Comparing Low and High Recovery Expectations Among Those with Lumbar Spinal Stenosis: A Network Analysis

Abstract

Objective

This study aimed to (1) compare networks of factors for patients with low and high recovery expectations in individuals with lumbar spinal stenosis (LSS) using biopsychosocial measures and duration of symptoms and (2) identify the most influential factors within each group.

Methods

We conducted a cross-sectional study using baseline data from 598 patients with LSS. Participants were recruited through Duke University and the University of Washington spine clinics between 2021 and 2024. Recovery expectations were assessed using a 0 (lowest) to 10 (highest) numerical rating scale. We stratified participants into low and high expectation groups. Network analysis was used to assess relationships among factors for each group. Networks included chronicity and PROMIS-29+2 Profile. Partial correlations, centralities, network comparison measures, and network stabilities were calculated.

Results

There were 226 participants with low and 108 with high recovery expectations. The two networks did not have significant structural differences and strength centralities were stable. Chronicity was not significantly associated with other factors in both networks. Pain Interference and Fatigue strength centralities show greater importance in both networks. Anxiety had significantly higher strength in the low expectation group, suggesting greater relevance for individuals with low expectations.

Conclusion

Pain Interference and Fatigue were influential for both groups, while Anxiety was more influential for those with low expectations. Future research and patient management should consider the influence of anxiety for those with low recovery expectations and pain interference and fatigue for all individuals with LSS on other associated factors.

Introduction:

Lumbar spinal stenosis (LSS) is a progressive degenerative disorder that leads to pain and disability affecting about 11% to 38% of the population.¹ It generally starts in the late fourth decade of life with increasing prevalence every consecutive decade.^1,2 It is a structural narrowing of the spinal canal and foramina compressing nerves, causing pain, weakness, and abnormal sensations in the back and legs, predominantly when standing and walking.³ An individual’s prognosis is difficult to predict due to the complex interactions among physical, psychological, and social factors contributing to the disorder.^2,3

Recovery expectations are an individual’s estimated probability of symptom resolution or functional improvement after a period of time.^4,5 They are a predictor of outcomes, with one study showing individuals with low recovery expectations being more than twice as likely to be disabled due to a work-related injury compared to those with high expectations.^4–6 Individuals with musculoskeletal pain conditions who have high recovery expectations have lower associated pain and disability scores compared to those with low expectations of recovery, and vice versa.^5,6 Expectations are determined by perceptions derived from patient beliefs and experiences that stem from both psychosocial and physical factors.⁷ Identifying which factors that most shape an individual’s recovery expectations may help improve health-related outcomes.

Patient-reported outcomes (PROs) have been used to assess factors that influence recovery expectations in musculoskeletal pain populations, including pain intensity and severity, pain interference, disability, self-efficacy, distress, fear avoidance, anxiety, depression, catastrophizing, treatment satisfaction, health-related quality of life, and impression of health change.^4,5,8–10 Although psychological and physical factors are associated with recovery expectations, identifying which of these factors most strongly influence one another among those with LSS at very low or very high recovery expectations is challenging due to the complexity of their interrelationships.

Network analysis is increasingly utilized in psychological and medical research to quantify and qualify the complex interplay between multiple biological, psychological, and social factors.^11,12 It provides an opportunity to overcome this hurdle and fill the knowledge gap that exists for those with LSS.^11,13 Network analysis characterizes relationships among factors, often employing graphical representations to help describe the experience of a health condition.¹³ Network analysis has been used to study populations with pain conditions (i.e., low back pain, fibromyalgia, and chronic pain), but has not been used to further our understanding of recovery expectations for those with LSS.^14–18 A framework illustrating the most influential factors can help researchers and clinicians better understand the relationship between components that contribute to recovery expectations for those seeking care.

