Physical activity trajectories and associated factors in patients with lung cancer undergoing chemotherapy: A longitudinal study

Abstract

Objective

To characterize heterogeneity in physical activity (PA) trajectories among patients with lung cancer during early chemotherapy and examine associated demographic, clinical, and psychosocial factors.

Methods

This longitudinal study enrolled 155 patients initiating chemotherapy at the First Affiliated Hospital of Anhui Medical University in Anhui Province, China. PA was assessed using the International Physical Activity Questionnaire-Short Form at baseline and before the second, third, and fourth cycles. Symptom burden, exercise self-efficacy, outcome expectations for exercise, and social support were assessed exclusively at baseline. Latent class growth analysis was utilized to identify PA trajectories. Factors associated with trajectory membership were analyzed using a two-stage strategy combining least absolute shrinkage and selection operator regression and Firth’s penalized logistic regression.

Results

Among 122 completers, two distinct trajectories were identified. The majority (77.9%) followed a “Moderate-Sustained Decline” trajectory, showing a significant drop at cycle 2 with persistently low activity. Conversely, 22.1% exhibited a “High-Decline-Partial Recovery” trajectory, characterized by a behavioral rebound after the initial decline. Multivariate analysis revealed that higher symptom burden (OR = 1.105, P = 0.028) and more pessimistic outcome expectations (OR = 1.723, P = 0.005) increased the likelihood of sustained decline. Higher exercise self-efficacy (OR = 0.855, P = 0.003) was significantly associated with the partial recovery profile.

Conclusions

PA patterns are heterogeneous, though a decline prior to the second cycle is universal. Lower symptom burden and stronger cognitive resources may support more favorable physical activity trajectories. Proactive nursing interventions integrating symptom management with cognitive-behavioral empowerment targeting the critical window before the second cycle are recommended to prevent functional decline.

Introduction

Lung cancer remains a critical global health challenge, accounting for the highest incidence and mortality rates among all malignancies worldwide. In 2022, approximately 2.48 million new cases were diagnosed, leading to an estimated 1.82 million deaths.¹ Due to the disease’s insidious onset, most patients are diagnosed at advanced stages, for whom platinum-based combination chemotherapy remains the primary first-line treatment.² While clinically effective, chemotherapy-related toxicities (e.g., nausea, vomiting, fatigue, neurotoxicity, and bone marrow suppression)3, 4, 5 often accumulate across cycles and can substantially impair physical functioning and daily activity.^3,4

Physical activity (PA) is widely recognized as a safe and cost-effective non-pharmacological strategy for cancer care. Accumulating evidence suggests that appropriately tailored PA can help mitigate symptom burden,⁵ prevent functional decline,⁶ and improve treatment tolerance.⁷ Current clinical guidelines from the American College of Sports Medicine (ACSM) and the American Cancer Society (ACS) advocate that patients maintain ≥ 150 minutes of moderate-intensity or ≥ 75 minutes of vigorous-intensity PA weekly throughout treatment.⁸ Despite these benefits and recommendations, PA levels during chemotherapy remain frequently low.9, 10, 11 A systematic review of 5665 patients confirmed generally low PA levels both before and during treatment.¹² Unfortunately, evidence suggested that this sustained inactivity can perpetuate a vicious cycle of accelerated functional decline, exacerbated treatment-related fatigue, and poorer prognosis.13, 14, 15

However, most existing studies have relied on cross-sectional designs or have summarized longitudinal change using a single average trajectory. Such variable-centered approaches assume population homogeneity, implying that all patients follow a broadly similar behavioral pattern over time.⁷ This assumption overlooks the clinical reality of chemotherapy, where patients face fluctuating symptom burdens and possess varying degrees of physiological reserve and adaptive capacity.16, 17, 18 It is clinically plausible that distinct subgroups of patients exhibit qualitatively different PA patterns. Ignoring this heterogeneity may hinder the identification of high-risk patients who require prioritized intervention. First, relying on average trajectories may conceal a subgroup of patients who experience rapid and sustained functional decline, thereby delaying necessary intensive support. Second, uniform exercise prescriptions—assuming all patients respond similarly—may prove ineffective or even discouraging for those lacking sufficient psychological resources. Therefore, identifying distinct PA trajectories is a prerequisite for precision nursing that tailors interventions to individual risk profiles and adaptive capacities. To capture this population heterogeneity, person-centered trajectory methods are needed. Latent class growth analysis (LCGA) is a specific framework within latent class growth modeling where within-class variability is constrained.¹⁹ This approach allows for the identification of unobserved subgroups that share similar longitudinal PA patterns. By distinguishing these unique trajectories, clinicians can better target resources, patient education, and timely supportive care toward those most in need.

