BACKGROUND
Thyroid cancer has an excellent overall prognosis with a low recurrence rate and very few patients actually die from the disease. This is because the cancer is very slow growing and we have very effective treatments, including surgery and radioactive iodine therapy. The latter acts as a magic bullet to seek out and destroy thyroid cancer cells. Because of the excellent prognosis, the treatment of thyroid cancer has changed to a more conservative approach involving more limited surgery, selective use of radioactive iodine therapy and less frequent follow-up. It is important to differentiate between the majority of patients with low-risk thyroid cancer and those with more aggressive features to recommend adequate treatment and follow-up. The ATA Risk Stratification System is a widely used method to estimate the prognosis and recurrence risk based on specific features and helps to guide treatment and follow-up for thyroid cancer patients. However, more recent research has showed that additional factors may affect the recurrence rate in thyroid cancer, while questioning the importance of several previously reported factors.

The goal of this study was to develop a comprehensive data-driven model to predict the risk of thyroid cancer persistence/recurrence by including all available patientand cancer-related features at the time of initial treatment. The performance of this model, which determines the impact of each feature on prognosis was compared to the ATA Risk Stratification System.

THE FULL ARTICLE TITLE
Grani G, et al. A data-driven approach to refine predictions of differentiated thyroid cancer outcomes: a prospective multicenter study. J Clin Endocrinol Metab 2023;108(8):1921-1928; doi: 10.1210/clinem/dgad075. PMID: 36795619.

SUMMARY OF THE STUDY
The study evaluated 4773 consecutive thyroid cancer patients with at least one follow-up registered in the Italian Thyroid Cancer Observatory (ITCO) database. This database collects data on thyroid cancer patients followed at 40 different academic and non-academic healthcare settings across Italy. The study was designed to provide a picture of real-world practice, where the participating centers managed the thyroid cancer patients independently without any general guidance or restrictions within the network.

The study data included demographic data, family history of cancer, how the cancer was discovered, cancer pathology data, surgery and radioactive iodine therapy data, and results of the follow-up tests. The study patients were followed for an average of 26 months (range: 6–84 months). The initial treatment consisted of total thyroidectomy and radioactive iodine therapy in 51% of patients, total thyroidectomy alone in 45%, and lobectomy alone in 3% of patients. The response to treatment and the risk of persistent/ recurrent disease was calculated using the ATA Risk Stratification System. Among the 4773 study patients, 52% were classified as having a low risk, 39% as having an intermediate risk, and 9% as having a high risk of persistent or recurrent cancer.

A decision-tree risk prediction model was used to assign a risk index for persistence/recurrence to each patient. This statistical model allows the investigation of the contribution of different patient- and cancer related features to the thyroid cancer recurrence risk. Two models were created, the first algorithm including all available variables, while the second algorithm excluded the variables used by physicians to decide whether to proceed with radioactive iodine treatment, since these variables were taken into consideration when recommending this treatment.

The two decision-tree models showed better performance as compared with the ATA Risk Stratification System. The second decision-tree model increased the sensitivity to detect structural disease from 37% to 49% in high-risk patients and improved the capacity to rule out the presence of persistence/recurrence by an additional 3% in low-risk thyroid cancer patients. Several factors not included in the ATA risk stratification system, such as age, gender, body-mass index (BMI), circumstance of cancer diagnosis, family history of thyroid cancer, surgical method, presurgical cytology result from thyroid nodule biopsy were found to affect the prediction of thyroid cancer persistence or recurrence.

WHAT ARE THE IMPLICATIONS OF THIS STUDY?
This study’s results suggest that machine learning systems using large databases can improve prediction of thyroid cancer persistence or recurrence. Inclusion of additional variables than those used in current risk-stratification systems can improve the risk assessment. This represents an important step towards precision medicine in predicting thyroid cancer recurrence.

—Alina Gavrila, MD, MMSC

Table of Contents | PDF File for Saving and Printing