Stroke stands among the leading causes of death and adult-acquired disability with 13.7 million new strokes every year worldwide (Collaborators, 2019). Post-stroke pain is one of the most poorly understood complications, arising either in the acute, but mainly in the subacute or chronic stages (i.e., often within 6 months) of stroke (Merskey, 1994) . The prevalence of post-stroke pain varies largely depending on the definition of pain; the musculoskeletal pain appears to be the most common being reported in up to 72% of stroke patients (Harrison and Field, 2015). While post-stroke pain syndromes in general are estimated to affect up to 30–40% of stroke survivors (Paolucci et al., 2016), central post-stroke pain (CPSP) is more rare: definite CPSP was found in 3.5%, definite/probable in 5.8% and CPSP-like pain or dysesthesia in 6.7% of patients in a specific population-based study of post-stroke pain (Klit et al., 2011). Pain after stroke can remarkably reduce the quality of life, causing depression, anxiety and sleep disorders making rehabilitation more difficult.

Pain after stroke is often under-reported, being diagnosed only if actively searched by the clinician (Harrison and Field, 2015). There are multiple types of post-stroke pain syndromes that can also occur in combination, with both neuropathic and nociceptive features. The most common types of pain after stroke include CPSP, pain secondary to spasticity, shoulder pain, complex regional pain syndrome (i.e., CRPS), and headache (O’Donnell et al., 2013). Dysesthesia and allodynia often occur and the symptoms generally develop within the area corresponding to the lesion with frequent involvement of face, hand and foot, but sometimes also of thigh and shoulder (Kim, 2014). In particular, CPSP is often characterized by dysesthesia, constant or intermittent pain and hyperalgesia/allodynia (Harrison and Field, 2015). CPRS is of type I when nerve lesion is not identifiable, while of type II when there is a definite nerve lesion.

Treatment of post-stroke pain is made challenging by the lack of universally accepted guidelines (Kim, 2014), due to the paucity of high quality evidence from controlled clinical trials guiding pharmacological management and, expecially, for non neuropathic syndromes despite their high frequency (Hansson, 2004). In neuropathic pain, tricyclic antidepressants (e.g., amytriptiline), serotonin and norepinephrine reuptake inhibitors (e.g., duloxetine) and calcium channel α2δ ligands (e.g., gabapentin or pregabalin) are recommended as first-line agents, but data supporting their use is based on studies in peripheral neuropathic pain, while the evidence in central neuropathic pain is very limited (Mulla et al., 2015). A single study suggested lamotrigine to have a moderate effect on CPSP (Klit et al., 2009). Botulinum toxin injections represent the gold standard for the treatment of post-stroke spasticity and related pain (Hillis, 2020). Given their potential for misuse and other adverse effects (McNicol et al., 2013), opioids stand among the third-line therapy and evidence on their effectiveness for post-stroke pain syndromes is even more limited.

The aim of manuscript is to conduct a systematic review and meta-analysis of evidence on the efficacy of opioid and opioid antagonist medications, important and useful under the recommended conditions (Morrone et al., 2017), for reducing post strokepain and improving pain-related symptoms. Agonists and antagonists at opioid receptors were included in the search. In the brain area subjected to stroke, altered perfusion (Strahlendorf et al., 1980) and changes in opioid neurotransmission (Baskin and Hosobuchi, 1981; Willoch et al., 2004) were suggested to be positively affected by naloxone; incidentally, opioids can reduce blood flow during cerebral ischemia, through inhibition of the release of noradrenaline in the locus coeruleus (Budd, 1985). Moreover, naloxone and kappa opioid receptor antagonists were tested in acute ischemic stroke showing in some cases benefit and improvement of neurological conditions (Fallis et al., 1984; Jabaily and Davis, 1984; Perey et al., 1984; Adams et al., 1986; Czlonkowska and Cyrta, 1988; Federico et al., 1991; Czlonkowska et al., 1992; Clark et al., 1996; Lyden, 1996; Clark et al., 2000). Levorphanol, an opioid agonist with high affinity for all the mu, delta and kappa opioid receptors, reported to interact with both N-methyl-D-aspartate (NMDA) receptors and serotonin and norepinephrine uptake (Codd et al., 1995), was included because of its favourable pharmacodynamic and pharmacokinetic characteristics and it showed efficacy in neuropathic pain (Le Rouzic et al., 2019). Oliceridine, a novel mu opioid agonist, was included because it can confer analgesia with less respiratory depression (Dahan et al., 2020).

