Establishment and Verification of a Prediction Model for Identifying Pathologic Infections Based on the Clinical Characteristics of Epidermoid Cysts.

Background: To construct a prediction model based on the clinical characteristics of epidermoid cysts to identify pathologic infections, evaluate the diagnostic accuracy of the model, and conduct preliminary verification. Patients and Methods: We conducted a retrospective analysis of 314 patients diagnosed with epidermoid cysts that had been removed surgically. The clinical and pathologic data of all patients were collected. The patients were divided randomly into modeling group and verification group in a 75:25 ratio. In the modeling group, the multifactor logistic regression method was used to construct a prediction model for identifying epidermoid cyst pathologic infection, and the receiver operating characteristic (ROC) curve was used to evaluate the diagnostic accuracy of the model, which was then validated in the verification group. Results: All 314 patients with epidermoid cysts were divided into non-infected group (183 cases) and infected group (131 cases) according to the pathologic results. Logistic regression analysis showed that the disease course, growth trend, redness, and texture of epidermoid cysts were independent factors affecting pathologic infection. The above four indicators were selected to construct the prediction model of epidermoid cyst pathologic infection. In the modeling group, the prediction model showed an area under the curve (AUC) of 0.898, with the sensitivity of 0.830, specificity of 0.890, positive likelihood ratio of 7.523, and negative likelihood ratio of 0.191. The AUC of the prediction model in the verification group was 0.919, which was not significantly different from that of the modeling group (p = 0.886). Conclusions: The prediction model based on the clinical characteristics of epidermoid cysts had good diagnostic accuracy and high specificity; it can be used to identify pathologic infections of epidermoid cysts.