Machine learning (ML), a type of artificial intelligence, could be used to predict how patients with acromegaly respond to first-generation somatostatin receptor ligands (fg-SRLs), according a study recently published in The Journal of Clinical Endocrinology & Metabolism.
Researchers led by Mônica Gadelha, MD, PhD, of the Endocrine Unit and Neuroendocrinology Research Center at the Medical School and Hospital Universitário Clementino Fraga Filho – Universidade Federal do Rio de Janeiro, in Rio de Janeiro, Brazil, point out that fg-SRLs are the mainstay of acromegaly medical treatment, but control rates are about 40%. “Diverse factors have been investigated as possible biomarkers of biochemical response/control to fg-SRL in acromegaly,” the authors write. “Mostly, they have been individually evaluated in small series. The combined use of different biomarkers enhance their accuracy to predict biochemical response to fg-SRL. Artificial intelligence (AI), more specifically machine learning (ML), is a useful tool to study the impact of different biomarkers on the therapeutic response simultaneously and to design accurate prediction models.”
For this study, the researchers included 153 patients from 16 Brazilian reference centers with acromegaly who had not been cured by primary surgical treatment who had adjuvant therapy with fg-SRL for at least six months. The team evaluated six AI models, randomly splitting patient data
into training and test sets (4:1 ratio) until there was no significant difference between features from both sets. These models included logistic regression, k-nearest neighbor classifier, support vector machine, gradient-boosted classifier, random forest, and multilayer perceptron. “The features included in the analysis were age at diagnosis, sex, GH and IGF-I levels at diagnosis and at pretreatment, somatostatin receptor subtype 2 and 5 (SST2 and SST5) protein expression and cytokeratin granulation pattern (GP),” the authors write.
The researchers found that the model that performed best was the support vector machine with the features SST2, SST5, GP, sex, age, and pretreatment GH and IGF-I levels. It had an accuracy of 86.3%, positive predictive value of 83.3% and negative predictive value of 87.5%. “In conclusion,” the authors write, “we developed an ML-based prediction model with high accuracy using biomarkers that are routinely available in all patients (age, hormone levels) or are affordable (SST2 and CAM5.2 IHC). It can improve medical management of acromegaly by helping make rational therapy choices, reducing acromegaly morbidity and mortality, the therapeutic burden on patient’s quality of life and health services costs.”