We conducted this cross-sectional study from October 2022 to January 2023. We obtained approval from the Bioethics Committee at Wroclaw Medical University on 27 October 2022 (No. KB-777/2022). The study was part of a larger project, Not Alone in the Crowd, which addressed the issue of loneliness and was funded by the Student Activity Fund (FAST) provided by the City of Wroclaw. We employed a convenience sampling strategy. Our interviewers recruited participants in public locations across Wroclaw, including three shopping malls, two refugee assistance centers, the Department for Foreigners at the Lower Silesian Voivodeship Office, a university campus, two senior citizens’ clubs, and during a university lecture that formed part of the Not Alone in the Crowd project. We selected these sites based on their anticipated high traffic of both Ukrainian refugees and Polish citizens living in Wroclaw. We invited individuals at these sites to complete a paper-based questionnaire. In addition to direct recruitment at designated locations, some participants were recruited using the snowball sampling method, whereby initial respondents referred others within their social networks. All participants provided informed consent, which we included in the survey instructions. The researchers were on site and addressed all questions regarding the items in the psychometric scales in cases of any doubts.

The study included adult participants aged 18 years or older. The sample consisted of two groups: Polish citizens residing in Wroclaw and Ukrainian citizens who had lived in Ukraine prior to the outbreak of the war and had subsequently sought refuge in Wroclaw. The inclusion and exclusion criteria were defined separately for each group:

Polish citizens:

Ukrainian war refugees:

To ensure accurate identification of war refugee status, a Ukrainian-speaking co-author (RB) actively participated in participant recruitment within refugee-focused locations. During initial contact, she conducted a brief conversation in Ukrainian to confirm the participant’s background and refugee status. Following confirmation, she provided all study-related information both in writing and, if needed, orally in Ukrainian to ensure full comprehension and support informed consent. To minimize selection bias, interviewers were instructed to recruit participants with the greatest possible diversity in terms of visually identifiable characteristics such as age, gender, and ethnicity. They consciously avoided preferential selection based on appearance or perceived approachability. Additionally, interviewers changed their position within each recruitment site at regular intervals (e.g., within shopping malls or refugee centers) to reduce environmental and temporal biases. Due to the nature of the selected sites, including high foot traffic, dynamic interpersonal interactions, and variation in timing and staffing across locations, we did not systematically collect data on response rates.

We distributed the questionnaires in both Ukrainian and Polish language versions. A native Ukrainian-speaking author (RB) translated the Polish version of the questionnaire using psychometric tools, and an independent native speaker performed the back-translation. The survey collected data on sociodemographic characteristics, including age, sex, place of residence, citizenship, war refugee status, education level, occupation, receipt of social benefits, and personal or family psychiatric history. The following variables were assessed using binary (yes/no) questions: citizenship, war refugee status, receipt of social benefits, personal psychiatric history, and family psychiatric history. Other variables were assessed using categorical or continuous response formats: age (open-ended), sex (categorical), education level (primary/secondary/higher), and occupation (employed/unemployed). We also administered three standardized instruments: the GHQ-28 (35), the Revised UCLA Loneliness Scale (36), and the Courage Social Network Index (37).

The General Health Questionnaire (GHQ-28) is a widely recognized psychometric tool for screening and assessing current nonpsychotic psychopathology in adults (38). It primarily reflects symptoms of depression and related affective disturbances. The scale has been validated for use in Polish occupational health settings (35). It includes 28 items rated on a 4-point Likert scale, yielding a total score ranging from 0 to 84, with higher scores indicating greater psychological distress. A score of 24 or above is considered clinically significant. The GHQ-28 comprises four subscales, each containing seven items: Somatic Symptoms (Subscale A), Anxiety and Insomnia (Subscale B), Social Dysfunction (Subscale C), and Severe Depression (Subscale D). In the current sample, the GHQ-28 demonstrated high internal consistency (Cronbach’s α = 0.93). The reliability of individual subscales was also satisfactory: α = 0.82 for Somatic Symptoms, α = 0.85 for Anxiety and Insomnia, α = 0.87 for Social Dysfunction, and α = 0.90 for Severe Depression.