While identifying these trajectories is a necessary first step, it is equally critical to explain why patients diverge into different pathways. Bandura’s social cognitive theory (SCT) provides a robust framework for this purpose. SCT posits that behavior is shaped by the reciprocal interaction between personal, behavioral, and environmental factors.^20,21 In the context of chemotherapy, personal factors include not only demographic and clinical characteristics but also cognitive constructs—particularly exercise self-efficacy (confidence in one’s ability to be active despite barriers like fatigue) and outcome expectations (beliefs about the benefits or harms of activity). Meanwhile, environmental conditions (e.g., support from family and healthcare providers) may affect PA both directly and by modifying patients’ cognitive resources. Integrating SCT with trajectory analysis thus provides a theoretical lens to understand the psychological mechanisms driving differential PA patterns.

Accordingly, this study aimed to apply LCGA to identify distinct PA trajectories across the first four chemotherapy cycles in patients with lung cancer, and to examine the demographic, clinical, symptom-related, and psychosocial factors associated with trajectory membership. By characterizing these heterogeneous patterns and their associated factors, this study seeks to inform risk stratification and the development of tailored nursing interventions to prevent functional decline during chemotherapy.

Methods

Study design and setting

This was a prospective longitudinal observational study. Participants were recruited from the Department of Respiratory Medicine at The First Affiliated Hospital of Anhui Medical University in Anhui Province, China, between October 2024 and July 2025.

Study participants

Patients scheduled to receive first-line chemotherapy were recruited using a convenience sampling method. The chemotherapy regimens were primarily platinum-based doublets, with a small proportion receiving other guideline-recommended regimens, administered in 28-day cycles. Patients were included if they met the following criteria: (1) pathologically confirmed diagnosis of primary lung cancer; (2) scheduled to undergo 4–6 cycles of chemotherapy; (3) aged ≥ 18 years; (4) aware of their diagnosis; and (5) possessed sufficient cognitive and linguistic abilities to complete questionnaires. Patients were excluded if they: (1) had physical impairments precluding PA (e.g., severe joint disease, uncontrolled pain); (2) had a history of other malignancies, or severe psychiatric disorders; or (3) were concurrently participating in other clinical trials.

Given the absence of established sample size guidelines for latent class growth analysis in this population, a cross-sectional formula was used to provide a baseline estimate. Sample size was calculated using the formula: N = [U²_α/2π(1 − π)]/δ². Based on the preliminary survey results indicating that 80% (π = 0.80) of patients with lung cancer undergoing chemotherapy failed to meet ACSM physical activity guidelines,⁸ with alpha = 0.05 (Z = 1.96) and a margin of error delta = 0.1π (0.08), the required sample size was calculated as 97. Anticipating a potential attrition rate of approximately 30% due to the longitudinal nature of the study, the target enrollment was set at 139 participants.

Measurements

Demographic and clinical characteristics

A structured questionnaire was used to collect demographic data, including age, gender, marital status, residence, occupational status, education level, monthly family income (CNY), medical expenses payment method, and smoking history. Clinical data (e.g., pathological type, disease stage according to the eighth edition of the AJCC TNM classification, metastasis, comorbidity, surgical treatment, and chemotherapy regimen) were extracted from electronic medical records.

Physical activity

The Chinese version of the International Physical Activity Questionnaire-Short Form (IPAQ-SF) was used to evaluate patients’ physical activity levels. Developed by the International Physical Activity Measurement Working Group,²² this scale comprises seven questions that assess the frequency and duration of vigorous-intensity activity, moderate-intensity activity, and walking over the preceding 7 days, as well as sedentary behavior.²³ Following the standard IPAQ scoring protocol, weekly PA volume was calculated by weighting each activity type by its energy requirements defined in metabolic equivalent of task (MET): walking (3.3 METs), moderate activity (4.0 METs), and vigorous activity (8.0 METs). The total PA score was expressed as MET-minutes/week. In 2004, Qu et al. validated the Chinese version, demonstrating satisfactory test-retest reliability and criterion validity.²⁴

Symptom burden

Symptom burden was assessed at baseline using the Chinese version of the M.D. Anderson Symptom Inventory (MDASI)²⁵ augmented with the Lung Cancer Module (MDASI-LC).²⁶ This instrument combines 13 core cancer symptoms (e.g., pain, fatigue, nausea) with 6 lung-cancer-specific symptoms (e.g., coughing, shortness of breath), totaling 19 symptom items. Additionally, six interference items evaluate the impact of symptoms on daily functioning. Items are rated on an 11-point numeric scale ranging from 0 (“not present”) to 10 (“as bad as you can imagine”). The score of the symptom items represents the overall symptom burden, with higher scores indicating greater symptom severity. Cronbach’s alpha for the MDASI and MDASI-LC in this study was 0.841 and 0.815, respectively.

