Gabapentin in procedure-specific postoperative pain management – preplanned subgroup analyses from a systematic review with meta-analyses and trial sequential analyses

The search was planned by a trial search coordinator using the Cochrane Library’s CENTRAL, PubMed, EMBASE, and Science Citation Index Expanded databases. Previous reviews, reference lists and Google Scholar were hand-searched for eligible trials. Www.clinicaltrials.​gov; www.​controlled-trials.​com; www.​centerwatch.​com; www.​eudraCT.​com, and at the homepage of the US Food and Drug Administration (FDA) were searched for unpublished trials. Non-English articles were translated to English. The electronic search (Additional file 1: Appendix 1 - search strategies) was last updated April 12th, 2016.

After removal of duplicates, titles and abstracts were screened by two authors (MLF, AG). MLF and one other independent author (AG, MSH, PLP, LN) assessed full texts, extracted data and assessed bias. The following characteristics were extracted from the trials using a data extraction form: Year of publication, number of participants, type of surgery, follow-up period and dose regimen, consumption of opioid, and non-opioid medication, pain intensity, and any adverse events described in the trials, including serious adverse events (SAEs).

The corresponding author was contacted whenever data were insufficiently reported, and contact was repeated after 14 days. In case of no response, the involved bias domains were classified as unclear.

Risk of bias was assessed using The Cochrane Handbook guidelines. All trials were classified as low, unclear or high risk of bias using the following domains: Random sequence generation, allocation concealment, blinding, incomplete outcome data, selective outcome reporting and other bias, including funding and confirmatory bias.

Disagreements between authors on study selection, data extraction or bias assessment were solved by OM, JBD or JW.

We decided a priori to report and conclude based primarily on results from trials classified as low risk of bias.

All trials were evaluated in this post hoc analysis and allocated to the corresponding group according to the numbers of participants included in the analyses. Small trials were defined as trials with less than 50 patients included in each group. Trials were allocated to the remaining two groups if they included either more than 50 patients, or more than 200 patients [12].

The present subgroup analyses of surgical procedures were predefined in the protocol investigating the effect of different surgical procedures: Cholecystectomy, hysterectomy, mastectomy, orthopedic arthroplasty surgery, spinal and thoracic surgery on the primary and secondary outcomes. Analyses of thoracic surgery and orthopedic arthroplasty surgery have been added post hoc [13]. The surgical procedures were chosen to represent a wide range of surgical interventions and patient populations. Cholecystectomy is a minor procedure, often performed as day-case surgery, whereas thoracic surgery is a major procedure, which may be associated with intensive care in the immediate postoperative period. Mastectomy and hysterectomy are procedures with moderate to severe pain postoperatively. Orthopedic arthroplasty and spinal surgery often represents patients with chronic pain preoperatively. We chose to add trials investigating thoracic surgery and orthopedic arthroplasty surgery post hoc, in order to broaden the range of surgical interventions.

The planning and interpretation of the subgroup analyses followed the direction of the Cochrane Handbook [14].

The primary outcomes were difference in 24-h postoperative opioid sparing effects, and reported serious adverse events (SAE) between surgical procedures. SAE’s were defined according to the International Conference of Harmonization – Good Clinical Practice (ICH-GCP) definitions: Medical events being either life threatening, resulting in death, disability or significant loss of function, or causing hospital admission or prolonged hospitalization [15].

Secondary outcomes were differences in early (6-h) and late (24-h) pain postoperatively, both at rest and during mobilization, and all other adverse events, between surgical procedures.

All opioids were converted to intravenous morphine based upon equivalency as presented in Additional file 1: Appendix 2. Various scales were used to report pain intensity in the trials. All pain intensity scales reporting pain levels between 0 and 10 were converted to the Visual Analogue Scale (VAS) 0 to 100 mm.

Review Manager (RevMan) [Computer program], Version 5.1.6, Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014, and Trial Sequential Analysis (TSA) software (version 0.9.5.5 Beta), Copenhagen Trial Unit, Denmark, was used for statistical analyses as predefined in the protocol.

In trials with more than one active treatment arm, including trials testing doses delivered pre- and immediate postoperatively, means and standard deviations were combined for the intervention groups.

