Complete blood count–based risk stratification to support lung cancer screening in resource-limited settings: a retrospective cohort analysis

Lung cancer remains the leading cause of cancer-related mortality worldwide, largely due to late-stage diagnosis. Although low-dose computed tomography (LDCT) reduces mortality, its large-scale implementation is constrained by cost, infrastructure demands, and unequal access, particularly in low- and middle-income settings. We developed and evaluated a deployable risk stratification model based on routine complete blood count (CBC) parameters to prioritize individuals for LDCT screening. In this retrospective study, 83,904 individuals aged ≥ 50 years who underwent chest CT or lung biopsy up to five years after a CBC collection were analyzed. The model was trained using Lung-RADS 4 CT findings and evaluated exclusively on biopsy-confirmed lung cancer cases in an independent test set, with low- and intermediate-risk CTs and negative biopsies serving as controls. Lung-RADS 4 denotes radiological suspicion of malignancy rather than confirmed cancer, and control status was radiological for all but 307 individuals with negative histology. Logistic regression with L2 regularization (ridge) was compared with gradient boosting, and feature selection retained red cell distribution width (RDW), neutrophil count, and mean corpuscular volume as predictors. The ridge model achieved an area under the receiver operating characteristic curve (AUC) of 0.62 (95% CI, 0.58–0.66), comparable to LightGBM. Risk stratification demonstrated modest enrichment: the high-risk group (22.46% of the population) captured 37.32% of lung cancer cases (relative risk [RR] = 2.34), and combining high- and moderate-risk categories identified 59.33% of cancers while prioritizing 42.84% of individuals for imaging. At the high-risk threshold, the number needed to screen fell from approximately 120 to 72 examinations per cancer detected, and decision-curve analysis showed net benefit over the better of the screen-all and screen-none strategies between threshold probabilities of 0.68% and 1.07%. The model requires only three CBC parameters, with no imaging data, questionnaire, or smoking history. Prospective evaluation against current eligibility criteria is required before CBC-based prioritization can be recommended.