Exercise self-efficacy

The Self-Efficacy for Exercise Scale (SEES) was used to measure patients’ confidence in their ability to exercise amidst cancer-related barriers.²⁷ The scale consists of 9 items rated on a 0–10 scale (0 = “not confident”, 10 = “very confident”). Total scores range from 0 to 90, with higher scores reflecting greater self-efficacy. Cronbach’s alpha in this study was 0.946.

Outcome expectations for exercise

The Outcome Expectations for Exercise Scale (OEES) assesses beliefs regarding the consequences of exercise.²⁸ It comprises nine items covering both physical and psychological outcomes. Items are rated on a five-point Likert scale from 1 (“strongly agree”) to 5 (“strongly disagree”). In this study, higher scores indicate more negative (pessimistic) expectations regarding the benefits of exercise. Cronbach’s alpha in this study was 0.880.

Social support

The Perceived Social Support Scale (PSSS) was applied to measure support from family, friends, and significant others.²⁹ It consists of 12 items rated on a 7-point Likert scale (1–7). Total scores range from 12 to 84, with higher scores indicating higher levels of perceived social support.³⁰ Cronbach’s alpha in this study was 0.831.

Data collection

The data collection schedule was strictly synchronized with the standard chemotherapy treatment timeline. Given that participants received platinum-based doublet regimens administered in 28-day cycles, assessments were conducted across four consecutive time points aligned with the commencement of each cycle. The baseline assessment (T1) was scheduled 1–3 days prior to the first cycle. Subsequent longitudinal assessments were performed on Day 1 of the second (T2), third (T3), and fourth (T4) cycles, occurring at 28-day intervals immediately preceding chemotherapy administration. At baseline (T1), after obtaining written informed consent, demographic data and all five scales (IPAQ-SF, MDASI-LC, SEES, OEES, PSSS) were collected via face-to-face interviews. Patients’ physical activity levels were dynamically assessed at the three follow-up time points (T2-T4). To minimize attrition, follow-up data were collected in person when patients returned to the hospital for their scheduled chemotherapy. For those unable to attend in person, structured telephone interviews were conducted.

The study implemented rigorous quality control throughout the investigation. All researchers received standardized training on research protocols and assessment methods prior to participation. During face-to-face interviews, researchers were present to answer questions and verify questionnaire completeness. For telephone interviews, key responses were reiterated to participants for confirmation to ensure accuracy. All data underwent double-entry verification to minimize entry errors.

Statistical analysis

Statistical description and preprocessing

Data analysis was performed using SPSS (v27.0), Mplus (v8.3), and R (v4.4.3). Continuous variables were presented as mean ± standard deviation (SD) or median (interquartile range, IQR) based on normal distribution. Categorical variables were expressed as frequencies and percentages. Baseline characteristics did not differ significantly between participants who completed the study and those who dropped out (Supplementary Table S1), suggesting that the missing data pattern was broadly consistent with the missing-at-random (MAR) assumption. Accordingly, complete-case analysis was used as the primary approach for trajectory identification among participants who completed all four assessments (n = 122). To examine the robustness of the identified trajectories against attrition, a sensitivity analysis using full information maximum likelihood (FIML) estimation was conducted on the entire enrolled cohort (n = 155). Two-sided analyses with P < 0.05 were considered significant.

Trajectory analysis

LCGA was conducted in Mplus to identify distinct PA trajectories. LCGA characterizes population heterogeneity by partitioning the population into mutually exclusive latent subgroups, with variance within classes fixed to zero.³¹ Model fit was assessed using multiple indices: the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and sample size-adjusted BIC (aBIC), where lower values indicate better fit. Classification accuracy was evaluated using entropy (range 0–1, with values > 0.80 indicating good classification).³² The optimal number of classes was determined by the Bootstrapped Likelihood Ratio (BLRT) test and the Lo-Mendell-Rubin (LMR) test, with P < 0.05 indicating significant improvement for k-class vs. k − 1 class. Model selection also considered clinical interpretability and ensured that each subgroup comprised ≥ 5% of the sample.

Analysis of factors associated with trajectory membership

Baseline demographic and clinical characteristics were compared between trajectory groups using independent samples t-tests, Mann–Whitney U tests, or Chi-square tests, as appropriate. To identify factors associated with trajectory membership while addressing potential overfitting, a two-stage analytical strategy was adopted. Initially, the least absolute shrinkage and selection operator (LASSO) regression with 10-fold cross-validation was employed to screen and select the most robust variables from the set of univariate-significant variables. Subsequently, these selected variables were entered into a multivariable Firth’s penalized logistic regression model to reduce small-sample bias and provide more stable parameter estimates (odds ratios [ORs] and 95% confidence intervals [CIs]). The discriminative performance of the final model was evaluated using the area under the curve (AUC) based on leave-one-out cross-validation (LOO-CV).

