medwireNews: Applying a machine learning technique to hormone measurements can help determine whether a girl with early signs of puberty has nonprogressive premature thelarche (PT) or may require treatment for idiopathic central precocious puberty (ICPP), suggest study findings published in The Journal of Clinical Endocrinology & Metabolism.
Although no single biochemical or anthropological measure was found to accurately differentiate between the two conditions, the principal component analysis (PCA) of six hormones was able to predict ICPP in girls with high accuracy, sensitivity and specificity.
“This offers clinicians an alternative approach in the early outpatient setting, where diagnostic uncertainty exists between nonprogressive PT and CPP before resorting to a [gonadotropin-releasing hormone; GnRH] stimulation test”, say Amanda Cleemann Wang, from Copenhagen University Hospital in Denmark, and co-authors.
The team collated data from 474 girls who were referred to the Rigshospitalet in Copenhagen between 2009 and 2019 with early puberty, 143 of whom were diagnosed with ICPP, defined as having breast stages (BS)2–5 before age 8 years and pubic hair stages (PH)1–5 alongside a basal luteinising hormone (LH) level above 0.3 IU/L and/or a pubertal GnRH stimulation test peak LH above 5 IU/L.
A further 91 girls were diagnosed with premature thelarche, defined as BS2–5 and PH1 with a basal LH of less than 0.3 IU/L and/or a peak prepubertal LH of less than 5 IU/L.
The remaining girls were diagnosed with premature adrenarche (n=224), organic CPP (n=11) or peripheral PP (n=5).
Cleemann Wang and colleagues say that despite a “marked increase” in the annual incidence of CPP and PT reported in Denmark in previous years, both the number of referrals per year and the proportion of girls with ICPP, PT and other diagnoses were stable during the 10-years of the current study.
“This stabilization of numbers may be due to a permanent change in referral patterns from general practitioners, a larger acceptance of early puberty among parents, or a true cessation in the previously reported secular trend toward earlier puberty”, they comment.
The investigators also note that 34% of the girls referred in their cohort had either developed breasts after age 8 years and were considered to have normal early changes or no longer had signs of puberty and were deemed to have transient thelarche.
Girls with ICPP did not significantly differ from those with PT by median age at diagnosis (7.19 vs 7.14 years) or by height, weight or BMI. But girls with ICPP had a significantly higher median height velocity than their PT counterparts (7.49 vs 6.19 cm/year) and greater bone age advancement (median 1.22 vs 0.32 years). They also had significantly higher concentrations of insulin-like growth factor (IGF)-I, basal follicle-stimulating hormone (FSH), oestradiol, inhibin B, testosterone, dehydroepiandrosterone sulphate and androstenedione.
Finding that none of these biomarkers could adequately differentiate between girls with the two conditions, the researchers created two PCAs – the first including bone age advancement, IGF-I, basal LH, basal FH, testosterone and androstenedione, the second including only bone age advancement, IGF-I, and basal levels of LH and FSH.
The first PCA including androgens was 87% accurate for ICPP diagnosis, with 90% specificity and 84% sensitivity, and positive and negative predictive values of 89% and 86%, respectively.
The corresponding values for the second PCA excluding androgens were poorer, at 83%, 82%, 85%, 73% and 90%.
“[C]ombining multiple parameters in a clinical and hormonal profile may improve the diagnostic differentiation of ICPP and PT, leaving the GnRH test redundant”, the authors suggest.
“Importantly, the 2 PCA models did not include parameters derived from the GnRH test, which until now has been considered the gold standard in diagnosing PP”, they write. “In the future, machine learning algorithms such as PCA may be a valuable diagnostic tool to minimize costly and time-consuming examinations, like the stimulation test in healthy children.”
By Lynda Williams
medwireNews is an independent medical news service provided by Springer Healthcare Ltd. © 2023 Springer Healthcare Ltd, part of the Springer Nature Group