Abstract The incidence of moderate to severe pain after chemotherapy with primary hepatic carcinoma (PHC) patients is high. Although standardized treatment can effectively relieve pain, the control effect is poor. More attention should be paid to the prevention of pain at the beginning of symptoms, so as to reduce the incidence of pain and promote the health of patients. However, there are lack of a prospective design to predict pain before it occurs. The study is a prospective case‒control study. Population was PHC patients who received chemotherapy from April to August to 2024 in three grade 3 and first-class hospital. Data were collected in two periods (on the day of admission and within 24 h of chemotherapy). According to the Brief Pain Inventory, the patients were divided into case group and control group. Then the patients were randomly divided into a training group and an internal validation group at a 2:1 ratio. Single-factor logistics regression was used to analyze the risk factors, and the back-propagation artificial neural network (BP-ANN) model was constructed and verified. A total of 467 patients consisting of 312 training samples and 155 validation samples. BP-ANN model showed the AUC, sensitivity, specificity, and accuracy of prediction were 0.808, 70.6%, 81.7%, 93%, respectively. Internal verification also indicated these indicators were 0.783, 78.8%, 70.8%, and 94.2%, respectively. Significant predictors identified were age > 57.5, BMI > 19.9, symptoms of insomnia prior to illness, worker, Renvastinib, Child–Pugh = B, glutamic oxalacetic transaminase, other platinum drugs, cancer staging of IV, ECOG = 2, NRS-2002 = 3, Oxaliplatin, and Donafenib. The BP-ANN model holds high predictive value for the moderate to severe pain of PHC patients after chemotherapy. In the future, the model can be further visualized to facilitate clinical screening and to provide a basis for subsequent intervention.