Lung metastasis (LM) in patients with thyroid cancer (TC), including medullary thyroid cancer (MTC), is independently associated with age, histopathological type, tumor size, and lymph node infiltration, according to a new study published in Cancer Medicine. The random forest machine learning algorithm showed promising results in accurately predicting LM in patients previously diagnosed with TC.

This retrospective study conducted by Liu et al used data from 9950 patients with TC, out of which 2.1% had LM. When comparing the LM group vs the nonlung metastasis (NLM) group, age was significantly higher, with an average of 54.52 vs 46.87, respectively. Regarding the tumor, node, and metastasis staging system, the majority of the patients in the NLM group were classified as T1a with 31% and T1b with 23.5%, while most (56.6%) patients in the LM group had a T4b stage.

Read more about MTC prognosis

Continue Reading

Additionally, 6 machine learning models were programmed with algorithms to predict LM in the patients included in the study. Decision tree, random forest (RF), k-nearest neighbor, support vector machine, logistic regression, and extreme gradient boosting were included, and while most of them performed well on accuracy with 90% on average, the precision resulted in 50%. However, RF showcased an unmatched accuracy of 0.99 with an above-average precision of 0.61.

“In addition, the clinical and practical importance of machine learning lies in the detection of risk factors that are closely associated with LM,” the authors wrote. “According to the permutation importance of feature variables, T stage, grade, histological type, age, and N stage were critical to LM.” This technology proposes a useful, noninvasive method to predict which patients currently diagnosed with TC may develop LM.

Even though the computerized tomography (CT) scan remains the most sensitive diagnostic tool for LM in TC patients, this prediction system could aid in its early detection by selecting patients at higher risk.

“We also noticed that people under the age of 60 have a reduced risk of LM from TC with age and a minimal risk of developing LM from the ages of 20 to 40 years. Therefore, the frequency of CT scans in TC patients older than 60 years should be higher than that of younger TC patients with earlier detection of LM,” they concluded.


Liu W, Wang S, Ye Z, Xu P, Xia X, Guo M. Prediction of lung metastases in thyroid cancer using machine learning based on SEER database. Cancer Med. Published online February 22, 2022. doi:10.1002/cam4.4617