Validation of a mortality composite score in the real-world setting: overcoming source specific disparities and biases