Studies eligible to be included in this systematic review and meta-analysis were required to meet the following criteria:

clinical trials assessing the effects of opioids on pain in post-stroke patients. No restrictions were placed on the publication date, study duration or follow-up; - patients of any age or ethnicity with post-stroke pain; - interventions include opioids.

Studies meeting the following criteria were excluded from the review:

in vitro and in vivo animal studies, narrative or systematic reviews and meta-analysis, abstracts and congress communications, proceedings, editorials and book chapters; - clinical trials assessing the effects of other pharmacological treatments (e.g., tricyclic antidepressant, pregabalin) or non - pharmacological management strategies (e.g., neurostimulation techniques); - studies not published in English.

Primary outcomes of interest were changes in objective measures of pain intensity (e.g., pain visual analog scale VAS) and secondary outcomes of interest were changes in pain-related outcomes (e.g., quality of life and physical functioning).

The literature search was conducted on PubMed/MEDLINE, Scopus, Web of Science and Cochrane Library databases for peer-reviewed studies on opioid medications for the treatment of post-stroke syndromes and published from databases inception until August 31^st, 2020 (date of last search). The search strings consisted in a combination of the following keywords: “stroke,” “post-stroke pain,” “pain after stroke,” “central post-stroke pain,” “CPSP,” “shoulder post-stroke,” “thalamic pain syndrome,” “central pain syndrome,” “shoulder hand syndrome,” “complex regional pain”; “Dejerine Roussy,” “facial pain,” “headache,” “facial neuralgia,” “trigeminal autonomic cephalalgia,” “temporomandibular joint disorders,” “allodynia,” “pain secondary to spasticity,” musculoskeletal pain,” “myofascial pain,” “neuropathic pain,” “opioids,” “methadone,” “tramadol,” “codeine,” “morphine,” “buprenorphine,” “oxycodone,” “fentanyl,” “tapentadol,” “loperamide,” “oxymorphone,” “hydrocodone,” “levorphanol,” “sufentanil,” “remifentanil,” “R-dihydroetorphine,” “Morphine-6-glucuronide,” “oliceridine,” “naloxone,” “naltrexone.”

Two authors independently screened titles and abstracts of the studies in agreement to the previously established inclusion and exclusion criteria. The reference lists of relevant papers were inspected for additional studies potentially missed in the database search. Any disagreement was planned to be solved by consensus or by consulting a third Author.

Two authors independently extracted the following data, according to the PICOS framework discussed above: study design, sample size, subtype of post-stroke pain syndrome, interventions, route of drugs administration, comparators, outcomes of interest (primary and secondary), drop-out rates, adverse effects.

A systematic and descriptive analysis of the results was provided with information presented in the text and tables. The narrative synthesis has been carried out according to the Cochrane Consumers and Communication Review Group guidelines (Ryan, 2013). Risk of bias and quality of the studies have been assessed, considering study limitations including lack of allocation concealment, lack of blinding, selective outcome reporting bias, inadequate sample or lack of sample size calculation. The revised Cochrane risk of bias tool for randomized trials RoB2 (Sterne et al., 2019) has been used. only the randomized clinical trials included were subjected to meta-analysis to assess imprecision. Indeed, the quality of the body of evidence for both outcomes was rated through the GRADE (Grading of Recommendations, Assessment, Development and Evaluations) system providing the evidence profile including the quality assessment and the summary of findings (Guyatt et al., 2011). Absolute and relative risk with 95% confidence intervals (CI) were calculated using the Cochrane Review Manager 5.3 (RevMan5.3; Copenhagen: The Nordic Cochrane Center, The Cochrane Collaboration). The random effect model (DerSimonian and Kacker, 2007) was used to manage eventual heterogeneity of the studies and to assess intra- and inter-study variation. In particular, for the assessment of inconsistency in results, since the retrieved studies number is small, the Higgins I² value was calculated to assess the heterogeneity of the studies (Higgins and Thompson, 2002). Relative risk below one favors the intervention (opioids) rather than the control/other treatment. Subgroup analysis, sensitivity testing and meta-regression have been performed to evaluate the impact and the causes of heterogeneity and publication bias has been assessed through Egger’s linear regression test to measure funnel plot asymmetry, adjusted through “trim and fill” method (Egger et al., 1997; Duval and Tweedie, 2000; Sterne and Egger, 2001).

