Pituitary volume as a permissive morphological threshold for hypothalamic-pituitary-gonadal axis activation in children: a comprehensive diagnostic prediction study with relative importance decomposition

@article{cai2026pituitary,
  title = {{Pituitary volume as a permissive morphological threshold for hypothalamic-pituitary-gonadal axis activation in children: a comprehensive diagnostic prediction study with relative importance decomposition}},
  author = {Jianjian Cai and Ling Zhang and Weiyin Vivian Liu and Qin Liu and Jun Lu and Dong Liu},
  journal = {Frontiers in Endocrinology},
  year = {2026},
  month = oct,
  doi = {10.3389/fendo.2026.1981603},
  url = {https://www.frontiersin.org/articles/10.3389/fendo.2026.1981603/full},
  note = {CC BY 4.0},
}

TY  - JOUR
TI  - Pituitary volume as a permissive morphological threshold for hypothalamic-pituitary-gonadal axis activation in children: a comprehensive diagnostic prediction study with relative importance decomposition
AU  - Jianjian Cai
AU  - Ling Zhang
AU  - Weiyin Vivian Liu
AU  - Qin Liu
AU  - Jun Lu
AU  - Dong Liu
PY  - 2026
DA  - 2026/10/22
JO  - Frontiers in Endocrinology
DO  - 10.3389/fendo.2026.1981603
UR  - https://www.frontiersin.org/articles/10.3389/fendo.2026.1981603/full
AB  - Objective: The gonadotropin-releasing hormone (GnRH) stimulation test, the reference standard for confirming hypothalamic-pituitary-gonadal axis (HPGA) activation, is invasive and costly. Pituitary volume (PV) on MRI has been proposed as a non-invasive surrogate, yet whether PV constitutes a permissive morphological threshold or is redundant with clinical parameters remains undefined. We evaluated PV as a morphological threshold of HPGA activation and quantified its relative importance, unique information content, and clinical utility.  Methods: We retrospectively enrolled 896 children who underwent pituitary MRI and GnRH stimulation testing. PV was measured using the ellipsoid approximation. Out-of-fold cross-validated AUC was the primary performance metric. Predictor importance was quantified by Shapley value decomposition, permutation importance, and Lindeman–Merenda–Gold (LMG) variance partitioning. Morphological-threshold analyses examined HPGA activation across PV deciles. Statistical mediation tested whether insulin-like growth factor 1 (IGF-1) mediated the PV–HPGA association, with formal power analysis for incremental AUC detection.  Results: PV exhibited a threshold pattern: HPGA activation was 28.9% (95% CI 20.5–39.0) in the lowest PV decile (≤115 mm³) versus 92.2% (95% CI 84.8–96.2) in the highest (≥335 mm³; OR 4.38, 95% CI 2.60–7.36; per-IQR OR 2.04, 95% CI 1.59–2.62). PV ranked second by Shapley decomposition (+0.065 AUC, 22.2%), carrying unique information (LMG 20.4% of pseudo-R²). At the 203.9 mm³ cutpoint, the diagnostic odds ratio was 4.78, positive likelihood ratio 2.44, and number needed to screen 2.73. PV tertile AUC was 0.693; combined with IGF-1, AUC rose to 0.748. IGF-1 statistically mediated 37.4% of the PV–HPGA association. The full model OOF AUC was 0.801 (95% CI 0.772–0.828) versus 0.798 for the clinical model (ΔAUC +0.003); power was ~5% at N = 896, requiring N ≈ 25360 for 80% power. The age-sex-adjusted PV z-score AUC was 0.592 (95% CI 0.554–0.630).  Conclusion: PV serves as a permissive morphological threshold for HPGA activation; while it’s not a strictly necessary condition, HPGA activation is physiologically improbable below a critical PV. Although PV is not a standalone predictive tool and its incremental value over clinical parameters is minimal, it functions as a complementary morphological checkpoint that contextualizes hormonal interpretation when pituitary MRI has been conducted to exclude intracranial lesions as per guidelines.
SN  - 1664-2392
ER  -