Evaluating the Efficacy of Targeted Muscle Reinnervation (TMR) in Extremities Amputations of Pedis Squamous Cell Carcinoma: A Comprehensive Systematic Review, Meta-Analysis, and Meta-regression of Functional and Pain Outcomes

Eligibility criteria

This review was done following the PRISMA guidelines for systemic reviews and meta-analyses. We comprised original research papers published between 2013 and 2024 with the last search being done on 20th May 2024. The studies that were selected had to meet the inclusion criteria for randomized controlled trials (RCTs) and clinical trials. However, exclusion was made for studies such as technical reports, editor responses, narrative reviews, systematic reviews, meta-analyses, non-comparative studies, in silico studies, in vitro studies or in vivo researches, scientific posters, research proposals or conference abstracts. Besides that, other exclusions included those not published in English language; incomplete data; or failing to look at the relationship between miRNA and PTEN gene expression in OSC in association with survival rates. The selected articles followed PICO guidelines as follows: i) amputees with indications for amputation of the limbs; ii) introduction of targeted muscle reinnervation (TMR) during amputation surgery; iii) conventional surgical methods as comparison; iv) VAS scores for assessment of phantom pain and its occurrence rate.

Data search and selection

The study on which this research is based entailed an extensive search across various databases, such as PubMed, Cochrane Library, Scopus, Springer and Epistemonikos. We used a combination of keywords to maximize the relevant literature we would be able to retrieve: (“Targeted Muscular Reinnervation” OR “TMR”) AND (“Limb Amputation” OR “Extremity Amputation”). The searches encompassed studies from the beginning of the databases up to May 20th, 2024, emphasizing research done in the last ten years. Boolean operators were employed for refining results based on Medical Subject Headings (MeSH) as per National Institutes of Health (NIH) National Library of Medicine browser guidelines. To manage all these studies, the authors’ library used Mendeley Group Reference Manager. All databases were filtered for only randomized controlled trials and clinical trial research articles.

The selection process was done by five independent authors (KCT, INSS, NADR, MS) who conducted initial literature screening articles were first assessed by title and abstract followed by removal of duplicates. Then they performed a second round of full-text reviews to check if eligible criteria for inclusion and exclusion were met. Any discrepancies during this process were resolved via discussions with another author (RN, KCT, NA).

Data extraction

In a Google Spreadsheet, after the last screening, selected studies were datas were extracted and combined. The collected information comprised the author, year of publication, country, design of the study, sample size, mean age of the respondents, VAS score and incidence of no-phantom pain. Five authors (RN, KCT, INSS, NADR and MS) independently carried out the process of data extraction. The main focus was on the post-surgical effectiveness related to VAS scores and incidences of no phantom pains comparison between TMR outcomes with conventional surgical techniques.

Study risk of bias assessment (qualitative synthesis)

The risk of bias assessment was independently conducted by five authors using established tools. For the randomized controlled trials (RCTs), we applied the Cochrane Collaboration’s Risk of Bias version 2 (RoB-2) tool, which evaluates five key methodological areas: (a) the randomization process; (b) deviations from intended interventions; (c) missing outcome data; (d) measurement of the outcome; and (e) selection of reported results. The assessment outcomes are presented in Figures 5 – 7, with bias categorized as “low risk,” “high risk,” or “some concerns.”

For non-randomized studies, the risk of bias was assessed using the ROBINS-I and ROBINS-E tools. The ROBINS-E tool considered seven domains, including: (a) bias due to confounding; (b) bias from exposure measurement; (c) bias in participant selection; (d) bias from post-exposure intervention; (e) bias due to missing data; (f) bias from outcome measurement; and (g) bias in the selection of reported results. Similarly, the ROBINS-I tool evaluated seven domains, such as: (a) bias from confounding; (b) bias in participant selection; (c) bias in intervention classification; (d) bias from deviations from intended interventions; (e) bias due to missing data; (f) bias in outcome measurement; and (g) bias in the selection of reported results.

Any disagreements during the assessment process were resolved through discussion. If consensus could not be reached, RN would make the final decision.

