Identification of multivariable predictors for prediabetes: biochemical integration and pathophysiological mechanism analysis based on health examination data
Background Pre-diabetes is a critical, reversible intermediate metabolic state bridging normal glucose tolerance and overt type 2 diabetes mellitus (T2DM). Early clinical detection is essential for timely intervention, yet single-biomarker screenings often fail to capture the systemic metabolic and inflammatory alterations characteristic of this state. This study aimed to identify independent clinical and biochemical risk factors for pre-diabetes and to construct and validate a robust, low-cost multivariable predictive model leveraging routine health examination data, with a specific focus on dissecting the underlying molecular and biochemical pathways. Methods A retrospective cross-sectional analysis was performed on clinical registry data from 2,376 adult participants (1,242 pre-diabetes patients and 1,134 normoglycemic controls) who underwent health examinations in 2023. Comprehensive demographic, anthropometric, and biochemical profiles were analyzed. Multivariable logistic regression was utilized to isolate independent risk and protective factors. Receiver Operating Characteristic (ROC) curve analysis evaluated the discriminative capacity of individual parameters versus the integrated multivariable model. Results Univariate analyses demonstrated widespread systemic alterations across metabolic, inflammatory, hepatic, and lipid domains in pre-diabetic patients (P<0.05). Multivariable logistic regression identified gender(OR = 1.635, 95% CI: 1.177–2.274, P = 0.0034), age (OR = 1.105, 95% CI: 1.093–1.117, P<0.001), BMI (OR = 1.12, 95% CI: 1.079–1.163, P<0.001), MPV (OR = 1.103, 95% CI: 1.009–1.205, P = 0.0306), total protein (OR = 1.233, 95% CI: 1.076–1.416, P = 0.0029), γ-GT(OR = 1.007, 95% CI: 1.004–1.012, P<0.001), UA(OR = 1.002, 95% CI: 1.001–1.003, P = 0.0047), and apolipoprotein A1 (OR = 2.736, 95% CI: 1.200–6.273, P = 0.017) as independent risk factors. Serum creatinine, and high-density lipoprotein cholesterol (HDL-C) were identified as independent protective factors. While single parameters showed poor diagnostic capacity (AUCs <0.76), after bootstrap internal validation, the 8-parameter model achieved a bias-corrected AUC of 0.806, indicating good discriminative ability and robust generalizability. Conclusion Pre-diabetes is associated with a complex, interconnected network of subclinical inflammation, hepatic stress, dyslipidemia, and altered skeletal muscle dynamics. Integrating these routine biomarkers into a unified predictive model offers a highly accessible, biologically sound, and cost-effective strategy for early diabetes screening in primary care settings.