This is the first study that explored the prognostic potential of nivolumab clearance on early time points. Our analyses revealed that nivolumab clearance at first dose, as well as the subsequent changes in nivolumab clearance over 6 and 12 weeks, hold significant prognostic potential for survival among NSCLC patients undergoing nivolumab treatment. Moreover, our findings suggest that changes in nivolumab clearance over 6 and 12 weeks do not confer superior prognostic potential when compared to nivolumab clearance at first dose.

Previous research has demonstrated that high nivolumab clearance is linked with shorter survival among lung cancer patients in contrast to those with low nivolumab clearance [3, 14, 15]. Our study supports these findings and provides valuable information regarding the sensitivity and specificity required for predicting non-response to treatment, which is critical for evaluating the clinical potential of prognostic biomarkers. The use of immune checkpoint inhibitor clearance as an early survival biomarker offers practical advantages, as quantitative assays for these monoclonal antibodies are widely available [16,17,18,19]. These biomarkers are determined through blood sampling, which can be easily combined with routinely performed blood sampling in daily clinical practice to minimise invasiveness.

Our study had some limitations. We aimed for high specificity (≥ 0.95) to enable physicians to stop immune checkpoint inhibitor treatment for non-responding patients. However, high specificity was not reached, which may be due to two decisions in our used method. Our definition of non-response (see Methods) was based on the OS of patients receiving BSC as second-line treatment who may have been unfit for systemic treatment [10]. Although the majority of our study population also consisted of patients who received nivolumab as second-line treatment, they were in a generally good condition, as reflected by their ECOG performance score of 0–1. Thus, our study participants had a median survival longer than 6 months, which resulted in a high false-positive rate (Supplementary Material 3) and low specificity. Additionally, our cut-points of nivolumab clearance were calculated by maximally selected rank statistics based on OS and not specifically on “non-response”, contributing to low specificity. However, high sensitivity may also be valuable for treatment decisions, as it indicates responders with high certainty, which enables the treating physician to reduce intensity of follow-up. The generalisability of our findings to the first-line setting with pembrolizumab remains to be investigated, but we expect similar potential to distinguish treatment non-response because the pharmacokinetic characteristics overlap. Furthermore, the retrospective study design may have resulted in missing data and smaller sample size, potentially contributing to selection bias and lower accuracy. Our dataset lacked complete information regarding PD-L1 expression and tumour histology for some participants, which may have contributed to variations in treatment response. However, we did not identify any genetic alterations that influence the response to nivolumab treatment. Therefore, we do not anticipate that the absence of tumour histology data compromises the validity of our results. The retrospective design of our study could also have contributed to the selection of subjects in a relatively good condition as their follow-up was more complete.

One could also argue that the underlying mechanism for differences in nivolumab clearance i.e., cachexia, could directly explain differences in non-response or survival. Cachexia, characterised by lower body weight, worse performance scores and hypoalbuminaemia due to increased catabolism [20, 21], among other symptoms, could theoretically account for differences in nivolumab clearance and possibly explain variations in non-response or survival. Although in our population albumin levels were not significantly different between patients with high and low nivolumab clearance at first dose, we cannot rule out that cachexia is an independent marker for survival. Moreover, the selection of subjects with a relatively good performance score in our study could result in relatively low nivolumab clearance, as cachexia is related to high nivolumab clearance via increased catabolism [22]. However, median nivolumab clearance estimated in our dataset of 9.4 mL/h is comparable to previously reported values (9.5 mL/h) [23]. The retrospective study design also forced to estimate nivolumab clearance largely (975) on trough concentrations. Even though trough concentrations provide better information regarding clearance than peak concentrations, richer sampling would help to estimate the individual nivolumab clearance.

Despite these limitations, multiple advantages of this study can be noted. First, OS as response measure is an objective endpoint, which facilitates straightforward interpretation and easy reproducibility for clinical validation, in contrast to endpoints like best overall response, which is based on imaging using RECIST version 1.1 [8]. Also, survival data were known for the majority of our subjects and a long-term date last known alive (3 months to 6 years after start nivolumab treatment) was known for the remainder, which resulted in thorough survival analyses.

Because a considerable number of NSCLC patients treated with immune checkpoint inhibitors do not respond to therapy, there is an urgent need for tools that can help us detect non-response early during treatment. Our results show that nivolumab clearance early in treatment is indicative of primary immunotherapy resistance in NSCLC patients. These results are promising for the prognostic potential of pembrolizumab clearance, which is currently used as first-line treatment of NSCLC.