Modeling the PHQ-15: The factor structure of somatic symptoms in a large community sample

Abstract

The Patient Health Questionnaire-15 (PHQ-15) is widely used to assess somatic symptom burden, but its latent structure remains uncertain. This study tested competing PHQ-15 factor models and their associations with demographic and psychological variables in a large, nationally representative UK sample (N = 1405). Confirmatory factor analyzes using WLSMV evaluated alternative structures reported in the literature. Several multifactor models showed acceptable fit, but a four-factor model comprising pain, cardiopulmonary, gastrointestinal, and fatigue domains provided the best overall fit and outperformed a one-factor solution. Inter-factor correlations were high, indicating substantial overlap between domains and supporting the utility of the total PHQ-15 score. Males reported slightly lower symptom burdens, and younger adults reported more somatic complaints. Findings support multidimensionality alongside a strong general somatic distress tendency, though the PHQ-15 does not fully align with ICD-11 bodily distress disorder criteria.

Introduction

Somatization describes the manifestation of psychological distress through physical symptoms (Stekel, 1924). The concept was advanced within psychosomatic medicine, which highlighted the interaction between emotional and physical processes (Alexander, 1950). Diagnostic criteria first appeared in the DSM-III (American Psychiatric Association, 1980), identifying somatic complaints lacking medical explanation. However, this symptom-based model faced criticism (De Figueiredo et al., 1980; Verhaest and Pierloot, 1980), leading to broader recognition of psychosocial influences (Lipowski, 1988). The ICD-10 (World Health Organization [WHO], 1992) included Somatoform Disorders, while the DSM-IV-TR (American Psychiatric Association, 2000) emphasized functional impairment. To address earlier limitations, the DSM-5 (American Psychiatric Association, 2013) introduced Somatic Symptom and Related Disorders, focusing on distress, preoccupation, and maladaptive behaviors. The ICD-11 (WHO, 2019) similarly defined Bodily Distress Disorder, highlighting persistent somatic symptoms with excessive attention and distress. These revisions align with Engel’s (1977) biopsychosocial model and Lipowski’s (1988) contextual approach to illness. Supporting this integrated view, biological studies have linked somatic disorders to genetic variants (Zannas and Chrousos, 2017), altered brain function (Borsook et al., 2012), and dysregulated stress systems (Wingenfeld et al., 2008). Psychological contributors include negative affectivity, maladaptive coping, and early trauma (Asmundson et al., 2010; Barsky et al., 2001; Henningsen et al., 2003). Cultural values (Kirmayer, 2004), limited social support (Lixia et al., 2022; Muntingh et al., 2016), and structural inequalities (Creed et al., 2012; Hinz et al., 2017; Konnopka et al., 2013) further influence symptom expression and reporting.

The Patient Health Questionnaire-15 (PHQ-15; Kroenke et al., 2002) is the most widely used measure of somatization, aligning with the DSM-5′s emphasis on distress/impairment and the ICD-11′s focus on persistent bodily symptoms (Terluin et al., 2022). It comprises 15 common physical complaints (Liu et al., 1997), each rated on a three-point scale (0–2) assessing the degree of bother over the past 4 weeks, producing a total score from 0 to 30. Although originally designed to assess somatic symptom severity, it has also been used to examine physical health problems and multiple mental health conditions (Grover et al., 2012; Jeong et al., 2014; Na et al., 2007; Shi et al., 2023; Yoon et al., 2014). PHQ-15 scores are strongly associated with depression, with higher somatic distress often correlating with greater depressive severity (Grover et al., 2012; Shi et al., 2023). There is also significant overlap with PTSD, where physical symptoms predict post-traumatic stress outcomes (Paredes-Echeverri et al., 2022; Soumoff et al., 2022; Sullivan et al., 2016). Somatic distress has also been linked to attachment anxiety and childhood trauma (Barends et al., 2022; Le et al., 2021; Sun et al., 2022). Distinguishing PHQ-15 subfactors is theoretically and practically important because different symptom clusters may reflect partially distinct mechanisms and have different implications for research and applied use. In clinical and public health settings, a total score can index overall somatic burden, but subscale profiles can clarify whether complaints are primarily pain-related, gastrointestinal, cardiopulmonary, or fatigue-related, which may support more precise communication, screening, and hypothesis testing. Subscales can also improve comparability across studies by clarifying whether associations with external variables are driven by broad somatic distress or by specific domains, and whether the same pattern replicates across populations and contexts. For these reasons, establishing the most defensible latent structure is necessary before interpreting domain-level scores or using them in downstream analyses.