The Revised UCLA Loneliness Scale (R-UCLA) is based on the original version developed by Russel et al. (39). For this study, we used the Polish version validated by Kwiatkowska et al. (36). The scale consists of 20 items rated on a 4-point Likert scale: 1 = never, 2 = rarely, 3 = sometimes, and 4 = often. Following the recommendation of Kwiatkowska et al., we excluded item 4 (“I do not feel alone”) from the analysis due to its direct wording and poor correlation with the total score. As a result, the maximum total score was 76. The R-UCLA includes three conceptually distinct subscales: Belonging and Affiliation, Intimate Others, and Social Others, each contributing to the total loneliness score. We calculated the total score as the sum of responses (excluding item 4), yielding a possible range of 16 to 76 points. For analytical purposes, we categorized loneliness levels into four groups: low (16–30 points), moderate (31–45), moderately high (46–60), and high (61–76) (40, 41). Cronbach’s alpha indicated high internal consistency for the total R-UCLA score (α = 0.89) and acceptable reliability for the three subscales (α = 0.70, 0.83, and 0.80, respectively).

The Courage Social Network Index (CSNI) is a multidimensional instrument developed to assess both the structure and function of individuals’ social networks. This tool was validated in a large European study, including a Polish population, and showed good reliability and content validity (37). It evaluates three key functional dimensions across each structural component: the closeness of social ties, the perceived availability of general support, and the frequency of face-to-face contact. There are 8 questions in each element of the social network corresponding to 5 structural components (spouse or partner, parents, other family members, neighbors, friends and co-workers). The score was obtained by the Item Response Theory procedure and the results were interpreted as social network saturation, where 0 indicated the lowest level of saturation and 100 the highest.

The CSNI presented satisfactory internal consistency (α = 0.77 overall; α = 0.92, α = 0.86, α = 0.82, α = 0.76, α = 0.80, respectively, for five structural components).

Internal consistency of all instruments and their subscales was evaluated separately for the Ukrainian versions used to assess the refugee group and the Polish versions administered to the Polish residents of Wroclaw. The reliability of all tools was satisfactory across both language groups. Cronbach’s alpha coefficients are summarized in Table 1.

Descriptive statistics are reported as means ± standard deviations (SD), absolute frequencies, and percentages. Little’s MCAR test indicated that missing data were missing completely at random (p = 0.411). Therefore, missing values were imputed using the Expectation–Maximization (EM) algorithm, applied at the subscale level for each questionnaire. Sum scores were calculated to evaluate general health (GHQ-28), loneliness (R-UCLA), and weighted social network saturation (COURAGE-Social Network Index; see below). The Shapiro–Wilk test was used to determine the normality of data distribution, and the Levene’s test was applied to assess homogeneity of variances. A chi-square test (or Fisher’s exact test, where appropriate) with Bonferroni correction was used to compare categorical variables (42). Differences between groups were tested using the Kruskal–Wallis test and one-way analysis of variance (ANOVA) when more than two groups were compared. For two-group comparisons, the independent samples t-test or Mann–Whitney U-test was used, as appropriate. Multiple linear regression was used to analyze potential associations between dependent variables: depressive symptoms (GHQ-28), loneliness (R-UCLA), and multiple independent variables. Dichotomous variables were dummy coded. The statistical significance level was set at p < 0.05. All analyses were performed using IBM SPSS Statistics, Version 27.

We applied Item Response Theory (IRT) modeling to evaluate latent traits within individuals across five distinct components of the COURAGE-SNI framework: spouse/partner, parents, other family members, neighbors, friends/co-workers. Following the methodology proposed by Zawisza et al. (37), we conducted all analyses in R (version 4.2.2),¹ using the mirt package (version 1.41) for IRT modeling and tidyverse package (version 2.0.0) for data manipulation. To estimate individual latent trait positions, we used the Generalized Partial Credit Model (GPCM), developed by Muraki (43). This model produced factor scores that captured each participant’s placement along the latent trait continuum. We then standardized these scores within each structural component to compute a global social network score, expressed on a 0 to 100% scale. Consistent with Zawisza et al.’s findings, we observed that varying structural components contribute differentially to the assessment of the total score for the COURAGE-SNI. Consequently, we derived Total Component Information (TCI) from the summary of values provided by the Item Information Function (IIF) (44) for items across all components. These TCIs served as weights in the calculation of the weighted total Social Network Score (WTSNS), thereby facilitating the determination of the COURAGE-SNI total score for each study participant.

