**4. Discussion**

Regional lymph node metastasis is common in NSCLC patients without distant metastasis (M0 disease) and is associated with a worse survival prognosis [32]. Disease recurrence within 5 years after initial curative treatment occurs in over half of lung cancer patients with nodal metastasis, and these patients eventually die of recurrence [9,10,32]. The survival outcomes of M0 patients with regional nodal metastatic NSCLC vary widely, despite initial aggressive treatment [8,9,11]. Therefore, a more reliable prognostic stratification tool is an unmet need. The prognostic value of 18F-FDG PET parameters derived from the primary tumor or the regional lymph node has been reported in patients with NSCLC [18–20,33,34]. However, the predictive power of combining the 18F-FDG PET parameters from both primary tumor and metastatic nodes has not been well investigated. Although some studies have combined the 18F-FDG PET volumetric parameters from the primary tumor and metastatic lesions, these study cohorts mixed locoregional disease cases with distant metastatic cases [35–38]. Thus, the results of these studies cannot be applied to patients with locoregional disease due to the diverse prognoses of patients with M0 or M1 diseases. In this study, we demonstrated that the total TLG derived from 18F-FDG PET, a combination of TLGs from both the primary tumor and regional lymph nodes, is an independent risk factor for PFS and OS in patients of locoregional NSCLC. Incorporating the total TLG with traditional clinical risk factors improved survival stratification.

Although the AJCC staging system is currently the mainstay of decision making regarding treatment strategies in NSCLC, it does not simultaneously assess tumor biological activity and burden. In patients with regional nodal metastatic NSCLC without distant spreading, the 18F-FDG PET metabolic parameters derived from primary tumor or metastatic nodes have been shown to be associated with survival outcomes in previous reports [18,33]. Because the primary tumor and metastatic nodes usually show different glucose metabolic profiles in 18F-FDG PET images, a combination of the two may provide comprehensive biological information for predicting prognosis. In this study, we found that combining the TLG of the primary tumor and the metastatic nodes into a total TLG resulted in an independent risk factor with a higher prognostic significance. The TLG is calculated by multiplying the MTV by the mean SUV, which weights the volumetric burden and metabolic activity of tumors. Kim et al. and Park et al. showed that larger primary tumor MTV was associated with a higher likelihood of occult nodal metastases [39,40]. Accordingly, larger metabolic tumor burden in patients with nodal metastatic NSCLC may also be expected to bear a higher risk of occult metastases in the more remote lymph node stations or even in distant organs. These occult lesions may escape from the most intensive treatment in the locoregional area. In addition to describing the tumor burden per se, the TLG also depicts the viability and the glycolytic activity of the tumor. The glycolytic pathway elicits diverse non-glycolytic mechanisms related to the promotion of cancer survival, proliferation, invasiveness, and adaptation to therapeutic agents, which are associated with unfavorable prognoses in patients with cancer [41,42]. Therefore, being a surrogate marker for both disease burden and vicious tumor behavior, total TLG may facilitate the stratification of patients into different risk groups.

Older age (>75.5 years) was an independent prognosticator for poor OS and PFS in our study. ECOG was also a prognostic factor for OS. Age and performance status are associated with survival outcomes in lung cancer as well as other malignancies, such as aerodigestive tract and gynecologic cancers [10,43–46]. The aged population has more medical comorbidities. In addition, aged patients may experience more toxicities from anti-neoplastic agents and suffer more perioperative complications, which may increase treatment-related mortality [47]. Moreover, the function of the T-cell-mediated immune system declines with age and limits the cellular immune response against tumor cells in the elderly population, further explaining the unfavorable survival outcomes in patients with advanced age [48–51]. We also analyzed the effects of different histopathological types on survival. Similar to other reports, squamous cell pathology was associated with worse survival outcomes compared with outcomes in patients with other histopathological types in the univariate analysis, whereas no statistical significance was found in the multivariate Cox regression analysis [20]. The histopathological type of NSCLC may vary according to age [9,46], in line with the age distribution in our cohort (Supplementary Table S3). Thus, the survival differences according to histopathologic types in our study appear to depend on patient age. Nevertheless, whether histopathological types impact the outcome of nodal metastatic NSCLC requires a more uniform patient cohort for verification.

