First-morning urinary luteinizing hormone and bone age decomposition for diagnosis of central precocious puberty in girls: a two-step prediction model study

Objective The diagnosis of central precocious puberty (CPP) in girls relies on the invasive gonadotropin-releasing hormone (GnRH) stimulation test. Urinary luteinizing hormone (LH) is a noninvasive candidate biomarker, but evidence from Chinese populations and its integration into multivariable prediction frameworks remain limited. We evaluated the diagnostic accuracy of first-morning urinary LH for CPP, established sex- and pubertal stage–specific reference intervals, and developed a two-step prediction model based on a bone age decomposition. Methods In this single-center retrospective diagnostic accuracy study, we enrolled 808 children: 295 girls with early breast development (146 CPP, 149 non-CPP), 271 healthy girls, and 242 healthy boys (reference intervals only). First-morning urine was assayed for LH, follicle-stimulating hormone (FSH), and creatinine. Diagnostic performance was evaluated primarily in the 295 symptomatic girls, with an extended analysis in 566 girls including healthy controls; multivariable logistic regression decomposed bone age into chronological age (CA) and the BA–CA difference as separate predictors. Results Urinary LH showed moderate accuracy for CPP in the primary symptomatic cohort (AUC 0.692, 95% CI 0.630–0.752), peaking in girls aged 7–8 years (AUC 0.708, 95% CI 0.621–0.785); the extended cohort gave similar results (AUC 0.712; secondary, spectrum-extended estimate). The LH-to-creatinine ratio offered no improvement. Basal serum LH was the superior single marker (AUC 0.868 vs 0.692, P < 0.001). The noninvasive Model 1 (CA + BA–CA + urinary LH + breast Tanner stage + uterine length) achieved a C-statistic of 0.853 (sensitivity 73.3%, specificity 83.2%), with CA, BA–CA, and urinary LH as independent predictors. Adding basal serum LH and FSH (Model 2) improved the C-statistic to 0.909; urinary LH retained independent significance (aOR 1.280, P = 0.024), whereas basal FSH was inversely associated with CPP. Bootstrap-corrected C-statistics were 0.838 (Model 1) and 0.893 (Model 2). Pubertal-onset cutoffs were >1.02 mIU/mL in girls and >0.95 mIU/mL in boys. Conclusions Decomposing bone age into CA and BA–CA enables tiered CPP risk stratification: the noninvasive model supports initial screening without blood sampling, and the complete model refines risk once serum gonadotropins become available. Urinary LH is a clinically useful component of tiered diagnostic algorithms for CPP.