Simpson's Paradox and the implications for medical trials