Abstract 4468
Background
APD1, while capable of restoring immunity, does not benefit all patients. While molecular tests like PD-L1 expression and TMB help in enriching response in respective subsets, a test identifying patients showing primary resistance to APD1 which does not require tissue samples could help in optimizing treatment regimens.
Methods
Outcome data (PFS/OS) were correlated with protein profiles from mass spectrometry of the circulating proteome of pretreatment serum from 116 2nd line NSCLC patients treated with nivolumab (development set S) using multivariate machine learning methods related to deep learning. The resulting test stratified patients into three groups: group A having very poor outcomes, group B having intermediate outcomes, and group C having very good outcomes. Development results were obtained using out-of-bag estimators. Two additional patient cohorts treated with nivolumab, V1(N = 58) and V2(N = 75), were used for validation.
Results
The proportions of patients in A, B, and C were 41:43:32 in S, 23:18:17 in V1, and 32:19:24 in V2. Median PFS/OS in the poor prognosis group A was 43/132 days in S, 105/189 days in V1, 90/278 days in V2, and in the good prognosis group C 276/528 days in S, 192/459 days in V1, and 155/not reached days in V2. In a comparison with historical controls treated with single agent chemotherapy and analyzed with the same technique, nivolumab appeared substantially superior in the good prognosis group C, while there was no evidence of superiority in the poor prognosis group A. In multivariate analysis including performance status, smoking history, and histology, the test remained an independent predictor of outcome. The patterns of protein expression related to poor prognosis in group A patients were associated with elevated complement, wound healing, and acute phase reactants.
Conclusions
We developed and validated a test stratifying patients into three groups with significantly different outcomes on nivolumab. The poor prognosis group showed little benefit from nivolumab, and other treatments may be needed, while in the good prognosis group outcomes were very good for a 2nd line population. These results emphasize the importance of the host immune response in the prediction of APD1 efficacy.
Clinical trial identification
Legal entity responsible for the study
Biodesix.
Funding
Has not received any funding.
Editorial Acknowledgement
Disclosure
J.G. Aerts: Advisory boards: BMS, MSD, Roche, AstraZeneca, Eli-Lilly, Boehringer Ingelheim, Amphera; Stock owner: Amphera. H. Roder: Officer, salary, stock options: Biodesix. C. Oliveira, J. Roder: Employment, stock options: Biodesix. All other authors have declared no conflicts of interest.