Menu

Almost All Articles On Cancer Prognostic Markers Report Statistically Significant Results

Almost All Articles On Cancer Prognostic Markers Report Statistically Significant Results: Commentary from F1000

Kyzas PA, Denaxa-Kyza D, Ioannidis JP

Eur J Cancer 2007 Nov 43(17):2559-79

Commentary from Faculty Member Willi Sauerbrei


This is a landmark review illustrating the severity of reporting bias in prognostic marker research. The article impressively illustrates one of the most severe weaknesses of cancer prognostic marker research. Altogether more than 1900 articles were assessed, of which only 26 articles were fully 'negative'.

 


It shows that p-values are important criteria to select results for publication and discusses the biases from selective reporting transfer to meta-analyses. The results indicate a research culture that is probably driven by the pursuit of statistical significance. Without a culture change towards sound scientific principles, e.g. severe improvement in statistical analyses and reporting with adherence to reporting guidelines, prognostic marker research will never be able to produce scientifically valid information. This is probably an important reason that only a tiny fraction of the initially proposed 'significant' prognostic markers have found their way into clinical application. Here, empirical evidence is illustrated for cancer only but, most likely, problems are similar in the whole area of prognostic marker research. As soon as possible, the community has to derive rules for systematic research and assessment of a new marker from first 'Phase I' studies to a meta-analysis as a basis for an evidence-based medicine-based assessment after several years. Candidate markers without a 'sufficient' effect have to be abandoned. Searching for 'interesting results' in high-dimensional (-omics) data will dominate this type of research in the next years. However, analyzing these types of data adds many critical issues. Much closer collaboration between disciplines and study groups is required to improve research.






Reviewed by Ramaz Mitaishvili, MD
BLOG COMMENTS POWERED BY DISQUS
Amount: 

Color I Color II Color III

Log In or Register