Table 3.
Ordinal regression model examining prescriber factors and willingness to terminate treatment for concerns of diversion (N=1,174)
| AOR (95% CI) | |
|---|---|
| Diversion-related practices | |
| Assesses all patients for diversion (vs. assesses <100% of patients) | 1.229 (0.917 1.647) |
| Frequency of office visits in early treatment | 0.920 (0.797, 1.062) |
| Always tests urine screen for buprenorphine (vs. does not always test) | 1.143 (0.877, 1.490) |
| Percentage of patients with medication counts | 1.006*** (1.003, 1.009) |
| Magnitude of the diversion problem in the community | 1.212** (1.073, 1.369) |
| Buprenorphine practice characteristics | |
| Delivers buprenorphine treatment in individual medical practice (vs. not individual practice) | 1.067 (0.784, 1.452) |
| Payment typology | |
| Only accepts cash | Reference |
| Accepts private insurance but not Medicaid | 0.934 (0.693, 1.261) |
| Accepts Medicaid | 0.670*+ (0.474, 0.947) |
| All other types of payment | 0.723 (0.406, 1.286) |
| Current number of buprenorphine patients | 0.999 (0.997, 1.001) |
| Years of prescribing buprenorphine | 0.980 (0.940, 1.021) |
| Waivered to treat up to 100 patients (vs. 30 patients) | 0.933 (0.712, 1.223) |
| Prescriber characteristics | |
| Medical specialty | |
| Addiction (psychiatry or medicine) | 0.526*** (0.406, 0.682) |
| Psychiatry | 0.714** (0.558, 0.914) |
| All other specialties | Reference |
| Member of ASAM/AAAP (vs member of neither) | 0.859 (0.670, 1.100) |
| Age | 1.007 (0.995, 1.018) |
| Female (vs. male) | 1.199 (0.942, 1.527) |
| Non-white (vs. white) | 1.280 (0.963, 1.702) |
| Threshold 1 | −3.700 (−4.499, −2.900) |
| Threshold 2 | −2.115 (−2.806, −1.425) |
| Threshold 3 | −1.209 (−1.870, −0.549) |
| Threshold 4 | 0.259 (−0.451, 0.970) |
Notes.
p<.05,
p<.01,
p<.001 (two-tailed tests).
p-value exceeds the false discovery rate (FDR) value for statistical significance when adjusting for multiple comparisons.
AOR = adjusted proportional odds ratio. CI = confidence interval. Responses to this dependent variable were 1=strongly disagree, 2=disagree, 3=neither agree nor disagree, 4=agree, and 5=strongly agree. The model was estimated using the logit version of the ordinal regression model with robust standard errors to adjust for the clustering of prescribers within states. Thresholds represent estimated cutpoints on the underlying latent variable of willingness to terminate patients. Threshold 1 is the estimated cutpoint used to differentiate “strongly disagree” responses from all other categories when values of the independent variables are zero. Threshold 2 is the estimated cutpoint that differentiates the groups “strongly disagree” and “disagree” from the remaining three groups when the independent variables are set at zero. Similar logic applies to Threshold 3 (“strongly disagree, “disagree” and “neither” vs. “agree” and “strongly agree) and Threshold 4 (“strongly disagree, “disagree,” “neither,” and “agree vs. “strongly agree). Pooled estimates were calculated from models estimated using 20 imputed datasets (n=1,174).