This study utilised a cross-sectional design. The reporting of this study follows the list of reports of observational studies (STROBE).

Recruitment took place from September 1, 2023, to July 31, 2025, across three hospitals in Zhejiang Province, China. Our sampling strategy aimed to capture both acute and rehabilitative care phases by including two public tertiary hospitals—one in an eastern coastal city and another in a western inland city—alongside one private rehabilitation hospital. We used convenience sampling to recruit dyads of stroke survivors and their primary family caregivers. The latter were defined as unpaid family members (e.g., spouses, adult children, parents, or other relatives) identified by the survivor as the principal source of daily support during hospitalization (Denham et al., 2018). A total of 773 eligible inpatient dyads were ultimately enrolled.

Inclusion criteria for stroke survivors were: (1) diagnosis of stroke confirmed according to the Chinese Guidelines for the Main Subtypes of Cerebrovascular Diseases (2019); (2) age 18 years or older; (3) clinical stability, defined as no transfer to the intensive care unit, requirement for emergency interventions (e.g., intubation, vasopressor medication), or neurological deterioration within 72 h prior to enrollment; and (4) provision of voluntary informed consent.

Exclusion criteria for stroke survivors included: (1) presence of severe systemic comorbidities (e.g., metastatic cancer) or altered consciousness; and (2) withdrawal or refusal during the study process.

Inclusion criteria for primary family caregivers were: (1) identified as an unpaid family member (spouse, adult child, parent, or other relative) by the stroke survivor as the main source of daily support during hospitalization; (2) being actively involved in assisting with or supervising activities of daily living (e.g., feeding, positioning); (3) age 18 years or older; and (4) provision of voluntary informed consent.

Exclusion criteria for caregivers were: (1) hired professional caregivers or individuals providing only occasional, non-primary support; and (2) presence of a serious physical or mental health condition that could impede their caregiving capacity.

To ensure robust latent profile models, we conducted an a priori sample size estimation. Methodological guidelines for latent variable models with multiple continuous indicators recommend a sample exceeding 500 for reliable parameter estimation and profile identification (Sinha et al., 2021). Given the 20 observed indicators in our model, the final sample of 773 dyads satisfies this recommendation and exceeds the common 10:1 participant-to-variable heuristic, supporting the analytic adequacy of our sample.

Trained assessors administered structured questionnaires during the survivors’ hospitalization. Demographic and clinical characteristics for both patients and caregivers were collected via a self-developed instrument capturing gender, age, education, time since diagnosis, employment status, primary medical payment method, and household income.

Functional independence among survivors was assessed using the Activities of Daily Living scale, which evaluates essential tasks like personal hygiene, dressing, feeding, and mobility. Scores range from 0 to 100, with higher values denoting greater independence. Following established cut-offs, scores of 61–99 indicate mild dependence (requiring occasional help), 41–60 moderate dependence (requiring substantial assistance), and ≤40 severe dependence (requiring full assistance for most activities) (Mlinac and Feng, 2016). Assessments were completed by a nurse during the hospital stay.

We measured family resilience using the 20-item FRQ developed by Bu and Liu (2019). This Chinese-specific scale comprises four dimensions: Perseverance (6 items), Harmony (5 items), Openness (5 items), and Supportiveness (4 items). All items use a 5-point Likert scale (1 = strongly disagree to 5 = strongly agree), yielding a total score between 20 and 100; higher scores reflect greater resilience. The scale has demonstrated strong psychometric properties, with a reported Cronbach’s α of 0.91 for the full scale (Bu and Liu, 2019). In the present sample, Cronbach’s α was 0.96.

Caregiver competence was evaluated with the 25-item FCTI, which assesses domains such as adapting to the caregiver role, responding to patient needs, managing personal emotions, assessing resources, and making life adjustments. Items are rated on a 5-point Likert scale. Notably, higher FCTI scores signify lower caregiver competence, reflecting greater difficulties in performing caregiving tasks. The Chinese version has shown high reliability (Cronbach’s α = 0.93) (Lee and Mok, 2011), which was consistent in our sample (α = 0.93).

