The over-reliance on p-values can lead to misinterpretation of data and a $150 million bet on a subgroup with scant evidence.