A diagnostic modeling study has demonstrated that an interpretable three-variable machine learning model using only routine pre-stimulation measurements can distinguish central precocious puberty from benign premature thelarche in girls presenting with early breast development. Developed using data from a tertiary hospital in northwest China, the parsimonious model relies on patient age, basal follicle-stimulating hormone, and the basal luteinizing hormone to follicle-stimulating hormone ratio. The retrospective analysis indicates that applying a pre-specified two-threshold triage strategy could potentially allow an estimated 60.0 percent of evaluated girls to defer invasive diagnostic testing, although external multicenter prospective validation is required before the tool can be considered for clinical use.
The study analyzed a cohort of 130 girls presenting with early breast development between 2020 and 2026 who had confirmed final diagnoses of either central precocious puberty (51 patients) or premature thelarche (79 patients). These individuals were identified from a broader screening population of 15,385 consecutive pediatric outpatients treated at the tertiary facility.
To construct the predictive framework, the researchers initially evaluated 17 candidate pre-stimulation variables, applying least absolute shrinkage and selection operator (LASSO) regression to eliminate redundant features. The team compared six distinct algorithmic approaches using repeated stratified five-fold cross-validation. To minimize overoptimistic performance estimates, feature selection, data imputation, and standardization procedures were refitted entirely within each training fold in a fully nested validation architecture. The resulting models were subsequently evaluated through out-of-fold area under the receiver operating characteristic curve (AUC), calibration curves, decision-curve analysis, and Shapley additive explanations (SHAP).
Across validation folds, patient age, basal follicle-stimulating hormone, and the basal luteinizing hormone to follicle-stimulating hormone ratio emerged as the most consistently selected predictors. A penalized logistic regression model incorporating these three metrics achieved a nested validation AUC of 0.813 and a pooled out-of-fold AUC of 0.805 (95% CI, 0.729 to 0.876), performing on par with complex non-linear algorithms. While patient age alone yielded an AUC of 0.818, the addition of the baseline hormonal markers increased the overall discriminatory AUC to 0.855. When matched at a clinical benchmark of 90 percent sensitivity, adding the hormonal variables improved diagnostic specificity from 0.569 to 0.627, accompanied by a moderate calibration slope of 0.600.
When applying a pre-specified two-threshold triage protocol to the validation cohort, 38.5 percent of the patients fell below the rule-out threshold, registering a negative predictive value of 0.900, which corresponded to five girls with central precocious puberty among 50 assigned to initial non-invasive surveillance. Meanwhile, 21.5 percent of patients scored above the rule-in threshold, exhibiting a positive predictive value of 0.750. Combined, these thresholds suggested that an estimated 60.0 percent of girls could potentially avoid immediate stimulation testing. The investigators explicitly emphasized that this figure reflects a retrospective mathematical estimate rather than a prospectively observed reduction in clinical testing.
Diagnostic accuracy showed variation across specific clinical subgroups, with model discrimination attenuating to an AUC of 0.707 among girls aged 6 to 8.99 years, while remaining relatively stable at an AUC of 0.787 in a hardened reference-standard subset of 101 patients. Because the model demonstrated moderate calibration and initially deferred definitive testing in five patients who had true central precocious puberty, the authors concluded that the hand-computable tool cannot be implemented in clinical workflows without rigorous prospective validation across multiple independent medical centers and careful local recalibration.
Subject of Research: Diagnostic machine-learning model differentiating central precocious puberty from premature thelarche using pre-stimulation clinical data
Article Title: Interpretable machine learning for the non-invasive differentiation of central precocious puberty from premature thelarche in girls: a diagnostic modeling study to reduce reliance on the gonadotropin-releasing hormone stimulation test
Article References: Da, Z., Chai, J., Liu, Y., Gao, Y., Zhu, Y., Li, J., Huang, L., Gao, Y., Lu, D., Li, T., Wang, W., Xie, X., & Nan, N. (2026). Interpretable machine learning for the non-invasive differentiation of central precocious puberty from premature thelarche in girls: a diagnostic modeling study to reduce reliance on the gonadotropin-releasing hormone stimulation test. Frontiers in Endocrinology, 17, Article 2008437. https://doi.org/10.3389/fendo.2026.2008437
Image Credits: AI Generated
DOI: 10.3389/fendo.2026.2008437
Keywords: central precocious puberty, premature thelarche, gonadotropin-releasing hormone stimulation test, machine learning, diagnostic modeling, LASSO regression, penalized logistic regression, follicle-stimulating hormone, luteinizing hormone, pediatric endocrinology, area under the curve, clinical triage, pre-stimulation predictors, SHAP analysis, nested cross-validation, breast development, endocrinology
Cite Scienmag News
Ophelia Keating. (November 17, 2026). Three-Variable Pre-Stimulation Model Identifies Central Precocious Puberty in Girls. Scienmag. https://scienmag.com/three-variable-pre-stimulation-model-identifies-central-precocious-puberty-in-girls/
Ophelia Keating. "Three-Variable Pre-Stimulation Model Identifies Central Precocious Puberty in Girls." Scienmag, 17 November 2026, https://scienmag.com/three-variable-pre-stimulation-model-identifies-central-precocious-puberty-in-girls/. Accessed 17 November 2026.
Ophelia Keating. "Three-Variable Pre-Stimulation Model Identifies Central Precocious Puberty in Girls." Scienmag. November 17, 2026. https://scienmag.com/three-variable-pre-stimulation-model-identifies-central-precocious-puberty-in-girls/








