LAR/NMLR stratify mortality risk before and after checkpoint inhibitor pneumonitis in NSCLC: multi-state and time-dependent analyses.

backgroundCheckpoint inhibitor pneumonitis (CIP) is a clinically important immune-related adverse event in non-small cell lung cancer (NSCLC), but its time-varying occurrence complicates prognostic assessment.

objectiveThis study evaluated whether baseline lactate dehydrogenase-to-albumin ratio (LAR) and neutrophil-monocyte-to-lymphocyte ratio (NMLR) were associated with CIP onset and mortality risk before and after CIP.

methodsThis retrospective cohort included adults with advanced or recurrent NSCLC treated with immune checkpoint inhibitors (ICIs) at Zhongshan Hospital (Xiamen), Fudan University, between 2021 and 2024. CIP incidence was estimated using cumulative incidence functions with death without prior CIP as a competing event. Transition-specific associations were evaluated using an illness-death multi-state model: CIP-free at-risk state after ICI initiation (state 0), post-CIP state (state 1), and death as the absorbing state (state 2). Overall survival (OS) was assessed using a time-dependent Cox model with CIP modeled as a time-varying exposure. Restricted cubic splines (RCS), calibration analyses, penalized sensitivity analyses, and benchmark comparisons with the lung immune prognostic index (LIPI) were performed.

resultsAmong 202 patients, 24 developed CIP and 119 died during follow-up. In parsimonious transition-specific Cox models, cause-specific hazard ratios (csHRs) showed that zLAR and zNMLR were not associated with CIP onset, but were associated with death without prior CIP (0→2: zLAR csHR 1.22, 95% CI 1.00-1.50; zNMLR csHR 1.28, 95% CI 1.09-1.51) and death after CIP (1→2: zLAR csHR 2.47, 95% CI 0.93-6.57; zNMLR csHR 3.14, 95% CI 1.32-7.45). In the time-dependent Cox model, CIP was not significantly associated with mortality, whereas zLAR (HR 1.25, 95% CI 1.01-1.55) and zNMLR (HR 1.25, 95% CI 1.01-1.54) remained prognostic. Adding zLAR and zNMLR to the clinical base model increased the C-index from 0.623 to 0.682. The dual-outcome risk map showed partial separation between predicted CIP and death risks.

conclusionBaseline LAR and NMLR were associated with mortality risk rather than the occurrence of CIP in ICI-treated NSCLC patients. These simple blood-based indices may help identify patients who require closer survival-oriented monitoring, while CIP risk should be assessed together with clinical pulmonary factors.