In our study, the total TLG is an independent risk factor depicting the disease, while the age and ECOG status characterize the host vitality. Survival outcomes in patients with cancer result from the complex interplay between the tumor and the host. Robust patient conditions with limited total TLG would have a higher chance of attaining disease-free status after curative treatment. On the other hand, vulnerable patient status and sizable total TLG are likely to experience treatment failure and eventually succumb to recurrence or disease progression (Figure 4). Therefore, incorporating both disease and host factors into one survival prediction model refines the prognostic stratification. Our survival stratification models also showed predictive value for survival outcomes in subgroups receiving different initial treatments. Because therapeutic decisions may vary from patient-to-patient based on clinical factors such as age, the baseline survival risk in subgroups receiving different initial treatment may vary as well [9,46]. For example, patients receiving curative surgery tend to be younger; thus, the surgical group has a lower baseline survival risk according to age. Nevertheless, the results of our study showed that our survival prediction model could be applied to different subgroups receiving different initial treatments, suggesting a wide utility of our survival stratification models in different treatment scenarios.

Despite the current therapeutic advances, treatment responses and survival times in nodal metastatic NSCLC patients are quite heterogeneous, and a reliable prognostic model for this patient group is still lacking [9–11]. For selected patients, salvage surgery for persistent or recurrent disease has been shown to improve disease control and may improve OS [52,53]. Furthermore, new therapeutic strategies have emerged that have improved disease control and prolonged survival in nodal metastatic NSCLC. For instance, adding adjuvant tyrosine kinase inhibitor in this patient group postpones recurrence in patients with an actionable epidermal growth factor receptor (EGFR) mutation [54]. Recently, preliminary data have suggested that neoadjuvant immunotherapy or chemoimmunotherapy may improve resectability and increase the pathological complete response rate [55–58]. However, sophisticated patient stratification before implementing these novel treatments is essential. Adding novel neoadjuvant or adjuvant therapy in patients with excellent survival outcomes after standard curative treatment may show little benefit and may result in undesirable adverse effects. Therefore, the survival prediction model in our study may aid in stratifying patients into different risk groups for tailored treatment decisions.

There were several limitations in our study. First, the patient cohort was not large and the study was conducted in a retrospective manner. In addition, heterogeneous patient characteristics such as the histopathological type of NSCLC and therapeutic strategies employed may introduce biases when analyzing survival data. Second, we did not include the EGFR mutation status in our survival analysis. Nonetheless, the meta-analysis by Zhang et al. showed that the EGFR mutation status was not predictive of the OS or disease-free survival in NSCLC with locoregional disease [59]. Furthermore, only 39 (43.8%) patients in our retrospective cohort were tested for EGFR mutation. Thus, we could not draw a clear conclusion on this issue. Finally, this study was performed in a single center and we only internally validated our results. Although external validation would be ideal before clinical implementation, external validation of a prognostic model requires a minimum of 100 events and ideally 200 events to produce reliable results [60]. Therefore, the generalizability of our survival prediction model warrants external validation in a larger prospective cohort.

**Figure 4.** Survival stratification according to the independent risk factors in our study. The 18F-FDG PET/CT images for a 50-year-old woman with adenocarcinoma in the left upper lobe and subaortic nodal metastasis, indicated by arrows in the panels (**<sup>a</sup>**–**<sup>c</sup>**). The clinical staging was cT2aN2M0, stage IIIA. The total TLG was 14.9 and the ECOG status was 0. The patient had no poor survival risk factor (hazards were both 1 for an unfavorable OS and poor PFS) and she underwent lobectomy of the left upper lobe and mediastinal lymph node dissection. The pathological staging was pT2aN2M0, stage IIIA. She underwent adjuvant chemotherapy and is now alive without recurrence (**d**). The OS and PFS were 116 and 115 months, respectively. A 75-year-old man with adenocarcinoma in the right lower lobe and ipsilateral hilar nodal metastasis, indicated by arrows in the panels (**<sup>e</sup>**–**g**). The clinical staging was cT2bN1M0, stage IIB. The total TLG was 246.6 (>81) and the ECOG status was 1 (the hazards for unfavorable OS and poor PFS were 16.8 and 3.3, respectively). The patient received definitive CCRT (2 Gy/fraction daily to a targeted dose of 66 Gy) and marked tumor shrinkage was observed (**h**). However, the patient experienced progression of the primary tumor 15.9 months after definitive CCRT, indicated by the arrow in the panel (**i**). The patient eventually died of lung cancer progression, with an OS of 24.0 months and PFS of 15.9 months. TLG, total lesion glycolysis; OS, overall survival; PFS, progression-free survival; CCRT, concurrent chemoradiotherapy.