The survey included 209 participants: 68 Ukrainian refugees and 141 Polish citizens residing in Wroclaw. Women constituted the majority of the sample (n = 149; 71.29%). The mean age of all participants was 40.22 ± 18.76 years. Ukrainian refugees were significantly older than Polish citizens (44.01 ± 15.58 vs. 38.38 ± 19.91 years; Mann–Whitney U test: Z = −2.925, p = 0.003). Regarding education, 95 participants (45.45%) reported having higher education. Tables 2, 3 present the demographic characteristics and comparisons on depression and loneliness measures for Ukrainian refugees and Polish citizens, respectively.

With respect to sociodemographic comparisons, we found no significant differences in gender distribution between Ukrainian refugees and Polish citizens. However, the groups differed in educational attainment: 56.03% of Polish participants had completed secondary education (vs. 36.76% of Ukrainians; p = 0.027), while 60.29% of Ukrainian participants held higher education degrees (vs. 38.30% of Poles; p = 0.008). A larger proportion of Ukrainian participants were unemployed (52.94%) compared to Polish citizens (37.59%). The chi-square test indicated a significant difference (p = 0.035), but this result did not remain significant after Bonferroni correction (adjusted p = 0.071). In addition, 30.88% of Ukrainian refugees reported receiving social benefits, compared to 11.35% of Polish participants (p < 0.001). Polish participants more frequently reported a personal history of psychiatric disorders (22.70% vs. 2.94%, p < 0.001) and a family history of psychiatric conditions (36.17% vs. 1.47%, p < 0.001). Table 4 summarizes these between-group sociodemographic differences.

Based on the GHQ-28 analysis, 121 participants (57.89%) scored above the clinical cut-off for psychological distress (Table 4). Ukrainian refugees reported significantly higher total GHQ-28 scores than Polish citizens. Among refugees, 51 individuals (75%) exceeded the clinical threshold, with a mean score of 32.73 ± 12.20 (p < 0.01; Tables 4, 5). In relation to GHQ-28 subscales, Ukrainian refugees scored significantly higher on Somatic Symptoms and Anxiety/Insomnia, with mean scores of 10.64 ± 4.22 and 9.88 ± 4.51, respectively (p < 0.001 and p < 0.003). We found no significant differences in GHQ-28 total scores between males and females in either group. Among Polish citizens, employed participants had significantly higher GHQ-28 scores than those unemployed (p = 0.049). Polish participants with a psychiatric history and those with a family history of psychiatric disorders also reported significantly higher scores (p < 0.001 and p = 0.003, respectively). In the Ukrainian refugee group, individuals with a psychiatric history scored significantly higher than those without (p = 0.03). Associations between GHQ-28 scores and sociodemographic characteristics are presented in Table 2 for Ukrainian refugees and in Table 3 for Polish citizens.

We found no statistically significant differences in either the categorical levels of loneliness (Table 4), the total R-UCLA score, or any of the subscale scores between Polish citizens and Ukrainian refugees (Table 5). One Polish participant had a total R-UCLA score of 19, which fell outside the defined cutoff ranges for loneliness classification. As a result, we excluded this individual from the chi-square test of independence. Among Polish citizens, those with a positive psychiatric history reported significantly higher levels of loneliness compared to those without (p < 0.001). Associations between R-UCLA scores and sociodemographic variables are presented in Table 2 for Ukrainian refugees and Table 3 for Polish citizens.

Ukrainian refugees demonstrated significantly higher mean total scores on the CSNI compared to Polish citizens (p = 0.001). Among the five structural components, only the “other family members” domain showed a significant difference between the groups, with refugees scoring higher than Polish participants (p = 0.001). We found no significant group differences in the remaining components: spouse/partner, parents, neighbors, or friends/co-workers. Table 5 presents the mean total CSNI scores and the mean scores for each of the five components in both groups.

We observed a strong positive correlation between the GHQ-28 total score and the R-UCLA total score (p < 0.01, r = 0.502). The GHQ-28 total score was also significantly associated with all R-UCLA subscale scores (p < 0.01; r = 0.521, 0.389, and 0.349, respectively). Higher loneliness scores (R-UCLA total) were moderately and inversely correlated with the CSNI total score (p < 0.01, r = −0.303). Similar negative correlations were found between R-UCLA subscales and the CSNI total score (p < 0.01; r = −0.265, −0.358, and −0.195, respectively). Age was positively associated with the CSNI total score (p < 0.01, r = 0.305), and with the “Other family members” and “Neighbors” subscales (p < 0.01, r = 0.665 for both). In contrast, age was negatively correlated with the “Parents” (p < 0.01, r = −0.577) and “Friends and Co-workers” subscales (p < 0.05, r = −0.137). We also observed negative correlations between age and the GHQ-28 total score (p < 0.05, r = −0.146) and the Severe Depression subscale (p < 0.01, r = −0.185). Table 6 presents all correlation coefficients among GHQ-28, R-UCLA, and CSNI scores, as well as their associations with age, calculated across all participants. See Supplementary Tables S1, S2 in the appendix for group-specific correlations.

