Predictions of AI-powered models strictly trial-specific, have no generalisability: Study