Abstract: Objective: To develop the prediction models for begin and malignant parotid neoplasms by analyzing multiple examination indexes. Methods: A retrospective analysis was performed on 242 patients with parotid neoplasms who were treated with parotid neoplasms resection in the First Hospital of Lanzhou University from March 2013 to March 2019(group A). The independent predictors of malignancy in parotid tumor patients were screened out through logistic regression analysis, then a prediction model was built. Other 42 parotid neoplasm patients(group B)with definite pathological diagnosis from April 2019 to September 2020 were used to validate this model. Results: Through multivariate logistic regression analysis on group A, the prediction model was built as follows: Y=e^x/ (1+e^x), X=-0.18347+ (-1.29435×tumor shape) + (-1.44877×tumor boundary) + (4.34121×surrounding tissues), in which “e” was natural logarithm. The validity of this model was validated by data from group B, and the results showed that the accuracy was 95.2%, the sensitivity was 1.00, and the specificity of the model was 93.4%. Conclusion: The establishment of logistic regression prediction model can help clinicians diagnose benign and malignant parotid tumors by combining the presence or absence of regular tumor shape, clear tumor boundaries, and involvement of surrounding tissues.

Key words: parotid neoplasms, differential diagnosis, Logistic regression, prediction model