The Joint Model of the Logistic Model and Linear Random Effect Model - An Application to Predict Orthostatic Hypertension for Subacute Stroke Patients

Abstract

Stroke is a common acute neurologic and disabling disease. Orthostatic hypertension (OH) is one of the catastrophic cardiovascular conditions. If a stroke patient has OH, he/she has higher chance to fall or syncope during the following courses of treatment. This can result in possible bone fracture and the burden of medical cost therefore increases. How to early diagnose OH is clinically important. However, there is no obvious time-saving method for clinical evaluation except to check the postural blood pressure.This paper uses clinical data to identify potential clinical factors that are associated with OH. The data include repeatedly observed blood pressure, and the patient’s basic characteristics and clinical symptoms. A traditional logistic regression is not appropriate for such data. The paper modifies the two-stage model proposed by Tsiatis et al. (1995) and the joint model proposed by Wulfsohn and Tsiatis (1997) to take into account of a sequence of repeated measures to predict OH. The large sample properties of estimators of modified models are derived. Monte Carlo simulations are performed to evaluate the accuracy of these estimators. A case study is presented.