Abstract: In order to analyze a medical sample while avoiding error contributions due to hemolysis, prior to the main reaction for a component present in the sample, a pre-reaction is produced and measured by which the degree of hemolysis in the sample is determined and the result of the sample to be determined that is subsequently obtained is corrected by this error contribution using the relationship (correlation) that has been found between the degree of hemolysis and the error contribution.

Abstract: A method for the analysis of a component of a medical sample with the aid of an autoanalyzer, in which the sample is reacted with a reagent system and a physically measurable quantity X resulting from the reaction of the sample with the reagent system is measured in order to determine at least one measured value R for a specified sample and in which this at least one measured value R is further processed with the aid of a processing unit in a processing stage in order to determine an analytical result A. In the processing stage of the invention, use is made of the results of a neural net training, in which, for a number of standard samples for which the analytical result A is known, at least one measured value R, or a measurement result derived from several measured values R.sub.i, is applied to the input of a neural net.

Abstract: Method for the analytical determination of the concentration of a component of a medical sample, in which a reaction of the sample with reagents leads to a time-dependent change S(t) in a measured quantity S and the concentration C correlates according to an evaluation curve C(X) with an input variable X derived from S(t), in which the calibration curve is ambiguous for at least a portion of the possible X values. In order to assign an input variable X to one of the sub-sections and thereby to obtain an unambiguous correlation to a particular concentration C, a training run and an analysis run are performed. In the training run, a discrimination algorithm is performed at least once, in which a discriminator set is generated from measurements of S(t), a score is generated in each case from the latter with a multivariate statistical technique and it is checked whether the scores can be divided into separate subsets, in which the concentrations are correctly assigned to the sub-sections of the calibration curve.