However, factor-analytic studies of the PHQ-15 have yielded inconsistent findings regarding its dimensional structure. Schlechter et al. (2023) identified four factors pain, gastrointestinal, cardiopulmonary, and fatigue each strongly associated with depression and anxiety. Leonhart et al. (2018) also supported a four-factor solution, though model fit varied by cultural context, with some samples supporting a dominant single factor. Bifactor models (Witthöft et al., 2013) have converged on similar structures, although item loadings vary (Leonhart et al., 2018; Schlechter et al., 2023). For example, Terluin et al. (2022) labeled one factor musculoskeletal rather than pain, and Zhou et al. (2020) placed the menstrual cramps item within the pain domain. Exploratory analyses typically find three- or four-factor solutions (Kliem et al., 2014; Liao et al., 2016; Zhang et al., 2016), while Comellas et al. (2015) supported a multidimensional model. Overall, cultural variation and item-specific inconsistencies suggest the PHQ-15’s dimensional structure is multifaceted and context-dependent.

This study uses Wave 7 data from the COVID-19 Psychological Research Consortium Study (C19PRC-UK; McBride et al., 2022), including 1405 participants representative of the UK population. Because the pandemic was associated with widespread disruption to physical and mental health (O’Mahony et al., 2022; Ran et al., 2020), these data offer a valuable context to examine the PHQ-15’s latent structure under heightened psychosocial stress. Using this large, diverse sample, we tested whether PHQ-15 items replicate previously proposed multifactor structures and clarified how somatic symptoms cluster in the general population. Accordingly, we compared competing factor-analytic models of the PHQ-15 and examined whether domain scores show differentiated associations with demographic characteristics and psychological distress beyond the total score.

This study tested three hypotheses about correlates of somatic symptom reporting. First, we expected women to report higher somatic symptom levels than men, consistent with evidence of more frequent and intense complaints among women, potentially linked to symptom perception, healthcare-seeking behavior, and psychosocial stress exposure (Atasoy et al., 2022; Ladwig et al., 2001). Second, we expected younger adults to report more symptoms than older adults, as somatic complaints appear more strongly related to psychological distress than to age-related physical decline (Hilderink et al., 2013). Third, we expected higher somatic symptom levels to correlate with psychological distress, including PTSD, sleep disturbances, GAD, and depression. PTSD is commonly accompanied by somatic symptoms (Gupta, 2013; Zhang et al., 2015), sleep disturbance often co-occurs with and may exacerbate somatic distress (Ionescu et al., 2021; Jiang et al., 2017), and robust links with GAD and depression likely reflect shared physiological and affective mechanisms (Bekhuis et al., 2016; Hoehn-Saric et al., 2004).

Method

Participants and design

This study drew on data from Wave 7 of the COVID-19 Psychological Research Consortium Study (C19PRC-UK-7), a longitudinal investigation of the pandemic’s psychosocial impacts (McBride et al., 2022). Full details on sampling and data collection are described elsewhere (McBride et al., 2020). Briefly, Wave 1 recruited 2025 adults (18+ years) via Qualtrics, using quota sampling for age, gender, and household income. By Wave 7, a total of 1405 participants completed the survey. All respondents were recontacts from earlier waves of the study, including the original baseline sample and top-up samples recruited during Waves 3, 4, and 6. Gender was assessed via self-report (“What is your gender?”: Male, Female, Transgender, Prefer not to say, Other); in the present analytic sample, two participants selected the “Transgender” response option and one selected “Prefer not to say”; due to very small cell sizes, analyses focus on male and female categories only. Of these, 49.8% identified as female (n = 702) and 50.2% as male (n = 700). We report and interpret results using the gender variable available in this dataset, and we use SAGER guidance to frame reporting and limitations where applicable (Heidari et al., 2016). Participants ranged in age from 18 to 65+ years, with a mean age of 48.85 years. The most common age groups were 45–54 (22.6%), 55–64 (19.6%), and 35–44 (18.2%). A full sociodemographic breakdown is available in Supplemental Table 2.

