Transformer-based AI approach to unravel long-term, time-dependent prognostic complexity in patients with advanced NSCLC and PD-L1 ≥50%: insights from the pembrolizumab 5-year global registry

Abstract

Background: With nearly one-third of patients with advanced non-small cell lung cancer (NSCLC) and PD-L1 Tumor Proportion Score≥50% surviving beyond 5 years following first-line pembrolizumab, long-term outcomes challenge traditional paradigms of cancer prognostication. The emergence of non-cancer-related factors and time-dependent trends underscores the need for advanced analytical frameworks to unravel their complex interplay.

Methods: We analyzed the Pembro-real 5Y registry, a global real-world dataset of 1050 patients treated across 61 institutions in 14 countries with a long-term follow-up and a large panel of baseline variables. Two complementary approaches were employed: ridge regression, chosen for its ability to address multicollinearity while retaining interpretability, and not another imputation method (NAIM), a transformer-based artificial intelligence model designed to handle missing data without imputation. Endpoints included risk of death at 6, 12, 24, 60 months and 5-year survival.

Results: The ridge regression model achieved a c-statistic of 0.66 (95% CI: 0.59 to 0.72) for the risk of death and an area under the curve (AUC) of 0.72 (95% CI: 0.65 to 0.78) for 5-year survival, identifying Eastern Cooperative Oncology Group Performance Status (ECOG-PS)≥2, increasing age, and metastatic burden as primary risk factors. However, wide CIs for some predictors highlighted statistical instability. NAIM demonstrated robust handling of missing data, with a c-index of 62.98±2.11 for risk of death and an AUC of 60.52±3.71 for 5-year survival. The comprehensive SHapley Additive exPlanations analysis revealed dynamic, time-dependent patterns, with early mortality dominated by acute factors (eg, ECOG-PS, steroids) and long-term outcomes increasingly influenced by systemic health markers (eg, absence of hypertension, increasing body mass index). Unexpected insights included the protective role of dyslipidemia (but not statins) and the nuanced impact of smoking status, reflecting evolving disease dynamics and host-tumor interplay.

Conclusions: Our integrative framework illuminates the complexity of long-term outcomes in patients with NSCLC treated with pembrolizumab, uncovering dynamic, non-linear prognostication trends. This analysis provides insights into patient trajectories, emphasizing the need for holistic, long-term management strategies.

Keywords: Immune Checkpoint Inhibitor; Immunotherapy; Lung Cancer; Survivorship.

Figure 1. Visual summary of the methodological pipeline used to analyze long-term outcomes of patients with advanced NSCLC (PD-L1 ≥50%) treated with first-line pembrolizumab in the Pembro-Real 5Y Registry. Two parallel approaches were adopted: Conventional modeling using penalized ridge regression for both 5-year survival (logistic) and overall survival (Cox), producing interpretable ORs and HRs with associated CIs, and model performance metrics (eg, C-statistic, AUC). NAIM (AI model) employing a transformer-based deep learning architecture to model time-dependent survival dynamics using SHAP values for feature interpretability and multiple performance metrics (AUC, F1-score, Matthews correlation coefficient [MCC], etc). Outputs were compared side-by-side. Shared predictors (eg, ECOG-PS, corticosteroid use) reinforced robustness, while NAIM identified additional time-sensitive or non-linear predictors (eg, BMI, dyslipidemia). This integrative approach supports both baseline prognostication and longitudinal survivorship strategies. AUC, area under the curve; BMI, body mass index; CV, cardiovascular; ECOG-PS, Eastern Cooperative Oncology Group Performance Status; NAIM, Not Another Imputation Method; NSCLC, non-small cell lung cancer; SHAP, SHapley Additive exPlanations; TMB, Tumor Mutational Burden.

Figure 2. Histogram plot summarizing the cumulative SHAP values from the NAIM analysis for the risk of death. The length of each bar represents the SHAP value, indicating the relative importance of each variable within the model. Features were ordered by their absolute contribution. The c-index (% ±SD) was 79.76±2.44 for the training set and 62.98±2.11 for the overall model. Variable’s definition and categorization details are reported in online supplemental methods. ALK, anaplastic lymphoma kinase; BMI, body mass index; CNS, central nervous system; ECOG-PS, Eastern Cooperative Oncology Group Performance Status; EGFR, epidermal growth factor receptor; GLM, glucose- lowering medications; NAIM, Not Another Imputation Method; NOS, not otherwise specified; PPI, proton pump inhibitors; pred, Prednisone; SHAP, SHapley Additive exPlanations; TMB, tumor mutational burden; TPS, Tumor Proportion Score.

Figure 3. Paired dot plot and histogram and plot summarizing the SHAP values from the NAIM analysis for the 5-year survival. Features were ordered by their absolute contribution, with high values (red) and low values (blue) positioned to indicate their influence on outcomes. For instance, red dots on the right side of the plot imply a positive association with the probability of being alive at 5 years, while blue dots on the right side imply a negative association. Missing values were represented as gray dots. The length of each bar represents the SHAP value, indicating the relative importance of each variable within the model. The metrics for the model (% ±SD) were as follows: the training set achieved an AUC of 78.53±3.25, an accuracy of 74.41±2.56, an F1-score of 71.38±2.82, an MCC of 47.34±3.41, and a G-Mean of 64.89±3.17. In the evaluation set, the model demonstrated an AUC of 60.52±3.71, an accuracy of 53.65±2.97, an F1-score of 45.43±23.10, an MCC of 10.72±8.35, and a G-Mean of 36.71±7.61. Variables definition and categorization details are reported in online supplemental methods. ALK, anaplastic lymphoma kinase; AUC, area under the curve; BMI, body mass index; CNS, central nervous system; ECOG-PS, Eastern Cooperative Oncology Group Performance Status; EGFR, epidermal growth factor receptor; GLM, glucose-lowering medications; NAIM, not another imputation method; NOS, not otherwise specified; PPI, proton pump inhibitors; pred, Prednisone; Matthews correlation coefficient: MCC; SHAP, SHapley Additive exPlanations; TMB, tumor mutational burden; TPS, Tumor Proportion Score.