To early and surely detect a change generated in a patient's electrocardiogram from a feature quantity repeatedly presented in the electrocardiogram such as change of an interval between R and R.

A normal distribution calculator 3 inputs measured electrocardiogram data during a period INT1 and calculates the feature quantity repeatedly presented in its electrocardiogram as time series data, and generates a convergence C of differences between adjacent elements of the time series data. Further, the calculator 3 repeats m times a process for randomly extracting n pieces of the elements from the convergence C to obtain m pieces of partial convergences, and calculates a mean value μ and a standard deviation σ of its normal distribution with the convergence of mean values of the m pieces of the partial convergences to be regarded as being the normal distribution and decides the normal distribution as the normal distribution representing the normal state of the feature quantity change. A state detector 4 obtains a convergence CT for diagnosis similarly to the convergence C from the electrocardiogram data for a shorter period than the period INT1. A diagnosing unit 5 compares a difference between the mean value of the convergence C and the mean value of the normal distribution with the standard deviation σ, and diagnoses.