The PSSS measured participants’ perceptions of support from family, friends, and significant others (Zimet et al., 1990). Its 12 items are rated on a 7-point Likert scale (1 = very strongly disagree to 7 = very strongly agree). Previous studies have established its high reliability (e.g., Cronbach’s α = 0.914) (Liu et al., 2016), a finding replicated here (α = 0.91).

The primary outcome was family resilience profile membership, a latent categorical variable generated from our analysis. Latent profile analysis (LPA) applied to the 20-item Family Resilience Questionnaire (FRQ) identified these subgroups. Selecting the optimal profile solution involved evaluating statistical fit indices alongside the theoretical coherence and distinctiveness of each class. Following model selection, each dyad was assigned to the profile for which it had the highest posterior probability, capturing its unique configuration of resilience attributes. These empirically derived profiles then functioned as the categorical outcome for subsequent predictive modeling.

Analyses proceeded in two sequential stages: latent profile identification followed by predictor examination.

The first stage used Latent Profile Analysis (LPA) in Mplus 8.3 to identify unobserved subgroups based on responses to the 20 FRQ items. We estimated models specifying one through six profiles. Model selection involved evaluating statistical fit indices, with lower values on the AIC, BIC, and aBIC preferred, and higher entropy values indicating more distinct profiles. We also assessed classification precision using entropy, with values ≥ 0.80 deemed acceptable, and used the Lo–Mendell–Rubin adjusted likelihood ratio test (LMRT) and the bootstrap likelihood ratio test (BLRT) to statistically compare a k-profile model against a k-1 profile model. When statistical indices were inconclusive, final model choice emphasized theoretical interpretability and clinical relevance, favoring the more parsimonious solution.

The second stage, performed in SPSS 26.0, involved descriptive and inferential analyses. Descriptive statistics characterized the sample, with continuous variables reported as mean ± standard deviation or median (interquartile range) based on distribution, and categorical variables as frequencies (percentages). Bivariate associations between profile membership and candidate predictors were tested using chi-square tests for categorical variables. To identify factors independently associated with profile membership, we performed multinomial logistic regression. For this model, continuous predictors—specifically scores on the Family Caregiver Task Inventory (FCTI) and the Perceived Social Support Scale (PSSS)—were converted to Z-scores to facilitate comparison of effect magnitudes. Statistical significance was set at p < 0.05 (two-tailed) for all tests.

A total of 773 patient-caregiver dyads completed the study. The patient cohort was predominantly male (57.70%), whereas caregivers were mostly female (61.06%). Full demographic and clinical characteristics for both groups are detailed in Table 1.

Analysis of responses to the Family Resilience Questionnaire using latent profile analysis (LPA) identified distinct subgroups. Table 2 presents model fit indices for one- through five-profile solutions, with descriptive statistics for all FRQ items available in Table 3.

A four-profile solution was ultimately selected. Although information criteria (AIC, BIC, aBIC) continued to decrease up to five profiles, the non-significant Lo–Mendell–Rubin and Bootstrap Likelihood Ratio Tests (p = 0.8046) for the five-profile model indicated that the added complexity was not statistically justified. The four-profile model provided excellent classification certainty (entropy = 0.984) and represented a significant improvement over models with fewer classes (p < 0.001), while yielding subgroups that were both theoretically interpretable and clinically meaningful (see Table 4).

These four profiles corresponded to distinct patterns of family adaptation (Figure 1):

Initial bivariate tests revealed significant associations between profile membership and every demographic and clinical variable examined (p < 0.01; see Table 1). As shown in Table 5, key study variables were interrelated: family resilience scores correlated negatively with caregiver difficulty (r = −0.686, p < 0.01) and positively with perceived social support (r = 0.638, p < 0.01), while caregiver difficulty and social support were themselves inversely correlated (r = −0.265, p < 0.01).

