A Kalman Filter Technique, Generalized from Renal Transplantation to Other Clinical Problems

Abstract

Patients with Kidney disorders can be monitored prospectively using a conputer program incorporating an adaptation of the multiprocess Kalman filter. This method was succesfully used to detect rejection in transplanted kidneys (British Medical Journal 1983. 286, 1695–1699). In order to develop the method for a range of other clinical monitoring, modifications and extensions to that methodology were needed. The method can now consider series with data provided at unequally spaced intervals, conmon in clinical work as there is often less frequent sampling when “risk” is considered to be low, and there are also “missing” values as a result of logistic or organisational problems. The method now also has the ability to consider other conventional time series models, other than the linear growth model previously reported, including quadratic, cubic growth and simple auto-regressive-moving average models.