The framing of time-dependent machine learning models improves risk estimation among young individuals with acute coronary syndromes