Relevant randomized controlled trials were identified by performing a Medline 10, a Cochrane Library 11 and a Google Scholar search 12, without language restrictions. Free text combinations including the search terms: "gabapentin", "post-operative pain" and "post-operative analgesia" were used [see Additional file 1]. Additional papers where sought by reviewing the reference list of retrieved reports and relevant reviews. Last search was performed January 2007. The QOURUM guidelines for reporting meta-analyses were followed 13.

Reports were considered if they were double-blind, randomized controlled trials of gabapentin (experimental intervention group) for postoperative pain relief compared with placebo (control intervention group) in adult patients (> 18 years) undergoing a surgical procedure. Only studies, in which data on either pain (visual analogue scale (VAS) or verbal score (VRS)) or supplemental postoperative analgesic consumption were stated, were included. Studies with less than 10 patients in treatment arms were not included 14.

Each identified study was read and scored independently by two authors (OM + JBD), using a 5-point scoring system as described by Jadad et al 15. If the reports were described as randomized, one point was given. One point was added if the randomization was described and appropriate (random number of tables, computer generated, etc), and likewise one point was subtracted if the randomization was described and inappropriate. If the study was described as double-blind one point was given, and an additional point was added if the method of blinding was described and appropriate. (identical placebo, active placebo etc.) For inappropriate blinding one point was subtracted. Finally a point was given if withdrawals and dropouts were appropriately described. Disagreement between the authors was solved through discussion.

Data from the studies were extracted onto a datasheet by one of the authors (OM). This included type of surgery; number of patients in intervention and control groups; time of administration and regimen of gabapentin treatment; mean VAS pain scores at rest and during mobilization early (at 4 or 6 hours) and late (at 24 hours) after surgery; supplemental analgesic regimen; type of and amount of supplemental analgesic consumption; and possible side-effects (nausea, vomiting, dizziness and sedation). Side-effects reported as somnolence or drowsiness were grouped under sedation, and reports of light-headedness and vertigo were grouped under dizziness. Pain reported on a 0 – 10 scale was converted to a 0 – 100 scale. In dose-finding studies, we extracted data from each dose-group onto the data sheet. When data in a study was only shown graphically, we extracted data from graphs. We contacted eight authors to get supplemental data for analysis, and received requested information from all.

Qualitative analysis of postoperative effectiveness was evaluated by assessment of significant difference (P < 0.05 as reported in the original paper) in pain relief using consumption of supplemental analgesic and pain scores between study groups, and by an assessment of clinical importance of observed findings. In addition, internal sensitivity was evaluated by an assessment of pain scores. It has been recognised that adequate sensitivity in trials of analgesics for acute pain, may only be achieved when patients are experiencing at least moderate pain (VAS pain score > 30 mm) with placebo, as it is difficult to detect an improvement with a low degree of pain 1617.

Quantitative analyses of combined data were only performed with data from similar procedures (e.g. hysterectomy), but not across trials with different surgical procedures.

Quantitative analyses of combined data from similar procedures were performed by calculation of the weighted mean difference (WMD) of the 24-hour cumulated opioid use between study group, and by calculation of WMD of VAS pain scores between study groups early (4 – 6 h) and late (24 h) after surgery, whenever sufficient data were provided in original papers (e.g. standard deviation (SD), number of included patients in each study group and the relevant mean value).

Opioids other than morphine were converted to their morphine equivalents, based on the equivalence of 100 μg fentanyl, 5 mg ketobemidone, and 100 mg tramadol respectively to 10 mg morphine.

From papers where data were available, dichotomous data were extracted for calculation of the relative risk (RR) of side-effects (nausea, vomiting, dizziness and sedation).

Quantitative analyses were performed using the Review Manager (RevMan) software (version 4.2 for Windows, Copenhagen, The Nordic Cochrane Centre, The Cochrane Collaboration, 2003). A random effect model was used if the statistical test for heterogeneity was positive, and a fixed effect model if the test came out negative.

An overview of the included trials is presented in Table 1. The retrieved studies were generally of high quality (median quality score: 5; range 1 – 5).

The surgical procedures were abdominal hysterectomy in five studies 2728293031, spinal surgery in four studies 24323334, breast surgery in two studies 3536 and a variety of different surgical procedures in the remaining twelve studies 252637383940414243444546. Median number of patients included in the studies was 50 (range 40 – 306). Three studies 253538lasted for only 4 h postoperatively, and one study 34 for 8 hours. The remaining studies lasted for 24 hours or longer.

