Longitudinal Dynamics of Pain, Fatigue, Anxiety-Depression, Sleep, and Functional Limitation in Older Adults with Rheumatoid Arthritis: A Cross-Lagged Panel Network Analysis

Abstract

Purpose

To investigate the longitudinal dynamics of five symptoms –pain (P), fatigue (F), anxiety-depression (A), sleep disturbance (S), functional limitation (FL)- in older patients with rheumatoid arthritis (RA), to identify longitudinal predictive relationships among these symptoms, and to inform individualized symptom management.

Methods

A prospective longitudinal study enrolled eligible RA patients from August 2025 to February 2026. Five symptoms were assessed at hospitalization (T1) and one month after discharge (T2). Cross-sectional networks compared inpatient and outpatient symptom covariation, and Cross-Lagged Panel Network (CLPN) analysis evaluated longitudinal predictive associations and network stability.

Results

A total of 177 participants completed data collection. All edges in the cross-sectional networks were positive. The most central symptoms at both T1 and T2, as indicated by expected influence, were functional limitation and fatigue. CLPN analysis revealed that fatigue was the only symptom with significant longitudinal predictive effects: it positively predicted subsequent sleep disturbance (β = 0.250, 95% CI [0.038, 0.621]) and functional limitation (β = 0.045, 95% CI [0.016, 0.068]) at T2. Anxiety-depression had the highest in-expected influence (In‑EI = 0.819), indicating that it was most susceptible to being predicted by other symptoms. Network stability was satisfactory (CS coefficients > 0.6 for both Out‑EI and In‑EI).

Conclusion

During the transition from inpatient to outpatient care in older RA patients, fatigue emerges as the primary longitudinal predictor of subsequent sleep disturbance and functional limitation, whereas anxiety-depression exhibited the highest susceptibility to being predicted by other symptoms. These findings may inform future symptom management strategies and generate hypotheses for interventional research.

Introduction

Rheumatoid arthritis (RA) is a chronic autoimmune disease characterized by symmetric polyarthritis, presenting with joint swelling, pain, morning stiffness, and impaired joint function, which may eventually lead to disability.¹ The disease imposes a substantial health and economic burden globally. According to the Global Burden of Disease Study, there are approximately 18.5 million RA patients worldwide.² Arthritis or rheumatic diseases rank among the top five chronic diseases in the elderly³ and have become a global public health concern.⁴

Older adults with RA face a multidimensional symptom burden: pain, fatigue, mood disturbance, sleep problems and functional limitation, which often co-occur and mutually reinforce one another. However, even with active pharmacological intervention, many patients still report high levels of joint pain, fatigue, and other symptoms.⁵ Joint swelling and pain,⁶ fatigue,⁷ anxiety and depression,⁸ functional limitation⁹ and sleep disturbance¹⁰ are the core symptoms in RA patients. Building on this, clinical practice and research¹¹ have confirmed that these symptoms are not isolated but rather interrelated and mutually reinforcing, compounding their impact on patients’ quality of life.¹² However, although the interconnections among symptoms have been widely recognized, current RA symptom management research remains largely focused on single symptom dimensions^13,14 or used total scores as intervention targets, often overlooking the dynamic interactions among symptoms.

As a chronic disease, RA has a long and often unresolved course. Patients’ medication regimens, psychological status, and activity levels differ between flare and remission phases, leading to dynamic changes in symptom associations over time. Previous cross‑sectional studies have revealed symptom interactions,^12,15,16 but they are inherently limited to contemporaneous associations and cannot capture how symptoms evolve over time. Consequently, longitudinal predictive relationships among symptoms in RA remain under-investigated. Clarifying these directional relationships is essential for developing precise, stage-specific intervention strategies.

In recent years, network analysis has emerged as a powerful approach for studying symptom systems in chronic diseases. It preserves the heterogeneity of individual symptom profiles and can identify which specific symptoms exert the greatest influence on the overall symptom network.^17,18 Several cross-sectional network studies in RA have been published to date,^19,20 however, these cross‑sectional designs capturing contemporaneous co‑occurrence but cannot capture directional or temporal relationships—information that is critical for developing targeted clinical interventions.

