, whereas PB dominated in 2022—a shift consistent with crisis‐related social restrictions and economic pressures that heighten appraisals of being a burden. Full wave‐specific mediation outputs are provided in Supporting Information S1: tables S3 and S4. INTERPRETATIONS AND IMPLICATIONS Across three serial cross‐sectional waves spanning pre‐pandemic and pandemic periods (2017–2022), we found that loneliness was robustly linked to SI among transgender women in China through stable interpersonal pathways. PB and TB consistently and partially mediated this association; prioritising the 2022 pandemic wave as the primary analysis, we observed markedly higher reported SI with strong indirect effects via PB and TB, which together accounted for approximately half of the total effect after covariate adjustment. Although reported ideation rose sharply during the pandemic wave—consistent with crisis‐related exacerbation—the underlying interpersonal mechanisms held, underscoring their central role in suicide vulnerability for this marginalised group. PB emerged as the dominant mediator during the pandemic. This extends prior syntheses positioning PB and TB as proximal determinants of suicidal desire and pathways through which social adversity confers risk. 12 Under system‐wide strain, loneliness may heighten self‐appraisals of being a liability to others in a chronically stigmatised population, fostering self‐blame and the belief that one's presence burdens family or community. This pattern accords with eusocial accounts of suicide, which describe PB as a maladaptive inference that self‐removal benefits close others when resources and reciprocity are disrupted. 13 TB provided a complementary pathway. Loneliness can sharpen vigilance to rejection and negative social cues, reinforcing perceived barriers to social embeddedness. 14 Although loneliness reflects dissatisfaction with relational realities, TB reflects meaning‐making around blocked belonging within one's social milieu. Temporal contrasts suggested TB's linkage to ideation was relatively stronger before the pandemic, whereas PB became more salient during the crisis period, aligning with evidence that PB can be especially proximal to suicidal desire when external stressors intensify. 15 These findings translate into concrete, durable targets for prevention. First, psychotherapies that explicitly reduce PB should be prioritised. 16 These include cognitive behavioural therapy (CBT) to restructure ‘I am a liability’ appraisals, behavioural experiments to challenge burden‐related predictions, structured problem‐solving to enhance mastery and efficacy and acceptance‐ or compassion‐based approaches (e.g., self‐compassion training) to mitigate shame and self‐criticism that fuel PB. Second, programmes that strengthen belongingness and counteract TB are essential. 17 These include peer‐led support groups for transgender women, family‐inclusive counselling to improve acceptance and communication, digitally mediated communities to sustain connection during mobility restrictions and co‐designed anti‐stigma initiatives that promote social inclusion and reduce TB. Together, these approaches offer actionable and sustainable directions for intervention. System enablers are essential to embed these interventions within routine care. Practical steps include routine PB/TB screening in community clinics, rapid referral pathways for acute suicide risk and integration with gender‐affirming and human immunodeficiency virus (HIV)‐related services to leverage trusted points of contact. Telehealth delivery should be standardised to maintain continuity during service disruptions, alongside investment in peer workforce development and culturally responsive training. Together, these components provide scalable infrastructure for a resilient suicide prevention architecture within Western Pacific health systems. Building on this, our findings indicate two durable targets: CBT that explicitly reduces PB and community programmes that strengthen belonging, integrated with gender‐affirming and HIV services and supported by routine PB/TB screening and telehealth. This study has limitations. The serial cross‐sectional design precludes causal inference and precise temporal ordering of mediators. SI measures differed across waves, constraining direct prevalence comparisons and possibly affecting effect sizes; we therefore emphasised pathway direction and relative contributions. Urban, community‐based snowball sampling may limit generalisability and underrepresent rural or less‐networked individuals. Self‐reported sensitive constructs are vulnerable to social desirability bias in stigma‐laden contexts, and modest sample sizes constrained subgroup analyses examining heterogeneity by age, income, migration or sexual orientation. Strengths include multi‐city sampling across distinct regions, repeated assessments spanning crisis periods, validated Chinese instruments with good internal consistency and a consistent mediation framework with bootstrap CIs. In summary, PB and TB are resilient interpersonal mechanisms linking loneliness to SI among Chinese transgender women, with PB particularly salient under crisis conditions. This indicates actionable priorities for prevention: therapies that reduce perceived burden and community programmes that strengthen belonging, supported by routine screening, rapid referral, service integration and telehealth. Future work should employ longitudinal designs to establish temporal ordering, test PB/TB‐focused intervention packages in pragmatic hybrid trials, evaluate scalability and cost‐effectiveness across urban–rural settings and refine implementation strategies in partnership with transgender communities to ensure cultural fit and sustained impact. AUTHOR CONTRIBUTIONS Yong Cai and Lulu Xu conceived and designed the study. Yong Cai, Ying Wang and Huachun Zou oversaw the project, supervised data acquisition, interpreted the results and provided critical feedback on the manuscript. Lulu Xu, Jin Xu and Fengyi Wang performed data analysis and drafted the initial version of the manuscript. Lulu Xu, Jin Xu, Fengyi Wang, Yawen Zheng, Shan Qiao and Yong Cai participated in revising the manuscript. Chen Xu, Shangbin Liu, Jiachen Lu, Ruijie Chang and Ying Wang supervised data collection and contributed to data interpretation. All authors reviewed the manuscript and approved the final submitted version. Lulu Xu and Yong Cai serve as the guarantors, accepting full responsibility for the overall content. FUNDING This work was supported by the Key Discipline Projects of Shanghai Three‐Year Action Plan for Public Health under Grant (GWVI‐11.1‐29), the Key Project of Medical–Engineering Interdisciplinary Programme, Shanghai Jiao Tong University (AF7320007, YG2025ZD29), Public Health Research Project from Shanghai Disease Control and Prevention Administration (2024GKM30), Science and Technology Commission of Shanghai Municipality (20JC1410204) and National Natural Science Foundation of China (71603166, 71673187). ETHICS STATEMENT The study was approved by the Ethics Committee of the School of Public Health, Shanghai Jiao Tong University (Protocol 2016022). Written informed consent was obtained from all participants prior to enrolment. PARTICIPANT CONSENT FOR PUBLICATION Consent was obtained directly from participant(s). Supporting information Supporting Information S1