Predictions of diseases require suitable laboratory tests, mathematical methods for analyzing data and true normal ranges. Normal ranges are to be defined as 95% range of test results obtained on healthy subjects. "Healthy" subjects here refers to the persons who are free of the particular disease now and previously. However, to confirm the disease-free condition, laboratory test result must be evaluated on the basis of a normal range. This is a dilemma. The only way to overcome this problem is to follow a cohort for 10 or 20 years. Some of the "healthy" subjects will be suffering from diseases, while the others will still remain free of disease. That the latter persons would be the truly "healthy" subjects. Therefore, by their laboratory data and/or specimens such as serum every year we can determine the differences between the laboratory data of the patients and the true healthy subjects. In this respect, the development of laboratory database systems and serum banks are essential. We report a mathematical method for selecting significant data items for such a database and decision-making.