Cross-lagged panel network (CLPN) analysis addresses this gap by leveraging longitudinal data to estimate how symptoms predict each other over time. For example, in colorectal cancer patients, fatigue predicted subsequent pain and walking ability, suggesting that early fatigue intervention might alleviate other symptoms.¹⁸ In RA, however, longitudinal network studies remain scarce, and the temporal dynamics of symptom interactions during the critical transition from inpatient to outpatient care have not been systematically examined. This transition period—typically the first month post-discharge—is clinically significant, it coincides with medication adjustments and rehabilitation initiation,²¹ is when the largest reduction in disease activity occurs,²² all of which may reshape symptom interrelationships. Understanding these longitudinal dynamics could inform more precise, stage-specific symptom management strategies for older RA patients.

Therefore, three research gaps remain: (1) how symptom networks change during the critical inpatient‑to‑outpatient transition in older RA patients is unknown; (2) the temporal order among pain, fatigue, sleep, mood, and body function has not been established; and (3) CLPN methodology has not been applied to RA transitional care. Therefore, this study employs CLPN analysis to investigate the dynamic evolution of five core symptoms (pain, fatigue, anxiety/depression, sleep disturbance, and functional limitation) in older RA patients from inpatient (T1) to outpatient (T2, one month later). We aim to identify structural differences in cross‑sectional networks between the two phases and to evaluate longitudinal predictive relationships using the CLPN model.

Methods

Study Design and Sample

A prospective longitudinal design was adopted, with data collected from August 2025 to February 2026. Inclusion criteria were: (1) age ≥ 55 years;²³ (2) meeting the 2010 ACR/EULAR classification criteria for RA;²⁴ (3) no hearing impairment and ability to communicate normally. Exclusion criteria: (1) severe organ dysfunction or malignancy; (2) history of psychiatric illness or severe cognitive impairment; (3) other immune-mediated diseases such as systemic lupus erythematosus or Takayasu arteritis. Loss to follow-up or voluntary withdrawal from the study during the follow-up period was considered as dropout.

This study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of Beijing Hospital (2025BJYYEC-KY038-01). Patients were recruited from the rheumatology and immunology department of a tertiary hospital in Beijing Hospital. Data collection was performed by a postgraduate student who had undergone rigorous standardized training (including study objectives, confidentiality protocols, and questionnaire quality assessment). On the day of data collection, eligible participants all signed informed consent forms. The researcher conducted face-to-face data collection using translated Chinese instruments. One month after discharge, participants were followed up by telephone.

Sample size was informed by comparable CLPN studies²⁵ examining 5 nodes symptom networks (sample size range: 150–225). Given that 20 directed cross-lagged paths were to be estimated (N×(N−1)=5×4=20),²⁶ the target was set at a minimum of 150 completers. A total of 195 participants were enrolled, of whom 177 provided complete data (effective follow-up rate: 90.77%), consistent with the sample sizes reported in prior studies.

Assessment Tools

Pain and fatigue: These were assessed using the Rheumatoid Arthritis Symptom Questionnaire (RASQ),²⁷ which was developed by Banderas et al in 2017. The questionnaire contains 7 items. The items for joint pain and fatigue were used, rated on a scale from 0 (not at all) to 10 (most severe). Higher scores indicate more severe joint pain. In this study, a score ≥ 3 was defined as the presence of the symptom. A score ≥ 3 was used to define symptom presence for prevalence calculations; continuous scores (0–10) were used for all network analyses. Pain and fatigue were selected as they are the most commonly reported symptoms in RA patients.²⁸

Anxiety and depression: The Hospital Anxiety and Depression Scale (HADS) was used. This scale was developed by Zigmond and Snaith in 1983,²⁹ and translated into Chinese by Ye Weifei.³⁰ HADS consists of 14 items (7 for anxiety, 7 for depression), each rated on a 4-point scale from 0 to 3. Higher scores indicate more severe symptoms. Previous studies have supported the use of the total score as an indicator of overall distress,³¹ with a Cronbach’s α coefficient of 0.73–0.87 for the total scale.³² Referring to the study by Hinz et al on the German general population, a total score ≥ 13 was used as the cut-off for screening positive for psychological distress.³³

Sleep disturbance: This was assessed using the Pittsburgh Sleep Quality Index (PSQI)³⁴ developed by Buysse et al in 1989. The scale contains 18 items and 7 dimensions, with a total score ranging from 0 to 21. Higher scores indicate worse sleep quality, and a score > 5 indicates the presence of sleep disturbance.