Results

Participant characteristics

A total of 155 patients were initially enrolled at baseline (T1). Of these, 122 participants completed all four assessments (T1-T4), resulting in an attrition rate of 21.3% (Fig. 1). Comparison of baseline data between the retained sample (n = 122) and those lost to follow-up (n = 33) revealed no statistically significant differences in demographic or clinical profiles (Supplementary Table S1), indicating that data were missing at random.

Fig. 1.

Flow chart of data collection of participants. PA, physical activity; IPAQ-SF, International Physical Activity Questionnaire-Short Form; MDASI, M.D. Anderson Symptom Inventory; LC, Lung Cancer; SEES, Self-Efficacy for Exercise Scale; OEES, Outcome Expectations for Exercise Scale; PSSS, Perceived Social Support Scale.

The demographic and clinical characteristics of the final analytical sample (n = 122) are detailed in Table 1. The majority of participants were male (82.8%), aged ≥ 60 years (72.1%), married (97.5%), and resided in rural areas (54.1%). Clinically, most patients were diagnosed with stage IV disease (55.7%) and had metastases (94.3%). The most common chemotherapy regimen was Taxol plus platinum (47.5%).

Table 1.

Demographic and clinical characteristics of the study participants (N = 122).

Variables	n (%)
Age (years)
< 60	34 (27.9)
≥ 60	88 (72.1)
Sex
Male	101 (82.8)
Female	21 (17.2)
Marital status
Married	119 (97.5)
Never married/Divorced/Widowed	3 (2.5)
Residence
City	23 (18.9)
County/Town	33 (27.0)
Rural	66 (54.1)
Occupational status
Employed	28 (23.0)
Retired	30 (24.6)
Unemployed	64 (52.5)
Education level
Elementary school and below	56 (45.9)
Junior high school	47 (38.5)
High school/secondary school and above	19 (15.6)
Monthly family income (CNY)
< 3000	4 (3.3)
3000–5000	63 (51.6)
5001–8000	40 (32.8)
> 8000	15 (12.3)
Medical expenses payment method
Self-payment	2 (1.6)
Basic medical insurance for urban workers	22 (18.0)
New rural cooperative medical insurance	98 (80.3)
Smoking history
Never smokers	48 (39.3)
Current/Former smokers	74 (60.7)
Pathological type
Adenocarcinoma	38 (31.1)
Squamous carcinoma	51 (41.8)
Small cell carcinoma	33 (27.0)
Disease stage
Ⅰ	3 (2.5)
II	6 (4.9)
III	45 (36.9)
IV	68 (55.7)
Metastasis
No	7 (5.7)
Yes	115 (94.3)
Comorbidity
No	83 (68.0)
Yes	39 (32.0)
Surgical treatment
No	108 (88.5)
Yes	14 (11.5)
Chemotherapy regimen
Taxol and platinum	58 (47.5)
Etoposide and platinum	29 (23.8)
Pemetrexed and platinum	28 (23.0)
Gemcitabine and platinum	4 (3.3)
Non-platinum regimens	3 (2.5)

N, number of cases; CNY, Chinese Yuan.

Longitudinal changes in physical activity

PA levels exhibited a distinct temporal pattern: median PA was highest at baseline [T1: 1409.00 (IQR 1120.00, 1866.00) MET-min/w], followed by a sharp decline prior to the second cycle [T2: 1032.00 (IQR 816.75, 1359.00) MET-min/w], and remained suppressed throughout T3 and T4 (median ≤ 990 MET-min/w). Adherence to clinical PA guidelines dropped notably from 32.0% at T1 to 8.2% at T2, with a marginal recovery to 14.8% by T4.

Identification of physical activity trajectories

LCGA was performed to explore heterogeneous PA trajectories, evaluating models with 1–5 latent classes (Table 2). Although information criteria indices (AIC, BIC, aBIC) continued to decrease with additional classes, the 2-class solution was selected as the optimal model, balancing statistical fit with distinct clinical interpretability. This model demonstrated high classification accuracy, indicated by an entropy of 0.928 and high average posterior probabilities (Class 1 = 0.961; Class 2 = 0.987; Supplementary Table S2). Notably, despite a non-significant Lo-Mendell-Rubin (LMR) test (P = 0.431), the 2-class structure was supported by the significant BLRT test (P < 0.001), which offers superior robustness in finite mixture modeling.³³ This final model partitioned the population into two distinct subgroups (Fig. 2). The first subgroup, labeled the “High-Decline-Partial Recovery” trajectory (22.1%), exhibited a high baseline PA level that dropped sharply at T2 (slope = −879.71, P < 0.001) but showed a partial rebound in subsequent cycles. In contrast, the second subgroup, labeled the “Moderate-Sustained Decline” trajectory (77.9%), began with moderate baseline activity, experienced a significant decline at T2 (slope = −326.23, P < 0.001), and remained suppressed throughout the chemotherapy period.