Results from the multinomial logistic regression, which used Profile 1 as the reference category, identified several independent predictors of profile membership (Table 6). Caregiver competence (operationalized by FCTI scores) and perceived social support (PSSS) were among the strongest predictors. Patient gender, primary payment source, caregiver relationship to the patient, daily hours of care, and patient functional status also contributed significantly to the model (all p < 0.05). This multivariate analysis suggests that these factors offer unique explanatory power in differentiating between family resilience profiles, beyond their initial bivariate associations.

Latent profile analysis identified four distinct family resilience profiles among stroke patient-caregiver dyads. This result directly challenges the notion of adaptation as a uniform construct distributed along a simple high-low continuum, confirming instead the multidimensional and heterogeneous nature of family resilience (Dong et al., 2021; Herbers et al., 2020). The emergence of these qualitatively different profiles supports a person centered perspective, suggesting that families configure their unique strengths and vulnerabilities into specific patterns of adaptation.

The four profiles reflect distinct configurations of resilience strengths and vulnerabilities, patterns which align closely with established theories of family adaptation.

Profile 1 (Low-Functioning Families) presents a complex clinical picture where preserved foundational capacities coexist with critical deficits in key processes. While these families maintain basic abilities in problem-solving (Item 1: 3.315) and role recognition (Item 2: 3.353), they struggle profoundly with emotional communication and conflict management (Item 8: 1.012). This pattern illustrates how core relational processes can falter even when some instrumental functions remain intact, consistent with Walsh’s emphasis on communication as the essential conduit for adaptive transformation under stress (Walsh, 2016).

Profile 2 (Moderately Resilient-Low Cohesive Families) reveals a pronounced dissociation between practical function and emotional connection. Competence in collaborative tasks (Item 1: 3.144; Item 4: 3.049) contrasts sharply with marked difficulties in expressing affection (Item 6: 2.41) and managing interpersonal dynamics (Item 8: 1.048). This fragmentation mirrors what Henry et al. describe as “functional fragmentation” (Henry et al., 2015), where families maintain operational competence while experiencing emotional disengagement, potentially compromising long-term adaptation.

The Chinese cultural context offers a lens through which to interpret this profile. Traditional values emphasizing harmony and filial obligation often drive pragmatic, task-focused caregiving during health crises (Yang et al., 2025). However, these same norms can suppress the open expression of distress to maintain surface harmony (Li Q. et al., 2024). In such settings, families may thus demonstrate functional caregiving competence while internalizing significant emotional strain. This culturally informed pattern of “high pragmatism, low emotional expressiveness” helps explain the coexistence of moderate resilience and low cohesion seen in Profile 2.

Profile 3 (Highly Resilient-Well-Functioning Families) demonstrates robust and integrated functioning across most domains. Relative to Profile 4, however, this group appears more inwardly focused, exhibiting comparatively lower scores on growth-through-adversity (Item 10: 2.509) and external social engagement (Item 20: 2.607). This configuration aligns with Black and Lobo’s observation that some well-functioning families prioritize strong internal cohesion while maintaining more bounded external connections (Black and Lobo, 2008).

Profile 4 (High-Functioning-Optimistically Resilient Families) distinguishes itself through exceptional performance across all dimensions, particularly excelling in proactive growth (Item 10: 4.584) and active social resource utilization (Item 20: 4.458). This profile embodies Walsh’s concept of transformative resilience, in which families not only withstand crisis but also experience growth and re-organization (Walsh, 2016). Their capacity to integrate pragmatic functioning with emotional attunement and optimistic outreach represents the most comprehensive realization of adaptive family processes identified in this sample.

The analysis positions caregiver competence and perceived social support as central, distinguishing factors among the resilience profiles. Notably, membership in Profile 2 relative to Profile 1 was strongly associated with lower caregiver competence. This link resonates with the Family Stress Model (Pearlin et al., 1990), where difficulties in performing core caregiving tasks can deplete a family’s internal resources, fostering an adaptation pattern that sustains practical functioning at the expense of emotional cohesion.

