There are several important limitations that should be mentioned. First, administrative claims rely on diagnosis codes, rather than structured evaluations, to identify patients with bipolar disorder, which may result in disease misclassification. In an attempt to minimize bias from misclassification, the sample was restricted by including only children with more than one bipolar diagnosis and/or an inpatient stay, which has been shown to increase the specificity of the diagnosis.(39) Related to this, a large proportion of our sample received bipolar disorder not otherwise specified diagnoses (bipolar NOS). This diagnosis may represent diagnostic uncertainty on the part of the clinician. It is unclear whether children with these diagnoses would convert to a bipolar-I or bipolar-II diagnosis, but evidence suggest that conversion from bipolar NOS to a more specific diagnosis occurs frequently among youth.(40) Importantly, inclusion of children with less severe behavioral diagnoses would likely result in conservative estimates of treatment costs and utilization for the sample. Second, it is unclear whether diagnoses made in childhood would persist as the child ages into adulthood, although some suggest that bipolar I and bipolar
