Estimation of transition rates in a multi-state proportional hazards model

An algorithm using the Interactive Scientific Processor is proposed for est imating parameters of the multi-state model for reverse and repeated transi tions. The algorithm was applied to longitudinal data on diabetes mellitus collected at a diabetes hospital in Bangladesh. On the basis of blood gluco se level, three distinct types of transitions from one state to another wer e detected. These are Transition, Reverse Transition, and Repeated Transiti on. Four variables are included in the models for the transitions: Sex, Bod y Mass Index (BMI), Age, and Area. For transition from state 2 to state 1, only sex has a significant association, indicating a higher rate of transit ion for male patients than that of female patients (p-value < 0.05). The va riable BMI is significantly associated (p-value < 0.05) with the blood gluc ose level, implying that a lower BMI accelerates transition from a lower le vel of blood glucose to a higher level (p-value < 0.05). (C) 1999 Elsevier Science Ltd.