[Calculation of normal range using a nonlinear least squares method].

We developed a method to calculate the normal range of laboratory tests using nonlinear least squares method. Our method decomposes a histogram of laboratory test into constituent distributions assuming these constituents to be gaussian distributions. The proposed method has been applied to the histograms of serum total protein and serum lactate dehydrogenase, and good fittings were obtained assuming the histograms consist of 2 and 3 constituents respectively. Accuracy and precision were calculated for the stimulated data of known distribution pattern with contamination of abnormal values to examine the robustness of our method. With the contamination rate of 10%, accuracy less than 1% and precision less than 3% have been shown for the sample size more than 2,000. And the estimation was not markedly affected by the contamination rate up to 50%. Based on the parameters estimated by our method, we proposed a new measure for laboratory tests named abnormality index, which gives a probability to be abnormal.