Summary of main results

In this update we added another class of drugs, botulinum neurotoxins, in particular botulinum toxin A (BoNT/A), to the list of pharmacologic interventions for treating established phantom limb pain. However, the results of the lone eligible investigation on BoNT/A did not substantially change the main results of the original version of the review.

Firstly, the short‐ and long‐term effectiveness of most pharmacologic interventions in established PLP remains unresolved for clinically relevant outcomes that include pain, function, mood or depression, sleep, quality of life, satisfaction, and safety. Botulinum toxin A was ineffective in phantom limb pain (based on a pilot study) in both short‐ and long‐term (up to six months) time frames. Of the remaining six drug classes reviewed, only morphine consistently demonstrated short‐term analgesia (based on two studies: n = 12 (Huse 2001); n = 31 (Wu 2002)), although only one study was adequately powered. The various NMDA receptor antagonists had differential efficacy, in that ketamine and dextromethorphan provided pain relief, and memantine did not. Gabapentin was shown to be beneficial with the pooled results, but these results should be interpreted with caution as computations were based on approximations. The studies on calcitonin and local anaesthetics had conflicting results.

Secondly, there was extensive variation in the methods, interventions, outcomes, outcome measures/scales, follow‐ups, data analyses, and reporting and presenting of results. This limited the pooling of results. Clinically relevant outcomes that include function, mood, sleep, quality of life, and satisfaction were missing in the majority of the studies, including the newly identified study.

Overall completeness and applicability of evidence

Botulinum toxin A injections did not lower phantom limb pain intensity assessed monthly for six months. Botulinum neurotoxins comprise a group of nerve‐blocking biologic agents that exert their blocking effect at the neuromuscular junction by preventing the release of acetylcholine (Aoki 2001a; Aoki 2001b). Of the existing seven serotype botulinum toxins, BoNT/A and botulinum toxin B have been studied in amputation‐related complications such as pain, hyperhydrosis, and involuntary movements (Charrow 2008; Jin 2009; Kern 2003; Kern 2004a; Kern 2004b; Kern 2011). The rationale for using BoNT/A in phantom limb pain relates to the peripheral mechanisms contributing to PLP. Phantom limb pain is often associated with neuroma, excessive muscle tightness, and spasm. Botulinum toxin A reduces muscular activity as a result of the neuromuscular blockade (Aoki 2001b; Brin 1997; Silberstein 2001). The negative results of this pilot study, Wu 2012, are not congruent with earlier studies (Jin 2009; Kern 2003; Kern 2004a; Kern 2004b). The authors of the study cited the small sample size, low baseline VAS scores, and the heterogeneous patient population included in the study. As this was only a preliminary study, it is difficult to make a definitive conclusion regarding the effectiveness and clinical applicability of BoNT/A in phantom limb pain.

Both oral and intravenous forms of morphine significantly reduced pain intensity. The percentage of participants responding to oral morphine in Huse 2001 was comparable to that of another study on morphine for postamputation pains (Wu 2008). As for adverse events, reviews on opioids for chronic non‐cancer and neuropathic pain also found constipation, along with nausea, vomiting, dizziness, and drowsiness as the most common and significant adverse events (Furlan 2006; Furlan 2011; McNicol 2013; Moore 2005).

The rationale for the use of opioids in PLP stems from the observed efficacy of these medications in neuropathic pain states. As with neuropathic pain, peripheral and central neural mechanisms have been implicated in the pathogenesis of PLP. Also, reviews investigating the prescription and use of opioids, in Hall 2013, as well as the effectiveness of opioids in neuropathic pain conditions (e.g. diabetic neuropathic pain, postherpetic neuralgia) included evidence from PLP trials, albeit results on their effectiveness including long‐term adverse events in these conditions have not been strong and conclusive (Finnerup 2015; Furlan 2006; Furlan 2011; Kalso 2004; McNicol 2013). The reasons why opioids might work in phantom pain are not well‐understood. Peripherally and at the spinal level, opioids act via presynaptic nerve terminals and postsynaptic neurons involved in pain transmission. Centrally, these drugs may decrease cortical reorganisation, a phenomenon where the topographic representation of lost extremity is shifted to other areas of the cortex and taken over by sensory input from other areas of the body, leading to perceptual remapping following the amputation (Birbaumer 1997; Elbert 1994; Flor 1995; Ramachandran 1992).

