3087 — Comparison of Alternative Methods for Identifying DSS Pharmacy Prescription Records
Schutte KK (Palo Alto COE), Hu KU
(Palo Alto COE), Schmitt SK
(HERC), Phibbs CS
Decision Support System (DSS) Pharmacy Datasets contain multiple variables per record that identify prescribed medications. Available documentation is unclear about which variable(s) should be used. We approached this ambiguity by empirically evaluating the amount and nature of disagreement among three key variables: The National Drug Code (NDC) and Internal Entry Number (IEN), both of which are embedded within the DSS “feeder-key”, and the Intermediate Product Number (IPNUM). We focused on AntiDepressant medication prescription records (ADRx) among patients with bipolar disorder. Identification of ADRxs among this cohort is critical to larger project goals: evaluating if ADRx portends worse outcomes among individuals with bipolar disorder.
Our national sample comprised 268,774 patients who received at least one bipolar disorder diagnosis in FY2002 to FY2009. We selected all FY2008 pharmacy records containing any NDC, IEN, or IPNUM evidence of ADRx and determined the number of records with (in)consistent ADRx information. Next, we compared four alternative methods for interpreting pharmacy record information to determine if use of different methods would yield different conclusions about whether and for how many days ADRx was prescribed: (1) IEN-Only; (2) NDC-Only; (3) IEN+NDC = use IEN to identify ADRx records except when IEN fails to identify a specific medication; then rely on NDC; and (4) two-variable method = 2/3 variables agree there is a ADRx; alternatively, one variable indicates ADRx while the other two identify no specific medication.
Of 1,223,074 ADRx records, 1,214,525 (99.3%) contained NDC, IEN, and IPNUM information in full agreement about the presence of an ADRx and about the specific antidepressant medication prescribed. Methods utilizing multiple pieces of DSS pharmacy information yielded modestly higher estimates of whether, and for how many days, an ADRx was prescribed than did those relying on IEN- or NDC-Only. Pharmacy records with nonspecific NDC, IEN, and IPNUM information were clustered by facility.
Agreement among three key pharmacy record variables was excellent, but use of multiple-variable methods to identify ADRxs yielded more complete information.
When ADRx identification is critical to study goals, multiple-variable methods should be used to identify ADRxs. Site-specific studies should consider how facility-level clustering of nonspecific pharmacy information influences ADRx assessment.