The primary aim of this study is to compare the network of factors for patients with LSS who have either low or high recovery expectations. We hypothesize that the networks will be structurally different due to differences in prognostic factors and outcomes for those reporting low versus high recovery expectations. The secondary aim of the study is to identify the most influential factors across the two networks. The findings from this study will help guide our understanding of recovery expectations and guide future research on individuals with LSS.

Methods:

Design and Participants:

This is a cross-sectional study using baseline data from the Lumbar Stenosis PROgnostic Subgroups for PErsonalizing Care and Treatment Study (PROSPECTS) cohort.¹⁹ Institutional Review Boards approved the data collection at the University of Washington (#STUDY00011262) and Duke University (#Pro00107101). PROSPECTS is a prospective cohort study of patients with LSS. Participants were aged 50 years or older and were seeking nonsurgical care for a new visit for symptomatic LSS at the Duke University (Durham, NC) and University of Washington Medicine (Seattle, WA) spine clinics from 2021 through 2024. Initial visits were defined as a lumbar spine-related visit with no specialty care visits or procedures for the lumbar spine in the previous 6 months. Participants were included if they had symptomatic LSS, based on reporting at least 3 of 4 criteria that are diagnostic for LSS.²⁰ Participants were excluded if they had been diagnosed with inflammatory spondyloarthropathy, spinal malignancy, spinal infection, vascular claudication, developmental spine deformities, severe vascular disease that may limit ambulation, severe pulmonary disease that may limit ambulation, severe coronary artery disease that may limit ambulation, severe osteoporosis, prior spine surgery, or cancer in the prior 5 years (excluding non-melanomatous skin cancer). The recruitment process and eligibility criteria are described in full detail elsewhere.^19,21

Measures:

Baseline data were collected electronically using REDCap software. Age was reported in years. Pain level was measured using an 11-point numeric rating scale (NRS) with 0 being “no pain” and 10 being “pain as bad as you can imagine”.¹⁹ Participants were asked to rate the average pain in their back and legs over the past week. Options for reported sex were male, female, or prefer not to answer.

The primary measures used for the network analysis were recovery expectations, chronicity, and the Patient-Reported Outcomes Measurement Information System (PROMIS-29+2 Profile v2.1) short-form questionnaires. Recovery expectations were measured using a single question: “How confident are you that your back and/or leg pain will be completely gone or much better 3 months from now?”¹⁹ This is rated on an 11-point NRS with 0 representing “not confident at all” and 10 representing “extremely confident”. A single-item recovery expectation question is as valid as multi-item questionnaires in predicting outcomes for low back pain.⁶ To separate the sample into subgroups based on expectations, scores were trichotomized into low (0-3), medium (4-7), and high (8-10) recovery expectations. We chose a cut-off point of three for low recovery expectations since it has the highest odds of a poor recovery compared to splitting the NRS into two equal halves based on a review of the findings by Carrière et al. (2023).⁶ Additionally, the measurement error for a single-item recovery expectation NRS is reported to be greater than 1 point for individuals with low back pain.²³ Therefore, we dropped the medium recovery expectations group from this analysis to serve as a buffer for this potential error and to compare groups at the extremes of recovery expectations, where we expected the most difference between the networks.²⁴

Chronicity was measured using a six-point ordinal scale with varying time increments. We consolidated Chronicity into a five-point ordinal scale by combining symptoms <1 month and 1-3 month categories. The final Chronicity categories were: <3 months, 3-6 months, 6-12 months, 1-5 years, and >5 years. Chronicity was included in the analysis since it is associated with recovery expectations – greater chronicity is associated with lower recovery expectations.^25,26

The PROMIS-29+2 measures are valid and reliable when assessing individuals with back pain.²⁷ The domains include: Anxiety, Cognition, Depression, Fatigue, Pain Interference, Physical Function, Sleep Disturbance, and Social Participation.²⁷ They represent physical, mental, and social health and well-being.^27,28 The PROMIS measures use T-scores where the mean in the general population is 50 and the standard deviation is 10. For Cognition, Physical Function, and Social Participation, lower scores represent greater difficulty and higher scores represent lower difficulty. For the measures of Anxiety, Depression, Fatigue, Pain Interference, and Sleep Disturbance, lower scores represent less difficulty, while high scores represent greater difficulty.