Functional limitation: This was assessed using the Health Assessment Questionnaire–Disability Index (HAQ-DI).³⁵ The scale contains 8 dimensions with a total of 20 items. Each item is rated from 0 (“without any difficulty”) to 3 (“unable to do”). The total score is the mean of the dimension scores. A score between 0 and 1 indicates mild functional impairment, 1–2 indicates moderate functional impairment, and 2–3 indicates severe functional impairment.

General Information

Demographic and social characteristics (age, sex, height, weight, education level) and disease-related information (family history of RA, disease duration) were collected. All of the above information was extracted from patients’ medical records.

Statistical Analysis

Only participants who completed both T1 and T2 assessments were included in the final analysis. Statistical significance for cross-lagged coefficients was determined by 95% confidence intervals excluding zero.

Descriptive statistics were calculated using SPSS 27.0 to summarize participants’ baseline demographic and clinical characteristics. Categorical variables (eg, sex, education level) were presented as frequencies and percentages (%). Continuous variables (eg, age, symptom scores) were described as mean ± standard deviation (Mean ± SD). The distribution of continuous variables was examined using the Shapiro–Wilk test; all symptom scores were within acceptable ranges for network estimation, and no transformation was applied.

Statistical analyses were performed using R software version 4.4.3. First, cross-sectional networks at baseline and follow-up were estimated using Gaussian Graphical Models (GGM) with the R package “qgraph”.³⁶ Prior to estimation, multicollinearity was examined using the Variance Inflation Factor (VIF < 5), and the assumptions of multivariate normality (Shapiro–Wilk test) and linearity (scatterplot matrices) were assessed. In the networks, nodes represented symptoms and edges represented partial correlations between symptoms. To control for spurious connections, LASSO regularization with the Extended Bayesian Information Criterion (EBIC, gamma = 0.5) was applied for edge selection. The same average layout was used to facilitate visual comparison. Expected influence (EI) centrality (ie, the sum of edge weights connected to a node, preserving the sign) was calculated for each symptom, with In-EI indicating susceptibility to being predicted (sum of incoming edges) and Out-EI indicating influence on other symptoms (sum of outgoing edges).

Subsequently, Cross-Lagged Panel Network (CLPN) analysis was used to explore longitudinal predictive relationships among symptoms. Estimation was performed using a LASSO-regularized vector autoregressive (VAR) model implemented with the R packages glmnet and qgraph. For each symptom at T2, all five symptoms at T1 were entered as predictors; a Gaussian regression model was fitted using 5-fold cross-validation to select lambda.min, yielding a 5×5 directed weighted adjacency matrix (autoregressive effects on the diagonal). In the network graphs, directed edges indicated the predictive direction from a symptom at T1 to a symptom at T2, and edge width was proportional to the absolute value of the regression coefficient. For the directed CLPN, Out-EI (sum of edge weights from a node to other nodes) and In-EI (sum of edge weights from other nodes to the node) were calculated. Bridge centrality was calculated to identify symptoms that connect different symptom clusters within the network. The R² decomposition was used to quantify the proportion of variance in each T2 symptom explained by all T1 symptoms; the autoregression-excluded predictability specifically refers to the variance explained by other symptoms at T1 after excluding the autoregressive effect of the symptom itself.

To examine the robustness of the primary findings, a covariate-adjusted analysis was performed. First, univariate analyses (Pearson correlation, independent t-test, or ANOVA, as appropriate) were used to screen for covariates significantly associated with each symptom. Covariates with p< 0.10 were selected for adjustment. The identified covariates were then regressed out from the symptom scores using linear regression, and the residuals were retained for all subsequent network analyses. This approach ensures that the estimated symptom associations reflect relationships net of the covariates’ influence.

For all networks, non-parametric bootstrapping (500 resamples) with bias correction was used to estimate 95% confidence intervals for edge weights. To account for multiple comparisons and control the false discovery rate, the Benjamini-Hochberg procedure was applied to the cross‑lagged edge weights. Case-drop bootstrapping (200 resamples, dropping proportions ranging from 10% to 70%) was performed to assess the stability of centrality indices, and the correlation stability (CS) coefficient was calculated, with CS ≥ 0.5 considered acceptable.

