IMPE Abstracts

Background: Gonadotropin-releasing hormone agonists (GnRHa) are the treatment of choice for girls with central precocious puberty (CPP). Several adult height prediction models by using deep learning based on growth measurements have been presented. But there is no widely used machine learning in Korean population to predict adult height. The aim of this study was to evaluate the effects of GnRHa therapy on final height outcomes from the start of the treatment to after menarche in girls with idiopathic CPP and estimate final adult height based on initial anthropometric data using deep learning algorithm.

Methods: We retrospectively reviewed the medical records of 200 girls with idiopathic CPP who reached menarche after GnRHa treatment. The change in height standard deviation score (SDS) and predicted adult height (PAH) SDS were assessed using the Bayley-Pinneau method from the start of the treatment to the end of treatment and after menarche. In this study, we developed a predicting adult height model by using features at the start of treatment to develop deep learning algorithm with Python, Tensorflow, Numpy.

Results: Mean chronological age (CA) at the start of GnRHa treatment was 8.24 ± 0.73 years. Mean treatment duration of GnRHa was 3.12± 0.81 years. Menarche had occurred at 12.73 ± 0.56 years of age, which was 1.43 ± 0.46 years after discontinuation of GnRHa therapy. The PAH(SDS) after menarche was significantly increased to 163.53 ± 5.05 cm (0.48 ±0.99) from 154.58 ± 6.72 cm (-1.33 ± 1.46) before treatment (P<0.001). The difference in initial PAH SDS and post-menarche PAH SDS was compared according to the age at onset of GnRHa treatment: <7 years, 7–8 years, and 8-9 years. Younger ages at start of treatment resulted in more PAH SDS gains. (-2.6 ± 1.71, -1.40 ± 1.49, and -1.19 ± 1.37, respectively; P< 0.05). The average difference between prediction adult height estimated by deep learning and PAH at post-menarche was 1.55 cm.

Conclusion: GnRHa treatment in girls with CPP improved the final adult height outcomes. Lower chronological age at the onset of treatment lead to greater final height. The developed model by using deep learning algorithm can be usefully applied in prediction adult height based on features at start of treatment. The prospective validation of the adult height prediction by this deep learning model should be conducted in the future.