A method for early detection of abnormal trends in serial clinical examination results over time

In this paper we describe a method for the early detection of changing health conditions which may eventually lead to serious disease. The method makes use of health screening test results accumulated for individuals and can be applied to those who have taken at least four health examinations. The degree of abnormality is calculated using a logistic regression equation. The variables defining the equation are selected by factor analysis and a stepwise variable selection method based on the likelihood ratio criterion. Statistical estimates include linear regression coefficients, isotonic regression probabilities and their standard deviations. These are used to represent trends in health screening results over time. The method is illustrated using a sample of 308 persons with gastric cancer and 3002 healthy persons. Cross-validations were also performed.