CIPHER-Lung: retrospective exploratory internal evaluation and clinical benchmarking of an interpretable peripheral blood immune score for lung cancer prognosis
Background Clinical Immune Phenotyping and Explainable Risk Modeling for Lung Cancer (CIPHER-Lung) is a rule-based peripheral blood immune score that organizes measured biomarkers into clinically readable concepts. We assessed whether these concepts outperform constituent markers, remain stable under assay-like computational perturbations, and identify discordance with standard disease burden. Methods This retrospective exploratory internal evaluation used 260 sample records from de-identified participants in an existing institutional cohort. Overall-survival analyses included 162 patients with lung cancer and 73 deaths, and only 31 had all six markers. The primary available-concept score combined the measured subset of a six-marker panel into four phenotypic proxies: loss of T-cell surveillance, checkpoint activation/terminal differentiation, cytokine inflammation, and hematology/myeloid inflammation. The fixed formula was applied without outcome-based coefficient refitting. Analyses included matched marker comparisons, clinical benchmarking, inverse-probability-of-censoring weighted (IPCW) evaluation, concept ablation, limited adaptation, computational jitter dropout, and 24-month discordance phenotyping. Results The available-concept score had a Harrell C-index of 0.697 and IPCW cumulative/dynamic AUROCs of 0.765 and 0.764 at 12 and 24 months. On matched participants, the composite exceeded CD3, CD8+PD-1+, CD8+CD57+, and the cytokine concept, whereas the NLR/hematology concept performed similarly in its smaller subset. The supplied fixed weighting did not outperform unweighted aggregation. Training-fold calibration and a four-concept linear probe produced only modest gains over the fixed score. Under five-point concept jitter plus 10% dropout, the median C-index was 0.691, rank correlation was 0.939, and high-risk agreement was 0.912. Descriptive point estimates were higher for clinical models. The combined model showed no material point-estimate gain over the clinical model. Paired intervals for clinical-model comparisons were unavailable. Conclusions CIPHER-Lung provides an interpretable immune-state representation supported by this retrospective exploratory internal evaluation. The score is not supported for stand-alone diagnosis, treatment selection, or replacement of stage and performance status. Candidate uses in longitudinal phenotyping, enrichment, or point-of-care reporting require complete-marker multicenter validation, documented reference standards, analytical qualification, external calibration, and prospective impact studies.
Authors
- Chung‐Yu Chen (ORCID: https://orcid.org/0000-0002-1725-0079)
- Shih-Chieh Chang (ORCID: https://orcid.org/0009-0006-4470-1783)
- Cheng-Yu Chang
- Yu-Feng Wei
- Ying-Yin Chen
Institutions
- National Yang Ming Chiao Tung University (TW)
- Far Eastern Memorial Hospital (TW)
- National Taiwan University Hospital (TW)
- National Cheng Kung University Hospital (TW)
- National Yunlin University of Science and Technology (TW)
- National Cheng Kung University (TW)
- I-Shou University (TW)
- Yuan Ze University (TW)
Publication Details
- Journal
- Frontiers in Immunology
- Published
- 2026-11-12
- DOI
- https://doi.org/10.3389/fimmu.2026.1981712
- Primary Topic
- Inflammatory Biomarkers in Disease Prognosis
- Type
- article
- Field-Weighted Citation Impact
- 0.00