Procedures

Data collection was managed by Qualtrics, which partnered with over 20 online panel providers to recruit participants. Anonymized data were obtained from the C19PRC-UKW-7 repository in accordance with approved data use guidelines. The dataset was examined for missing values and outliers; where applicable, listwise deletion or mean/median imputation was applied. Distributions of the 15 PHQ-15 items were assessed for normality, and the menstruation item was excluded to ensure applicability across the full sample. Ethical approval for the study was granted by the University of Sheffield Research Ethics Committee (Reference number: 033759). Participants provided informed electronic consent prior to participation, were advised of their right to withdraw, and received debriefing information, including publicly available health resources. Personal identifiers were removed prior to deposit in secure open-access repositories.

Measures

Patient health questionnaire-15

The PHQ-15 (Kroenke et al., 2002) consists of 15 common physical complaints (Liu et al., 1997) rated for frequency over the past 4 weeks on a three-point scale from 0 (“not bothered at all”) to 2 (“bothered a lot”). Total scores range from 0 to 30, with higher scores indicating greater somatization. This measure aligns with DSM-5′s emphasis on distress/impairment and the ICD-11′s focus on persistent somatic symptoms (Terluin et al., 2022). It exhibits good internal consistency (Cronbach’s α = 0.80; Cao et al., 2022; Sitnikova et al., 2017) and correlates strongly (r = 0.74) with the Somatic Symptom Scale-8 (Gierk et al., 2014). In this study, the 14-item version was used, with the menstruation item removed so that all participants could complete the measure.

Patient health questionnaire-9

The PHQ-9 (PHQ-9; Kroenke et al., 2001) is a nine-item self-report measure assessing depressive symptoms over the past 2 weeks. Items are rated from 0 (“not at all”) to 3 (“nearly every day”), with total scores ranging from 0 to 27. It shows excellent internal consistency (α ≈ 0.89) and strong test–retest reliability (r = 0.84; Kroenke et al., 2001).

Generalized anxiety disorder scale-7

The GAD-7 (Spitzer et al., 2006) assesses core symptoms of generalized anxiety over the past 2 weeks using 7 items scored from 0 to 3. Total scores range from 0 to 21, with higher scores indicating greater anxiety severity. The scale demonstrates excellent internal consistency (α = 0.92) and strong validity (Spitzer et al., 2006).

International trauma questionnaire

The ITQ (Cloitre et al., 2021) includes 18 items assessing PTSD and complex PTSD symptoms in line with ICD-11 criteria. Symptoms are rated from 0 (“not at all”) to 4 (“extremely”), capturing both core PTSD and disturbances in self-organization. It has strong construct validity and internal consistency (Cloitre et al., 2021).

Loneliness scale

This brief scale assesses perceived social isolation through three items rated on a three-point scale (Hughes et al., 2004). Total scores range from 3 to 9, with higher scores indicating greater loneliness. It has acceptable internal consistency (α ≈ 0.72) and correlates well with longer loneliness measures (Hughes et al., 2004).

Sleep disorders symptoms checklist-17

The SDS-CL-17 is a 17-item checklist screening for six common sleep disorders over the past year, including insomnia, apnea, and narcolepsy. Items are rated by frequency and grouped by disorder subtype. The measure has demonstrated good diagnostic accuracy and validity (Klingman et al., 2017).

Health service use scale

This custom measure was created from three items assessing healthcare use since the pandemic in the C19PRC-UKW-7 questionnaire, with yes/no questions about attending outpatient services, inpatient stays, or GP visits. Items are scored 0 (“No”) or 1 (“Yes”), with total scores from 0 to 3. Though not validated, it provides a brief index of recent health service use.