Statistical Analysis:

All analyses were conducted using R software (version 4.3.2, R Foundation for Statistical Computing, Vienna, Austria). We used descriptive statistics of number and percent or mean and standard deviation to characterize each group. (Table 1)

Table 1.

Baseline demographics and PROMIS-29+2 measures

DEMOGRAPHIC VARIABLES	LOW EXPECTATIONS GROUP (N=225)	HIGH EXPECTATIONS GROUP (N=108)
AGE (MEAN, SD)	65.9 (9.1)	67.7 (8.4)
MISSING (N, [%])	6, [2.7%]	0, [0%]
FEMALE (N, [%])	127, [57.0 %]	69, [64.5%]
MISSING (N, [%])	2, [0.9 %]	1, [0.9%]
BACK PAIN LEVEL (MEAN, SD)	4.8 (2.5)	4.5 (2.6)
LEG PAIN LEVEL (MEAN, SD)	5.4 (2.5)	5.5 (2.5)
CHRONICITY (N, [%])
<3 MONTHS	10, [4.4%]	39, [36.1%]
3-6 MONTHS	13, [5.8%]	12, [11.1%]
6 MONTHS - 1 YEAR	14, [6.2%]	13, [12.0%]
1-5 YEARS	72, [32.0%]	16, [14.8%]
>5 YEARS	116, [51.6%]	28, [25.9%]
PROMIS MEASURES (MEAN, SD)
ANXIETY	52.9 (9.4)	49.7 (8.3)
COGNITION	51.1 (7.3)	53.8 (7.4)
DEPRESSION	51.4 (8.7)	47.1 (6.8)
FATIGUE	55.4 (9.0)	50.1 (8.9)
PAIN INTERFERENCE	62.4 (7.1)	61.2 (6.3)
SOCIAL PARTICIPATION	43.1 (7.7)	45.9 (7.2)
PHYSICAL FUNCTION	37.6 (5.9)	38.6 (5.6)
SLEEP DISTURBANCE	54.5 (8.7)	50.5 (7.2)

N = number of participants; % = percent of sample; SD = standard deviation; No missingness for back and leg pain intensity, chronicity, and all PROMIS measures.

We used the qgraph, bootnet, NetworkComparisonTest, and corrplot packages to conduct network analysis and generate visualizations. In network analysis, nodes represent factors while edges are the connections between the nodes (Figure 2). The nodes in our network were chronicity and the PROMIS-29+2 measures for Anxiety, Cognition, Depression, Fatigue, Pain Interference, Physical Function, Sleep Disturbance, and Social Participation. The edges are the regularized pairwise partial correlations calculated between each node using the cor_auto command. Cor_auto is a function in qgraph that identifies variable type to execute polychoric, polyserial, or Pearson correlations as appropriate.²⁹ The thickness of the edges represents the weighted partial correlations between the nodes with thicker edges depicting stronger correlations. The colors of the edges represent either positive (purple) or negative (orange) correlations. (Figure 2). Networks for the low and high expectations groups were created using the EBICglasso (Extended Bayesian Information Criterion graphical least absolute shrinkage and selection operator) algorithm in the estimateNetwork command which is ideal for relatively small datasets.¹² The EBICglasso algorithm incorporates a penalty to limit the sum of absolute parameter values, thereby reducing the small weak edges toward zero and eliminating spurious connections between nodes to identify the optimal network structure.^12,29 We plotted the networks using the “spring” layout which is a force-directed method that positions the nodes based on a balance between nodes without edges repelling, nodes with edges attracting, and optimizing visual interpretability.

Figure 2.

Network visualizations. (A) Low recovery expectations network. (B) High recovery expectations network. Purple represents positive correlations and orange represents negative correlations. Each node was labeled at the center for Chronicity, Anxiety, Cognition, Depression, Fatigue, Pain Interference, Physical Function, Sleep Disturbance, and Social Participation and color-coded by community membership.