Table 2.

Model fit indices for latent class growth analysis of physical activity trajectories (N = 122).

Profile	K	AIC	BIC	aBIC	Entropy	P value		Profile prevalence
						LMR	BLRT
1	7	7602.607	7622.235	7600.102	–	–	–	–
2	11	7288.461	7319.305	7284.525	0.928	0.431	< 0.001	0.221/0.779
3	15	7098.581	7140.641	7093.214	0.925	0.211	< 0.001	0.123/0.484/0.393
4	19	7003.101	7056.377	6996.303	0.917	0.669	< 0.001	0.115/0.434/0.213/0.238
5	23	6876.594	6941.086	6868.365	0.950	0.002	< 0.001	0.115/0.410/0.221/0.246/0.008

K, number of free parameters; AIC, Akaike information criterion; BIC, Bayesian information criterion; aBIC, sample-size adjusted BIC; LMR, Lo-Mendell-Rubin likelihood ratio test; BLRT, bootstrapped likelihood ratio test.

Fig. 2.

Heterogeneous trajectories of physical activity in patients with lung cancer during four cycles of chemotherapy. PA, physical activity, expressed as MET-minutes per week (MET-min/w); T1, T2, T3, and T4 denote assessment time points prior to the commencement of the first, second, third, and fourth chemotherapy cycles, respectively. Class 1 (blue line): High-Decline-Partial Recovery (22.1% of participants), characterized by a high baseline, a rapid decline at T2, followed by a partial recovery in later cycles; Class 2 (red line): Moderate-Sustained Decline (77.9% of participants), characterized by a moderate baseline, a significant decline at T2, with PA levels remaining low thereafter without recovery. MET, metabolic equivalent of task.

Sensitivity analysis revealed that the trajectory patterns and class proportions obtained using the primary deletion method were consistent with those derived from the FIML method on the entire enrolled cohort (n = 155, Supplementary Table S3 and Fig. S1), suggesting that the findings are robust against attrition bias.

Factors associated with trajectory membership

Univariate analysis indicated that participants in the “Moderate-Sustained Decline” trajectory were significantly older, had a higher proportion of males, higher symptom burden, lower exercise self-efficacy, more pessimistic outcome expectations, and lower social support compared to the “High-Decline-Partial Recovery” trajectory (Table 3). In the two-stage multivariate analysis, LASSO regression first retained four optimal variables: age, symptom burden, exercise self-efficacy, and outcome expectations (Table 4; Supplementary Fig. S2). Subsequently, the Firth’s penalized logistic regression model (Table 5) revealed that higher symptom burden (OR = 1.105, 95% CI: 1.010–1.247, P = 0.028) and more pessimistic outcome expectations (OR = 1.723, 95% CI: 1.158–3.146, P = 0.005) were significantly associated with an increased likelihood of belonging to the “Moderate-Sustained Decline” trajectory. Conversely, higher exercise self-efficacy was significantly associated with a decreased likelihood of membership in this trajectory (OR = 0.855, 95% CI: 0.737–0.953, P = 0.003). Although younger age (<60 years) demonstrated a similar trend towards reduced odds (OR = 0.157), this association did not reach statistical significance in the final penalized model (P = 0.058). This reflects the different aims of the two methods: LASSO selects predictors to minimize cross-validated prediction error, while the subsequent Firth regression tests statistical significance; a variable retained for prediction may not remain significant when its effect is modest. The final multivariable model demonstrated satisfactory discriminative performance, yielding a LOO-CV AUC of 0.981 (95% CI: 0.958–1.000).

Table 3.

Univariate analysis of factors associated with physical activity trajectory membership (N = 122).

