A systematic review and meta‐regression analysis of prophylactic gabapentin for postoperative pain

Abstract

SummaryWe searched MEDLINE, Embase, CINAHL, AMED and CENTRAL databases until December 2014 and included 133 randomised controlled trials of peri-operative gabapentin vs placebo. Gabapentin reduced mean (95% CI) 24-h morphine-equivalent consumption by 8.44 (7.26-9.62) mg, p < 0.001, whereas more specific reductions in morphine equivalents were predicted (R 2 = 90%, p < 0.001) by the meta-regression equation: 3.73 + (À0.378 9 control morphine consumption (mg)) + (À0.0023 9 gabapentin dose (mg)) + (À1.917 9 anaesthetic type), where 'anaesthetic type' is '1' for general anaesthesia and '0' for spinal anaesthesia. The type of surgery was not independently associated with gabapentin effect. Gabapentin reduced postoperative pain scores on a 10-point scale at 1 h, 2 h, 6 h, 12 h and 24 h by a mean (95% CI) of: 1.68 (1.35-2.01); 1.21 (0.88-1.55); 1.28 (0.98-1.57); 1.12 (0.91-1.33); and 0.71 (0.56-0.87), respectively, p < 0.001 for all. The risk ratios (95% CI) for postoperative nausea, vomiting, pruritus and sedation with gabapentin were: 0.78 (0.69-0.87), 0.67 (0.59-0.76), 0.64 (0.51-0.80) and 1.18 (1.09-1.28), respectively, p < 0.001 for all. Gabapentin reduced pre-operative anxiety and increased patient satisfaction on a 10-point scale by a mean (95% CI) of 1.52 (0.78-2.26) points and 0.89 (0.22-1.57) points, p < 0.001 and p = 0.01, respectively. All the effects of gabapentin may have been overestimated by statistically significant small study effects.

Baseline Morphine Consumption May Explain Between-Study Heterogeneity in Meta-analyses of Adjuvant Analgesics and Improve Precision and Accuracy of Effect Estimates

“…Future reviews could also investigate whether heterogeneity in other outcomes such as pain scores or length of stay can be explained by baseline risk. Our previous work has indicated that this may indeed be the case 10 and other research has shown this may apply to other outcomes such as depression. 142 Such explanation may have clinical implications, as well as allow use of the methods presented in this paper to improve reporting of effect estimates.…”

“…137 In our previous meta-analysis with gabapentin, morphine consumption varied even within procedure-specific subgroups and type of surgery was a small determinant of heterogeneity between studies in relation to morphine consumption and pain scores. 10 Our results suggest that expected postoperative morphine consumption (as a surrogate for pain, participant characteristics and concurrent analgesia) is a large determinant of heterogeneity between studies.…”

“…We identified these studies from 8 narrative reviews , 34-41 25 systematic reviews and 72 meta-analyses [10][11] (Figure 1). Of the included reviews that conducted a meta-analysis, 78% investigated heterogeneity.…”

“…10 Due to the problems with analyzing baseline risk using frequentist meta-regression, we additionally undertook Bayesian meta-regression using Markov Chain Monte Carlo (MCMC) with Gibbs sampling following recently developed methodology that incorporates the uncertainty of the covariate estimates, which avoids the problems of regression to the mean (Supplementary Text Document 1). 6 We used vague prior distributions and burnt in the MCMC chains for 10,000 iterations and then used a sample of 50,000 iterations on which to base inferences.…”

“…Previous primary research has shown that the pain level experienced by a participant determines analgesic efficacy, with higher pain levels resulting in higher absolute pain score reductions following analgesic administration. 8,9 We have previously demonstrated that using control group morphine consumption (baseline risk), we were able to explain a large degree of between-study heterogeneity, 10,11 which may improve the quality of evidence as per Grades of Recommendation, Assessment, Development and Evaluation Group (GRADE). 2 Moreover, if baseline risk is found to explain statistical heterogeneity, any one effect estimate from a meta-analysis may be confounded by the variable baseline risk in the included trials and therefore quoting meta-regression estimates from a fixed value may help reduce confounding.…”