Changing diagnosis coding routines may confound the results of longitudinal childhood pneumonia studies