A corresponding correlation matrix was created to visualize the estimates of the edges in the network using the corrplot package. The correlation matrix color scheme is the same as the network visualization, with purple representing positive associations and orange representing negative ones. Spurious correlations that were dropped when estimating the network were not represented as edges in the graphs. Also, values between -0.01 and 0.01 were not represented in the correlation matrix but kept in the network visualization due to differences in the qgraph and corrplot package formatting. Correlations were graded from none (0), poor (>0 to ±0.2), fair (±0.2 to ±0.5), moderate (±0.5 to ±0.7), very strong (±0.7 to ±0.9), and perfect (±0.9 to ±1).³⁰ We then applied community detection using leiden_cluster in igraph to identify communities of nodes, in which nodes within a community may be more closely related to one another than to nodes in other communities, thereby providing insight into the relationships among the variables.³¹

A network comparison test was conducted to assess the differences between the two networks. The NCT command in the NetworkComparisonTest package was used to analyze the network structure invariance test and the global strength invariance test using 1000 permutations as a default.²⁹ These two tests are metrics to determine if the networks are statistically different. The network structure invariance test estimates differences in how the nodes are connected between the networks.³² The global strength invariance test estimates the difference in the absolute sum of the edge strengths between the networks.³² A significance level of p<0.05 was set to determine if there is a difference in the structure of the networks when comparing the low recovery expectations and high recovery expectations groups for both comparison tests.

For our secondary aim, we estimated the network strength centrality for the low and high recovery expectations groups. Strength quantifies how much a node is connected to other nodes in the network by calculating the absolute, weighted sum of edges connected to each node.^29,33 This centrality measure is the most suitable for our study as we aim to understand the potential influence of the nodes.³³ The nodes with high strength centrality have a greater propensity to influence or be influenced by changes in the network.³³ The centralityPlot and centralityTable commands were used to calculate the z-score-standardized estimates and to visualize the strength of the nodes in the network.

The bootnet command was used to measure the stability and edge-weight accuracy. A correlation-stability coefficient, also known as a CS-coefficient, was used to assess the stability of the strength centrality estimates, and is based on how the removal of cases or nodes affects the estimates.^29,34 Non-parametric bootstrapping was used to create 5000 samples at 95% CIs. CS-coefficient estimates greater than 0.7 are considered optimal, greater than 0.5 are preferred, greater than 0.25 are acceptable, and anything below is deemed not acceptable or unstable.²⁹ Also, the robustness of the edge-weights was estimated using the edge value and associated bootstrapped confidence intervals. Confidence intervals that overlapped were not considered significantly different from other edges.

Finally, we completed a sensitivity analysis using the total sample and included Recovery Expectations with Chronicity and all PROMIS measures to provide added context to Recovery Expectations’ influence on the rest of the network.

Results:

The study consisted of 598 participants, of which six were removed due to missing recovery expectation data. This resulted in a total of 592 participants, with 226 (38%) in the low recovery expectations group and 108 (18%) in the high recovery expectations (Table 1). There were 259 (44%) participants in the medium recovery expectation group who were removed from the analysis, leaving a total of 334 participants analyzed. There was no missing data for the variables of interest – PROMIS measures and chronicity. The demographic and health characteristics of each group are described in Table 1. A greater proportion of participants with symptoms >1 year were in the lower recovery expectation group compared to the high recovery expectation group.

Correlations among the factors included in the network analysis are presented in Figures 1A and 1B. Only the association between Anxiety and Depression was moderate in the low recovery expectations group (Figure 1A) but was fair in the high recovery expectations group (Figure 1B). Fair associations were found between Physical Function and Pain Interference, Physical Function and Social Participation, and Pain Interference and Social Participation in both groups. The association between Fatigue and Sleep Disturbance was fair in the low recovery expectations and poor in the high recovery expectations group. The association between Fatigue and Cognition was fair in the high recovery expectations group and poor in the low recovery expectations group. Chronicity was poorly associated with Anxiety, Cognition, and Depression in the low recovery expectations network (Figure 1A) and poorly associated with Fatigue in the high recovery expectations network (Figure 1B).

