ScienceInsider has the story here. Money quote:
One major problem, the report says, is “overfitting”: Because the studies often look for patterns in hundreds of biomolecules using a relatively small number of patient samples, it is easy to find correlations that do not reflect the biology of patients’ disease. The report recommends a set of steps to validate the tests, such as repeating the test on blinded samples from a different institution. Journals and funders should also require that data and models from papers be made freely available so that other researchers can check the results.
Discoverers and Discoveries 