Heterogeneity in systematic reviews on spinal surgery: A meta-epidemiological study

Tags: Poster
Jacobs W1, Kruyt M2, Verbout A2, Oner C2
1Leiden University Medical Center, The Nethrlands, 2University Medical Center Utrecht, Netherlands

Background: Methodological design characteristics of trials in meta-analyses have been shown to influence pooled effect sizes and are therefore recognised as potential sources of heterogeneity in several medical fields. However, for spinal surgery, the presence and the strength of the effect of effect modifiers have not been assessed.

Objectives: The goal of this study was to identify the influence of design characteristics on effect size in studies on effectiveness of spinal surgery.

Methods: Searches were performed in MEDLINE, CDSR and EMBASE. Methodological quality of the included meta-analyses was assessed with the Amstar tool. The effect sizes of trials included in the meta-analyses were assessed. The differences in effect sizes were calculated as risk differences (RD). Relation of RDs to sponsoring, randomization, allocation concealment, blinding and study size was assessed with meta-regression adjusted for multiple testing.

Results: Seven reviews consisting of 76 studies and 8191 patients were included. Data provided by the systematic reviews alone was insufficient to analyse heterogeneity. Meta-regression analysis of the individual trials though, showed that sample size, strict randomization and outcome blinding were significant factors influencing study effect. Validly randomised studies showed an increase in RD of 5.4% (95% CI 1.2 to 9.6; p = 0.044) compared to not validly randomised and observational trials. Studies with adequate observer blinding showed a 7.2% decrease in RD (95% CI 0.8 to 13.7; p = 0.049). For each increase of 100 patients, the RD decreased 3.6% (95% CI 0.4 to 6.8; p = 0.098).

Conclusions: Contrary to basic methodological assertions, formal and strict randomization appeared to increase the risk difference significantly in spinal surgery research. Sufficient sample size and observer blinding, on the other hand, can decrease this risk difference. This may imply that by trying to limit bias, strict randomization may actually cause bias.