DD and PSM estimates from unadjusted and adjusted models with RSD and the control regions
| Lumbar MRI rates | ||||
| DD | PSM† | |||
| Unadjusted | Adjusted model‡ | Quarterly data§ | ||
| DD (RSD* post-treatment years from 2010 to 2013) | 1.83* (1.37 to 2.29) | 1.39* (0.93 to 1.85) | 0.43* (0.32 to 0.53) | 3.80* (2.67 to 4.94) |
| Constant | 9.04* (8.85 to 9.23) | 2.60 (−5.28 to 10.48) | 2.24* | – |
Four models showing the main outcome lumbar MRIs per 1000 enlisted with a GP.
†The PSM model use conventional options; nearest neighbour with (calliper = one-fourth of the SD on the propensity scores). We use the MRI rates in 2008 and 2009 as matching covariates to control for unobservable selection.31 32 Furthermore, we control for clustering at GP level.34
‡Adjusted model include covariates for comorbidity score 2+, full time job, vocational education, marital status single, income DKK 0–399.999, male, 18–59 years of age, visit at physiotherapist, visits at chiropractor, visits at office-based spine specialist, use of pain medication.
§The DD model with quarterly data use the same covariates. However, in addition, we included quarter dummies (Q1, Q2 Q3 and Q4) to take away the obvious seasonality shown in the online supplementary figure A1.
*P<0.001.
DD, difference in difference; GP, general practitioner; PSM, propensity score matching; RSD, Region of Southern Denmark.