Objective To construct a prediction model for radiodermatitis (RD) associated with breast cancer radiotherapy based on frequent pattern growth (FP-Growth) algorithm, and to evaluate its prediction efficiency for acute and advanced RD, providing a reference for clinical RD risk warning.

Methods A retrospective analysis was conducted on the clinical data of 1000 female patients (age (50.6±18.4) years) who received radiotherapy after modified radical mastectomy for breast cancer from January 2010 to January 2024 at the First Affiliated Hospital of Hebei North University. The patients were divided into modeling and validation groups using simple random sampling method. The FP-Growth algorithm was used to perform the association rule analysis on the baseline data of the patients in the modeling group, and a prediction model for acute and advanced RD was established. The model was internally validated using the consistency index (CI) and calibration curve. Meanwhile, external validation was realized using the area under the curve (AUC) of the receiver operating characteristic (ROC) curve. Comparison between the groups of count data and rank data was conducted using the chi square test and the Wilcoxon rank sum test, respectively. The Delong test was also used to compare differences in AUC between the modeling and validation groups, and the Hosmer-Lemeshow test was used to evaluate the goodness-of-fit of the ROC curves.

Results The modeling and validation groups comprised 512 and 488 patients, respectively. The RD prediction model ultimately selected two effective strong association rules. (1) The combination of effective strong association rules for acute RD: body mass index (≥35 kg/m²), tumor size (4–5 cm), history of chemotherapy, and albumin level (<35 g/L). The incidence of acute RD was 79%. (2) The combination of effective strong association rules for advanced RD: age (60–69 years), Karnofsky Performance Status score (<70 points), history of chemotherapy, history of diabetes, and albumin level (<35 g/L). The incidence of advanced RD was 69%. Internal validation results showed that the CI of the RD prediction model for acute RD and advanced RD in the modeling group was 0.863 (95%CI: 0.646–0.932, P=0.011) and 0.812 (95%CI: 0.669–0.892, P=0.023), respectively. The calibration curve indicated good consistency between the predicted probability and the actual probability of the model. External validation results revealed that the AUC for acute RD prediction in the modeling and validation groups was 0.882 and 0.876, respectively, revealing no statistically significant difference (Z=0.334, P=0.205). AUC for advanced RD prediction in the modeling and validation groups was 0.673 and 0.668, respectively, revealing no statistically significant difference (Z=0.982, P=0.092). The Hosmer-Lemeshow test results showed that the prediction models for acute and advanced RD were well fitted (χ²=4.921, 5.039; P=0.125, 0.327).

Conclusion The efficiency of the RD prediction model for acute and advanced RD based on the FP-Growth algorithm can meet clinical requirements, serving as a reference for risk warning and clinical intervention for RD associated with breast cancer radiotherapy.