Variables	High-Decline-Partial Recovery trajectory (n = 27)	Moderate-Sustained Decline trajectory (n = 95)	Statistic	P value
Age (years), n (%)			25.964	< 0.001a
< 60	18 (66.7)	16 (16.8)
≥ 60	9 (33.3)	79 (83.2)
Sex, n (%)				0.019b
Male	18 (66.7)	83 (87.4)
Female	9 (33.3)	12 (12.6)
Marital status, n (%)				>0.999b
Married	27 (100.0)	92 (96.8)
Never married/Divorced/Widowed	0 (0.0)	3 (3.2)
Residence, n (%)			3.365	0.186a
City	8 (29.6)	15 (15.8)
County/Town	8 (29.6)	25 (26.3)
Rural	11 (40.7)	55 (57.9)
Occupational status, n (%)			3.984	0.136a
Employed	10 (37.0)	18 (18.9)
Retired	6 (22.2)	24 (25.3)
Unemployed	11 (40.7)	53 (55.8)
Education level, n (%)			1.408	0.492b
Elementary school and below	15 (55.6)	41 (43.2)
Junior high school	8 (29.6)	39 (41.1)
High school/secondary school and above	4 (14.8)	15 (15.8)
Monthly family income (CNY), n (%)			2.738	0.397b
< 3000	1 (3.7)	3 (3.2)
3000–5000	15 (55.6)	48 (50.5)
5001–8000	6 (22.2)	34 (35.8)
> 8000	5 (18.5)	10 (10.5)
Medical expenses payment method, n (%)			1.698	0.405b
Self-payment	0 (0.0)	2 (2.1)
Basic medical insurance for urban workers	7 (25.9)	15 (15.8)
New rural cooperative medical insurance	20 (74.1)	78 (82.1)
Smoking history, n (%)			1.126	0.289a
Never smokers	13 (48.1)	35 (36.8)
Current/Former smokers	14 (51.9)	60 (63.2)
Pathological type, n (%)			3.622	0.164a
Adenocarcinoma	11 (40.7)	27 (28.4)
Squamous carcinoma	7 (25.9)	44 (46.3)
Small cell carcinoma	9 (33.3)	24 (25.3)
Disease stage, n (%)			1.917	0.557b
Ⅰ	1 (3.7)	2 (2.1)
II	0 (0.0)	6 (6.3)
III	10 (37.0)	35 (36.8)
IV	16 (59.3)	52 (54.7)
Metastasis, n (%)				1.000b
No	1 (3.7)	6 (6.3)
Yes	26 (96.3)	89 (93.7)
Comorbidity, n (%)			2.884	0.089a
No	22 (81.5)	61 (64.2)
Yes	5 (18.5)	34 (35.8)
Surgical treatment, n (%)
No	25 (92.6)	83 (87.4)		0.733b
Yes	2 (7.4)	12 (12.6)
Chemotherapy regimen, n (%)			6.064	0.157b
Taxol and platinum	8 (29.6)	50 (52.6)
Etoposide and platinum	9 (33.3)	20 (21.1)
Pemetrexed and platinum	9 (33.3)	19 (20.0)
Gemcitabine and platinum	1 (3.7)	3 (3.2)
Non-platinum regimens	0 (0.0)	3 (3.2)
MDASI-LC,Mean ± SD	21.30 ± 8.70	41.21 ± 11.32	−8.499	<0.001c
SEES, M (P25, P75)	47.00 (38.00, 55.00)	24.00 (19.00, 30.00)	−7.235	<0.001d
OEES, M (P25, P75)	12.00 (11.00, 14.00)	19.00 (16.00, 21.00)	−6.893	<0.001d
PSSS, Mean ± SD	66.37 ± 5.51	59.13 ± 5.29	6.225	<0.001c

MDASI-LC, Chinese version of the M.D. Anderson Symptom Inventory with the Lung Cancer Module; SEES, Self-Efficacy for Exercise Scale; OEES, Outcome Expectations for Exercise Scale; PSSS, Perceived Social Support Scale; SD, Standard Deviation; CNY, Chinese Yuan. ^a χ² test; ^b Fisher’s exact probability; ^c Independent samples t test; ^d Mann–Whitney U-test.

Table 4.

Variable selection via LASSO regression for associated factors of trajectory membership (N = 122).

Variables	Coefficient	Selection
Age < 60 years (vs. ≥ 60)	−0.384	Selected
MDASI-LC	0.053	Selected
SEES	−0.102	Selected
OEES	0.130	Selected
Female (vs. Male)	0.000	Excluded
PSSS	0.000	Excluded

Coefficients shown are the non-zero coefficients retained by the LASSO regression under the optimal lambda selected by the one-standard-error rule (λ_1se) via 10-fold cross-validation. Positive coefficients indicate factors associated with an increased likelihood of membership in the “Moderate-Sustained Decline” trajectory; negative coefficients indicate factors associated with a decreased likelihood. LASSO, Least Absolute Shrinkage and Selection Operator; MDASI-LC, Chinese version of the M.D. Anderson Symptom Inventory with the Lung Cancer Module; SEES, Self-Efficacy for Exercise Scale; OEES, Outcome Expectations for Exercise Scale; PSSS, Perceived Social Support Scale.

Table 5.

Multivariate analysis of factors associated with physical activity trajectory using Firth’s penalized logistic regression (N = 122).

Variables	Moderate-Sustained Decline Trajectory
	B	SE	OR	95% CI	P value
Age, years
< 60	−1.853	0.994	0.157	0.011–1.061	0.058
≥ 60	Reference
MDASI-LC	0.100	0.047	1.105	1.010–1.247	0.028
SEES	−0.157	0.057	0.855	0.737–0.953	0.003
OEES	0.544	0.211	1.723	1.158–3.146	0.005

The “High-Decline-Partial Recovery” trajectory served as the reference group. B, regression coefficient; SE, Standard Error; OR, Odds Ratio; CI, Confidence Interval; MDASI-LC, Chinese version of the M.D. Anderson Symptom Inventory with the Lung Cancer Module; SEES, Self-Efficacy for Exercise Scale; OEES, Outcome Expectations for Exercise Scale. Model performance: AUC(LOO-CV) = 0.981. Likelihood ratio test: χ²(4) = 94.22, P < 0.001.