The results for the NMDA receptor antagonists as a group were at best equivocal. Blocking the NMDA receptors in the dorsal horn, which play a significant role in central sensitisation, hyperexcitability, and wind‐up phenomenon, can decrease pain manifestations. Among the NMDA receptor antagonists, dextromethorphan and ketamine had short‐term analgesic effects compared with placebo, but these findings were based on underpowered studies. On the other hand, memantine did not have the same positive effects (Maier 2003; Wiech 2004). The issues raised were low drug dosage, short run‐in period, other probable mechanisms that maintain phantom pain aside from the NMDA receptor activation (Maier 2003), time‐dependent effect of memantine on neural transmission via NMDA receptor pain maintenance, and the differential affinity of the various NMDA receptor antagonists (Wiech 2004).

The differences in the results for ketamine and memantine are not easily explained. Firstly, the type of neuropathic pain involved could be a factor. Ketamine significantly decreased pain evoked by mechanical stimulation and increased pressure pain thresholds (Nikolajsen 1996). Conversely, memantine in neuropathic pain after amputation, surgery, and postherpetic neuralgia neither decreased mechanical and cold allodynia, mechanical hyperalgesia, and wind‐up‐like pain nor increased thresholds to mechanical pressure (Eisenberg 1998; Nikolajsen 2000). However, in the included memantine studies, the detailed characteristics of the neuropathic pain were not explicit nor were there outcomes on allodynia, hyperalgesia, wind‐up‐like pain and pressure pain thresholds. Secondly, the timing of the intervention might be important. Memantine given in combination with brachial plexus blockade in the early postoperative stage significantly decreased the intensity and prevalence of PLP at four weeks and six months (Schley 2007). On the other hand, memantine given in established PLP in the current review led to negative findings. A characteristic of the populations was continuous pre‐existing pain of at least 12 months. Pre‐existing pain results in the formation of a somatosensory “pain memory” that indicates long‐term changes in the central nervous system (Katz 1990). When this pain memory is in place, functional and structural changes in nociceptive structures have already occurred (Lei 2004). NMDA antagonism might thus no longer be useful in longstanding phantom pain where the neuroplastic changes are already fixed. Thirdly, the route of administration might also be an issue. In this review, intravenous ketamine during the chronic phase of phantom pain altered the pain intensity favourably. On the other hand, all three memantine studies used the oral preparation and demonstrated consistent negative findings.

The results for the analgesic efficacy of gabapentin were contradictory. An earlier study, Bone 2002, indicated positive findings in favour of gabapentin, while a later study, Smith 2005, showed otherwise. Combining the results of the two studies entailed estimating certain parameters that include the treatment effect and its standard error (SE) from each study (Bone 2002; Smith 2005). We approximated the SE in one study given the standard deviations of the pain intensity differences and n for treatment and placebo (Bone 2002). In the later study, we computed the SE by dividing the mean difference between placebo and treatment by the given value of the t statistic (Smith 2005). We performed the generic inverse variance method for pooled analysis. The pooled estimate suggested a trend favouring gabapentin, but this should be interpreted with caution for reasons mentioned above. A recent review on gabapentin for chronic neuropathic pain that included PLP found that gabapentin was better than placebo and that 34% to 38% of participants experienced at least 50% pain relief with the drug (Moore 2014). Adverse events were also significantly more frequent with gabapentin in this review (Moore 2014), compared to the Bone study.

The negative results for amitriptyline (based on only one study) were attributed to a missed a small treatment effect, insufficient duration of treatment (six weeks), and the type of participants included (Robinson 2004). Tricyclic antidepressants including amitriptyline have been considered first‐line drugs for neuropathic pain (Finnerup 2015; Moulin 2014; Tan 2010). However, a recent meta‐analysis that focused on amitriptyline in neuropathic pain showed a lack of good‐quality studies to support its beneficial effects or lack of effect (Moore 2015). Furthermore, the review suggested that it may benefit a few and select group of patients, but not the majority (Moore 2015).

