Development and validation of a mortality risk prediction model for patients with bone tumors using the MIMIC-IV database: a retrospective cohort study

Background: Bone tumors (BT) are neoplasms that occur in the skeletal system or its affiliated tissues and can be categorized into primary and secondary malignant tumors. Currently, malignant bone tumors are associated with poor prognosis and high mortality rates. The central motivation for this scholarly research was to construct a mortality risk prediction model for BT patients. Methods: In this study, data on BT patients traceable to Medical Information Mart for Intensive Care IV (MIMIC-IV) were utilized to predict the mortality risk of BT patients. A total of 3,061 adults participated in the study, and through ICD-9 and ICD-10, they were categorized into two groups. In order to assess the impact of potential confounders on outcome BT, This study selected some important variables as covariates. Then, identified through the proportional hazards (PH) assumption test, univariate Cox analysis, and machine learning. Subsequently, a nomogram was constructed and assessed. Results: In this study, the subjects included 303 survivors and 82 deceased. An analysis of the two groups of subjects found that there were significant differences in 7 covariates, such as hemoglobin, sodium, and prothrombin time (PT). Further laboratory - index tests showed that there were significant differences in 11 covariates between the outcome - variable groups, among which were bicarbonate and blood urea nitrogen (BUN) scores. The results of the independent prognostic tests indicated that race, Simplified Acute Physiology II (SAPS II), platelets, blood sodium, and PT could serve as independent prognostic indicators for BT patients. Finally, based on these five independent prognostic factors, a nomogram with good predictive ability was constructed. Conclusion: Based on five independent prognostic factors (race, Simplified Acute Physiology II (SAPS II), platelets, blood sodium, and PT), this study successfully constructed a nomogram with good predictive ability. It provided a valuable new perspective for understanding the model of predicting the in - hospital death risk of BT patients.