From Data to Decisions: Tailoring Treatment for Pituitary Tumors

Pituitary neuroendocrine tumors (PitNETs) account for approximately 10–15% of primary brain tumors, making them among the most common intracranial neoplasms. Although most PitNETs are histologically benign, they can have significant clinical consequences. Patients may experience hormone hypersecretion, leading to conditions such as acromegaly or Cushing syndrome, hormone deficiency, visual disturbances, or symptoms related to tumor invasion of adjacent tissues.

Notably, the clinical behavior of PitNETs is highly variable. While some tumors remain stable for decades, others can recur, progress, or develop resistance to conventional therapies. Advances in the identification of transcription factors and molecular markers aid in refining the diagnosis of pituitary tumours, with the World Health Organization (WHO) recently updating its classification in 2017, 2020, and 2022. Despite these improvements in diagnostic precision, a major clinical challenge remains unresolved: reliable prediction of clinical outcomes.

Tumours with similar histological and molecular profiles can exhibit markedly different patterns of growth, recurrence, and response to treatment. Consequently, clinicians lack dependable tools to predict prognosis accurately or to individualize therapy beyond standardized treatment algorithms. To assess the ability of computational methods to address this gap, scientists conducted a PRISMA 2020- compliant systematic review of studies applying machine learning (ML) methods to omics data in PitNETs.  

These reviewed studies examined a variety of omics data, such as genomics, transcriptomics, epigenomics, proteomics, and liquid biopsies. Most of the unsupervised clustering or regularised regression methods were used, which are highly appropriate when the dimensions are large, and the samples are small, as in the case of PitNET research. Conversely, deep learning methods were rarely used, primarily because of the lack of large, clinically well-annotated datasets required for effective model training. 

One of the most significant reviews was the limited availability and standardization of data. Most datasets were not publicly accessible, and many lacked critical clinical annotations such as treatment outcomes and long-term follow-up. This variability poses a significant challenge for developing predictive models that are both verifiable and clinically applicable. To support future research, the investigators compiled a comprehensive collection of publicly available PitNET omics datasets, categorizing them based on accessibility and potential clinical utility. A limited percentage of them were high-value resources, which included open access, modern technologies, and annotations of clinical importance. 

Machine learning has the potential to integrate molecular, pathological, and clinical data into predictive models that move beyond descriptive classification. In principle, such models could enable earlier diagnosis, improved risk stratification, and more personalized treatment strategies for patients with PitNETs. However, the review highlights that most existing ML models remain proof-of-concept and lack extensive external validation. Even advanced algorithms struggle to generalize across patient populations in the absence of large, harmonized, multicenter datasets.

In conclusion, the researchers emphasize that realizing the full potential of machine learning in PitNET management will require coordinated data-sharing initiatives and standardized clinical harmonization. International and national collaborations, supported by appropriate ethical and regulatory frameworks that would make sure that ML tools are transparent, reliable, and clinically significant. While machine learning in PitNETs is a promising field, it remains in an early stage and can only transform personalized endocrine oncology if challenges related to data quality, accessibility, and standardization are effectively addressed.

References: Gil J, de Pedro-Campos P, Carrato C, et al. Assessing the value of data-driven frameworks for personalized medicine in pituitary tumours: a critical overview. Mach Learn Knowl Extr. 2026;8(1):16. doi:10.3390/make8010016