A Novel Joint Longitudinal Model for Predicting Post-ICU Anemia

Patients are known to be at an elevated risk of severe hemoglobin deficiency (anemia) following an Intensive Care Unit (ICU) admission. By using a patient's full hemoglobin history, we create a novel functional predictor of hemoglobin levels for the year following discharge. Because readmission to the hospital is associated with rapid changes in hemoglobin levels due to surgeries and blood transfusions, we incorporate a multistate joint longitudinal model so that patients can change between hospitalized and at home states throughout the period of interest. This model allows for the state of hemoglobin to influence readmission and discharge hazards, and, in a significant innovation in joint longitudinal modeling, readmission events impact a patient's hemoglobin trajectory via an explicit semiparametric model.