Abstract:Objective To construct a predictive model to analyze the diagnostic value of serum alpha fetoprotein (AFP), carcinoembryonic antigen (CEA), heat shock protein 90a (Hsp90a) for primary liver cancer (PHC). Methods A retrospective study was conducted on 72 patients with PHC (PHC group), 68 patients with chronic liver disease (chronic liver disease group), and 50 healthy subjects (healthy control group) admitted to The No.2 Hospital of Baoding from March 2020 to December 2021. Serum AFP and CEA levels were detected using chemiluminescence, serum Hsp90a levels were detected using enzyme-linked immunosorbent assay. Receiver operating characteristic (ROC) curve and area under curve (AUC) were used to evaluate the diagnostic efficacy of each indicator for PHC. Multivariate Cox risk regression analysis was used to identify independent factors for the occurrence of PHC. A predictive model was constructed and validated by R software. Results The differences in AFP, CEA and Hsp90a levels among the three groups were statistically significant, with the highest in the PHC group and the lowest in the control group, and the difference was statistically significant in a pairwise comparison (P＜0.05). AFP (OR=1.770, 95%CI: 1.137-2.756), CEA (OR=2.651, 95%CI: 1.168-6.015), and Hsp90a (OR=1.402, 95%CI: 1.141-1.723) were independent risk factors for PHC (P＜0.05). The AUC of AFP, CEA, and HSP90 predicting PHC separately were 0.929, 0.947, and 0.899, respectively. The C-index of the constructed joint prediction model was 0.669 (0.647-0.831). Conclusion The model based on AFP, CEA and HSP90a has high diagnostic value for patients with PHC, and it can be used for early PHC screening in clinic.