The findings on the analgesic action of calcitonin were contrasting. The earlier study involved participants with acute phantom pain (developing within seven days after amputation) (Jaeger 1992), whereas the later study included participants with years of history of phantom pain (Eichenberger 2008). The mechanism involved in the analgesic action of calcitonin is unclear, although its direct central action is likely the main mechanism, as suggested by its inhibitory effect on the neuronal firing in response to peripheral stimulation and the finding of its receptors in the central nervous system structures (Azria 2002). The ineffectiveness of calcitonin in one study was attributed to its possible lack of effect on central sensitisation processes, which are important in phantom pain pathophysiology (Eichenberger 2008).

The results for the local anaesthetics (lidocaine, bupivacaine) were variable. While both are sodium channel blockers, they were administered via different routes. Lidocaine at 4 mg/kg was administered systemically as an infusion, whereas bupivacaine at 2.5 mg was given locally as an injection. Lidocaine infusion was ineffective in PLP, as the mechanisms responsible for PLP are primarily considered to be central, although peripheral inputs are important in maintaining pain. Lidocaine is considered to act for the most part in the periphery by decreasing ectopic discharge following peripheral nerve injury (Wu 2002). Conversely, bupivacaine injection to contralateral myofascial hyperalgesic areas decreased phantom pain intensity (Casale 2009). The mechanism of pain relief from local anaesthetic contralateral injection is not clear. In animal experiments, blocking afferent inputs on the contralateral side can decrease the spontaneous hyperactivity and after discharges following noxious evoked responses in the wide dynamic response neurons in the  ipsilateral (injured) side (Bileviciute‐Ljungara 2001).

Aside from the type of medication, the dosing, route, and ease of administration are also of clinical importance. While some of the drugs were given for a period of days to weeks (oral medications), others were administered as a single dose via relatively more invasive manner (e.g. infusion, injection) such as ketamine, calcitonin, intravenous morphine, and bupivacaine injection. Thus far, outcome assessment, including adverse events, for these single‐dose drugs given intravenously or as injections, have been within short time frames (e.g. at the end of infusion, within 30 minutes to 48 hours). Also, the duration of their analgesic effects has not been documented. This may put some degree of uncertainty on the clinical value of such drugs at this stage for this type of pain, which is generally considered chronic.

Other issues that can impact on applicability of evidence are the characteristics of the population and the phantom limb pain. The majority of studies included populations with mixed amputation aetiologies, although predominantly traumatic. In the United States, dysvascular disease accounts for most amputation cases, followed by trauma (motor vehicular accidents) (Sheehan 2014; Ziegler‐Graham 2008). Also, as aftermath of the Afghanistan and Iraq wars, major limb amputations constitute 7.4% of major limb injuries, and about 88% of these are attributed to explosive device (Stansbury 2008). Phantom limb pain is more likely in traumatic amputees (Ephraim 2005).

Phantom limb pain chronicity (four months to 12 years) was a feature of the included studies, except for one that dealt with acute phantom pain. None of the studies provided analyses that explored the relationship between chronicity and treatment response, although in general, people with chronic pain usually have a more protracted course of treatment, as there are other associated problems such as mood and sleep disorders. Furthermore, none of the studies described the phantom pain in detail (e.g. frequency, quality, mechanism, severity, etc.). The association between characteristics of the phantom limb pain and response to particular drugs needs further investigation.

The second main finding in this review relates to heterogeneity. The study populations varied, from acute PLP to chronic phantom pain, although most were of the chronic type. Seven groups of interventions examining 10 individual drugs of dissimilar doses (e.g. gabapentin at 2.4 g/day or 3.6 g/day; single intravenous morphine infusion), of differing routes (e.g. oral morphine versus morphine infusion; intravenous lidocaine versus myofascial injection of bupivacaine), and of variable duration of administration (minutes versus weeks or months) were studied. A wide range of pain scales and measures were utilised (e.g. 11‐point numerical rating scale, 0‐to‐100 VAS, NNTB for 30% or 50% pain relief, McGill Pain Questionnaire). A variety of scales were used to assess secondary outcomes as well (e.g. Functional Independence Measure or Barthel Index for function; the Hospital Anxiety and Depression Scale or Center for Epidemiologic Studies Depression Scale for depression). Outcomes were measured at different time points ranging from 30 minutes to 6 months. The majority of studies had a cross‐over design that differed in number of treatment arms, phases (blinded and open), and analyses of the data. The results for outcomes were reported and presented in various ways.

