Although therapeutic plasma exchange (TPE) can be associated with bleeding, there are currently no known strategies to reliably predict bleeding risk. This study developed a TPE bleeding risk prediction model for hospitalized patients. To develop the prediction model, we undertook a secondary analysis of public use files from the Recipient Epidemiology and Donor Evaluation Study-III. First, we used a literature review to identify potential predictors. Second, we used Multiple Imputation by Chained Equations to impute variables with <
30% missing data. Third, we performed a 10-fold Cross-Validated Least Absolute Shrinkage and Selection Operator to optimize variable selection. Finally, we fitted a logistic regression model. The model identified 10 unique predictors and seven interactions. Among those with the highest odds ratios (OR) were the following: >
10 TPE procedures and antiplatelet agents (OR 3.26)
nephrogenic systemic sclerosis (OR 3.15)
and intensive care unit stay (OR 3.08). Among those with the lowest OR were the following: albumin-only TPE (OR 0.50)
male sex (OR 0.82)
and heart failure (OR 0.85). The model indicated an acceptable performance with a C-statistic of 0.71 (95% CI 0.699-0.717). A model to predict bleeding risk among hospitalized patients undergoing TPE identified key predictors and interactions. Although the model achieved acceptable performance, future studies are needed to validate and operationalize it.