Model Predicts Presence of Occult Nodal Metastasis Better Than Current Standard in Early-Stage Oral Cavity Squamous Cell Carcinoma

For many patients with early-stage oral cavity squamous cell carcinoma (OCSCC) and clinically negative nodes, elective neck dissection (END) is standard practice because of the high risk of subclinical nodal metastasis. “However, approximately 70% to 80% of patients who undergo END will have pathologically negative lymph nodes, potentially resulting in unnecessary surgical morbidity and increased health care costs for these patients,” wrote Andrés Bur, MD, Department of Otolaryngology–Head and Neck Surgery, University of Kansas Medical Center, Kansas City, KS, and colleagues regarding their study of machine learning models they developed to predict the presence of occult nodal metastasis in these patients (JAMA Network Open. 2022:5(4):e227226. doi:10.1001/jamanetworkopen.2022.7226).

To develop their models, the researchers used clinicopathological variables available after surgical removal of the primary tumor that were collected from 7 tertiary care academic medical centers throughout the United States. Their study aimed to validate the models and to compare their predictive performance with the performance of depth of invasion, the currently accepted standard for predicting occult nodal metastasis in patients with OCSCC after surgical removal of the primary tumor.

The study enrolled 634 adult patients with early-stage OCSCC without nodal disease who had surgical removal of the primary tumor, either with or without END. The mean age of the patients in the study group was 61.2 years (standard deviation, 13.6 yr). Initial evaluation of these patients occurred between 2000 and 2019. The main study outcome was occult nodal metastasis, identified either at END or on the basis of regional recurrence within 2 years of the initial surgery.

The researchers found that 114 of the 634 patients, or 18%, had occult nodal metastasis and that patients with occult nodal metastasis were more likely to have lymphovascular invasion, perineural invasion, and margin involvement by invasive tumor than those without pathological lymph node metastasis. More specifically, for patients with occult nodal metastasis compared with patients without them, the frequency of lymphovascular invasion was 26.3% vs. 8.1% (P<0.01), of perineural invasion, 40.4% vs. 18.5% (P<0.01), and of margin involvement by invasive tumor, 12.3% vs. 6.3% (P=0.046). 

Moreover, patients with occult nodal metastasis were more likely to have a poorly differentiated primary tumor (20.2% vs. 6.2%) and greater depth of invasion than patients without them (7.0 vs. 5.4 mm; P<0.01 for both comparisons). 

The study also revealed that the predictive model the team constructed with XGBoost architecture outperformed the standard, 4-mm depth of invasion threshold; the model’s area under the curve was 0.84 (95% confidence interval [CI], 0.80 to 0.88) vs. 0.62 (95% CI, 0.57 to 0.67) for the standard depth of invasion threshold. The model had a sensitivity of 91.7%, a specificity of 72.6%, a positive predictive value of 39.3%, and a negative predictive value of 97.8%.

“Results of this study showed that machine learning models that were developed from multi-institutional clinicopathological data have the potential to not only reduce the number of pathologically node-negative neck dissections but also to accurately identify patients with early OCSCC who are at highest risk for nodal metastases,” concluded Dr Bur and colleagues.