Data analysis

Analyzes proceeded in three steps. First, descriptive statistics were computed for all PHQ-15 items. The menstruation item was excluded so that all participants could be included in analyses. Second, we tested competing PHQ-15 factor structures using confirmatory factor analysis (CFA), specifying models previously reported in the literature. Third, to examine external validity, we estimated associations between PHQ-15 total and domain scores and relevant demographic and psychological variables.

CFAs were estimated in Mplus version 8.1 (Muthén and Muthén, 2017) using WLSMV with THETA parameterization and a probit link, treating the 14 items as ordinal indicators. Missing data on PHQ items were handled using the WLSMV pairwise present approach for polychoric correlations (Mplus default for categorical outcomes). Candidate models were drawn from published studies (2013–2022) identified via a targeted peer-reviewed search and retained where item to factor mappings were clearly reported for the 14-item set (see Supplemental Table 3). Models originally derived via EFA were re-specified as CFA models for comparability; any cross-loading item reported in the source study was assigned to its strongest factor in the present specification. No EFA was conducted in the current dataset.

Model fit was evaluated using χ² (interpreted cautiously given sample size; Tanaka, 1987), CFI, TLI, RMSEA with 90% CI, and SRMR. We report both CFI and TLI for completeness, but prioritized RMSEA and SRMR alongside CFI (RMSEA ⩽ 0.06 and SRMR ⩽ 0.08 indicating good fit; CFI ⩾ 0.95 good, ⩾0.90 acceptable). Because WLSMV is not likelihood-based, information criteria such as BIC were not used for model comparison. Supplemental Table 6 presents maximum likelihood CFA results for comparison only; all substantive conclusions are based on the WLSMV analyses reported in the main text.

As an additional sensitivity analysis, a bifactor CFA was also specified in which all items loaded on a general somatic factor alongside the domain-specific factors, with the general and specific factors specified as orthogonal. Local fit was examined by reviewing the largest standardized residuals and modification indices. Descriptive statistics and group comparisons (e.g. t tests and ANOVAs) were conducted in SPSS (v27), whereas all CFA models were estimated in Mplus, as indicated in the relevant table captions.

Results

PHQ-14 item responses

After excluding the menstruation item, response patterns for the remaining 14 PHQ items were analyzed. Trouble sleeping (14.2%) and feeling tired (14.0%) were the most frequently endorsed symptoms in the “bothered a lot” category, followed by pain in the arms, legs, or joints (10.8%) and back pain (9.2%). These findings suggest that sleep disturbance, fatigue, and musculoskeletal complaints were the most prominent issues in this sample. In contrast, fainting spells showed the lowest endorsement, with 90.8% of participants reporting they were “not bothered at all,” and a mean score of just 0.11 (SD = 0.38). Mean item scores ranged from 0.11 (fainting spells) to 0.61 (trouble sleeping), indicating notable variability in symptom severity. Items related to somatic pain (e.g. pain in arms, back pain, and headaches) also showed moderate endorsement, while cardiopulmonary symptoms (e.g. chest pain, shortness of breath) and sexual problems were among the least frequently reported. Full item-level response distributions are provided in, Supplemental Table 7.

Fit statistics

Supplemental Table 1 presents model fit indices for all candidate PHQ-14 structures estimated using WLSMV. Overall, several multifactor models showed excellent fit. The four-factor model proposed by (Schlechter et al., 2023), which is equivalent in item assignment to Witthöft et al. (2013), provided the strongest overall balance of fit and interpretability. For this model, χ²(59) = 328.56, p < 0.001, CFI = 0.982, TLI = 0.976, RMSEA = 0.057 (90% CI (0.051, 0.063)), and SRMR = 0.038. Other multifactor solutions, including the four-factor models reported by Zhou et al. (2020) and Leonhart et al. (2018), also demonstrated similarly strong global fit, although their factor compositions differed. In contrast, the one-factor model showed weaker absolute fit, χ²(77) = 656.74, p < 0.001, CFI = 0.966, TLI = 0.960, RMSEA = 0.073 (90% CI (0.068, 0.078)), SRMR = 0.059.

