Ethical approval has been obtained from the Ethical Committee of Guangdong Provincial People’s Hospital (2012124H(R2)). The clinical trial had been registered in the Chinese Clinical Trials Registry (ChiCTR-TRC-13003400). This study was conducted in accordance with the declaration of Helsinki ethical principles for human experimentation and all patients gave their informed consent to participate in the study.

A quantitative study was designed and Fig. 3 presents the flowchart of the study. Patients who underwent lobectomy were recruited according to criteria for non-small cell lung cancer staged T1, T2, and T3a without chronic obstructive pulmonary disease. Exclusion factors included: Stage T3b and T4 non-small cell lung cancer, patients with serious chronic diseases (i.e., coronary artery disease, heart failure, atrial fibrillation, hypertension, osteoporosis, infections), refused to participate in this study. Each patient had a routine full clinical assessment prior to inclusion, and all patients included in this study provided written informed consent before data collection.

Data including age, sex, and body mass index (BMI) were recorded, and assessments including forced vital capacity (FVC), FEV₁, 6MWD, anaerobic threshold (AT), maximal oxygen consumption (VO_2max), and the 36-item Short Form Health Survey V1 (SF-36) were performed one month after lung resection. Data were collected from January 2014 to December 2017. The same therapists conducted all evaluations throughout duration of the study. Data collection and analysis were not carried out by the same persons who completed the assessment.

According to the results of preliminary experiment, with power of 0.9, alpha of 0.05, ρ0=0, and ρ=0.5. Sample size was calculated using a formula for correlation study with PASS20.0 software (NCSS, LLC), and the total sample size is 37. To compensate non-compliance among subjects, the sample size was increased (+10%) to 41 patients.

Pulmonary volumes including FVC and FEV₁ were measured using a spirometer (Electgraph HI-101; CHEST), and corrected for temperature and barometric pressure, according to the American Thoracic Society recommendations [16]. Each patient performed at least three trials and the best results were used for analysis.

Physical activity was determined by 6MWD [17]. 6MWD was carried out under the same conditions. Patients were instructed to walk at their fastest pace to cover the longest possible distance over 6 minutes. The longest walk was taken to represent the value.

Before the cardiopulmonary exercise test, a physician examined the patients beforehand to ensure their suitability for the test. We assessed CRF by VO_2max and AT in incremental (30 W/3 min) cycle ergometer exercise (METAMAX 3B; CORTEX) until patient fatigue or dyspnea by pointing to a score on Borg scale category ratio 10. VO_2max was determined as the highest 1-minute average value and was normalized for body mass (mL/kg/min).

Health-related QoL was evaluated with SF-36 [18]. The survey is a valid and reliable 36-item questionnaire that is widely used to measure QoL. It yields two summary scores of physical (physical functioning, role-physical, bodily pain, and general health) and mental (vitality, social functioning, mental health, and role-emotional) health. Scales range from 0 to 100, with 0 indicating the worst situation and 100 indicating the best situation in each domain.

Data were expressed through descriptive and inferential statistical analysis. Firstly, we examined whether the variables were normally distributed with the Anderson-darling test, then we used Spearman’s correlation coefficients for non-normally distributed variables. For normally distributed variables, we used Pearson’s correlation coefficients. The dependent variables were SF-36 total score, and the independent variables were FEV₁, FVC, 6MWD, VO_2max, AT, and personal factors including age, sex, and BMI. The Statistical Package for Social Sciences (IBM SPSS 20.0; IBM Corp.) was used. Secondly, a structural model was estimated and structural equation modeling was used to examine the path coefficient between the measures, and the proposed hypotheses were confirmed. To test the proposed model, we adopt partial least squares structural equation modeling (PLS-SEM) using SmartPLS software (version 4) due to the small sample size. And the sample size of the PLS-SEM was determined in accordance with the principle that the sample size should be a minimum of 10 times the greater of: (1) the largest number of formative indicators employed to measure a single construct, or (2) the largest number of structural paths directed towards a specific construct within the structural model [19]. In our study, the largest number of structural paths directed towards a specific construct within the structural model were three, and the total sample size was at least 30. Hence, the designated sample size in this study was 41 which was sufficient to perform the PLS-SEM. When performing the confirmatory factor analysis, we ensured the original model of four factors and 8 items with factor loadings above 0.50, the composite reliability (CR) and convergent validity (average variance extracted, AVE) ranged from 0.6 to 1.00 and higher than 0.5, respectively [20]. In our study, we interrogated the robustness of the model using blindfolding to obtain cross-validated redundancy measures for each construct and using the bootstrapping procedure with 5,000 re-samples to determine the significance level of weights, factor loadings, and path coefficients [20]. A p-value of <0.05 was considered statistically significant.

Forty-one patients were enrolled in this study and the complete data were included in the analysis. Among these participants, the TNM classification of lung cancers was as follows: 26 stages 1, 11 stages 2, and 4 stages 3a. The types of lung cancer cells were as follows: 25 adenocarcinomas, 13 squamous carcinomas, and 3 others. The median age was 62 years (61.73±10.89 years), and 51.20% (21 out of 41) were male. The median BMI was 22. Table 1 summarized the characteristics, QoL, FVC, FEV₁, 6MWD, AT, and VO_2max of the patients.

In the study, the values of sex, BMI, 6MWD, VO_2max, and SF-36 had normal distribution. In relation to the SF-36 total score, Table 2 showed a significant linear correlation was detected between 6MWD (r=0.317, p=0.044), AT (r=0.442, p=0.004), and VO_2max (r=0.344, p=0.028). SF-36 total score was poorly correlated with FVC (r=0.144, p=0.370), FEV₁ (r=0.251, p=0.114), BMI (r=0.187, p=0.242), age (r=0.216, p=0.176), and sex (r=-0.054, p=0.739).

A reflective modeling approach was employed to test the hypotheses. CR and the AVE were used to assess the convergence of observed variables. As shown in Fig. 4, all items exceeded the 0.62 loading threshold, with CR ranging from 0.6 to 1.0 and higher than 0.5 with AVE. To establish discriminant validity, we selected the heterotrait-monotrait ratio, with all values in this study below 0.597 (Table 3), meeting the recommended cut-off of 0.85 [20]. In addition, collinearity was examined to ensure unbiased regression results, and the variance inflation factors (VIFs) for inner model paths were evaluated. The results revealed that VIF values ranged from 1.0 to 1.15, well below the recommended threshold of 5. With respect to model fit, the standardized root mean residual and normed fit index had values of 0.057 and 0.933, respectively, which were both within the recommended range. These data indicated that the constructs met the required standards for the present study, allowing us to adequately test the research hypotheses (Table 3).

As shown in Table 4, the path linking personal factors and physical activities to perceived SF-36 was positive and statistically significant, and the path linking FVC, and FEV₁ to SF-36 was not positive and statistically insignificant. The R² value for SF-36 is 0.360 which means that 36.0% of the changes in QoL were due to pulmonary function, physical activity, and personal factors in the model (Fig. 4). In general, the effect of a latent predictor variable is small at the structural level if f² is 0.02, medium if f² is 0.15, and large if f² is 0.35 [20]. We found that physical activity (f²=0.292) and personal factors (f²=0.197) had medium to large effect sizes on perceived QoL, while pulmonary function had no effect sizes on perceived QoL in patients with lung resection (Table 5).