Causal Machine Learning Methods for Estimating Personalised Treatment Effects -- Insights on validity from two large trials