The literature search retrieved a total of 83,435 results. The 83,435 references obtained have been searched for duplicates, leaving 34,285 articles to screen. After titles and abstract screening, not original articles like reviews, book chapters and conference proceedings have been eliminated leaving 24,950 titles and abstracts to screen. After elimination of in vivo and in vitro studies 2,736 have been screened to exclude observational and retrospective studies, thus leading to 2,531 clinical studies, among which 25 were obtained for full-text reading. One of these trials (Fallis et al., 1984) was not available in full text and one significant paper (Yamamoto et al., 1991) was further identified by the inspection of the reference lists of the relevant records. Eight studies met the inclusion criteria and were therefore included in qualitative synthesis. The four randomized controlled trials (RCTs) (Bainton et al., 1992; Attal et al., 2002; Maier et al., 2002; Rowbotham et al., 2003) were subjected to meta-analysis. The selection process is illustrated in the PRISMA flow diagram (Figure 1).

The 8 included articles (Budd, 1985; Yamamoto et al., 1991; Bainton et al., 1992; Yamamoto et al., 1997; Attal et al., 2002; Maier et al., 2002; Rowbotham et al., 2003; Saitoh et al., 2003) were clinical trials meeting the previously mentioned inclusion criteria. Studies were grouped according to the intervention (i.e., type of opioid medication), following the Cochrane Consumers and Communication Review Group guidelines (Ryan, 2013). Details of the included studies are summarized in Tables 1 and 2.

Two studies assessed the analgesic effect of morphine on CPSP (Attal et al., 2002; Maier et al., 2002). Attal et al. (2002) performed a double-blind, placebo-controlled, crossover study with the two-fold aim to evaluate the efficacy of intravenous morphine on spontaneous and evoked pain and the long-term benefit of oral morphine on neuropathic pain caused by spinal cord injury or stroke. They reported that the analgesic effect of intravenous morphine regarded only some components of evoked pain (i.e., the intensity of brush-induced allodynia) and that the effects of morphine on ongoing pain were not significantly different from those of the placebo and some patients were reported to receive other pharmacological treatment for pain. Regarding the long-term benefit of oral morphine, only three patients were reported to be still treated after 1 year with persistent pain relief, while the others dropped out before three months due to side effects. It is however unclear whether the patients still on oral morphine treatment at 1 year were those belonging to the group of neuropathic pain caused by stroke or by spinal cord injury. Maier et al. (2002) conducted a prospective, randomized, double-blind, placebo-controlled, crossover study on forty-nine patients with either neuropathic (of which only two had CPSP) or nociceptive pain syndromes, assessing the efficacy and effectiveness of 1 week of oral morphine administration. In fact, the MONTAS study assessed the efficacy of morphine on chronic non-tumor associated pain syndromes (Maier et al., 2002). An interdisciplinary consensus protocol on compulsory and optional treatments for pain, excluding strong opioids was followed before inclusion. The two patients with CPSP were classified as partial responders according to the reduction from 7.8 to 5.6 of mean pain intensity measured with an 11 points Numerical Rating Scale (NRS) and to the overall tolerability of adverse effects connected with opioid medications. Pain reduction was reported to correlate with improvement of physical function. Moreover, the Authors found a reduction of pain disability, depression score, mood and exercise endurance, secondary pain-related outcomes. The pharmacological background of intractable CPSP was characterized in three studies through morphine tests: two by Yamamoto and coworkers (Yamamoto et al., 1991; Yamamoto et al., 1997) and one by Saitoh and collaborators (Saitoh et al., 2003). The first study evaluated deafferentation pain using the morphine/thiamylal test enrolling twenty-five patients suffering from intractable deafferentation pain (thalamic/suprathalamic lesions n = 16; brainstem lesions n = 2; spinal cord lesions n = 2; peripheral nerve lesions n = 5) (Yamamoto et al., 1991). The morphine test consists in intravenously administering 3 mg morphine hydrochloride every 5 min up to reach 18 mg, followed by injection of naloxone to reverse thus confirming the effect of morphine, and assessing pain through a visual analog scale at 5 min intervals (Yamamoto et al., 1991). All the patients included were resistant to pharmacological therapy. According to the results, only two patients with thalamic or suprathalamic lesions were responding to morphine and thiamylal (Yamamoto et al., 1991). In the second study by Yamamoto et al. (1997) thirty-nine patients with intractable hemibody CPSP associated with dysesthesias and allodynia (twenty-five affected by a small thalamic infarct or hemorrhage and fourteen affected by infarct or hemorrhage in the posterior limb of the internal capsule or subcortical parietal area sparing the thalamus) were subjected to the morphine and thiamylal tests and only twenty-three recent cases were subjected to the ketamine test. All the patients had been received treatment with tricyclic and heterocyclic antidepressants, benzodiazepines and non-narcotic analgesics without satisfactory pain reduction. During this study, eight patients with CPSP were sensitive to morphine experiencing transient satisfactory pain reduction (Yamamoto et al., 1997). In the study by Saitoh and colleagues (Saitoh et al., 2003) nineteen patients with central and peripheral deafferentation pain (seven who had thalamic hemorrhage, one putaminal hemorrhage, one pontine hemorrhage, six brachial plexus injury, two phantom limb pain, one spinal cord injury and one pontine injury) of which eighteen underwent drug challenge test and pain was assessed through a visual analog scale and the McGill Pain Questionnaire. All the patients were treated with various medications including NSAIDs, anticonvulsants and antidepressants also used in combination, without sufficient reduction of pain. Among these patients, five were sensitive to morphine (Saitoh et al., 2003).

