Investigation of the effect of single nucleotide polymorphism and micronucleus formation on radiotoxicity in patients with head and neck cancer treated with curative radiotherapy

Objective: Radiotherapy (RT) is an important local treatment modality in the treatment of head and neck cancers. Although treatment-related toxicity rates have decreased with the development of radiotherapy techniques, many acute or late toxicities can be seen in patients. In addition to factors such as döşe and irradiated volume that affect RT toxicity, the personal radiosensitivity of patients has an important role in the development of these toxicities. Determination of personal radiosensitivity may affect the decision of patients to be treated with surgery or RT and may allow dose volume modifications. For this purpose, many studies have been conducted including genetic data. In this study, we aimed to investigate the effects of rs1805794 polymorphism in the Nibrin (NBN) gene and post-treatment micronucleus frequency on radiation toxicity and to develop an artificial intelligence algorithm that can predict patients at risk of developing serious side effects by using genetic parameters, dosimetric and clinical data. Materials and Methods: The study included 125 patients who received curative radiotherapy treatment for head and neck squamous cell carcinoma and who were disease-free at follow-up visits. Clinical and dosimetric data were collected, chronic side effect examinations were performed during routine controls and blood samples were collected. Micronucleus frequencies were determined by (Cytokinesis Blocked Micronuclei) CBMN method and NBN gene rs1805794 G>C polymorphism was analysed by PCR. Parameters effective on toxicity were determined by using statistical methods. Algorithms were created with four different machine learning methods using the parameters affecting toxicity. While creating the artificial intelligence algorithm, patients with more than 2 grade 2 and above toxicity were grouped as sensitive in order to simplify the model. Results: MN frequency above 10 was found to be associated with grade 2 and higher hearing loss. NBN1 gene rs1805794 G>C polymorphism was found to be associated with severe acute dermatitis. No significant association was found between MN frequency and NBN1 gene rs1805794 G>C polymorphism and other toxicities. Sixty- one patients were radiosensitive and 64 were radioresistant. Among the algoritms developed by using PTV 70 volume, addition of chemotherapy to the treatment, age, T and N stage, smoking, which are effective in the development of chronic toxicity, XGBoost algorithm was found to be the most accurate algorithm. It was found that adding NBN1 gene rs1805794 G>C polymorphism to this algorithm increased the accuracy. It was also determined that the machine learning algorithm including the SNP variable could predict patients with more than 2 grade 2 and above toxicity with an accuracy of 0.76 (AUC: 0.78; Sensitivity: 0.74; Specificity: 0.79). Conclusion: Radiotherapy-related toxicity is a multifactorial process. In this study, it was shown that artificial intelligence algorithms created with dosimetric, clinical and genetic data can predict toxicity and can be used in individualised treatment. Although the developed algorithm is not suitable for routine use, it can be improved by larger number of patients, by detailing the dosimetric data and by using more genetic data with a genome-wide approach. Keywords: Radiotherapy, toxicity, single nucleotide polymorphism, machine learning