WLSMV is not likelihood-based, so information criteria such as BIC were not used for model selection. Model comparisons were therefore based on CFI, RMSEA, and SRMR alongside theoretical coherence and interpretability. On this basis, the equivalent four-factor structure reported by (Schlechter et al., 2023; Witthöft et al., 2013) was retained as the preferred model. Local fit was inspected by reviewing residual correlations and modification indices; residual correlations ranged from −0.12 to 0.10, and the largest modification index was 90.88. No post hoc error covariances or cross-loadings were added, in order to preserve parsimony and maintain comparability with the published factor structures.

As a sensitivity analysis, a bifactor CFA was specified with a general somatic factor alongside four orthogonal domain-specific factors corresponding to the retained four-factor domains (Schlechter et al., 2023; Witthöft et al., 2013). This bifactor model did not yield an admissible solution under WLSMV estimation, with Mplus indicating a non-identification problem involving the two-indicator fatigue-specific factor.

Factor loadings

Supplemental Table 8 presents standardized factor loadings and factor inter-correlations for the best-fitting model, the four-factor CFA solution (Schlechter et al., 2023; Witthöft et al., 2013). All items showed statistically significant and substantial loadings on their intended factors. For example, gastrointestinal symptoms such as “Stomach pain” and “Nausea or indigestion” loaded at 0.83 and 0.85, respectively, on the Gastro factor, and even the PHQ’s less frequent symptom “Pain or problems during sexual intercourse” loaded strongly (0.83) on this factor. Fatigue-related items were also well represented: “Feeling tired or having low energy” had one of the highest loadings in the model (0.89 on the Fatigue factor), and “Trouble sleeping” loaded at 0.77. The Pain factor was defined by somatic pain complaints such as “Back pain” (0.74) and “Headaches” (0.78), while the Cardiopulmonary factor showed robust loadings from “Chest pain” (0.89) and “Heart pounding or racing” (0.87), among other items.

Inter-factor correlations were high, ranging from approximately 0.80 to 0.91 (Supplemental Table 8). The strongest associations were observed between the Gastro and Cardiopulmonary factors (r = 0.91) and between Pain and Cardiopulmonary (r = 0.89), indicating substantial overlap between those symptom domains. Even the lowest correlation (Gastro with Fatigue, r = 0.80) remained high. Overall, this pattern suggests that, although the four domains are empirically distinguishable, they are not independent, and a strong general somatic distress tendency likely underlies the PHQ-15 symptom clusters.

Subscale demographic scores

Supplemental Table 4 presents PHQ-15 subscale scores by gender and age group, alongside associated test statistics and effect sizes. Due to very small cell sizes for gender-minority categories (Transgender n = 2; Prefer not to say n = 1), subgroup comparisons are reported for male and female respondents only. Gender differences were statistically significant although small, with females reporting higher scores across all domains most notably for fatigue (test statistic = 8.51, effect size = 0.018). Age-related differences were more pronounced: younger adults (18–24) consistently reported higher symptom levels compared to older adults (65+), particularly for cardiopulmonary symptoms (test statistic = 13.93, effect size = 0.47), indicating a moderate to large effect.

Bivariate correlations

Supplemental Table 5 displays bivariate correlations between PHQ-15 subscale scores and psychological variables. Depression, GAD, PTSD, sleep disturbance, loneliness, and health service use were all significantly associated with greater somatic symptom severity across subscales and the total score. The strongest correlations were observed between depression and fatigue (r = 0.66, p < 0.001), and between total PHQ-15 scores and depression (r = 0.70, p < 0.001). However, this association should be interpreted cautiously because the PHQ-9 includes items assessing sleep disturbance and low energy/fatigue (items 3 and 4), which overlap directly with the PHQ-15 fatigue domain (items 14 and 15: trouble sleeping and feeling tired). This shared symptom content may inflate the observed correlation. Correlations between GAD and sleep disturbance were also strong (both r = 0.654, p < 0.001), supporting convergent validity.

