The clinical data of patients with NSCLC admitted to the Affiliated Hospital of Xuzhou Medical University between December 2022 and October 2024 were retrospectively analyzed. Patients who met the following criteria were included in the study: (1) diagnosed with advanced lung cancer by pathology and imaging; (2) received pembrolizumab in combination with chemotherapy and have data on pre-treatment biochemical and hematological indices; (3) received at least three treatment cycles; (4) had at least one measurable tumor; (5) aged 18 to 70 years when treated; (6) had an Eastern Oncology Cooperative Oncology Group (ECOG) score ≤1; (7) had complete clinical follow-up data.

Exclusion criteria: (1) received previous treatment with PD-L1 inhibitors or targeted drugs; (2) had severe infections or systemic inflammation; (3) had also hematological or immune system disorders, hepatic insufficiency, or hepatitis; (4) had a history of long-term steroids or immunosuppressant use; (5) had severe cardiac, hepatic, renal, or other organic pathologies; (6) had also other malignant tumors or non-primary NSCLC; (7) inaccurate or incomplete clinical data. Patients who received lipid-modifying drugs (e.g., statins and betas) during the study period were excluded from the analyses to reduce potential confounding effects on the TyG index. Based on the inclusion and exclusion criteria, the study included 243 patients ( Figure 1 ). The Ethics Review Committee of the Affiliated Hospital of Xuzhou Medical University approved the study (XYFY-), which adhered to the principles outlined in the Declaration of Helsinki. The requirement for written informed consent was waived due to the retrospective nature of the study.

The TyG index was calculated as follows: TyG index (mmol/L) = ln [fasting triglycerides (mmol/L) × fasting glucose (mmol/L)/2]. Fasting triglyceride and glucose levels were determined by an enzymatic method using an automated biochemical analyzer (AU2700, Olympus). The optimal TyG index cut-off value was calculated by the X-tile software using the survival time after the combined immunochemotherapy.

The patients were divided into high (≥1.61) and low (<1.61) TyG index groups based on the calculated cut-off value. Pembrolizumab (200 mg) was administered intravenously once every three weeks. The chemotherapy regimen comprised 260 mg/m² intravenous paclitaxel (albumin-based) once every three weeks and 75 mg/m² cisplatin. Patients with non-squamous histology received 500 mg/m² pemetrexed once every three weeks and 75 mg/m² cisplatin.

Patients underwent chest CT scanning every cycle and an enhanced CT or MRI examination every two cycles. The efficacy of treatment was assessed by applying the solid tumor efficacy evaluation criteria (RECIST version 1.1) and was classified into complete remission (CR), partial remission (PR), stable disease (SD), and disease progression (PD). Objective response rate (ORR) = (CR+PR)/total number of cases × 100%; Disease control rate (DCR) = (CR+PR+SD)/total number of cases × 100%. Progression-free survival (PFS) was defined as the time from the start of the 1st immunotherapy to tumor progression or death. Overall survival (OS) was defined as the time from the start of treatment to death from any cause (event). Survival time was calculated in months. Long-term treatment effects were assessed using OS and PFS. To ensure accuracy, all imaging data were evaluated by two independent radiologists with more than 10 years of experience. In cases of disagreement, a third radiologist with more than 10 years of experience assessed the information to help reach consensus and minimize inter-observer error. The final follow-up visit was in October 2024.

TYG index and tumor response were assessed by blinded independent reviewers at baseline and after three cycles of pembrolizumab plus platinum-based chemotherapy. PFS and OS were defined as primary endpoints. Follow-up was conducted through scheduled hospital visits or telephone interviews every 3 weeks, with a final data collection cutoff date of October 2024.

Data were analyzed using R software (version 4.4.1). The Kolmogorov-Smirnov test was used to assess normality prior to hypothesis testing for continuous variables. Continuous variables are presented as the mean ± SD, and categorical variables are described using frequencies (percentages). Normally distributed continuous data are expressed as means ± standard deviations (SD) (with Welch’s correction if variances were unequal based on Levene’s test), while non-normally distributed continuous data are expressed as medians with interquartile ranges (IQR). The baseline characteristics between groups were compared using the Fisher’s exact test, χ ² test, and t-test. The Kaplan-Meier method was used to calculate the median OS and PFS, which were used in the X-tile procedure to determine the optimal survival-related cut-off value for TyG. Cox proportional hazards regression was used for univariate and multivariate analyses. Multivariate analysis used stepwise backward selection based on the Akaike Information Criterion (AIC), with entry and removal thresholds of P<0.05 and P > 0.10, respectively. A nomogram was developed to predict the survival at 12, 24, and 36 months. The variables were selected according to their statistical significance and clinical relevance and weighted according to their hazard ratio (HR) derived from multivariate analysis. Internal validation of the model and the accuracy of the predictions were assessed by the bootstrap method with 1,000 resamples. Predictions generated by the model were compared with actual results to assess their calibration and reduce overfitting. All statistical tests were two-sided, and P<0.05 was considered statistically significant.

