Abstract: Background Gonadotropin-releasing hormone (GnRH) stimulation test is the gold standard for the diagnosis of central precocious puberty (CPP) at present, but multiple blood samples are needed. The truncation value of basic luteinizing hormone (LH) in diagnosing CPP varies greatly in different studies. The diagnostic prediction models of CPP that have been reported are either inconvenient to operate or unsatisfactory in their diagnostic performance.Objective To establish a predictive model which is convenient for clinical operation and has high diagnostic efficiency to assist the diagnosis of CPP in girls.DesignClinical data of girls with precocious puberty (PP) who completed the GnRH stimulation test were collected. Lasso regression analysis was used to screen the predictors of CPP, and Logistic regression was used to establish the prediction model.Methods Clinical data were collected from girls admitted to Fuzhou Children's Hospital of Fujian Medical University from January 2014 to April 2020 with PP who had undergone GnRH stimulation test and whose breast developed at the age of ≥4 to < 8. Through literature review and consultation with clinical experts in endocrinology, independent predictors were screened out preliminarily. After variable conversion, Lasso regression analysis was used to determine the final predictors. Logistic model was re-fitted to analyze its diagnostic performance and conduct internal validation.Main outcome measures The diagnostic efficacy of the model for CPP in girls.Results A total of 1,107 PP girls were included into the analysis, including 537 CPP and 570 non-CPP. Finally, five predictors were included-course of PP, breast Tanner stage, basic LH, bone age (BA), and uterine size-to establish a prediction model for CPP in girls: LN[P/(1-P)]=-5.508+1.579×basic LH("Middle")+2.861×basic LH("High")+1.191×uterine size+0.316×BA+0.371×course of PP("Middle")+0.430×course of PP("Long")+0.285×breast Tannar stage("B>2"). The AUC of the ROC of this model was 0.858. When the predicted cut-off point was 0.476, the Youden index was the highest, with sensitivity of 72.6% and specificity of 86.7%. When the prediction cut-off point was 0.75, the specificity and sensitivity were 95.1% and 50.5%. When the prediction cut-off point was 0.25, the sensitivity and specificity were 90.9% and 51.9%.Conclusion The diagnostic performance of the diagnostic prediction model for CPP in girls constructed in this paper is relatively satisfactory, and different diagnostic cut-off points can be used for clinical diagnosis or screening.

Key words: Girls, Central precocious puberty, Prediction model, Lasso regression