Background: The Swedish Lumbar Spine Register has been collecting patient-based data since 2000, and more than 80% of all spinal units in Sweden are now including their patients. In a few years, it will produce useful clinical information just as arthroplasty registers have, but to permit proper interpretation of data in the future, the reliability of the protocol must be tested.
Methods: Between January 2000 and March 2003, a sample of 122 patients was asked to fill in the questionnaire twice: 63 preoperatively and 59 postoperatively. Test-retest reliability was calculated with intra-class correlation coefficient (ICC) or weighted kappa when appropriate.
Results: Test-retest interval varied (range 0-235 days); in the "worst case scenario", the lowest ICC for SF-36 was 0.62 for the postoperative RE. Other values were above 0.70; for non-SF variables, ICC was in the range 0.79-0.89. Kappa values for the ordinal outcomes were high (0.74-0.91).
Interpretation: When separate reliability analysis was performed according to the time interval, a 0-2 days interval produced a significant memory effect; after 3 weeks, the reliability seemed to drop in the preoperative group, whereas results were reproducible up to 9 weeks postoperatively. The protocol studied can reliably detect postoperative improvements between large groups of patients such as in a register.