Table 2. Effect of the intervention on imaging referral, with and without discounting.
| Variable | No. practices (no. GPs) | Mean (SD) | Adj IRR1,2 (95%CI) | Incremental Effect3 (SE)4 | ||
| Rx | Control | Rx | Control | |||
| X-ray | 34 (44) | 37 (40) | 14.6 (12.1) | 19.2 (14.6) | 0.83 (0.61, 1.12) | −3.43 (3.10) |
| X-ray @ 5%discount | 34 (44) | 37 (40) | 13.9 (11.5) | 18.3 (13.9) | 0.82 (0.61, 1.11) | −3.27 (2.95) |
| X-ray @ 3%discount | 34 (44) | 37 (40) | 14.2 (11.7) | 18.6 (14.1) | 0.83 (0.61, 1.11) | −3.33 (3.01) |
| X-ray @ 7%discount | 34 (44) | 37 (40) | 13.7 (11.3) | 17.9 (13.6) | 0.82 (0.61, 1.11) | −3.21 (2.89) |
Adjusted rate ratios estimated from models fitted using xtgee, family(nbinomial “estimated heterogeneity parameter”) link(log) vce(robust) exposure(total Medicare patients) where intercept derived from nbreg. All models adjusted for the following design strata and pre-specified confounders: GP age (years), years since GP graduated, self-reported special interest in LBP, number of GPs per practice, practice method of billing, rural/metro practice.
IRR = incidence rate ratio. Estimate of intervention effect adjusted for design strata and potential confounders (specified prior to undertaking the analysis).
Incremental effect = change in referral per GP due to exposure to the intervention after controlling for design strata and pre-specified potential confounders. Here, incremental effects derived from model predicted values using method of recycled predictions [24].
Standard errors derived from bootstrap using bsmultiv.do [24].