BACKGROUND: Accurate identification of study populations and outcomes is critical to draw appropriate inferences in observational studies using health insurance claims databases. Traditional validation methods that rely on review of paper medical charts are costly and time-intensive. Curated patient-level chronological listings of claims data (“patient profiles”) may represent a more efficient alternative.
OBJECTIVES: To illustrate potential applications of chronological claims profile review using 3 study examples: (1) the development of an algorithm to identify a study population of patients with migraine; (2) rapid review of potential anaphylaxis and hemorrhagic stroke events among influenza vaccine recipients; and (3) refinement of a previously published stroke algorithm.
METHODS: The study examples were sourced from a large US health insurance claims database. De-identified claims profiles were created using diagnoses, services, procedures, and medications that generated insurance claims within a pre-specified time range surrounding the diagnoses of interest. In the first example, patients with migraine were identified based on the presence of ≥1 of 5 criteria using different combinations of diagnosis codes for migraine associated with inpatient, outpatient physician, emergency room, or neurologist visits, and/or claims for acute migraine treatment. Profile review was conducted on a random sample of 50 patients, 10 meeting each criterion. In the next 2 examples, profiles were created and reviewed for all events meeting a claims-based definition.
RESULTS: Example 1: 6/10 profiles meeting a criterion of ≥1 inpatient migraine diagnosis reflected patterns of medical care where the migraine diagnosis claim appeared to represent a more non-specific migraine variant or was part of an inpatient evaluation for conditions such as transient ischemic attack, where migraine might be one of the rule-out diagnoses. The algorithm was subsequently refined to exclude such patients. Example 2: 9/11 anaphylaxis and 11/12 hemorrhagic stroke events were classified as having potential alternate contributing factors or as likely miscodes. Example 3: 20/32 claims-identified stroke events were confirmed through profile review. Of these 20 confirmed cases, 19 were also confirmed by clinical adjudicators with access to the patient medical record.
CONCLUSIONS: These examples illustrate the use of chronological claims profile review as an efficient tool to develop and refine study population and outcome definitions.
