Article type
Year
Abstract
Background: Empirical evidence exists that inadequate concealment of treatment allocation and inadequate double-blinding are associated with bias of results of randomized trials. There is no evidence for other methodology criteria used by the Cochrane Back Group (CBG). Furthermore, this evidence is collected in fields other than back pain and it is unclear if it is also valid for back pain studies.
Objectives: To assess the effect of quality items and total quality scores on effect sizes of back pain trials.
Methods: We used randomized trials from Cochrane back pain reviews comparing an intervention to a placebo/usual care/control group. These spanned a broad range of interventions. Effect sizes (ES) were calculated for one outcome per study, using a hierarchy: 1. Short-term pain, 2. Short-term function, 3. Long-term pain, 4. Long-term function. As a measure of bias, we used the ratio of ES. Here, the pooled ES for studies that scored positively on the item were divided by the pooled ES for studies that scored negatively on the item. A ratio of less than one means that studies scoring positively on the item have a smaller ES. We used a random-effects model to pool study results. As a total quality score, we used a simple sum of the positive quality items, and compared the pooled ES of studies above and below a variety of thresholds. We used the bootstrap method to determine confidence intervals.
Results: There were 108 studies. The preliminary results are presented in the Table.
[table excluded]
Conclusions: These preliminary analyses provide evidence that most individual Cochrane Back Group items are associated with bias and a sum quality score threshold of four is associated with statistically significant bias. Additional planned analyses to test the robustness of these findings include stratification by type of intervention and tests for co-linearity among quality items.
Objectives: To assess the effect of quality items and total quality scores on effect sizes of back pain trials.
Methods: We used randomized trials from Cochrane back pain reviews comparing an intervention to a placebo/usual care/control group. These spanned a broad range of interventions. Effect sizes (ES) were calculated for one outcome per study, using a hierarchy: 1. Short-term pain, 2. Short-term function, 3. Long-term pain, 4. Long-term function. As a measure of bias, we used the ratio of ES. Here, the pooled ES for studies that scored positively on the item were divided by the pooled ES for studies that scored negatively on the item. A ratio of less than one means that studies scoring positively on the item have a smaller ES. We used a random-effects model to pool study results. As a total quality score, we used a simple sum of the positive quality items, and compared the pooled ES of studies above and below a variety of thresholds. We used the bootstrap method to determine confidence intervals.
Results: There were 108 studies. The preliminary results are presented in the Table.
[table excluded]
Conclusions: These preliminary analyses provide evidence that most individual Cochrane Back Group items are associated with bias and a sum quality score threshold of four is associated with statistically significant bias. Additional planned analyses to test the robustness of these findings include stratification by type of intervention and tests for co-linearity among quality items.