Serum small extracellular vesicle-associated snoRNA panel adds diagnostic value to CEA and CYFRA21-1 for non-small cell lung cancer: an exploratory single-center study.

BACKGROUND

Circulating small extracellular vesicle (sEV)-associated small nucleolar RNAs (snoRNAs) have emerged as candidate biomarkers because of their stability in biofluids. However, whether these molecular signals provide diagnostic information beyond established serum tumor markers remains uncertain. This study evaluated a four-snoRNA panel and investigated its incremental value when combined with carcinoembryonic antigen (CEA) and cytokeratin-19 fragment (CYFRA21-1) for NSCLC discrimination.

METHODS

This retrospective single-center case-control study included 240 Untreated patients with pathologically or cytologically confirmed NSCLC and 205 healthy controls. A discovery microarray was performed using RNA isolated from the sEV-enriched fraction of three NSCLC samples and three controls. Four candidate snoRNAs (AC007956.1, AC093523.1, AL358875.1, and SNORA51) were subsequently quantified by quantitative reverse-transcription PCR using ΔCt values. Logistic regression models were developed for the four-snoRNA panel, the clinical marker model (CEA plus CYFRA21-1), and the combined six-marker model. Model performance was assessed using repeated stratified 10-fold cross-validation (20 repetitions), training-fold-derived thresholds, 2,000 stratified bootstrap resamples, calibration metrics, paired DeLong comparisons, subgroup analyses, and age/sex-matched sensitivity analyses.

RESULTS

The discovery dataset contained 936 analyzable snoRNA probes. Although 134 probes showed nominal differential expression (P < 0.05), none remained significant after false-discovery-rate correction. NSCLC cases were older than controls (62.5 vs. 42.3 years; standardized mean difference, 1.73), whereas sex distributions were comparable. The four-snoRNA model achieved a cross-validated AUC of 0.924 (95% CI, 0.898-0.948), compared with 0.903 (95% CI, 0.874-0.931) for the CEA plus CYFRA21-1 model. The combined six-marker model showed the highest discrimination, with an AUC of 0.970 (95% CI, 0.955-0.983), sensitivity of 87.1%, specificity of 96.6%, calibration intercept of -0.018, calibration slope of 0.923, and Brier score of 0.064. Compared with the clinical model, the six-marker model increased AUC by 0.067 (95% CI, 0.043-0.092; paired DeLong P < 0.001). Model performance remained consistent in stage 0-II disease (AUC, 0.963; 95% CI, 0.940-0.980) and age/sex-matched analyses (AUC, 0.964; 95% CI, 0.941-0.983). No snoRNA marker retained significant associations with clinical characteristics after multiple-testing correction.

CONCLUSIONS

A serum sEV-associated four-snoRNA panel showed incremental diagnostic value when combined with CEA and CYFRA21-1 in this exploratory NSCLC case-control study. The resulting six-marker model maintained discrimination during repeated internal cross-validation, including analyses restricted to early-stage disease and age/sex-matched cohorts.