Table 2. Association of the Opioid Hospital Quality Improvement Program (O-HQIP) With Patient Treatment With Buprenorphine Within 30 Days of Index Emergency Department Visit, Full Sample (17 428 Observations)a.
| Model | β (95% CI) | ||
|---|---|---|---|
| Overall before to after, O-HQIP vs no O-HQIP implementation | Year-specific indicators for post–O-HQIP period | ||
| 2019 | 2020 | ||
| A: Main analysis | |||
| A1: Unadjusted models | |||
| Buprenorphine within 30 d | 0.025 (0.005 to 0.046)b | 0.016 (−0.013 to 0.045) | 0.038 (0.015 to 0.060)b |
| A2: Adjusted models | |||
| Buprenorphine within 30 d | 0.026 (0.005 to 0.047)b | 0.016 (−0.012 to 0.045) | 0.037 (0.015 to 0.059)b |
| B: Analysis with OLAN as an additional covariate | |||
| B1: Unadjusted models | |||
| Buprenorphine within 30 d | 0.025 (0.004 to 0.047)b | 0.016 (−0.012 to 0.044) | 0.035 (0.008 to 0.063)b |
| B2: Adjusted models | |||
| Buprenorphine within 30 d | 0.025 (0.003 to 0.046)b | 0.015 (−0.014 to 0.043) | 0.036 (0.008 to 0.063)b |
| C: Analysis with prior buprenorphine treatment as an additional covariate | |||
| C1: Unadjusted models | |||
| Buprenorphine within 30 d | 0.025 (0.005 to 0.045)b | 0.015 (−0.011 to 0.044) | 0.037 (0.015 to 0.060)b |
| C2: Adjusted models | |||
| Buprenorphine within 30 d | 0.026 (0.005 to 0.046)b | 0.016 (−0.013 to 0.044) | 0.038 (0.016 to 0.061)b |
Abbreviation: OLAN, Opioid Learning Action Network.
In the model that evaluates the overall association, the coefficient of interest (O-HQIP attesting) is an indicator for whether a patient’s index opioid use disorder emergency department visit occurred during or after January 2019 (the start date for the O-HQIP) at O-HQIP–attesting hospitals. In the model that evaluates the year-specific effects, the 2 coefficients of interest are 2 indicator variables: 1 for patients visiting attesting hospitals in 2019 and 1 for 2020. All analyses used linear regression at the patient level and included an indicator variable for whether the patient was treated at an O-HQIP–attesting hospital after January 1, 2019. All models also included hospital-level fixed effects to control for time-invariant hospital unobserved effects. The sample sizes of the overall and year-specific models are equal (ie, 17 428). Clustering was performed at the hospital level.
Statistically different from 0 at the 5% level. Adjusted regression models also controlled for patient characteristics, including age (at the time of index opioid use disorder visit), sex, race and ethnicity, eligibility category, rural residence, managed care organization region, number of inpatient episodes in prior 6 months, number of outpatient episodes in prior 6 months, number of opioids fills in prior 6 months, number of benzodiazepine fills in prior 6 months, number of unique prescribers of opioids in prior 6 months, hepatitis C virus, HIV, anxiety disorder, mood disorder, schizophrenia and other psychosis, posttraumatic stress disorder, abscess, osteomyelitis, endocarditis, and soft skin tissue infection.