Statistical analysis

We employed the standardized mean difference (SMD) with 95% confidence intervals (95% CI) to pool continuous outcome variables, using the inverse variance method. Due to the expected heterogeneity from variations in population characteristics and follow-up durations, random-effects models were chosen for this review.

Heterogeneity across studies was assessed using the I-squared (I²) statistic, with I² values over 25% indicating significant heterogeneity. For data reported as median and interquartile range (IQR) or as median, minimum, and maximum values, we applied formulas from Luo D et al. and Wan X et al. to convert these into means and standard deviations (SD) for pooling in the analysis.

The initial analysis compared the intervention and control groups based on follow-up time subgroups. While overall, the intervention group showed better functional outcomes, some studies and subgroup analyses revealed contrary results. As a result, we conducted a separate analysis of preoperative and postoperative values to assess the mean difference for each evaluation in both the intervention and control groups. The pooled mean differences from each evaluation were then compared between these groups.

Publication bias was assessed when more than ten studies were available for any given outcome. If asymmetry was detected in the funnel plot, we reviewed PICO and outcome characteristics to determine if it resulted from publication bias or other factors such as methodological variability between studies. All statistical analyses were performed using Review Manager 5.4 from the Cochrane Collaboration. We further assessed the robustness of the outcomes through sensitivity meta-analyses, which included only studies with a low overall risk of bias.

Regression analysis was conducted to explore the relationship between dependent variables and one or more independent variables. Using linear regression, we examined how variables like mean age and sex influenced the relationships found in the study. The selection of these variables was guided by theoretical considerations relevant to the analytical objectives. Jamovi software was used for the regression analysis, ensuring reliable data interpretation.

Study selection

A total of 920 studies were initially identified through database searches, including 155 from PubMed, 13 from Cochrane, 488 from Scopus, 249 from Springer, and 15 from Epistemonikos. After applying publication year and study type filters, 542 studies were excluded during the first round of screening. The remaining 378 studies were imported into the Mendeley Group Reference Manager for further review, applying the inclusion criteria.

Of these, 320 studies were excluded for being off-topic, and 5 were removed due to duplication. The remaining 53 studies were manually screened by the authors, leading to the exclusion of 48 studies for various reasons: 25 had the wrong study design, 15 reported the wrong outcome, 1 used the wrong intervention, 1 had an incorrect comparator, and 4 focused on the wrong procedural indication.

The full texts of the remaining seven studies were then thoroughly assessed, and all seven were deemed eligible for inclusion. The research selection process is visually summarized in the PRISMA flow chart (Figure 1). The eligibility of these seven studies was confirmed using the Cochrane ROB-2 tool, with all passing the bias assessment. The PRISMA flow chart provides a detailed account of the research selection process.

Figure 1. Prisma flow chart.

Study characteristic

Table 1 provides a summary of seven studies that explore different interventions for managing residual limb pain, phantom limb pain, and limb function in amputees. These studies originate primarily from the USA and the UK and include a mix of retrospective cohort studies, randomized clinical trials, and clinical trials. Sample sizes range from smaller cohorts of 8 participants to larger groups of up to 100, with both male and female subjects included. The interventions evaluated across these studies include Targeted Muscle Reinnervation (TMR), nerve insertion techniques, and Agonist-antagonist Myoneural Interface (AMI).

The age of participants varied, with mean ages ranging from the mid-40s to late 50s. The outcomes consistently suggest that TMR and other nerve-related interventions offer promising improvements in reducing pain and enhancing functional outcomes in amputees. However, the effectiveness of these interventions may differ depending on the specific conditions and methodologies used in each study. The diverse mix of study designs and sample populations offers a broad and comprehensive view of current strategies for addressing post-amputation pain and improving function.

Risk of bias in studies

All included trials in this study undergo a detailed evaluation with the ROB-2 risk of bias tool. Three studies raised concerns across multiple domains, particularly due to unclear randomization procedures, missing outcome data, and deviations from the intended interventions. The results of the risk-of-bias assessment are visually presented in Figures 1, 2, and 3, highlighting the methodological strengths and weaknesses identified in the reviewed studies.