Results

Participant Characteristics

Table 1 presents the characteristics of the participants (n = 177). The mean age of the participants was 67.52 years (range: 55–88 years). The majority were female (79.1%), with the highest proportion having an education level of college degree or above (36.1%), and married (94.4%). Patients with a family history of RA accounted for 20.3%, and the mean disease duration was 77 months.

Table 1.

Socio-Demographic and Clinical Characteristics of Participants at Baseline (N=177)

Variables	n/% or Mean ± SD
General demographic and sociological data
Age	67.52±7.627
Ethnicity/race
Han	171 (96.61%)
Minority	6 (3.39%)
Gender
Male	37 (20.90)
Female	140 (79.10)
Level of education
Primary school and below	20 (11.30)
Junior high school	51 (28.81)
Senior high school/Secondary vocational school	42 (23.73)
Junior college and above	64 (36.16)
Disease-related characteristics
Duration of illness (months)	77 (IQR 189.5)
Family history
Yes	36 (20.34)
No	141 (79.66)
BMI	23.44 (4.93)

Symptom Prevalence and Severity

Table 2 shows the prevalence and frequency of the five symptoms. At T1 (inpatient), fatigue had the highest prevalence (70.1%), followed by pain (61.0%). At T2 (outpatient, one month post-discharge), sleep disturbance became the most prevalent symptom (68.9%), followed by fatigue (63.3%).

Table 2.

Symptom Scores of Participants (N=177)

Symptom	T1			T2
	Score (Mean ± SD)	Median (IQR)	Prevalence	Score (Mean ± SD)	Median (IQR)	Prevalence
Pain	3.51±2.77	3.00 (1.00–6.00)	61.0%	2.94±2.54	3.00 (0.00–5.00)	52.5%
Fatigue	4.45±2.88	4.30 (2.00–7.00)	70.1%	4.10±3.00	4.00 (2.00–7.00)	63.3%
Anxiety-depression	6.80±6.96	5.00 (1.00–11.00)	20.3%	6.08±6.17	4.00 (2.00–8.00)	14.7%
Sleep disturbance	7.31±4.80	7.00 (3.00–10.00)	66.1%	7.84±4.90	7.00 (4.00–12.00)	68.9%
Body dysfunction	0.88±0.74	0.75 (0.25–1.38)	35.0%	0.85±0.66	0.75 (0.38–1.25)	38.4%

Cross-Sectional Network Structure

Figure 1 shows the cross-sectional symptom networks of pain, fatigue, anxiety-depression, sleep disturbance, and functional limitation in older RA patients at baseline (hospital admission) and follow-up (one month after discharge). All edges in both networks were positive (100% positive edges), with no negative associations observed. Supplementary Figure S1 presents the results of non-parametric bootstrapping; the 95% confidence intervals were narrow, indicating high precision in edge weight estimation. Case‑drop bootstrapping showed that the stability of strength centrality was acceptable for both time points (CS coefficients were above 0.5 at both time points, exceeding the recommended threshold of 0.5; see Supplementary Figure S2).

Figure 1.

Cross-sectional networks of symptoms in older RA patients at inpatient (T1) and outpatient (T2). (A) Inpatient network (T1); (B) Outpatient network (T2). Nodes represent symptoms, and edges represent partial correlations between symptoms. Edge thickness is proportional to the strength of the partial correlation.

Abbreviations: P, pain; F, fatigue; A, anxiety-depression; S, sleep disturbance; FL, functional limitation.

The centrality plot (Figure 2) showed that at T1, functional limitation had the highest strength (Expected Influence, 1.87), followed by fatigue (1.47) and pain (1.44). At T2, functional limitation remained the strongest (1.79), while the strength of sleep disturbance increased markedly from 1.14 at T1 to 1.49, and fatigue (1.72) and anxiety-depression (1.32) also increased. In the baseline network, the strongest edge was “pain–fatigue”; at T2, the connections “fatigue–sleep disturbance” and “fatigue–functional limitation” were notably strengthened.

Figure 2.