Figure 1.

Correlation matrices. (A) Low recovery expectations correlation matrix. (B) High recovery expectations network correlation matrix. Purple represents positive correlations and orange represents negative correlations. The correlation color gradient is depicted on the right of each correlation matrix. Chronicity and PROMIS measures for Anxiety, Cognition, Depression, Fatigue, Pain Interference, Social Participation, Physical Function, and Sleep Disturbance are presented in the x- and y-axes.

The network visualizations for the low and high expectations groups are presented in Figure 2. The low recovery expectations (Figure 2A) network had more edges, with 25 out of a total of 36 possible edges present, compared to 21 of 36 in the high recovery expectations network, indicating a greater network density (Figure 2B). The direction of association for each edge was the same in both networks. The visualization showed Pain Interference, Physical Function, and Social Participation formed a community of yellow nodes (Figures 2A & 2B) while Depression, Anxiety, Sleep Disturbance, Fatigue, and Cognition formed a separate community of blue nodes. Chronicity was shown in green as its own community and was positioned far from all other nodes with limited weak edges in both networks. Fatigue was centrally located and connected all nodes in both networks, except for a missing edge with Chronicity in the low recovery expectation group.

The discrepancy in the number of edges/partial correlations between the correlation matrix (Figure 1) and the network visualization (Figure 2A&B) was due to the differences in formatting between the qgraph and corrplot packages in R. The correlations between Anxiety-Physical Function (r=−0.004) and Cognition-Depression (r=−0.004) in the low recovery expectation correlation network (Figure 2A) and Cognition-Pain Interference (r=−0.003) in the high recovery expectation network (Figure 2B) were dropped in the correlation matrices.

The edge-weight accuracy visualization for all edges in the networks was presented in Figure 3. In the low recovery expectations group, the Anxiety-Depression, Pain Interference-Social Participation, Physical Function-Pain Interference, and Pain Interference-Social Participation edges were reliably the strongest edges (Figure 3A). In the high recovery expectations group, only Pain Interference-Social Participation and Physical Function-Pain Interference were reliably some of the strongest edges. (Figure 3B). Bootstrapped edge-weight mean estimates and confidence intervals can be found in Supplementary Table S1.

Figure 3.

Edge-weight accuracy. (A) Low recovery expectations edge-weight accuracy. (B) High recovery expectations edge-weight accuracy. The left column shows the node-to-node edge. The red lines are the sample partial correlation coefficient estimates. Black lines show the bootstrapped partial correlation coefficient mean estimate. The grey bars indicate the bootstrapped 95% confidence intervals.

The strength centralities of the factors in each network are presented in Figure 4. In the low recovery expectations network, the greatest strength centralities were Pain Interference (z-score=1.01), Social Participation (z-score=0.78), Anxiety (z-score=0.77), and Fatigue (z-score=0.55) (Figure 4A & Table 2). This differed in the high recovery expectations network as Fatigue (z-score=1.39) had the greatest strength, followed by Pain Interference (z-score=0.77) and Physical Function (z-score=0.54) (Figure 4B & Table 2). In both networks, Chronicity had very low influence (low expectations: z-score=−2.04; high expectations: z-score=−2.08). Strength CS-Coefficients for low recovery expectations and high recovery expectations were 0.67 (0.60, 0.75) and 0.36 (0.29, 0.44), representing preferred and acceptable levels of network stability, respectively (Table 2).

Figure 4.

Network centralities. (A) Low recovery expectations centralities. (B) High recovery expectations centralities. Chronicity and PROMIS measures for Anxiety, Cognition, Depression, Fatigue, Pain Interference, Physical Function, Sleep Disturbance, and Social Participation are presented on the left-hand side of the centrality plots with associated Z-standardized estimate plots.

