Addition of tumor microenvironment immune cell composition to improve the performance of a predictive model for oral squamous cell carcinoma

Background Conventional clinicopathological characteristics insufficiently predict prognosis in oral squamous cell carcinoma (OSCC). We aimed to assess the added predictive value of tumor microenvironment immune cell composition (TMICC) in addition to conventional clinicopathological characteristics. Methods Primary tumor samples of 290 OSCC patients were immunohistochemically stained for CD4, CD8, CD20, CD68, CD163, CD57, FoxP3 and Programmed cell Death Ligand 1. Additionally, clinicopathological characteristics were obtained from patients’ medical files. Predictive models were trained and validated by conducting Least Absolute Shrinkage and Selection Operator (LASSO) regression analyses with cross-validation. To quantify the added predictive power of TMICC within models, receiver operating characteristic (ROC) analyses were used. Results Recurrence occurred in 74 patients (25.5%). Conventional clinicopathological characteristics (tumor localization, pathological T-stage, pathological N-stage, extracapsular spread, resection margin, differentiation grade, perineural invasion, lymphovascular invasion) and treatment modality, were used to build a LASSO logistic regression-based predictive model. Addition of TMICC to the model resulted in a comparable AUC of respectively 0.79 (±0.01) and 0.76 (±0.1) in the training and test sets. The model indicated that high numbers of CD4+ T cells protected against recurrence. Lymph node metastasis, extracapsular spread, perineural invasion, positive surgical margins and reception of adjuvant treatment were associated with increased risk for recurrence. Conclusions The TMICC, specifically the number of CD4+ T cells, is an independent predictor , however, addition to conventional clinicopathological characteristics does not improve the performance of a predictive model for recurrence in OSCC.