In most studies gabapentin was administered one to two hours preoperatively, but in three studies 313436 a multiple (repeat) dosing regimen initiated the day before surgery was investigated.

In sixteen studies a single dose of gabapentin, varying from 300 mg to 1200 mg was administered and in seven studies 27303134363946 repeat dosing regimens were investigated. (Table 1)

The overall 24-hour reduction in opioid consumption following abdominal hysterectomy as well as spinal surgery, confirm the results from reviews with pooled data from a variety of surgical procedures 6789, and the postoperative opioid sparing effect of gabapentin compared to placebo seems unquestionable.

In our review the opioid sparing in abdominal hysterectomy (13 mg morphine), is somewhat different from that of spinal surgery (32 mg morphine), and several reasons for this result could be thought of. First, data for spinal surgery might have been skewed, since one trial 24 used very high postoperative doses of fentanyl, and consequently reported a larger opioid sparing effect (56 mg morphine). If this study was removed from the analyses, we could demonstrate a 24-hour WMD of – 20 mg morphine instead. Second, the pain scores at rest for patients in spinal surgery are higher than for abdominal hysterectomy both early (mean VAS 42 mm vs. 30 mm) and late (mean VAS 26 mm vs. 17 mm), and it is possible that the analgesic effect of gabapentin simply reflects this difference in pain states. Third, the patho-physiology of postoperative pain in spinal surgery might be of a different origin than for abdominal hysterectomy, e.g. that pre- and per-operative nerve damage in a higher degree contributes to postoperative pain in spinal surgery than in abdominal hysterectomy. Gabapentin is effective in the treatment of chronic neuropathic pain 4, and so a larger analgesic effect in postoperative pain after spinal surgery than abdominal hysterectomy is a possibility.

The relevance of the pain score improvement demonstrated in the quantitative analysis is debatable. A minimal clinical significant change in patient pain severity has previously been measured as 13 mm in trauma patients 4748, and another author 49 reports of a 33 % pain intensity difference being clinically meaningful. Using these definitions our 11 mm VAS pain relief at rest for abdominal hysterectomy patients are of minor clinical value, whereas the pain relief in spinal surgery patients (- 17 mm at rest early) could be of clinical importance, since it may improve patients ability to resume daily activity.

However, comparison of pain scores during PCA treatment is problematic since the patients can avoid uncontrolled pain in both treatment groups. Accordingly, our primary outcome parameter was morphine consumption.

Our analyses indicate, that preoperative gabapentin reduces postoperative pain scores more effectively in the early rather than the late postoperative period. Several explanations may be possible. Most studies used a single dose of gabapentin, and since T_max in plasma is 2–3 h, the plasma level of gabapentin accordingly is higher in the early postoperative phase. One study 27 has demonstrated a significant inverse association between plasma levels of gabapentin and morphine usage both at two and four hours postoperatively, indicating a dose-response relationship. Furthermore, the postoperative pain scores are generally higher in the early compared to the late postoperative phase, and consequently, the improvement in pain score by gabapentin could also be relatively larger in this phase. Finally, the absence of internal sensitivity (VAS < 30 mm) may also play a role in measuring the effect of gabapentin, since five of eight trials in the two procedures that were non-significant had low internal sensitivity.

In the most recent review 9 the authors reported of a significant reduction in both early (6 h) and late (24 h) VAS pain scores with a single preoperative dose of 1200 mg gabapentin (WMD -17 mm and -11 mm, respectively). The improvement in VAS-pain reported in the reviews by Hurley et al 8 and Seib et al 7 also correlate well with our findings.

One of the major challenges in postoperative pain treatment is to combine different treatment modalities, in order to improve patient analgesia and potentially reduce side-effects 1. A cornerstone in this area is to minimize the need for postoperative opioid analgesics and hopefully reduce side effects, especially postoperative nausea and vomiting (PONV). Zhao et al has reported a dose-response relationship between opioid use and adverse events for patients in ambulatory surgery. Every 3–4 mg increase of morphine usage was associated with 1 additional clinically meaningful opioid-related symptom 50. Marret et al 51 reported that the opioid sparing by NSAID and COX-2 inhibitors (approximately 30 %) was followed by a significant reduction in PONV and sedation (also by approximately 30%), but not in urinary retention and respiratory depression.