One study (Rowbotham et al., 2003) evaluated the efficacy of low and high doses of the opioid agonist levorphanol on eighty-one patients with neuropathic pain of different aetiology (patients with CPSP were ten). All patients had not achieved pain relief with previous non opioid medications and a trend towards previous use of low dose opioid was reported in the low-strength group. Compared to low ones, high doses of levorphanol resulted in higher rates of reduction in the intensity of neuropathic pain, considering the whole patient sample; however, high doses of levorphanol also resulted in more severe side effects that led to higher drop-out rates. Despite the additional outcomes of affective distress and interference with functioning were reduced, no difference between groups were observed. Moreover, pain relief was less frequent for patients suffering from CPSP. Pain effect on physical functioning was evaluated only in this study (Rowbotham et al., 2003).

Two studies assessed the effect of the opioid antagonist naloxone on CPSP (Budd, 1985; Bainton et al., 1992). Budd (1985) performed a single group study on thirteen patients with pain due to thalamic syndrome resistant to prior analgesic or other therapies and reported analgesia, assessed by direct questioning, for seven patients after twenty intravenous administrations of naloxone. The duration of the effects varied from 4 days to two and a half years. On the other hand, the placebo-controlled study by Bainton et al. (1992) failed to demonstrate the efficacy of intravenous administration of naloxone in alleviating CPSP.