Table 2:

Strength centrality z-standardized estimates, and network comparison

Nodes	Strength (z-standardized estimates)
	Low Expectations	High Expectations	p-value
Anxiety	0.77	−0.13	0.04*
Chronicity	−2.04	−2.08	0.64
Cognition	−0.51	−0.14	0.70
Depression	0.26	0.23	0.73
Fatigue	0.55	1.39	0.18
Pain Interference	1.01	0.77	0.36
Social	0.78	0.25	0.16
Participation
Physical Function	0.14	0.54	0.55
Sleep	−0.97	−0.84	0.93
Disturbance
Network Invariance Test		M=0.18	0.71
Global Strength Invariance Test		S=0.20	0.43

^*Statistically significant (p<0.05)

Table 2 presents the findings from the Network Comparison Test comparing the strength centralities of the two networks. The overall networks were not significantly different when assessed using the Network Invariance Test (p=0.71) and the Global Strength Invariance Test (p=0.43). When comparing node strengths between networks, a significant difference was observed between Anxiety, with Anxiety having a greater influence in the low than high recovery expectations network (p=0.04).

The correlation matrix, network visualization with community detection, edge-weight accuracy, and strength centralities visualizations and associated estimates can be found in Supplementary Materials (Figures S1–S4 & Table S2–S3). An interpretation of the sensitivity analysis was provided in Supplementary Table S4.

Discussion:

The networks of biopsychosocial factors for those with low and high expectations did not exhibit significant structural differences. The direction (positive or negative) and relative magnitude of associations were also similar between networks. Overall, this suggests that those with low and high recovery expectations have similar relationships among physical and mental health factors.

Visual inspection of both networks (Figure 2) revealed that Pain Interference, Social Participation, and Physical Function were in close proximity, with fair associations, and formed a community of more densely connected nodes, suggesting a close relationship between these nodes. This is consistent with the literature and coincides with the physical health factors of the PROMIS-29.^35,36 Anxiety, Cognition, Depression, Fatigue, and Sleep Disturbance formed a community in both networks, showing greater connections between these factors compared to physical health factors and Chronicity. Hays et al (2018) found that anxiety and depression primarily defined the mental health component of PROMIS-29 measure while fatigue and sleep disturbance contributed to both physical and mental health factors. Huang et al (2019) concluded that fatigue and sleep disturbance contributed more to mental health factors for older adults with chronic conditions.³⁷ We support Huang et al’s adaptation to mental health factors while adding Cognition from the PROMIS-29+2 v2.1 profile. Broadening the scope of mental health factors better represents the relationship between cognitive and physiological response factors for those with LSS.

It was surprising that Chronicity lacked influence in the multimodal networks of low and high recovery expectations when compared to other factors. Our findings indicated that Chronicity had minimal importance in either network, but our sensitivity analysis showed a reliable, fair association and formed a community with Recovery Expectations when included in a single network. The literature supports an association between chronicity and recovery expectations for individuals with low back pain.^25,26 The findings from the sensitivity analysis, with a single network, may be more useful in generating hypotheses about the role of recovery expectations moving forward.

Prior literature also indicates that chronicity is associated with anxiety, cognition, depression, fatigue, pain inference, social participation, physical function, and sleep disturbance, which were supported in our sensitivity analysis but less indicative in our primary analysis.^26,38–42 The relationships between Chronicity and mental health factors (i.e., Anxiety, Depression, Cognition, and Fatigue) at differing recovery expectations and their relationships with Recovery Expectations in the sensitivity analysis may indicate that the relationship between Chronicity and mental health factors are mediated by Recovery Expectations. This should be further investigated in future studies. Ultimately, these findings suggest that chronicity is not as directly related to how individuals with LSS experience other aspects of their condition. LSS is a chronic condition, as represented in our sample. The distribution of Chronicity may partly explain why it is less influential in the networks for patients with LSS.

