Applying regression discontinuity in epidemiology; a study on when to start HIV treatment