Mean and standard deviations were estimated from median and range values according to the method described by Hozo et al. [16] Standard deviations were calculated by dividing the difference in interquartile ranges with 1.35 [14].

Longer ordinal scales were analyzed as continuous data. For dichotomous data, RR with a 95% confidence interval was calculated.

We examined the heterogeneity between trials using chi-squared test. The heterogeneity was measured by I², which quantifies inconsistencies. If the I² was greater than zero the results were calculated using both a fixed effect model (FEM) and random effect model (REM) and the most conservative estimation was used [14, 17]. In the case of very few and rare events, Peto’s odd ratio was used to provide the best coverage of confidence intervals [18, 19].

Estimates were pooled in meta-analyses whenever more than one trial was included for the outcome. Tests for subgroup differences were carried out for all surgical procedures on all outcomes whenever a meta-analysis was possible. Using RevMan, the method to test for subgroup differences was implemented for all types of meta-analyses [14]. Our test for subgroup differences was performed on each of the subgroups testing them for subgroup differences against the compiled, remaining subgroups using the chi-squared test.

We used Trial Sequential Analysis (TSA) in post hoc analyses to adjust the confidence intervals for sparse data and repetitive testing. Minimal clinical relevant differences were defined as in our main review [12]. In the event that the accrued information size was less than 5% of the required information size, no TSA was reported, as the TSA program is unable to calculate trial sequential monitoring boundaries in this situation.

Trial characteristics are presented in Table 1. Eight trials were classified as overall low risk of bias, [39, 43, 46, 56, 59, 72, 73, 85] 18 trials were overall unclear risk of bias [21, 23, 24, 26, 28, 29, 31, 36, 38, 42, 52, 54, 55, 61, 65, 66, 68, 69, 80] and 48 trials were classified as high risk of bias, [20, 22, 25, 27, 30, 32‐35, 37, 40, 41, 44, 45, 47‐51, 53, 57, 58, 60, 62‐64, 67, 70, 71, 74‐79, 81‐84, 86‐94] (Fig. 2: Bias assessment). Allocation concealment, selective outcome reporting and “other bias” were the domains with most unclear or high risk of bias evaluations (Fig. 3: Risk of bias graph).

Sixty-six trials were classified as small trials [21‐27, 29‐31, 33‐45, 47‐55, 57, 58, 61‐64, 66, 67, 69, 71‐81, 83, 84, 86‐94], five had more than 50 participants in each group [20, 32, 46, 56, 85], and three included more than 200 patients [59, 68, 70].

The gabapentin dose in the included trials ranged from 100 mg to 1800 mg, and was mostly administered as a single dose (46 trials). [20, 22‐24, 28‐32, 34‐37, 41, 43, 45, 47‐51, 53, 55‐57, 61‐64, 67‐71, 75, 77‐79, 81, 82, 85‐90, 92, 94] In 30 trials, gabapentin was administered in combination with a basic, non-opioid/opioid analgesic regimen [20, 22, 23, 26, 28, 30‐32, 37‐41, 43, 46, 49, 54, 56, 57, 59, 61, 72, 73, 76, 79, 80, 82, 87, 90, 91]. In 44 trials, gabapentin was administered together with an opioid as the only analgesic [20, 25, 27, 29, 33‐36, 44, 45, 47, 48, 50‐53, 55, 58, 62‐71, 74, 75, 77, 78, 83‐86, 88, 89, 92‐94]. In five trials, gabapentin was administered in combination with a NSAID [27, 45, 75, 77, 83],, and in two trials, the postoperative analgesic regimen was not described [24, 81].

Eight trials were classified as overall low risk of bias. None from the mastectomy subgroup and one trial from the cholecystectomy group was overall low risk of bias [85]. In the subgroups hysterectomy, [39, 43] and thoracic surgery [46, 56] two trials were low risk of bias in each group, and three trials were classified as low risk of bias in the orthopedic arthroplasty subgroup [59, 72, 73].

Below, we present analyses from trials with low risk of bias. In addition, analyses from all trials are presented in Table 2: Primary outcomes from trials with low risk of bias and all trials estimates, and Table 3: Secondary outcomes from trials with low risk of bias and all trials estimate.

