Xiaoqiang Ye, Chengying Xu, Liyang Yu, Bicui Zhan
This study investigated the association between dynamic changes in D-dimer (D-D) during first-line systemic therapy and overall survival (OS) in patients with advanced lung cancer (LC). We also evaluated the independent prognostic value of these dynamic changes after adjusting for confounders and other coagulation markers. This retrospective study included patients with advanced LC who received first-line systemic therapy between January 2018 and December 2023. The percentage change in D-D from baseline to 3 months was calculated. A 3-month landmark analysis was used to reduce immortal time bias. We fitted multivariable Cox proportional-hazards models adjusted for age, sex, disease stage, histology, liver metastasis, and additional confounders (thrombosis history, hypertension, diabetes, chronic obstructive pulmonary disease [COPD], and anemia) to assess the independent association between D-D dynamics and OS. Combined models incorporating D-D, fibrinogen, and platelet data were also evaluated. Restricted cubic splines were utilized to validate the threshold selection. Model performance was evaluated using time-dependent receiver operating characteristic curves, calibration plots, and decision curve analysis. A nomogram for individualized prediction was derived from the combined model and internally validated by bootstrap calibration. Among 192 enrolled patients, the landmark cohort comprised 160 cases (median follow-up 32.5 months). The D-D increase group had a significantly shorter median OS than the stable groups (8.2 vs. 51.8 months) and the decrease group (median not reached; log-rank P < 0.001). After full adjustment, a marked D-D increase remained independently associated with a higher risk of death compared with the stable group (HR = 2.60, 95% CI: 1.44-4.71, P = 0.002), whereas a marked decrease showed a protective trend (HR = 0.56, 95% CI: 0.29-1.08, P = 0.084). The combined model that included D-D dynamics, fibrinogen changes, and platelet status achieved a C-index of 0.697, outperforming the D-D-only model (0.667) and the clinical-only model (0.592). Dynamic D-D changes are independently associated with OS in advanced LC during first-line treatment. The combined model, presented as a practical nomogram, offers a tool for individualized risk stratification; however, external multicenter validation is warranted.