Quality of the evidence

Studies were generally small and of short duration. Adequately powered studies yield more accurate and reliable estimates of treatment effect. The clinical applicability of interventions for phantom limb pain that is considered chronic would rely on the assessment of clinically relevant outcomes including adverse events over the long term. Also, the completeness of outcome data was questionable in some studies, as the attrition and exclusions from the analyses were not explicit. Another issue was that the majority of the studies had a cross‐over design, and the possibility of carry‐over effects could not be entirely eliminated. Evaluation of carry‐over effects is not straightforward, as statistical techniques to analyse such effects are far from satisfactory and rely for the most part on judgement (Higgins 2011). To reduce the risk of carry‐over effects, most of the studies utilised sufficient wash‐out periods between treatment periods, except for the dextromethorphan study, which did not report any wash‐out periods. Some studies also reported no significant differences in the baseline pain intensity levels between the start of each treatment period to indicate that carry‐over effect was unlikely.

Potential biases in the review process

This review has several limitations. Firstly, all of the included studies had small sample sizes, which made it problematic to form generalisations and conclusions. Secondly, we did not include studies where populations included participants with a diagnosis of postamputation pains with analyses that did not distinguish PLP from the other pains (e.g. postoperative pain, stump pain, etc.). One study found oral morphine to be more effective than placebo and mexiletine in improving pain but not function in people with postamputation pains (Wu 2008). The exclusion of such a study might have led to an underestimation of the analgesic effect of oral morphine in this review. Also, we did not include a negative study on another class of drugs being investigated for phantom limb pain, synthetic cannabinoids, due to the lack of information regarding methods, data, and actual study results, as it was published only as an abstract (Khanahmadi 2012). Thus, the possibility of publication bias thus cannot be entirely discounted. However, the general results of that study based on the abstract would not have changed the main conclusions of this update. Thirdly, we were not able to do a fully satisfactory meta‐analysis due to the variability in designs, outcomes and outcome measures, and analyses and reporting of the results. For example, for the opioid and anaesthetic studies, different routes of administration were studied. By combining the results of the memantine studies (Maier 2003; Wiech 2004), we incurred a unit of analysis error, as we treated the cross‐over study, Wiech 2004, as though it was a parallel study by taking all measurements from memantine periods and all measurements from placebo periods and analysing these data as in a parallel study. The analysis assumed that this study had 16 participants, when in reality it had only 8. We thus see in the forest plot that the confidence interval is very wide and that the study has very little weight (Analysis 1.1). But to start with, the trial has a very small sample size, so the presentation of forest plot here does not in fact add more to what the individual results show. The strategy of combining the two gabapentin studies was also not without shortcomings (Analysis 2.1). Ideally, the estimates of the treatment effect with their SEs are available in the study, but in this case, they were not; we therefore performed imputations. We noted, however, that the SE estimate of the positive study with the smaller sample, Bone 2002, was slightly smaller, and hence the weight was larger than that of the study with a slightly bigger sample size. It is possible that there was a little more variability in measurement in the other study (Smith 2005). These pooled results should therefore be interpreted with caution.

Agreements and disagreements with other studies or reviews

Several reviews have also identified and discussed various modalities of treatment in phantom pain but without firm recommendations on which is the best for clinical use (Foell 2011; Halbert 2002; Manchikanti 2004; Sherman 1980; Wolff 2011). A recent systematic review identified level 2 evidence (classed as "one or more well‐powered randomized, controlled trials") for the efficacy of intravenous ketamine and intravenous morphine in phantom limb pain in the short term (McCormick 2014). This shows some agreement with the findings of our review.

A recent meta‐analysis on medications for neuropathic pain of various causes including postamputation pain found moderate to strong evidence of efficacy for tricyclic antidepressants, serotonin‐noradrenaline reuptake inhibitors, pregabalin, gabapentin, tramadol, strong opioids, capsaicin patches, and BoNT/A (Finnerup 2015).