Development of Interactive Nomograms for Predicting Short-Term Survival in ICU Patients with Aplastic Anemia

Background:Aplastic anemia is a severe hematologic disorder marked by pancytopenia and bone marrow failure. ICU admission often reflects disease progression or complications requiring critical care. Predicting short-term survival in these patients is vital for individualized treatment and resource optimization. Nomograms provide a practical tool for integrating clinical parameters, offering accurate visualized survival predictions to guide decision-making in patients with aplastic anemia in the ICU. Methods:Using the MIMIC-IV database, we identified ICU patients diagnosed with aplastic anemia. From thousands of available variables, we extracted data across five dimensions: demographic, synthetic indicators, laboratory events, comorbidities, and drug usage. Based on existing studies of aplastic anemia, more than 400 variables were further refined and machine learning techniques were applied to identify the seven most effective predictors for modeling. Preprocessing was performed using machine learning approaches, and the feasibility of these predictors was validated through additional classification and regression models, the verification method is AUROC. Furthermore, external validation was performed using data from the eICU Collaborative Research Database to assess the generalizability of our models.The interactive nomograms were constructed using logistic regression (LR) to predict mortality rates at 7 days, 14 days, and 28 days in patients with aplastic anemia. Results: A total of 1,662 patients diagnosed with aplastic anemia were included in this study, with a 7:3 ratio split into training and testing cohorts. The logistic regression model demonstrated strong predictive performance, achieving AUC values of 0.8227, 0.8311, and 0.8298 for 7-day, 14-day, and 28-day mortality predictions, respectively. External validation using the eICU database further confirmed the model's generalizability, with AUC values of 0.7391, 0.7119, and 0.7093. These results highlight the model's stability and effectiveness in predicting short-term survival in aplastic anemia patients. Conclusion:A set of seven predictors, led by APS III, proved effective for modeling short-term survival in aplastic anemia patients. Using these predictors, Cox and logistic regression models generated nomograms that accurately predict 7-day, 14-day, and 28-day mortality. These tools can support clinicians in personalized risk assessment and decision-making.