Centrality plot of expected influence (EI) in the cross-sectional networks. (A) Inpatient (T1); (B) Outpatient (T2). Higher EI values indicate greater centrality.

Abbreviations: P, pain; F, fatigue; A, anxiety-depression; S, sleep disturbance; FL, functional limitation.

Bridge centrality analysis (Supplementary Table S1) indicated that anxiety-depression and functional limitation were key nodes connecting different symptom clusters (both bridge strength = 0.263). Comparison of overall network strength between T1 and T2 showed that the mean edge weight increased from 0.351 to 0.392 (see Supplementary Table S2), but the difference was not statistically significant (p > 0.05). While this cross-sectional bridge analysis identifies concurrent connectors across symptom domains (see Figure 1 and Supplementary Table S2), the CLPN analysis below provides complementary longitudinal evidence by revealing temporal directions between symptoms.

Cross-Lagged Longitudinal Network of Pain, Fatigue, Anxiety-Depression, Sleep Disturbance, and Functional Limitation

Figure 3 presents the CLPN of the five core symptoms (pain, fatigue, anxiety-depression, sleep disturbance, functional limitation) in older RA patients from the inpatient setting (T1) to the outpatient setting (T2). The weights of all autoregressive and cross-lagged edges are summarized in Supplementary Table S3. Among the 13 edges, 11 (84.6%) were positive and 2 (15.4%) were negative.

Figure 3.

Longitudinal cross-lagged network of symptoms of older rheumatoid arthritis patients during the transition from inpatient to outpatient settings.

Note: Green edges represent positive associations and red edges represent negative associations between nodes.

Abbreviations: P, pain; F, fatigue; A, anxiety-depression; S, sleep disturbance; FL, functional limitation.

All symptoms showed significant autoregressive effects (see Supplementary Table S2). The strongest autoregressive coefficients were observed for sleep disturbance (β = 0.477, 95% CI [0.371, 0.635]), functional limitation (β = 0.463, 95% CI [0.307, 0.634]), and fatigue (β = 0.462, 95% CI [0.303, 0.607]).

Fatigue was the only symptom with significant cross‑lagged predictive effects. It positively predicted subsequent sleep disturbance (edge weight = 0.250, 95% CI [0.038, 0.621]) and functional limitation (edge weight = 0.045, 95% CI [0.016, 0.068]) at T2. In addition, fatigue also showed positive predictions for pain (edge weight = 0.144, 95% CI [0, 0.270]) and anxiety‑depression (edge weight = 0.154, 95% CI [0, 0.467]), though these did not reach statistical significance. A negative cross‑lagged edge from anxiety‑depression to pain (β = –0.024, 95% CI [–0.079, 0]) was observed but was not significant.

Figure 4 shows the standardized centrality measures. Fatigue exhibited the strongest Out-EI (0.655), primarily predicting subsequent sleep disturbance (edge weight = 0.250) and functional limitation (edge weight = 0.045). In addition, fatigue also showed positive predictions for pain (edge weight = 0.144) and anxiety-depression (edge weight = 0.154), though the latter had a numerically larger coefficient but lower precision. Anxiety-depression had the highest In-EI (0.819) among all symptoms, mainly influenced by T1 functional limitation (edge weight = 0.634), fatigue (edge weight = 0.154), and pain (edge weight = –0.024, negative). Notably, the positive predictive effect of functional limitation on anxiety-depression was particularly prominent in magnitude. Supplementary Figure S3 presents the difference between Out-EI and In-EI for each symptom in the T1 to T2 longitudinal network.

Figure 4.

Standardized centrality indices of the longitudinal cross-lagged network. (A) Out-EI (predictive influence on other symptoms); (B) In-EI (susceptibility to being predicted by other symptoms).

Abbreviations: P, pain; F, fatigue; A, anxiety-depression; S, sleep disturbance; FL, functional limitation.

Further analysis of the proportion of variance explained (R²) for each symptom showed that functional limitation had the highest total R² (44.7%), with within-group effects contributing 25.7%, autoregressive effects 14.8%, and cross-group effects 4.3%. Sleep disturbance had a total R² of 35.9%, mainly driven by autoregressive effects (23.6%). Fatigue had an R² of 26.7% (autoregressive contribution 16.3%), followed by pain (21.5%) and anxiety-depression (17.8%). Notably, the autoregressive-excluded predictability of functional limitation was as high as 40.2%, suggesting that most of its variance at T2 could be explained by other symptoms at T1.

