Reliability of dose volume constraint inference from clinical data