On the use of Harrell's C for clinical risk prediction via random survival forests