[Criteria for biostatistical assessment in surgery of stomach cancer].

The problem of making valid statements about the prognosis of survival after radical operation of stomach cancer, based on retrospective or prospective studies, is investigated. Some explanations are given for contradictory or non-reproducible results in different subgroups. Especially the problem of searching for "significant" results by multiple testing of different subgroups (data snooping) is treated critically. In case of remaining open questions randomised studies appear to be inevitable.