Endocrine Abstracts

Background: The appropriate risk stratification of patients with differentiated thyroid cancer (DTC) is crucial because most cases have an indolent behavior and need a conservative approach. One of the most widely used tools is included in the American Thyroid Association (ATA) Guidelines, based on heterogeneous literature data derived by different populations, settings, and timeframes. Recent research focused on the inclusion of other features or questioned the clinical relevance of some of the included ones. In this prospective cohort study, we analyzed data of DTCs managed in 40 Italian clinical centers. The aims were to develop comprehensive, data-driven prediction models, able to capture all available features and to determine the weight of the potential predictors.

Methods: The Italian Thyroid Cancer Observatory (ITCO) web-based database (NCT04031339) now includes prospectively collected data of 10000 patients with histologically confirmed thyroid cancer. Each record contains information on patient demographics and biometrics, circumstances of the diagnosis, tumor pathology, treatments, and periodic follow-up examinations. We selected consecutive cases with DTC (n=4773) and at least early follow-up data. We built a decision tree, a relatively simple prediction model, to assign a risk index to each patient. The model allows to investigate the impact of different variables in the prediction of the risk level.Results. 2492 patients (52.2%) are classified as low, 1873 (39.2%) as intermediate, and 408 as high risk, according to the ATA risk estimation. Their response to treatment during their whole follow-up is excellent response in 2188 (45.8%), indeterminate in 1957 (41%), biochemical incomplete response in 250 (5.2%), and structural incomplete response in 378 (7.9%). The decision-tree model outperformed the ATA risk stratification system: the sensitivity of high-risk classification for structural disease increased from 37% to 49%, and the negative predictive value for low-risk patients also slightly increased by 3%), even without including information derived from radioiodine treatment (performed only in a subgroup of patients). The feature importance was estimated: several variables not included in the ATA system significantly impact the prediction of disease persistence/recurrence: age at diagnosis, gender, body-mass index, cytology, family history of thyroid cancer, surgical approach, pre-surgical cytology, and circumstances of the diagnosis.

Conclusions: The current risk stratification systems may be complemented by the inclusion of other demographic, clinical and anthropometric data, to improve the prediction of response to treatment. The use of a complete set of variables allows for a more precise clustering of patients, to predict their responses to treatment.