Evidence of “Repeated Admission Bias” Among Those Who Use Injection Drugs Across 2 Decades of US Treatment Admissions: 2000–2020

Abstract

Objectives Encounter-based datasets like the Treatment Episode Dataset—Admissions (TEDS-A) are used for substance use–related research. Although TEDS-A reports the number of previous treatment admissions, a limitation is this reflects encounters, not people. We sought to quantify the methodologic bias incorporated by using all encounters versus initial encounters and assess if this risk is evenly distributed across all routes of drug administration. Methods TEDS-A 2000–2020 dataset with nonmissing primary substance data was used. Of the data, 3.17% were missing the usual administration route, and 11.9% were missing prior admission data. Prior admissions are documented as 0 through 4, then binned for 5 or greater (5+). Risk of admission bias was defined as odds ratio (ORRAB): odds of total admissions relative to the odds of the first admission. Bootstrap confidence intervals were generated (5000 iterations) across administration routes and demographics; however, their widths were <0.0055 and not reported. Results There were 38,238,586 admissions over the 21 years, with 13,865,517 (41.2%) first admissions. Of all admissions, 15.7% indicated injection drug use (IDU); 26.3% of encounters reporting IDU were in the 5+ group. This resulted in an ORRAB of 1.77. White enrollees had an elevated ORRAB (1.05), whereas among Latinos, ORRAB was low (0.74). Conclusions Using encounter-based datasets can introduce bias when including all admissions versus exclusively initial treatment episodes. This report is the first to quantify this bias and shows that individuals with IDU are at highest risk for returning to treatment, thereby over-representing this method of use when all encounters are used.