A novel model for predicting hospitalization risk among hemodialysis patients based on blood test variables

Hemodialysis patients are at high risk of hospitalization. Predicting such risk in dialysis patients may be critical to maintaining quality of life and reducing costs to the healthcare system. In this paper, we present and fractional polynomial stepwise logistic regression model to specify how routinely collected blood test variables could be linked to a significant increase in hospitalization risk. We found that eight of nineteen variables were significantly able to predict hospitalization risk; albumin (p<0.05), creatinine (p<0.05), calcium (p<0.01), bicarbonate (p<0.01), hemoglobin (p<0.05), mean cell hemoglobin concentration (MCHC) (p<0.0001), mean corpuscular volume (MCV) (p<0.0001), and potassium (p<0.01). The model achieved accuracy, sensitivity, and specificity of 77.31%, 83.03%, and 69.05%, respectively.