The Use of Survival Analysis Modelling with Incomplete Data with Application to Breast Cancer
Mahdi Saber Raza
Department of Software and Informatics Engineering, College of Engineering, University of Salahaddin, Erbil, Iraq.
Mark Broom *
Department of Mathematics, City University London, Northampton Square, London EC1V 0HB, UK.
*Author to whom correspondence should be addressed.
Abstract
There are strong survival analysis methodologies for data sets which are complete, with accurate information on censoring. But what if they are not complete? In an earlier paper we built a methodology for estimating survival probabilities and hazard functions in a health setting, using breast cancer data from the Kurdistan region of Iraq, for censored and uncensored data when a substantial portion of individuals are lost to the study. In this paper we build on these models to consider further issues based upon the accuracy of the records of patient death, where deaths often occur beyond the hospital in family settings and patients ceasing treatment and contact with the hospital may or may not represent their death; thus the record of their time of death may not be accurate. We develop a new Markov chain-based methodology for generating survival curves and hazard functions, and demonstrate this using a different breast cancer dataset from the Kurdistan region of Iraq.
Keywords: Survival analysis, Markov model, breast cancer, censoring data