In this light it is noteworthy, that despite of the opioid sparing effect with gabapentin, our analyses of side-effects showed a just significant lower incidence of nausea in favour of gabapentin for abdominal hysterectomy, but not spinal surgery patients, while vomiting and sedation were non-significant between treatment groups. These disappointing results correspond with results from another review on opioid sparing with the use of COX-2 inhibitors and the lack of evidence for reduction of opioid related side-effects 52.

One of the reasons for this finding could be the relatively small number of patients and thereby low statistical power of our analyses. A post-hoc statistical power calculation of the non-significant quantitative analyses of adverse events, using a risk of Type I error on 5% and the observed cumulated difference between treatment groups, revealed a statistical power of less than 50% on analyses of most adverse events in both abdominal and spinal surgery.

Another reason could be the lack of quality of data on adverse events from the original papers, as in most reports adverse events were not well defined and only reported as present or absent. Furthermore, data of adverse events were a mixture of spontaneous reporting and specific questioning.

We did not find any sign of clinically limiting side-effect with the use of gabapentin, and especially the data for sedation and dizziness were non-significant. It is possible that the well-known sedative effect of gabapentin may be masked by the use of opioids in the studies and the fact that we monitor patients in the early phase after general anaesthesia.

The wide variability of gabapentin dosing regimens, and the differences in pain score- and side-effect evaluation, undoubtedly influences the outcome of this review, as do the pooling of different rescue analgesics (morphine, fentanyl and tramadol). By making this review procedure specific we tried to minimize variability in the pooled outcome parameters, but differences in e.g. surgical techniques among the different centres may still limit our analyses.

Another important limitation for the interpretation of the analgesic effect of gabapentin is the low VAS pain score in many of the included trials, both early and especially 24 hour postoperatively, since previous studies has emphasized the importance of moderate to high initial pain intensity in postoperative assessment of analgesic drugs 16.

In our quantitative analysis the primary outcome is PCA or "on demand" opioid consumption with simultaneous VAS-pain measurement. The assumption is that patients in the active and control treatment groups will titrate "down" to the same low VAS-pain, and the difference in opioid consumption at the same VAS-pain gives us the valid difference in opioid consumption. This only happens in three of nine trials which is a limitation when the analgesic effect of gabapentin is interpreted.

All trials in this review were randomized and the methodological quality satisfactory, the median Oxford quality score being 5, and so selection bias should not have been a problem. Most trials reported of a positive outcome of either opioid reduction or pain score improvement, and publication bias (skewed publication of only trials with a positive outcome) cannot be ruled out. Given the nature of abdominal hysterectomy gender bias is not the case here, whereas for the spinal surgery trials twice as many male patients were included as women (173 vs. 93). However, since the male/female distribution generally was equal among treatment groups, we find no tendency to gender bias in the trials. Language restriction to only articles published in English could be a potential bias, but we did not find any non-English published articles. Many analyses revealed heterogeneity among the included studies, and in these cases we used a random-effect model in the analyses to compensate. Finally almost all studies represent small sample sizes with the increased likelihood of Type 2 errors.

The development of chronic post surgical pain has attracted increased attention 53 and with central sensitization as a prerequisite, gabapentin with its ability to attenuate secondary hyperalgesia in pain models 5455, is an interesting "anti-hyperalgesic". Only one of the included trials investigated pain beyond the immediate postoperative period 31, and our review cannot make a conclusion on this subject. In two excluded studies by Fassoulaki et al 2122, gabapentin was investigated together with local anaesthetics as part of a multimodal analgesic regimen. In both abdominal hysterectomy and breast surgery, a reduction in postoperative morphine usage, as well as pain for more than one month postoperatively was reported.

In the qualitative analysis of the trials we found no obvious dose-response relationship for the use of perioperative gabapentin. One trial 24 investigated the effect of different preoperative gabapentin doses (300–600–900–1200 mg) and found no additional analgesic effect raising the gabapentin dose above 600 mg. It is the only study in this review addressing this subject and so, the material does not allow for any conclusions on this topic.

In a recent pilot trial by Leung and colleagues 19, preoperative gabapentin significantly reduced the incidence of postoperative delirium. This subject is sparsely addressed in the trials covered here. Two studies reported on lack of concentration 2426, two studies reported on visual disturbances 3539, and one study on hallucinations 43. All were non-significant on the subject.