Higher levels of perceived social support, conversely, were uniquely associated with membership in the higher-functioning profiles (Profiles 3 and 4). This finding aligns with the buffering hypothesis (Cohen and Wills, 1985), positing that robust external support networks provide essential resources that help maintain family integrity and facilitate the proactive, optimistic outlook observed in the most resilient profiles. Within our model, these two factors operated as related yet distinct components of the family’s resource ecosystem.

Several demographic and clinical characteristics further delineated profile membership. Spousal caregivers showed a different pattern of association than direct family, being less prevalent in the most resilient Profile 4. This result underscores the distinct emotional dynamics of spousal caregiving (Pinquart and Sörensen, 2011), wherein the relationship’s intensity and the risk of role engulfment (Panourgia et al., 2022; Skaff and Pearlin, 1992) may shape a family’s adaptive response, particularly in developing the optimistic, growth-oriented resilience epitomized by Profile 4.

Specific patient characteristics also proved influential. For instance, families of patients with moderate functional dependence were more frequently classified in Profile 2 than in Profile 1, a pattern consistent with the competence-press model (Rizeq et al., 2021). This indicates that moderate care demands might be sufficient to mobilize a family’s problem-solving capacities while simultaneously generating interpersonal strains that erode emotional unity. Furthermore, families with male patients were less represented in Profile 4, possibly reflecting gendered patterns in illness response and support-seeking that merit further investigation (Wang et al., 2025).

A strong bivariate association was observed between household monthly income and resilience profile membership (χ² = 142.04, p < 0.001). Notably, households with a monthly income of ≥8,000 CNY were almost absent from the two lower-resilience profiles (Profile 1: 0%; Profile 2: 0.54%) but constituted a markedly higher proportion of Profile 4 (17.58%). This distribution pattern indicates a close link between higher financial resources and more resilient family configurations.

Drawing on theoretical frameworks, economic capital may function as a critical contextual resource that can facilitate adaptation (Schüz et al., 2016). Beyond alleviating immediate financial strain, it may help stabilize a caregiver’s employment status and preserve opportunities for social engagement, thereby fostering a family climate more conducive to effective communication, collaborative problem-solving, and shared positive beliefs (Conger and Donnellan, 2007). These potential pathways are consistent with the Family Stress Model, which posits economic pressure as a salient factor interacting with family processes.

Nevertheless, the near-total separation of higher-income households from the low-resilience profiles rendered this variable unstable for estimation in our multivariate model. Future research should employ more refined economic indicators and larger, socioeconomically diverse samples to clarify the unique contribution of financial resources to family resilience trajectories.

This study applied latent profile analysis to reveal four distinct configurations of family resilience, demonstrating that structural heterogeneity characterizes how families adapt following a stroke. The identified profiles enable a shift toward precision family support, where interventions must be tailored to specific patterns of strengths and deficits. For “Low-Functioning Families” (Profile 1), interventions should initiate a crisis management and resource linkage protocol, prioritizing respite care and basic financial navigation to alleviate their overwhelming burden. For “Moderately Resilient but Low-Cohesive Families” (Profile 2), brief family counseling focused on emotional validation and safe expression is indicated, directly targeting their core deficits in specific emotional communication and conflict management items. For the highly resilient families (Profiles 3 and 4), the goal is to preserve robust functioning through preventive support. Specifically, for “High-Functioning-Optimistically Resilient Families” (Profile 4), their strengths can be leveraged by formally engaging them as peer support mentors within the care community. This stratified approach moves beyond one-size-fits-all support and is essential for effectively promoting sustainable adaptation and post-traumatic growth in stroke families (Liu et al., 2025; McCurley et al., 2019).

A recent longitudinal study investigating the trajectory of family resilience during the first 6 months post-stroke (Zhang et al., 2023) highlights the need to consider both temporal dynamics and structural heterogeneity for a comprehensive understanding of post-stroke family adaptation. Building on the profiles established in the present study, future longitudinal research should examine the stability of these resilience configurations over time.