Discussion

The primary aim of this study was to test alternative factor analytic models of the PHQ-15 and identify the best-fitting structure using data from a large, nationally representative sample in Wave 7 of the COVID-19 Psychological Research Consortium Study (C19PRC-UK; McBride et al., 2022). A secondary aim was to describe the distribution of subscale scores across gender and age groups and to examine the unique predictive validity of each subscale in relation to key psychological and health-related variables. Our descriptive data showed a well-balanced gender distribution and a diverse age range, enhancing the applicability of our findings to various sociodemographic groups.

Consistent with the findings of Schlechter et al. (2023) and Witthöft et al. (2013) findings, our analyses showed that a four-factor model (with Pain, Cardiopulmonary, Gastrointestinal, and Fatigue dimensions) provided the best fit to the PHQ-15 data. The clear superiority of this multifactor solution, coupled with the only moderate fit of a one-factor model (which had a higher RMSEA and worse misfit), indicates that somatic symptom complaints are better represented by multiple specific factors rather than a single general factor. Notably, when we attempted to explicitly model a general somatic factor alongside the four specific factors (bifactor model), the solution did not properly converge, likely due to the instability caused by the two-item Fatigue factor. This empirical issue underscores the difficulty in partitioning PHQ-15 variance into distinct orthogonal components, even though the high first-order factor correlations suggest a strong common somatic distress factor underpinning all items.

Findings from previous studies support the interpretation that a dominant general factor may underlie PHQ-15 responses. Terluin et al. (2022) reported that over 85% of total score variance was attributable to a general factor, indicating essential unidimensionality. Similarly, Leonhart et al. (2018) found that the PHQ-15 subscale structure failed to replicate in Chinese samples, suggesting that multidimensional models developed in European contexts may lack cross-cultural generalizability. Witthöft et al. (2013) also demonstrated strong general factor loadings in a bifactor structure, although formal indices of unidimensionality were not reported. Taken together, our findings and those of prior studies suggest that while the PHQ-15′s items can be grouped into multiple symptom domains, in practice it also functions largely as a unidimensional measure in applied and cross-cultural settings.

Overall, our findings suggest that while somatic symptoms may cluster in multiple ways, the selection of a scoring model should be guided primarily by the best-fitting factor analytic solution, as determined by statistical fit and conceptual clarity. However, it is important to note that the PHQ-15 does not fully align with ICD-11 criteria for diagnosing bodily distress disorder (BDD), as ICD-11 emphasizes distress and functional impairment rather than symptom frequency alone. Thus, development or adoption of a new scale specifically designed to align with ICD-11′s BDD diagnostic criteria may be warranted.

Notably, this study advances the literature by examining PHQ-15 factor structures within a large, diverse sample during a period of heightened psychosocial stress, namely the COVID-19 pandemic. Unlike many earlier studies, it draws on data collected amid significant societal disruption, contributing to the limited body of research on somatic symptom patterns during global health crises. However, without comparative pre-pandemic data, no causal conclusions can be drawn about the specific impact of COVID-19 on symptom presentation. This pandemic-era context provides unique insights into how gender and age in particular could be more vulnerable to stress-related somatic complaints.

Our first hypothesis posited that women would exhibit higher somatic symptom levels in pain, cardiopulmonary, gastrointestinal, and fatigue factors (Leonhart et al., 2018; Schlechter et al., 2023; Witthöft et al., 2013; Zhou et al., 2020). An Analysis of Variance (ANOVA; Supplementary Analysis 10) supported this prediction: women consistently showed elevated scores, particularly in fatigue. These findings align with prior work suggesting that group differences in somatic symptom reporting may reflect psychosocial factors (e.g. gendered norms around symptom appraisal and help-seeking) and, as discussed in the wider literature, sex-related biological processes (e.g. hormonal influences on some symptom domains; Barsky et al., 2001; Heitkemper and Jarrett, 1992; Osborne and Davis, 2022). However, sex assigned at birth was not measured in this dataset, so biological mechanisms cannot be evaluated in the present study. Effect sizes were modest, indicating that gender differences in somatic expression may be subtler in general population samples than in clinical cohorts (Lanzara et al., 2018; Marcus et al., 2005). It is also possible that gender category alone does not adequately explain variation in symptom reporting, as contextual factors such as caregiving burden, social stress exposure, and access to healthcare may contribute to disparities. Future research should measure sex assigned at birth and gender identity more explicitly to support clearer interpretation of potential mechanisms. Studies using structural equation modeling may also help to disentangle biological, psychological, and sociocultural influences on somatic symptom reporting.

