Longitudinal tumor size and NLR as predictive factors of individual survival compared to their baseline values in patients with non-small cell lung cancer treated with durvalumab.

Abstract

e20047 Background: Our ability to accurately predict survival of patients with non-small cell lung cancer (NSCLC) while on treatment is limited. Prognostic markers such as stage and tumor size are well established, while neutrophil-to-lymphocytes ratio (NLR) and other hemogram measurements have recently been studied. Gain in prognostic accuracy of these markers when measured longitudinally has not been established. Methods: There were 679 NSCLC patients (Stage 3 or 4, ECOG PS 0 or 1) from clinical studies of durvalumab 10 mg/kg every two weeks (NCT02087423 and NCT01693562). We developed three models of overall survival (OS) all with ECOG as covariate: a Cox proportional hazards model with baseline tumor sum-of-longest-diameters (SLD) and NLR as covariates (COX); a joint model of OS and longitudinal SLD and baseline NLR (JM SLD); and a joint model of OS and longitudinal SLD and NLR (JM SLD&NLR). We compared prognostic accuracy of these markers measured longitudinally vs. at baseline, using predicted probability of OS at 12 months after start of durvalumab as a prognostic score. We evaluated predictive performance of the models using area under the receiver-operating characteristic curve (ROC AUC) describing trade-off between true and false positives (i.e., survival past 12 months). The AUCs were calculated for patients in the dataset using longitudinal data up to different cut-offs. Results: The AUC for all patients starting durvalumab using baseline ECOG, SLD and NLR was 0.73, while it decreased to 0.64 for patients surviving to 6 months, compared to 0.50 for noninformative models. The AUC using longitudinal information for SLD and NLR was larger the more longitudinal data was used for prediction and was 0.81 using 6 months’ worth of data. Conclusions: Using longitudinal information for SLD and NLR increased individual predictive performance of these markers compared to only baseline information in NSCLC patients. [Table: see text]