Twelve RCTs (n=600) were included in the meta-analysis. All 12 reported blood glucose concentration as an outcome, and 11 reported the percentage of glycated haemoglobin.
Blood glucose control.
Glycaemic control was better during pump treatment. The standardised mean difference in blood glucose concentrations between insulin pump and optimised insulin injection therapy was 0.56 (95% confidence interval, CI: 0.35, 0.77). The estimate from the fixed-effect model was similar (0.53, 95% CI: 0.36, 0.70). The treatment effect in terms of absolute units was 1.06 mmol/L (0.88, 95% CI: 0.52, 1.24 mmol/L with unstandardised data).
The results of the two tests showed no significant heterogeneity among trials (P=0.17). There was no clear publication bias in a funnel plot, and the result of Egger's test was not significant (P=0.168). The trim and fill method gave an estimated corrected effect size of 0.39 (95% CI: 0.15, 0.63). Only duration of treatment was related to effect size in a regression analysis (regression coefficient 0.32, 95% CI: 0.06, 0.58). This model estimated the effect size as 0.46 (95% CI: 0.14, 0.77) at 6 months of treatment and 0.93 (95% CI: 0.30, 1.57) at 2 years.
Glycated haemoglobin.
The percentage of glycated haemoglobin was lower during pump therapy, the standardised mean difference being 0.44 (95% CI: 0.20, 0.69). This was equivalent to an effect size of 0.51% in original units, which was consistent with that seen in a meta-analysis with unstandardised data (0.45%, 95% CI: 0.20, 0.71). The fixed-effect model gave a similar standardised mean difference to the random-effects model (0.41, 95% CI: 0.23, 0.58). There was some evidence of heterogeneity (chi-squared P=0.07), and a funnel plot and Egger's test (P=0.02) revealed some possible publication bias. The trim and fill method gave an estimated effect size corrected for bias of 0.31 (95% CI: 0.15, 0.48). The regression analysis showed that none of the measured variables were significantly related to effect size.
Insulin dose.
The improved control during insulin pump therapy was achieved at a reduced total daily insulin dosage. The standardised mean difference in insulin dose was 0.58 (95% CI: 0.34, 0.83). This represents a mean dosage reduction of 14% during pump therapy. The effect size was 7.58 units/day in original units, which was similar to that seen in a meta-analysis with unstandardised data (7.33, 95% CI: 4.07, 10.59 units/day). The estimate from the fixed-effect model was similar to that of the random-effects model (0.53, 95% CI: 0.36, 0.71).
Analysis of insulin dose showed some evidence of heterogeneity (P=0.07). The funnel plot showed some bias, though the result of Egger's test was not significant (P=0.17). The effect size corrected for bias was 0.42 (95% CI: 0.25, 0.58). In the regression analysis, the duration of treatment was negatively related to effect size (regression coefficient -0.41, 95% CI: -0.66, -0.15). The model estimated the effect size to be 0.66 (95% CI: 0.33, 0.10) at 6 months of treatment and 0.05 (95% CI: -0.59, 0.70) at 2 years.
Variability in blood glucose concentration.
Using the SD of blood glucose concentration as a measure of glycaemic variability, the variability was found to be significantly higher with insulin injections than with pump therapy (weighted geometric mean of the SD ratios 1.27, 95% CI: 1.11, 1.47).