Four of the studies are randomized clinical trials (Bainton et al., 1992; Attal et al., 2002; Maier et al., 2002; Rowbotham et al., 2003), one is a single arm trial (Budd, 1985) and three (Yamamoto et al., 1991; Yamamoto et al., 1997; Saitoh et al., 2003) are drug challenge tests. Therefore, the included studies are very heterogeneous in terms of study design. Moreover, four studies (Budd, 1985; Yamamoto et al., 1991; Yamamoto et al., 1997; Saitoh et al., 2003) included one single group without control. In the study by Rowbotham et al. (2003) the groups compared are high and low strength. The lack of a control arm can rise some concerns in terms of bias as for concealment. The population enrolled is heterogeneous across the eight studies and the number of patients is small for all the trials except for Maier et al. (2002) and for Rowbotham et al. (2003); however, the CPSP patients are only two and ten, respectively. Moreover, in the MONTAS study, with crossover design, number needed to treat and number needed to harm are reported to have been calculated only with reference to first week since the results of the second week could feel the effect of opioid withdrawal symptoms. Compliance to treatment has been assessed by pill counts and repeated urine screening, revealing only minor protocol violations. Interestingly, double masking was applied to three trials (Bainton et al., 1992; Attal et al., 2002; Rowbotham et al., 2003), but for the study by Attal et al. (2002) it was reported that seven patients and the examiner (in ten cases) had identified the active treatment, thus impairing blindness. In the study of Maier et al. (2002), a random generator was used for patients randomization and the medication package was blinded. The summary of risk of bias assessment according to intention-to-treat analysis is reported in Figure 2.

The quality of evidence of the two selected outcomes, i.e. analgesic efficacy of opioids in post-stroke pain and effectiveness on pain-related domains, was rated through the GRADE system (Guyatt et al., 2008; Guyatt et al., 2011). The quality assessment was based on: Limitations; Inconsistency; Indirectness; Imprecision and Publication bias. For each outcome, the four retrieved randomized clinical trials (Bainton et al., 1992; Attal et al., 2002; Maier et al., 2002; Rowbotham et al., 2003) were subjected to meta-analysis (Figure 3), for the assessment of absolute and relative risk and width in the CIs to calculate imprecision, with funnel plot for the evaluation of publication bias (Figure 4). The GRADE assessment reveals very low quality of evidence for the outcome of pain reduction and low quality of evidence for pain-related outcomes. This meta-analysis follows the Initiative on Methods, Measurement, and Pain Assessment in Clinical Trials (IMMPACT) recommendations (Turk et al., 2003). The core outcome domains for clinical trials of chronic pain treatment efficacy and effectiveness have been identified as pain; physical functioning; emotional functioning; participant ratings of global improvement; symptoms and adverse events; participant disposition (including adherence to the treatment regimen and reasons for premature withdrawal from the trial) (Turk et al., 2003). A meaningful decrease in chronic pain representing a clinically important difference in pain intensity is determined as change of approximately 2.0 points of Numerical Rating Scale (NRS) or 30–36% (Dworkin et al., 2008). Therefore, administration of opioids (agonists or antagonists) is not associated to meaningful pain relief (Relative Risk RR 1.05; 95% CI 0.57–1.92; I² = 0%; p = 0.53; Figure 3) and data are influenced by the paucity and the design of the studies. Though in agreement with I², heterogeneity allows comparison of these RCTs, RR is not estimable for the study by Rowbotham et al. (2003), since there is not a real control arm, but a high- and a low-strength arm. This occurs also for the pain-related outcome, thus influencing the RR calculation (RR 1.00; 95% CI 0.49–2.05; heterogeneity not applicable; Figure 4), since only two RCTs evaluate this outcome. In this study (Rowbotham et al., 2003) no significant changes in total mood disturbance and in quality of life were reported in either treatment group. In the study by Maier et al. (2002), the improvement of pain-related outcomes exerted by morphine administration reached statistical significance (p ≤ 0.05) only for pain disability and sleep quality. Therefore, RR is based only on the study by Maier et al. (2002), for pain-related outcome, thus forest and funnel plots are not reported; evidence coming from a single trial is uncertain. According to the forest plot in Figure 3 the results do not favor the experimental treatment (opioid agonist or antagonist) rather than the placebo for the outcome of pain reduction.

The GRADE assessment is based on rating of the following four domains:

Limitations: lack of allocation concealment and/or of blinding, loss to follow-up, failure to adhere to an intention to treat analysis and failure to report outcomes. This key outcome was downgraded for failure of concealment and blinding (Attal et al., 2002), lack of control arm and large loss to follow-up (Rowbotham et al., 2003) and minor protocol violations (Maier et al., 2002) for pain reduction; for the same reasons, this outcome was downgraded for pain-related outcomes.

The EP with quality assessment and Summary of Findings (SoF) is reported in Table 3.