Consistent with our second hypothesis, younger adults reported higher somatic symptom levels than older adults across all four domains (pain, cardiopulmonary, gastrointestinal, and fatigue). This aligns with Hilderink et al. (2013), who suggested somatic symptoms are more strongly linked to psychological distress than to physical aging, and may therefore be more common among younger individuals under elevated distress. Although other studies have linked higher somatic burden to older age (Goulia et al., 2012; Rief et al., 2001), our reverse pattern may reflect pandemic-specific stressors. In our data, younger adults, particularly those aged 18–34, reported significantly higher depression, anxiety, and paranoia (Supplementary Analysis 9), all associated with increased somatic complaints (Schlechter et al., 2023; Terluin et al., 2022). They also experienced disproportionate disruption to employment, education, and social networks during lockdowns (Varma et al., 2021), likely intensifying distress and symptom reporting. By contrast, older adults may have shown greater resilience despite higher medical risk, and symptom levels did not differ significantly across older age groups. Clinically, these findings highlight the importance of interpreting somatic profiles in their sociohistorical context, and longitudinal research is needed to test whether younger adults’ elevated symptom burden persists beyond pandemic-related stress.

Aligned with our third hypothesis, each somatic domain of pain, cardiopulmonary, gastrointestinal, and fatigue showed strong positive associations with psychological variables including depression, generalized anxiety, PTSD, and sleep disturbances. These findings reinforce the conceptualization of somatic symptoms as closely intertwined with emotional distress, consistent with prior literature highlighting their overlap with affective and anxiety-related conditions (Shi et al., 2023). The strongest associations were observed for depression, anxiety, and sleep difficulties, suggesting that these variables may reflect a broader underlying distress construct captured by the PHQ-15 total score.

Loneliness was associated with somatic symptoms to a lesser extent, while health service use showed the weakest associations. This suggests that somatic complaints are closely tied to internal psychological states but do not always translate into help-seeking. Overall, the findings emphasize the importance of assessing psychological comorbidities when interpreting somatic symptoms, while domain scores may still offer clinically useful nuance. For instance, the fatigue domain showed the strongest association with depressive symptoms, but this should be interpreted cautiously because the PHQ-9 includes sleep disturbance and low energy items that overlap with the PHQ-15 fatigue items, which likely inflates this correlation. As a sensitivity analysis, we recalculated the PHQ-9 depression score excluding items 3 and 4 and re-examined its association with the PHQ fatigue domain. The correlation remained moderate-to-strong (r = 0.579, p < 0.001), suggesting that the observed association was not entirely driven by overlapping item content. Future studies could test this more rigorously by examining depression scores that exclude overlapping items or by using depression measures that do not include somatic symptoms.

Likewise, robust ties between cardiopulmonary complaints and PTSD suggest the potential utility of interventions addressing both psychological trauma and related physiological symptoms (Sullivan et al., 2016). Clinically, this distinction is significant: health professionals could use the total PHQ-15 as a broad screener, then focus on specific subscales (e.g. fatigue, gastrointestinal) to tailor interventions more precisely, especially for younger adults contending with pandemic-induced stress or women manifesting higher fatigue complaints.