We conducted two multiple linear regression models to examine predictors of GHQ-28 and R-UCLA total scores. Dichotomous variables were dummy-coded as follows: citizenship (0 = Polish citizen of Wroclaw, 1 = Ukrainian war migrant) and psychiatric history (0 = no psychiatric history, 1 = presence of psychiatric history). Both models were statistically significant based on F-tests [GHQ-28: F(5, 203) = 27.45, p < 0.001; R-UCLA: F(5, 203) = 26.12, p < 0.001], indicating that the included predictors explained a meaningful proportion of variance in the outcomes. The GHQ-28 model explained 38.9% of the variance (Adjusted R² = 0.389), while the R-UCLA model accounted for 37.6% (Adjusted R² = 0.376), suggesting acceptable model fit. In the GHQ-28 model, higher scores were associated with being a war migrant (β = 0.225), greater loneliness (β = 0.516), younger age (β = −0.169), and the presence of psychiatric history (β = 0.217). In the R-UCLA model, higher loneliness scores were associated with higher depressive symptoms (β = 0.526), lower social network scores (β = −0.285), and older age (β = 0.148). We excluded sex, psychiatric history in the family, employment status, and receipt of social benefits from the final models due to a lack of statistically significant associations with either dependent variable.

Table 7 presents the regression coefficients and model fit indices for both models.

Given the observed links between loneliness, social networks, and psychological distress, our results point to specific intervention priorities for supporting the mental health of Ukrainian war migrants. The strong association between loneliness and psychological distress suggests that reducing perceived social isolation may be a key target for improving well-being. At the same time, more developed social networks among refugees, particularly involving extended family, did not appear sufficient to mitigate psychological distress. This highlights that while bonding social capital may reduce loneliness, it does not necessarily protect against broader psychological consequences of displacement. Therefore, interventions should not only maintain existing social ties, but also enhance the perceived quality and supportiveness of those relationships, especially for individuals who may appear socially connected but still experience emotional distress. Zabłocka-Żytka et al. recommend actions such as disseminating reliable mental health information, improving access to diagnosis and treatment, and fostering cross-sectoral and institutional collaboration to support both refugees and those involved in their care (73). To provide effective support, mental health care for refugees should be embedded within a broader framework that addresses basic needs, ensures access to essential services, and reduces the risk of long-term psychological consequences. Establishing links with culturally and linguistically appropriate services across voluntary, charitable, and public sectors is essential (74, 75). Expanding access to basic mental health first aid by training non-medical personnel such as transport workers and caseworkers may help foster a supportive and trauma-informed environment. These individuals can also be trained to recognize common signs of distress requiring professional attention. Moreover, developing clear and accessible pathways to primary care and mental health services, including financial support where needed, is crucial for addressing war-related trauma among Ukrainian refugees (76). These actions correspond closely to the goals of social capital-based interventions, as described in prior studies (23, 30, 31). In particular, initiatives aimed at fostering trust, reciprocity, and social connectedness have been shown to reduce psychological distress, especially when they promote active participation in community life and meaningful interpersonal relationships. Unlike structural reforms alone, these approaches address the relational dimension of refugee well-being by supporting both bonding ties within families and communities, and bridging and linking connections with the host society and public institutions. Interventions such as peer support programs, culturally adapted group activities, and refugee-led networks may be particularly effective in enhancing social participation and resilience. Coordinated efforts between municipalities, NGOs, and grassroots actors can facilitate not only access to services but also the sense of belonging and agency that underpins cognitive and structural social capital. By focusing on the quality of social relationships and inclusion in communal life, these strategies complement formal systems of care and contribute to long-term psychological recovery. However, recent studies have highlighted a persistent mismatch between the needs of refugees and the actual support provided in Poland, both in the short and long term (77). Two major barriers to effective healthcare delivery include the lack of prior medical documentation and significant language obstacles, which complicate communication between refugees and medical personnel (78). While targeted systemic solutions could reduce the daily workload of healthcare providers, no comprehensive governmental strategy has been implemented to ensure sustainable support or integration. As a result, the prolonged inadequacy of state-level assistance contributes to negative social attitudes and growing tensions between host communities and refugee populations (77, 78).

