BACKGROUND: Health care databases are increasingly used for medication and device safety studies. Data sources with electronic health records (EHRs) have many advantages over administrative claims databases; among them, clinical notes in EHRs that provide additional insight into medical encounters, medical events, lifestyle factors, and health‐related behaviors. Methods to utilize these richer data sources must be developed and validated before use.
OBJECTIVES: To develop and validate automated algorithms using both structured and unstructured data from health care systems with EHRs to identify uterine perforations and intrauterine device (IUD) expulsions.
METHODS: Four sites in the United States participated: 3 Kaiser Permanente (KP) sites—Northern California (KPNC), Southern California (KPSC), Washington (KPWA)—and Regenstrief Institute (RI) in Indiana. The study population included all identified IUD insertions (n=325 582) among women aged ≤50 years (n = 282 028). The study period at each site was from the time the EHR was fully implemented at that site through September 30, 2015. Using structured (eg, International Classification of Diseases codes, Current Procedural Terminology codes, National Drug Codes) and unstructured (Natural Language Processing terms) data, site‐specific algorithms for uterine perforation and IUD expulsion were developed and validated by clinical experts via EHR review of a random sample of one‐third of the identified possible cases (up to 100). If sites further refined their algorithm, a second random sample of one‐third of the possible cases was selected for review. Positive predictive values (PPVs) of the algorithms were calculated.
RESULTS: The number of possible uterine perforations identified by algorithms at each site ranged from 67 to 444, and the number of possible IUD expulsions ranged from 268 to 4185. Two sites refined their algorithms and selected one or two additional possible‐case samples. PPVs for uterine perforation were KPNC, 77%; KPSC, 81%; KPWA, 82%; and RI, 47%. PPVs for IUD expulsion were KPNC, 77%; KPSC, 87%; KPWA, 68%; and RI, 37%.
CONCLUSIONS: These results suggest that a retrospective study using algorithms to identify the outcomes of uterine perforation and IUD expulsion can be successfully conducted in these sites with EHRs, although review of possible cases is recommended at the one site that relied solely on unstructured data and in certain subgroups at other sites, to reduce misclassification of outcomes
