Data sieving (L1) represents an inductive approach for the exhaustive information extraction from large multi-parametric data spaces in view of predictive or diagnostic goals. Hypothesis driven data collection (fig.1a) is followed by data sieving (fig.1b) and interpretation (fig.1c) of the resulting predictive data patterns for individualized classification of unkown patients
The data pattern may serve in repetitive rounds as starting point for new multi-parametric experiments followed by data sieving and interpretation of the resulting new data pattern.
Large multi-parametric data spaces can be investigated in this way in a comparatively short time. Besides the predictive data pattern the vast majority of excluded non informative parameters are equally of interest because they are positively excluded from further hypothesis or concept development. This limits substantially the amount of thinking possibilities which may otherwise very quickly become an innovation limiting problem.

The ultimately desired high statistical significance of results for clinical applications is initially in conflict with the search for individually predictive parameter patterns through the collection of large amounts of multi-parametric information from flow cytometry of heterogeneous cellular suspensions, bead arrays or DNA and protein expression arrays. A two phase strategy (L2) is therefore appropriate (fig.2). The initial pilot phase study (fig.2 phase 1) is performed at an acceptable minimum of statistical stringency such as a significance level of P<0.05 or P<0.10. The majority of uninformative parameters can be eliminated at this stage by data sieving.