Several limitations should be considered when interpreting the findings of this study. First, the cross-sectional design limits the ability to draw causal inferences or assess changes over time. Mental health status, loneliness, and social network structure are likely to evolve depending on factors such as length of stay and stage of integration. Longitudinal research would provide a more comprehensive understanding of these dynamics. Second, the study relied on self-reported measures of psychological distress, loneliness, and social network characteristics, which may be subject to social desirability or recall bias. Additionally, the Ukrainian version of the survey was self-translated. Although care was taken to ensure linguistic accuracy, replication in other Ukrainian samples using validated instruments is needed to confirm the reliability and clarity of the measures.

Third, the sampling method may have introduced bias in terms of representativeness. Due to the character of the project, which focused on direct engagement with participants, building interpersonal contact, and including individuals who might be excluded from online studies, we employed a convenience sampling strategy. Participants were recruited in public spaces and refugee assistance centers across Wroclaw, enabling real-life data collection. Although this approach aligned with the project’s goals and allowed for the inclusion of vulnerable populations, it may have limited sample diversity and generalizability. However, several measures were taken to reduce sampling bias, such as deliberately approaching individuals with diverse visible characteristics and regularly changing the interviewers’ positions within each site. Future studies should consider using randomized or stratified sampling techniques to further enhance representativeness and external validity.

Fourth, while elements of our results were discussed in light of the stress-buffering hypothesis and social capital theory, this was an exploratory, interpretative approach rather than a primary objective of the study. Our analyses were not based on instruments specifically designed to assess these theoretical frameworks. Although the CSNI and R-UCLA scales provide relevant insight into social connectedness and perceived support, future studies should incorporate validated tools that directly measure different dimensions of social capital and stress-buffering mechanisms. This would enable more precise theoretical testing and facilitate cross-study comparison.

Fifth, although several sociodemographic and clinical variables such as education level, receipt of social benefits, and personal or family history of psychiatric disorders significantly differed between the two groups, they did not show statistically significant associations with psychological distress or loneliness in our multiple regression models and were therefore excluded. Nevertheless, these between-group differences may have introduced residual confounding that could affect the results obtained from psychometric scales. For example, although personal psychiatric history was significantly associated with psychological distress in the final regression model, it was reported much less frequently by Ukrainian refugees than by Polish citizens. This discrepancy may reflect underdiagnosis, stigma, or limited prior access to mental health care. As a result, the observed group difference in psychological distress may underestimate the true burden among refugees and should be interpreted with caution.

Finally, the total sample size in our study was relatively small (N = 209), consisting of 68 Ukrainian refugees and 141 Polish citizens. Although this number allowed for valid statistical analyses, including regression models, it may limit the detection of more subtle effects and reduce generalizability. One of the reasons for the modest sample size was our methodological decision to limit recruitment to in-person, field-based assessments using paper-and-pencil questionnaires. This approach was not incidental but stemmed from the specific character of the broader project, which aimed not only to collect data but also to foster human connection, establish trust, and promote engagement, particularly with vulnerable populations such as refugees. By conducting surveys in person, we sought to reduce barriers to participation and create an inclusive environment in which participants could ask questions and receive clarification when needed. Establishing direct interpersonal contact was a deliberate element of our research design. Compared to online surveys, in-person data collection may offer several advantages. It allows for better sampling control, improves comprehension of questionnaire items, enhances participant motivation, and reduces the risk of inattentive or random responses. It also enables the inclusion of participants who might be excluded from online research due to lack of internet access, digital literacy, or technological trust. This is especially relevant in refugee populations, who often experience digital exclusion. Moreover, the presence of the researcher fosters engagement and allows for contextual observations that enrich data quality. These strengths have been emphasized in a previous analysis comparing field-based and online studies (79). Although our strategy may have contributed to a smaller sample size and reduced scalability, it helped ensure higher data reliability and greater inclusivity. Future studies may consider a mixed-mode approach to balance reach and data quality.