This retrospective study analyzed the data of 243 patients with advanced NSCLC who received pembrolizumab in combination with platinum-based chemotherapy between December 2022 and October 2024 at Xuzhou Medical University Hospital ( Figure 1 ). Among them, 111 were classified into the low TyG (<1.61) index group and 132 into the high TyG (≥1.61) index group. We excluded 334 patients because of incomplete documentation of clinical data (n=127), comorbid with other malignancies or non-primary NSCLC (n=31), loss to follow-up for more than six months (n=75), used alternative PD-1/PD-L1 inhibitors or targeted medications (n=62), and missing baseline TyG index data (n=39). Table 1 provides a comprehensive overview of patient characteristics, including clinicopathological features such as age, sex, BMI, smoking and drinking status, ECOG(Eastern Cooperative Oncology Group) score, history of COPD(Chronic Obstructive Pulmonary Disease), maximum pre-treatment tumor diameter, extrapulmonary invasion, and various laboratory parameters such as fasting blood glucose, fasting triglycerides, and albumin levels. The two groups differed significantly in BMI, TYG, PD-L1 expression level, and tumor diameter. In this study, 243 NSCLC patients were stratified by histological subtype. The majority had adenocarcinoma (n = 116, 47.7%) or squamous cell carcinoma (n = 107, 44.0%). Less common subtypes included large cell carcinoma (n = 9, 3.7%), adenosquamous carcinoma (n = 6, 2.5%), and sarcomatoid carcinoma (n = 5, 2.1%). This distribution aligns with established epidemiological patterns of NSCLC.

Table 2 describes the efficacy of the response to the oncological treatment. No patient achieved CR, 84 achieved PR, 81 achieved SD, and 78 showed PD. The ORR was 45.05% and 25.76% in the low and high TyG index groups, respectively (P=0.002), and the DCR was 85.59% and 53.03%, respectively (P<0.001).

The median PFS in the low TyG index group was significantly longer than in the high TyG index group [13.8 (9.2–14.4) vs. 8.2 (7.3–8.7) months; P<0.001; HR, 0.385; 95% CI, 0.278–0.534; Figure 2A ]. The median OS in the low TyG index group was significantly longer than in the high TyG group [26.1 (23.4–29.8) vs. 17.0 (13.9–18.3) months; P<0.001; HR, 0.403; 95% CI, 0.268–0.608; Figure 2B ].

Table 3 presents the results of univariate and multivariate Cox proportional hazards regression analyses for PFS. Multivariate regression analysis revealed several independent prognostic predictors for PFS in patients with NSCLC. Patients with a higher ECOG score had a shorter PFS (HR, 1.24; 95% CI, 1.12–1.55; P=0.021). Table 4 presents items related to OS. Good metabolic function was strongly correlated with long OS, with a lower TyG index indicating better metabolic function, specifically in the benign prognostic factors of PFS (HR, 4.05; 95% CI, 2.84–5.31; P<0.001) and OS (HR, 2.87; 95% CI, 1.43–4.52; P=0.003), suggesting that patients with better metabolic function responded more favorably to the treatment. Higher PD-L1 expression in tumor cells (TPS ≥ 50%) predicted longer PFS (HR, 0.51; 95% CI, 0.34–0.78; P=0.004) and OS (HR, 0.66; 95% CI, 0.44–0.89; P=0.011). Although BMI was an independent predictor for OS (HR, 0.78; 95% CI, 0.29–0.86, P=0.029), it did not predict PFS in the multivariate regression analysis (P=0.142), implying that higher BMI was associated with improved long-term survival but not with short-term treatment tolerance and PFS. Elevated CEA levels (≥3 ng/mL) predicted shorter PFS (HR, 1.21, 95% CI 0.94–1.81, P=0.039). Possibly due to overlapping effects with more influential factors such as ECOG score and PD-L1 expression, age, sex, extrapulmonary metastases, and pathological staging were insignificant predictors in the multivariate regression analysis. The nomogram model developed for 12, 24, and 36 months based on the independent prognostic predictors obtained from the multivariate regression analysis showed a consistency index of 0.772 for the training set and 0.786 for the validation set ( Figure 3 ), considered to have good predictive properties.

The patients were randomly assigned to two groups at a ratio of 7:3. We assigned 170 cases to the training set and 70 to the internal validation set. Table 5 lists the baseline characteristics of these two sets. The respective OS AUCs(Area Under Curve) for the training and internal validation sets were 0.811 and 0.817 for 12 months ( Figure 4A ), 0.737 and 0.720 for 24 months ( Figure 4B ), and 0.755 and 0.788 for 36 months ( Figure 4C ). The OS calibration curves for the training and internal validation sets at 12, 24, and 36 months are shown in Figure 5 . These curves indicate that the risk predicted by the model is in close agreement with the actual observed risk.