Adjusting contacts with observed infections: consequences on predictions about vaccine effectiveness