Discussion

This study moves beyond the traditional variable-centered approach that assumes population homogeneity by applying LCGA to visualize the dynamic evolution of PA during the early phase of chemotherapy. We successfully identified two distinct PA trajectories among patients with lung cancer. The majority of patients (77.9%) followed a “Moderate-Sustained Decline” trajectory, while a smaller but resilient subgroup (22.1%) exhibited a “High-Decline-Partial Recovery” trajectory. Both trajectories exhibited a precipitous drop in PA levels prior to the second chemotherapy cycle (T2). This universal decline underscores that the initial administration of chemotherapy acts as a profound physiological and psychological stressor that temporarily overwhelms the coping mechanisms of most patients.³⁴ However, the divergence observed between the second (T2) and third (T3) cycles is particularly compelling. While the sustained decline group continued to deteriorate, the partial recovery group demonstrated a significant behavioral rebound. This shift from decline to increase suggests that the period between the second and third cycles represents a pivotal window of adaptation. It indicates that for a subset of patients, the initial shock of treatment is manageable, and they are capable of re-engaging in physical activity as they physiologically adjust to the treatment regimen and psychologically accept their diagnosis. This study adds to the limited evidence using person-centered trajectory methods to examine longitudinal PA patterns in patients with lung cancer undergoing chemotherapy. The finding that over one-fifth of patients exhibit a partial behavioral rebound challenges the assumption of uniform deterioration and offers a novel target for resilience-enhancing nursing interventions.

Our multivariate analysis identified symptom burden, exercise self-efficacy, and outcome expectations as independent factors associated with trajectory membership. Social Cognitive Theory (SCT) provides a cohesive framework to interpret how these personal and cognitive factors synergistically shape patient behavior through reciprocal determinism. In the “Moderate-Sustained Decline” trajectory, the high baseline symptom burden was identified as a significant factor associated with increased likelihood of membership. Viewed through the SCT lens, this heavy physiological toll may create a context that erodes psychological reserves. Previous literature suggests that persistent symptoms can act as a barrier to self-efficacy, undermining a patient’s confidence in their ability to perform physical tasks.^35,36 Simultaneously, severe symptoms may shape pessimistic outcome expectations, leading patients to internalize the belief that activity will induce fatigue rather than recovery.³⁷ Consequently, the coexistence of high symptom burden and low cognitive resources (self-efficacy and outcome expectations) observed in this subgroup suggests a theoretical vicious cycle that suppresses behavioral engagement.

Interestingly, social support, despite being a core environmental construct in SCT and significant in our univariate analysis, was not retained in the final multivariable model. While this could potentially reflect limited statistical power given our modest sample size, its exclusion during the LASSO variable selection process offers insight into the hierarchical structure of behavioral determinants. Within the SCT framework, environmental factors like social support often function as “distal” antecedents that influence behavior indirectly by bolstering “proximal” cognitive regulators.²¹ The fact that LASSO retained exercise self-efficacy and outcome expectations while discarding social support suggests that the impact of environmental support on PA in our cohort was likely mediated through the enhancement of patients’ confidence or expectation.^38,39

The behavioral rebound observed in the “High-Decline-Partial Recovery” trajectory highlights a distinct phenotype of resilience. Although these patients experienced a pronounced initial decline at T2 consistent with the universal shock of treatment initiation, their subsequent trajectory demonstrates superior adaptive capacity. From an SCT perspective, this resilience appears to be anchored in lower symptom burden and robust initial cognitive resources. Specifically, these patients entered the treatment phase with significantly higher baseline levels of exercise self-efficacy and more positive outcome expectations, alongside a relatively lighter symptom load. This favorable physiological and psychological profile likely acted as a critical buffer, preventing the temporary physical setback at T2 from consolidating into permanent behavioral disengagement.⁴⁰ It is plausible that this foundational confidence enabled these patients to employ effective problem-focused coping strategies, such as self-regulating their activity intensity during nadir periods and actively re-engaging in exercise once acute toxicities subsided.¹⁸ Furthermore, the recovery trend from T2 to T3 suggests a successful psychological transition from the initial crisis of diagnosis to a phase of active disease management, where patients mobilize internal resources to adapt to the new normal of life during treatment.

