Incremental prognostic value of immune cell densities beyond clinical parameters in non-small cell lung cancer

Multiplex immunofluorescence imaging enables detailed characterization of the tumor immune microenvironment, but whether immune cell densities add prognostic value beyond established clinical factors in non-small cell lung cancer (NSCLC) remains unclear.

Tissue samples from an NSCLC cohort (n = 298) were stained with a multiplex immunofluorescence panel targeting immune cell markers (CD4, CD8, FoxP3, CD20), cancer cells (pan-cytokeratin), and cell nuclei (DAPI). We quantified immune cell densities, nuclear pleomorphism features, and clinical variables, and trained four machine learning models (logistic regression, random forest, support vector machine, and k-nearest neighbors) to predict overall survival.

Clinical parameters consistently demonstrated the strongest performance in predicting long and short-term survival (logistic regression mean accuracy 0.60 ± 0.01, AUC 0.66 ± 0.01). The addition of immune cell densities revealed a small, statistically significant improvement in survival prediction (accuracy 0.62 ± 0.01, p < 0.01, AUC 0.67 ± 0.01, p = 0.04), while nuclear pleomorphism features did not improve prediction. When combined with clinical parameters, immune cell densities also improved survival stratification in Cox regression analyses numerically (HR = 0.51 vs. 0.55 for clinical parameters alone). Model interpretation analyses showed that stage and performance status have the largest effect on model performance. Selected immune cell densities (tumor CD4-helper and stroma B-cells) have a limited but consistent effect.

Clinical parameters remain the dominant predictors of outcome in NSCLC, with immune cell densities providing only limited prognostic value for clinical stratification. The openly available code and datasets present a unique resource for method development or focused analysis.