Associations of derived inflammatory, lipid, and anthropometric indices with CT-detected pulmonary nodules: a hospital-based cross-sectional study with explainable machine learning.

Background: Pulmonary nodules are frequently detected during routine health-examination chest computed tomography (CT). From a systems-endocrinology perspective, routinely available indices integrating adiposity distribution, lipid metabolism, immune-cell balance, and low-grade inflammation may characterize cross-organ metabolic-inflammatory phenotypes, but their associations with CT-detected pulmonary nodules remain uncertain. Methods: This hospital-based cross-sectional study included 1,072 adults undergoing routine health examinations at the Second Affiliated Hospital of Shandong First Medical University from 2024 to 2025. The primary outcome was any CT-detected pulmonary nodule; the secondary outcome was an institutional Lung-RADS category ≥3. Right-skewed indices were natural-log transformed before standardization. We applied multivariable logistic regression, Benjamini-Hochberg false-discovery-rate (FDR) correction, a reduced non-redundant representative-index model, restricted cubic splines, repeated nested cross-validation, and machine-learning models with SHAP explanations computed only for held-out observations. Results: CT-detected pulmonary nodules were present in 381 participants (35.54%), and 109 participants (10.17%) had Lung-RADS category ≥3. In the mutually adjusted representative-index model, AIP (OR, 1.38; 95% CI, 1.16-1.64; q = 0.003) and SIRI (OR, 1.37; 95% CI, 1.14-1.64; q = 0.004) were associated with any pulmonary nodule. For the exploratory secondary Lung-RADS category ≥3 outcome (109 events; 6.41 events per model parameter), SIRI was associated with Lung-RADS category ≥3 (OR, 1.48; 95% CI, 1.15-1.91; q = 0.008), but this low-EPV estimate should be interpreted cautiously. In exploratory single-index analyses, several lipid-adiposity and monocyte-related indices remained significant after FDR correction. The basic risk model achieved a cross-validated AUC of 0.76; the derived-index model achieved an AUC of 0.77, without a statistically significant AUC increment (difference, 0.01; P = 0.100). Conclusions: AIP and SIRI were independently associated with CT-detected pulmonary nodules. An exploratory association between SIRI and Lung-RADS category ≥3 was also observed; however, the limited number of events and EPV of 6.41 require cautious interpretation and independent validation. These cross-sectional findings describe an exploratory metabolic-inflammatory phenotype rather than a causal mechanism or clinically ready diagnostic tool. Multicenter longitudinal validation with standardized radiological characterization is required.