These findings have direct and actionable implications for clinical nursing practice. The identification of the T2 time point as a universal nadir suggests that nursing interventions are often implemented too late if they wait for functional decline to become clinically apparent. Nurses should view the period before the second chemotherapy cycle as a critical window for proactive intervention. Assessment protocols should be expanded beyond the routine monitoring of physical symptoms to include the screening of cognitive risk factors, specifically low exercise self-efficacy and pessimistic outcome expectations. For patients identified as high risk for the sustained decline trajectory, nurses should implement targeted interventions that disrupt the vicious cycle described by reciprocal determinism of SCT. This includes aggressive symptom management to reduce the physiological burden, combined with cognitive-behavioral strategies designed to boost self-efficacy, such as setting attainable micro-goals and providing verbal persuasion regarding the safety and benefits of movement. For patients showing potential for partial recovery, nursing support should focus on reinforcing their adaptive coping strategies and guiding them to safely titrate their physical activity levels as they recover from the acute effects of each chemotherapy cycle.

Several limitations warrant consideration when interpreting these findings. First, the single-center design and convenience sampling method may restrict the generalizability of the results to broader lung cancer populations with varying socio-demographic backgrounds. Although advanced statistical methods, specifically the two-stage strategy combining LASSO variable selection with Firth’s penalized regression, were employed to enhance model robustness given the modest sample size, these distinct trajectory phenotypes require validation in larger, multi-center cohorts. Second, a primary limitation regarding the theoretical application is that symptom burden and cognitive variables (exercise self-efficacy and outcome expectations) were assessed exclusively at baseline. This measurement design precludes the examination of time-varying associations and limits our ability to fully capture the dynamic, reciprocal evolution between changes in cognition and behavioral fluctuations across chemotherapy cycles. Future research should incorporate repeated measures of these psychosocial constructs to map their concurrent trajectories with physical activity, thereby providing a more granular test of the Social Cognitive Theory framework. Third, detailed data on cycle-specific chemotherapy toxicities were not systematically collected, potentially omitting specific physiological confounders that contributed to the rapid decline observed at T2. Fourth, the use of mixed-mode data collection (in-person and telephone) might have introduced measurement variability. Finally, the study experienced a relatively high attrition rate over the longitudinal follow-up period, which poses a potential risk of selection bias. To rigorously address this issue, a sensitivity analysis was performed using full information maximum likelihood (FIML) estimation on the entire enrolled cohort. The trajectory patterns from this inclusive analysis were consistent with the primary results, confirming that the findings are robust and not substantially compromised by loss to follow-up.

Conclusions

In conclusion, this longitudinal study elucidates the significant heterogeneity in PA patterns among patients with lung cancer during the early phase of chemotherapy. We identified that while the majority of patients follow a “Moderate-Sustained Decline” trajectory, a resilient subgroup demonstrates a capacity for distinct behavioral recovery following the initial treatment shock. Physiological symptom burden and modifiable cognitive factors, including exercise self-efficacy and outcome expectations, were identified as significant factors associated with these divergent pathways. These findings highlight the relevance of social cognitive theory in understanding behavioral changes in this clinical context. Clinically, these findings underscore the necessity of shifting from reactive care to proactive, multidimensional risk stratification. The period prior to the second chemotherapy cycle represents a critical window for intervention. Nursing practice should integrate early screening of these psychosocial constructs to identify high-risk patients and implement tailored, theory-based interventions that concurrently address symptom management and cognitive empowerment. Early attention to these modifiable factors may help clinicians support patients in maintaining physical activity throughout treatment.

CRediT authorship contribution statement

Z.-Y.C.: Conceptualization, Data curation, Formal analysis, Methodology, Writing – original draft, Writing – review & editing, Visualization, Funding Acquisition. M.D.: Conceptualization, Data curation, Formal analysis. Q.X.: Conceptualization, Writing – Review & Editing. Y.-H.Z.: Data curation, Formal analysis. Z.-Z.W.: Formal analysis, Methodology, Writing – Review & Editing. J.-F.H.: Writing – Review & Editing, Supervision, Funding Acquisition. All authors have read and approved the final manuscript.

Ethics statement

The study was approved by the Ethics Committee of Anhui Medical University (Approval No. 82240175) and was conducted in accordance with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. All participants provided written informed consent.

Data availability statement

The data that support the findings of this study are available from the corresponding author, JH, upon reasonable request.

Declaration of generative AI and AI-assisted technologies in the writing process

No AI tools/services were used during the preparation of this work.

Funding

This study was supported by the National Natural Science Foundation of China (82272926) and the 2025 Anhui Medical University College of Nursing Graduate Student Youth Cultivation Project (Grant No. Hlqm12025019). The funders had no role in considering the study design or in the collection, analysis, interpretation of data, writing of the report, or decision to submit the article for publication.

Declaration of competing interest

The authors declare no conflict of interest.

Appendix. Supplementary material

The following is the Supplementary data to this article: