*3.1. Patient Characteristics*

Eighty-nine patients were eligible for analysis, whose baseline characteristics are summarized in Table 1. In total, 15 (16.9%), 53 (59.6%), and 21 (23.6%) patients were clinically classified as N1, N2, and N3 status, respectively. Thirty-seven (41.6%) patients received curative surgery or neoadjuvant CCRT and surgery as the initial treatment. Fifty-two (58.4%) patients received definitive CCRT or definitive radiotherapy. The median (interquartile range, IQR) time from the 18F-FDG PET/CT to initiation of treatment was 14 (12) days. The median follow-up period was 25.4 months (range, 1.7–130.2 months) for all patients and 48.7 months (range, 10.0–130.2 months) for the 34 surviving patients. Sixty (67.4%) patients experienced recurrence or disease progression after or during the initial treatment; 32 of these patients had locoregional recurrence or progression only and 15 patients developed distant metastases without locoregional failure. The remaining 13 patients had both locoregional failure and distant metastasis. By the time of the last follow-up, 55 (61.8%) patients died. The 5-year OS and PFS rates were 33.4% and 24.7%, respectively.


**Table 1.** Baseline characteristics for patients in this study (*n* = 89).

NSCLC, non-small cell lung cancer; ECOG, Eastern Cooperative Oncology Group; CCRT, concurrent chemoradiotherapy; SD, standard deviation; IQR, interquartile range; SUV, standardized uptake value; TLG, total lesion glycolysis; NTSUVR, nodal to primary tumor SUVmax ratio; NTTLGR, nodal to primary tumor TLG ratio; TNSU-Vproduct, product of primary tumor and nodal SUVmax. a Staging according to 8th edition of American Joint Committee on Cancer manual.

#### *3.2. Univariate and Multivariate Survival Analyses*

Our ROC curve analysis identified six semiquantitative 18F-FDG PET parameters that were associated with patient death. These parameters were included in the survival analysis (Supplementary Table S1), including the primary tumor SUVmax, primary tumor TLG, nodal SUVmax, nodal TLG, total TLG, and TNSUVproduct. The median OS and median PFS were 31.5 months (range, 1.7–130.2 months) and 15.1 months (range, 0.2–126.8 months), respectively. The results of the univariate and multivariate Cox regression analyses are outlined in Table 2. The univariate analysis of OS showed that

age > 75.5 year-old, squamous cell pathology, T2–T4 disease, ECOG status > 0, never received surgery, only received radiotherapy, primary tumor SUVmax > 8.05, primary tumor TLG > 42.5, nodal SUVmax > 2.94, nodal TLG > 18.3, total TLG > 81, and TNSUVproduct > 27 were associated with shorter OS. The univariate Cox regression analysis for PFS showed that age >75.5 year-old, never received surgery, only received radiotherapy, primary tumor SUVmax > 8.05, primary tumor TLG > 42.5, total TLG > 81, and NSUVproduct > 27 were predictive of shorter time to progression. The statistically significant variables in the univariate analysis were fitted into multivariate Cox regression models. Age > 75.5 year-old, ECOG status > 0, and total TLG > 81 independently predicted unfavorable OS, whereas age > 75.5 year-old and total TLG > 81 were independent risk factors for shorter PFS.


**Table2.**Univariateandmultivariateanalysesforsurvivalprognosticfactors.

OS, overall survival; PFS, progression-free survival; HR, hazard ratio; CI, confidence interval; ECOG, Eastern Cooperative Oncology Group; SUV, standardized uptake value; TLG, total lesion glycolysis; TNSUVproduct, product of primary tumor and nodal SUVmax; NA, not applicable. a Received curative surgery or neoadjuvant chemoradiation followed by curative surgery. b Only received radiotherapy as the initial treatment.

#### *3.3. Survival Model Construction and Validation*

The independent risk factors were used to develop prediction models for OS (age > 75.5 years, ECOG > 0, and total TLG > 81) and PFS (age > 75.5 years and total TLG > 81). The construction of our survival models is demonstrated in the Supplementary Figure S1. The risk for each patient was calculated by multiplying the hazard ratio of each risk factor. In the OS model, the HRs for age > 75.5 years, ECOG > 0, and total TLG > 81 were 2.6, 3.3, and 5.1, respectively; in the PFS model, the HRs for age > 75.5 years and total TLG > 81 were 2.7 and 3.3, respectively (Table 2). If a patient had all three independent risk factors for OS, the total hazard of poor OS for this patient was 43.8 (2.6 × 3.3 × 5.1). If a patient had no risk factor, then the total hazard was the baseline hazard. If a patient was > 75.5 years and had a total TLG > 81, then the risks of having a shorter OS and poor PFS compared to no risk factor would be 13.3 (2.6 × 5.1) and 8.9 (2.7 × 3.3), respectively. The resulting hazards ranged from 1 to 43.8 for the OS model and 1 to 8.9 for the PFS model. Patients with similar 5 year survival outcomes in the Kaplan–Meier curve analysis were re-stratified into one risk group. Finally, we obtained three separate risk categories (HR < 3, HR = 3–10, and HR > 10 for the OS model; HR < 3, HR = 3–5, and HR > 5 for the PFS model).

The bootstrap method was used to validate our survival analysis results. Supplementary Table S2 presents the results of the bootstrap validation. The β estimate of each independent prognosticator was statistically significant in predicting OS and PFS.

#### *3.4. Model Performance and Comparison to AJCC Staging System*

The Cox regression models in our study significantly stratified patients into different survival risk groups (Figure 2). The c-indices of our Cox regression model for OS and PFS were 0.732 and 0.672, respectively. The Cox regression model developed in our study cohort significantly outperformed the AJCC staging system. In addition, our models (combining the independent clinical prognosticators with total TLG) showed the highest c-indices compared with other models using the combination of independent clinical risk factors with primary tumor TLG or nodal TLG (Table 3).



NA, not applicable; AJCC, American Joint Committee on Cancer; TLG, total lesion glycolysis. a Staging according to 8th edition of AJCC manual. b Model constructed from the independent clinical risk factors and primary tumor TLG. c Model constructed from the independent clinical risk factors and nodal TLG. d In comparison with AJCC staging system.

**Figure 2.** The Kaplan–Meier curves for OS and PFS in patients with nodal metastatic NSCLC without distant metastasis. Survival was stratified according to the eighth edition of the AJCC system (**<sup>a</sup>**,**b**) and our survival prediction model (**<sup>c</sup>**,**d**). OS, overall survival; PFS, progression-free survival; AJCC, American Joint Committee on Cancer.

#### *3.5. Model Performance in Subgroups of Different Initial Treatments*

We also applied our survival prediction models to subgroups according to different initial treatments (curative surgery or definitive CCRT). Our models significantly stratified patients into different survival risks independent of the initial treatment strategy (Figure 3). The c-indices of our models were compared to the AJCC staging system in the subgroups. Our models significantly outperformed the AJCC staging system, except the c-index for PFS in patients who received initial curative surgery, which only showed a statistical trend (Table 4).

**Figure 3.** The Kaplan–Meier curves depicting OS and PFS stratified by our prediction model in subgroups of different initial treatments. The use of our model in subgroups that underwent curative surgery or neoadjuvant CCRT followed by surgery (**<sup>a</sup>**,**b**) and in initial non-surgical subgroup (**<sup>c</sup>**,**d**). OS, overall survival; PFS, progression-free survival; CCRT, concurrent chemoradiotherapy.


**Table 4.** A comparison of the c-indices of our model with the traditional cancer staging system in subgroups of different initial treatment strategies.

NA, not applicable; AJCC, American Joint Committee on Cancer. a Including initial curative surgery or neoadjuvant chemoradiation followed by curative surgery. b Staging according to 8th edition of AJCC manual.
