Objective To evaluate the accuracy and rationality of decision tree in predicting the risk of unavoidable pressure ulcers for cancer patients, and to provide reference for prevention of pressure ulcers. Methods Clinical data of 611 cancer patients with high Braden scores were collected, and the decision tree model predicting unavoidable pressure ulcers for cancer patients was constructed using the CHAID algorithm, then its prediction effect was compared with that of Braden score according to the area under the ROC curve, sensitivity and specificity. Results Totally, 46 patients (7.53%) developed unavoidable pressure ulcers. The decision tree model included three stratums and eleven nodes. Six classification rules were extracted and four high-risk populations were screened out as follows: Braden score ≤11 points, turning and repositioning plan could not be implemented; Braden score ＞11 points, existing or potential damage appeared in the skin; Braden score ≤11 points, turning and repositioning plan could be implemented, but special situations increasing risk of pressure ulcers existed; Braden score ＞11 points, there was no existed or potential damage in the skin, but the turning and repositioning plan could not be implemented. The area under the ROC curve of the decision tree model was 0.840. The sensitivity of the decision tree model was 0.848, and its specificity was 0.774. Conclusion The decision tree model enjoys good area under the ROC curve, sensitivity and specificity, making it suitable to screen and manage cancer patients at high risk of unavoidable pressure ulcers.