For our second aim, we found that Anxiety was significantly more influential in the low than in the high recovery expectations group when comparing strength centrality estimates. These findings are consistent with other studies that have utilized network analysis for chronic pain.^14,17 Anxiety may play a role in the perceptions of other health factors, especially in those with low recovery expectations. We recommend interpreting our findings with caution due to the exploratory nature of this study. Future studies should evaluate the effect of anxiety interventions on the rest of the network and vice versa.

Fatigue is noteworthy for its high strength and its orientation in both networks. In the literature, fatigue is associated with chronic low back pain and is a predictor of disability and mental well-being and often co-occurs with chronic pain.^43–45 Fatigue was centrally located in all networks and the only node connecting all physical and mental health nodes in both networks, supporting its role as a bridging node or mediating factor. Additionally, it had one of the highest strength centralities across all networks. The spatial orientation interpretation should be analyzed with caution given the nature of the force-directed algorithm implemented in this study. Our findings support the notion that, in those with LSS, fatigue is similarly associated with both physical and mental health factors, as demonstrated by community detection exhibiting a stronger association with mental health factors, providing a deeper understanding of its relationship to the two domains of the PROMIS-29. Future studies should investigate whether fatigue mediates the effects of interventions on outcomes across all individuals with LSS, regardless of recovery expectations.

A secondary finding of interest is related to Pain Interference. Current pain level and how it interferes with activities have been shown to have a negative association with recovery expectations.^4,5,46 Using network analysis, we found Pain Interference had the highest strength centrality in the low recovery expectations group and the second highest in the high recovery expectations group. Pain Interference may be an important contributor to the perceptions of other factors in the network for all individuals with LSS, regardless of recovery expectations, as our findings illustrate the complexity of the multidimensional pain experience. Future studies should investigate how interventions aimed at mitigating pain can impact the rest of the network and vice versa.

Limitations:

There are four primary limitations in this study. The first limitation is that this is a cross-sectional analysis, so our results cannot predict how these factors may affect future outcomes. Also, we do not know how individuals’ recovery expectation networks change over time. Following participants through the course of the disease can enhance our understanding of the evolution of these factors and how specific treatments impact the networks. Future research should incorporate longitudinal findings to better understand the changes in node influence and network structure over time.

The second limitation is that we used only eight modifiable factors to assess physical health, psychological factors, social impact, and symptoms. The PROMIS-29+2 v2.1 was initially chosen because it was validated for those with spine conditions and included an array of mental and physical health domains.^27,47 The LSS pain experience may be more complex than the PROMIS-29+2 v2.1 can account for alone. Including other clinical measures and psychosocial factors may enhance our holistic understanding of the LSS pain experience. Future studies should consider incorporating a broader range of factors when evaluating networks for individuals with pain conditions.

The third limitation is related to the characteristics of the participants, as mentioned earlier. LSS can have an onset as early as the third and fourth decades, whereas the age range in our study was primarily older, with a focus on those older than 50. Also, these participants had primarily symptoms >1 year, high pain interference, and greater disability overall. This sample limits our understanding of recovery expectations for those at earlier stages of the condition, more acute symptoms, milder symptoms, and greater physical capacity, thus affecting the generalizability of our findings.

The fourth limitation is the smaller sample sizes of the low- and high-recovery-expectation groups. We compared only low and high recovery expectations while removing the middle group. Future studies should replicate our study with a larger sample of individuals with LSS to confirm our findings.

Conclusion:

We found no significant differences in the network structures when participants are stratified based on recovery expectation ratings. Physical and mental health factors are two clusters found in the networks of those with LSS. Pain Interference and Fatigue are influential factors in both networks. Anxiety is significantly more influential in the low recovery expectations network compared to the high recovery expectations. Chronicity’s relationship with mental health factors may be mediated by recovery expectations. Future research and LSS management should consider the influence of anxiety for those with low recovery expectations and pain interference and fatigue for all individuals with LSS.