The accuracy plot (Supplementary Figure S4) indicated that the bootstrap-based Cross-Lagged Panel Network (CLPN) estimates had high precision. Moreover, case-drop bootstrap results showed that the stability of out-expected influence (Out-EI) and in-expected influence (In-EI) was acceptable (CS coefficients = 0.7 and 0.6, respectively; see Supplementary Figure S5). The observed predictive effects remained significant after FDR correction (q < 0.05).

Robustness Analysis After Covariate Adjustment

To examine whether the primary findings were influenced by potential confounders, we performed a covariate-adjusted analysis using the residual method. After adjusting for covariates including age, sex, marital status, disease duration, joint function, and family history, the core results of the CLPN model remained unchanged. The adjusted results showed that fatigue was still the only symptom that significantly predicted subsequent sleep disturbance (β = 0.209, 95% CI [0.044, 0.669]) and functional limitation (β = 0.044, 95% CI [0.017, 0.067]). Fatigue had the highest out-expected influence (Out-EI = 1.121), and anxiety-depression had the highest in-expected influence (In-EI = 0.865). Network stability was good (CS coefficients > 0.5 for both). These findings indicate that the results of this study are robust.

Discussion

Our study showed that in the longitudinal network, fatigue (F) had the highest Out-EI, indicating its strong predictive ability for other symptoms, while anxiety-depression (A) had the highest In-EI, suggesting that it was more susceptible to being influenced by other symptoms. Fatigue was the only driver that longitudinally predicted subsequent symptoms: fatigue at T1 significantly predicted sleep disturbance (β = 0.250) and functional limitation (β = 0.045) at T2. These findings have direct clinical implications: during hospitalization, systematic fatigue assessment and early non‑pharmacological interventions (eg, energy conservation, graded activity) may prevent the development of sleep problems and functional decline after discharge.

The finding that fatigue is the only symptom with a significant longitudinal predictive effect in older RA patients contrasts with previous cross-sectional studies in which pain occupied a central position. In a cross-sectional network analysis of 211 RA patients, Tung²⁰ found that pain had the highest degree centrality and low mood had the highest betweenness centrality, serving as key nodes connecting clinical indicators and mental health. Similarly, Akan a study of 190 patients³⁷ analyzed 190 patients and also found that pain occupied a structurally central position in the symptom network of RA patients. Several explanations may account for this discrepancy. First, the symptomatic experience of older RA patients shows age-specific characteristics. Pavlov³⁸ found that elderly-onset RA often presents with acute onset and prominent systemic symptoms. Gezer³⁹ compared symptom differences between patients over and under 65 years of age and found that pain, depression, fatigue, and sleep quality in older RA patients worsen with age, indicating that these symptoms interact more prominently in older RA patients. Second, the multifactorial nature of fatigue makes it an independent driver in the chronic phase. The biopsychosocial network model of RA fatigue proposed by Geenen⁴⁰ indicates that psychological and physical function, pain, sleep disturbance, as well as depression and anxiety, collectively sustain the patient’s fatigue state. Pope⁴¹ similarly pointed out that RA fatigue can arise from multiple sources (disease activity, chronic pain, sleep disturbance, low mood, etc). This means that fatigue is essentially the result of multiple interacting causes, while pain may be a more direct reflection of inflammation. A five-year longitudinal study⁴² also showed that the long-term prognosis of fatigue is mainly related to mood, sleep, and other factors rather than disease activity itself. Therefore, in the remission phase when fatigue has already developed, it continuously drives subsequent symptom changes through cross-interactions with multiple factors. Notably, a longitudinal study by Fazaa⁴³ revealed the temporal relationship between pain and fatigue: baseline pain was the only independent predictor of persistent fatigue 12 months later, suggesting that pain plays a key role in the initiation of fatigue. In contrast, the one-month follow-up of our study captures the phase of fatigue development and maintenance. At this stage, fatigue is driven by multiple factors rather than a direct effect of pain, and it simultaneously influences other symptoms. Therefore, in symptom management, pain control should be prioritized during the acute phase to reduce the likelihood of fatigue development; in the remission phase, fatigue should become the main target of intervention to minimize its impact on other symptoms and improve patients’ quality of life.

