Predicting Bone Metastasis in Thyroid Cancer: Can Machine Learning Play a Role?
Posted: Monday, May 3, 2021
Far better than a crystal ball or a mirror on the wall, machine learning may help predict the future for patients with thyroid cancer. Specifically, Jia-Ming Liu, MD, of Nanchang University, China, and colleagues have studied how machine learning models, with the help of demographic and clinicopathologic data about thousands of previous patients, may help predict which individuals with newly diagnosed thyroid cancer may develop bone metastases. In Cancer Medicine, they described how the random forest algorithm model they created, based on machine learning, may facilitate personalized diagnosis and refine clinical decision-making for this patient population, concluded Dr. Liu and colleagues.
Bone metastases tend to be rare and usually asymptomatic among patients with thyroid cancer, the investigators explained. Thus, testing for bone metastases during an initial thyroid cancer diagnosis usually is infrequent. In fact, bone scans are recommended “only in the presence of suspicious skeletal‐related events, and…the median time to develop [these events] is 5 months after bone metastasis,” they stated. “By then, many thyroid cancer patients may [have missed] out on the best treatment opportunities because they may have developed advanced disease or multiple metastases.”
The retrospectively analyzed data came from 17,138 patients diagnosed with thyroid cancer between 2010 and 2016 and documented in the Surveillance, Epidemiology, and End Results database. Just under 1% of patients in this cohort (n = 166) developed bone metastases, which was lower than the 4% the team expected, they noted, based on earlier prevalence reports.
In outperforming the logistic regression model, the random forest model generated promising statistics, the authors noted: area under the curve = 0.917; accuracy = 0.904; recall rate = 0.833; and specificity = 0.905. The most important bone metastasis predictors were grade, T stage, histology, race, sex, age, and N stage.
Disclosure: The study authors reported no conflicts of interest.