Our study underscores the complexity inherent in measuring and interpreting somatic complaints. While (Schlechter et al., 2023; Witthöft et al., 2013) four-factor model best fit our sample, the relatively small differences in correlations across domains point to considerable overlap among somatic symptoms. For clinicians and researchers, this dual finding implies that the PHQ-15 can be used both as a total score (capturing general somatic distress) and as separate subscales (highlighting domain-specific patterns). It should be noted that, by using an estimator appropriate for ordinal data (WLSMV), the present study provides a more accurate test of these factor structures than some prior analyses that treated the PHQ-15 as continuous. This approach may partially explain why multiple models showed generally good fit in our dataset. However, several limitations merit caution. First, although quota sampling is designed to mirror population distributions, it is not purely random and could skew participant characteristics in unanticipated ways. Second, few participants identified as transgender or gender diverse, limiting the generalizability of findings about gender-based differences. Third, because data was self-reported, recall bias and social desirability effects may influence responses. Lastly, the cross-sectional nature of Wave 7 data precludes definitive causal inferences. Observed relationships between somatic symptoms and psychological distress likely involve bidirectional or complex interactions that warrant longitudinal examination. Nevertheless, the novelty of capturing PHQ-15 data during the COVID-19 pandemic lends unique value. Such real-world crisis data help distinguish transient, high-stress influences on somatic reporting from more stable, trait-like expressions of somatic distress, offering potential guidance for clinicians managing pandemic-era (or other crisis-related) mental health concerns.

Given that older adults did not consistently report more somatic complaints and that younger individuals displayed elevated symptom levels, further research is needed to unravel how life stage, stress, and broader social contexts intersect. Longitudinal tracking pre- and post-pandemic would clarify whether the observed age patterns represent a stable trend or a unique result of COVID-19 lockdowns. Investigating other major crises, such as economic recessions or natural disasters, could identify shared factors that amplify somatic presentations in younger cohorts. Moreover, specific subscales of the PHQ-15 could be examined in clinical populations. If certain domains, such as fatigue, predict particular outcomes (e.g. severity of depression) more effectively, these targeted insights could refine diagnostic and therapeutic strategies. Integrating additional psychosocial variables (e.g. social isolation, caregiver responsibilities, economic strain) would also help clarify why some individuals exhibit higher somatic distress. Ultimately, a multifaceted and context-sensitive approach appears essential to advancing both theoretical and practical understandings of somatic symptomatology.

In conclusion, this research substantially refines our understanding of the PHQ-15′s latent structure by situating the analysis within the heightened stress of a global pandemic. The findings underscore that a four-factor model (pain, cardiopulmonary, gastrointestinal, and fatigue) best explains somatic symptom patterns, yet high inter-factor correlations confirm the utility of a total PHQ-15 score for capturing broad somatic distress. Nevertheless, the PHQ-15′s structure does not adequately correspond to the ICD-11 diagnostic framework for bodily distress disorder, suggesting the need for a new or adapted measure that better addresses ICD-11′s emphasis on distress and impairment. Moreover, results revealed only small effect sizes regarding gender differences, suggesting that while women exhibit slightly higher symptom levels, broader psychosocial influences may be more critical than gender alone. Notably, the trend of younger adults reporting higher somatic complaints challenges the conventional view that somatic symptom severity increases with age. Such results point to the significant impact of acute stressors, including lockdowns and financial uncertainties, especially for younger populations who may have fewer coping resources or support systems. Clinically, these findings suggest that practitioners could screen with the total PHQ-15 and then delve deeper into specific symptom clusters for more tailored interventions. From a public health perspective, the elevated distress in younger cohorts during the pandemic emphasizes the need for targeted mental health strategies. Future studies should investigate whether these patterns persist post-pandemic, offering insights into the long-term trajectories of somatic distress under evolving social and economic conditions.

Supplemental Material

sj-docx-1-hpq-10.1177_13591053261442550 – Supplemental material for Modeling the PHQ-15: The factor structure of somatic symptoms in a large community sample

Supplemental material, sj-docx-1-hpq-10.1177_13591053261442550 for Modeling the PHQ-15: The factor structure of somatic symptoms in a large community sample by James Cunningham, Mark Shevlin and Eoin McElroy in Journal of Health Psychology