What can the randomness of missing values tell you about clinical practice in large data sets of children’s vital signs?