Attributing medical spending to conditions: A comparison of methods