Another important finding of this study is the high in-degree centrality and low out-degree centrality of anxiety-depression. This suggests that in older RA patients, anxiety-depression is more likely to be a consequence of long-term inadequate management of fatigue, pain, and sleep problems. Ukolova⁴⁴ also found in a network analysis of depressive symptoms in RA patients that the network structure of depressive symptoms changes with health status, and that in patients with higher disease activity, depressive symptoms manifest more as a result of other symptoms than as a cause. Our findings echo this conclusion and further provide temporal evidence through CLPN: fatigue at T1 positively predicted anxiety-depression at T2, whereas the reverse path was not significant. This finding is consistent with the study by Bacconnier⁴⁵ in an early RA cohort, which showed that the HAQ disability index was the most important predictor of psychological distress, whereas biological markers and treatment regimens were not associated with psychological distress. These findings suggest that, when older RA patients present with anxiety or depression, clinicians should carefully evaluate the control status of upstream physical symptoms—such as fatigue, pain, and sleep—alongside emotional symptoms. Optimizing the management of these physical symptoms (eg, enhancing cognitive‑behavioral interventions for fatigue, improving sleep conditions) may be beneficial for alleviating emotional distress, which should be addressed in conjunction with appropriate psychological support.

Our study also found that the transition from inpatient to outpatient care specifically strengthened the edge between fatigue and sleep disturbance. A possible explanation is that the inpatient environment, with its regular daily routines, scheduled medication administration, and professional medical support, may reduce the vicious cycle between fatigue and sleep. After discharge, older patients at home may face irregular schedules, insufficient or excessive activity levels, and decreased medication adherence, which can exacerbate the interaction between fatigue and sleep disturbance. This finding is consistent with the study by Provan,⁴⁶ which showed that sleep disturbance is significantly associated with a persistent fatigue trajectory, suggesting that identifying sleep problems during hospitalization may help prevent worsening fatigue after discharge. Chen⁴⁷ have found that the re-engineered discharge (RED) care model can significantly improve patients’ anxiety, depression, and joint symptoms. Discharge planning could incorporate combined management of fatigue and sleep, such as providing sleep-related guidance, rational use of sleep medications, encouraging patients to record their sleep status and analyze potential influencing factors, as well as offering advice on physical activity and skills for emotional management to better help patients manage these two symptoms.

The strength of this study is the use of cross-lagged panel network analysis to identify core symptoms and examine longitudinal predictive relationships among symptoms in older RA patients, thereby providing theoretical support for future precise interventions. However, several limitations should be noted. First, this was an observational study with only two measurement time points; CLPN can identify predictive relationships but cannot establish causality. Second, disease activity (eg, DAS28) and treatment variables were not included in the network model, and unmeasured confounding cannot be ruled out despite covariate adjustment. Third, the single‑center design and use of self‑report measures may have introduced selection and reporting biases. Fourth, the one‑month follow‑up may not fully capture slowly changing symptoms (eg, anxiety‑depression). Future studies with longer follow‑up and inclusion of clinical and treatment variables are needed to further validate these findings.

Conclusion

This study reveals the dynamic evolution of symptom networks in older RA patients during the transition from inpatient to outpatient care. Fatigue emerged as the primary longitudinal predictor of subsequent sleep disturbance and functional limitation, whereas anxiety‑depression exhibited the highest susceptibility to being predicted by other symptoms. The inpatient-to-outpatient transition specifically strengthened the connection between fatigue and sleep. These findings may inform future symptom management strategies and generate hypotheses for interventional research—whether fatigue-targeted interventions during hospitalization or combined fatigue‑sleep management in discharge planning could improve patient outcomes.

Funding Statement

This work was supported by the Beijing Research Ward Excellence Program BRWEP2024W234050108, The Beijing High-Level Innovation and Entrepreneurship Talent Support Program young top talent projects (G202522083), National Natural Science Foundation of China (72304046; 72474032), National High Level Hospital Clinical Research Funding (BJ-2025-133; BJ-2024-199), National Key R&D Program of China (2020YFC2008504).

Disclosure

The authors report no conflicts of interest in this work.