These subgroup analyses have some strength. The analyses were planned in a PROSPERO published protocol, and were derived from a PRISMA compliant systematic review adhering to Cochrane standards in methodology and bias assessment. The trials have been critically assessed using the Cochrane bias assessment tools, and conclusions are based on trials with low risk of bias, which is unlike most of the previous systematic reviews. The TSA has been added to adjust for sparse data and repetitive resting, which is a risk when the vast majority of included trials are small, that is <50 patients in each group [95‐98].

The limitations of this analysis mirror those of the included trials, and the limitations of the general methodology in subgroup analyses. Subgroup comparisons are to be perceived as observational because we compare pre-existing non-randomized groups, and must be interpreted as such [14].

The critical assessment of the trial methodology shows a very small number of trials with overall low risk of bias. Eighty-nine percent of the included trials have unclear or high risk of bias in one of the bias domains or more, risking an overestimation of beneficial -, and underestimation of harmful outcomes.

According to Oxman and Guyatt [99], Xin Sun et al [100] and their criteria to evaluate the credibility of subgroup analyses, we have to consider further limitations such as: If the subgroup can be considered independent; no a-priori direction of the subgroup effect has been published; the subgroup effects found in our analyses does not seem to consequently manifest in closely related outcomes.

A number of systematic reviews investigating individual surgical procedures, or with a procedure specific approach, have been published [95‐98, 101‐103]. Overall, there are some general methodological differences that separates the present work from previously published systematic reviews, such as: The emphasis on trials with low risk of bias, subgroup analyses with test for subgroup differences, and the use of trial sequential analyses. The emphasis on trials with low risk of bias prevents any direct comparison with estimates from other systematic reviews. Several reviews have evaluated bias in the included trials, and report different overall bias evaluations and number of trials with overall low risk of bias, compared to the present review [97, 98, 104]. This may be explained by the fact that that we contacted every author whenever a bias domain was deemed unclear risk of bias, to give them a chance to describe their methods in greater detail. To our knowledge, this was not done in the previously published systematic reviews.

In a recent review, the effect of trials with high risk of bias were tested using sensitivity analyses, and no impact was demonstrated on the outcomes [104]. This is in contrast to the present analyses, and to the findings of our recently published systematic review [12, 104]. The different approach to author contact, and a different statistical approach, may explain some of the differences in the findings. Further, the systematic review by Doleman et al. [104] reports that the type of surgery was not independently associated with the effect of gabapentin, which may provide some confirmation of the results from the present analyses. However, the results are obtained with different statistical approaches and the number and types of surgeries are not stated in the findings from Doleman et al. Consequently, the confirmation must be interpreted with caution [104]. In another review, Mathiesen et al. demonstrated a greater beneficial effect (reduction in 24-h opioid consumption) in their hysterectomy and spinal surgery groups [95], which seems similar to the results from the all trials estimates in Table 2 in the present review. Mathiesen et al., however, did not test for subgroup differences [95], and we found no differences between subgroups in the all trials estimates of the present review.

Most previous systematic reviews report favorable results for gabapentin treatment, similar to the findings from the all trials estimates in the subgroup analyses of the present review. In comparison with the systematic reviews of gabapentin for hysterectomy, cholecystectomy and thoracic surgeries [95, 96, 101], more trials have been included in our subgroups. Due to different inclusion criteria and subgroup analyses in the different published systematic reviews, it is not possible to conduct a full comparison of estimates.

However, none of the systematic reviews above have investigated the risk of SAEs, limiting the ability to weigh the benefit and harm of gabapentin in perioperative pain management [95‐98, 101].

We observed no systematic differences in postoperative opioid consumption, pain intensity, or adverse- or serious adverse events between six different surgical procedures treated with peri-operative gabapentin.

SAEs were very poorly reported, and only half the subgroups reported this outcome. More than 80% of the SAEs were reported in the thoracic surgery trials, making it impossible to rely on the risk and subgroup differences between the surgical procedures. In the original review, excess SAEs were reported in the gabapentin versus control groups, and approximately twice as many SAEs were found in trials with low risk of bias, compared with all trials [13]. Most trials have a short follow-up period and only report on SAEs and adverse effects for a short period postoperatively, which seems insufficient for a full evaluation. The inconsequent and diverse reporting of SAEs and adverse events complicates any reliable evaluation of these outcomes.