Figure 2. Risk of bias of randomized studies.

Figure 3. Risk of bias of non-randomized intervention studies.

The risk of bias for all included studies was assessed using the Revised Tool for Risk of Bias in Randomized Trials (RoB-2.0) for randomized studies and the Risk Of Bias In Non-randomised Studies - of Interventions (ROBINS-I) for non-randomized studies. The findings indicated that all randomized and non-randomized interventional studies had a low risk of bias, while 50% of the non-randomized exposure studies were also classified as having a low risk of bias.

The bias assessment using the RoB-2.0 tool revealed that the randomized controlled trials (RCTs) were rated “low” across all five evaluated domains. The detailed assessment results are displayed in Figure 2.

The bias risk assessment using the ROBINS-I tool revealed that the non-randomized controlled trial (non-RCT) was classified as having a “low” risk of bias. However, among the six non-randomized exposure studies, three were found to have a “some concern” risk of bias. Specifically, the study by Lauren et al. (2020) raised concerns in domain 7. The study by O’Brien et al. (2019) had bias issues in domains 1 and 5, while Lu V et al. (2023) showed bias in domains 1, 2, 5, and 6. The remaining three non-randomized exposure studies were deemed to have a “low” risk of bias.

The bias assessment outcomes are depicted in Figures 3 and 4, providing a visual representation of the results.

Figure 4. Risk of bias of non-randomized exposure studies.

VAS score

Four studies, involving a total of 121 participants, reported outcomes using the Visual Analog Scale (VAS). As shown in Figure 5, the intervention group exhibited a statistically significant improvement, with a pooled mean difference (MD) of -1.74 [95% CI: -2.46 to -1.02, P<0.00001] compared to the control group. The analysis also revealed significant heterogeneity, although the funnel plot in Figure 6 does not show evidence of heterogeneity.

Figure 5. Forest plot of VAS score.

Figure 6. Funnel plot of VAS score.

No-phantom pain incidents

Five studies, involving a total of 346 participants, assessed outcomes related to the absence of phantom pain, which is described as pain that feels like it originates from a body part that is no longer present. As shown in Figure 7, the intervention group had a higher, though not statistically significant, proportion of patients without phantom pain after surgery compared to the control group, with a pooled risk ratio (RR) of 1.58 [95% CI: 0.61 to 4.11, P=0.35]. The analysis also revealed significant heterogeneity, as evidenced by the funnel plot in Figure 8, which indicates true heterogeneity among the studies.

Figure 7. Forest plot of phantom pain.

Figure 8. Funnel plot of phantom pain.

Meta-regression

The meta-regression data provided examines the influence of several predictors (age, sample size, gender) on two outcome variables: Visual Analog Scale (VAS) scores, which measure pain intensity, and the incidence of phantom pain. Here is an interpretation of the results for each predictor within these outcome variables.

VAS score

The coefficient for age is negative, suggesting that an increase in age might be associated with a slight decrease in VAS scores, indicating lower pain levels. However, the p-value of 0.581 indicates this relationship is not statistically significant. The wide confidence interval further suggests considerable variability, indicating that age may not be a reliable predictor of VAS scores in this context.

The negative coefficient implies that larger sample sizes might be associated with slightly lower VAS scores. Nevertheless, the p-value of 0.645 shows this result is not statistically significant, and the confidence interval includes zero, further confirming the lack of a significant effect.

The positive coefficient suggests that males might report slightly higher VAS scores than females. However, with a p-value of 0.193, this association is not statistically significant. The confidence interval also includes zero, indicating the effect is not robust.

For the female gender, the coefficient is negative, implying that females might report lower VAS scores than males. Still, the p-value of 0.129 is not statistically significant. The confidence interval includes zero, suggesting this effect is also not substantial.

No phantom pain incidence

The coefficient is close to zero, indicating the minimal impact of age on phantom pain incidence. The high p-value of 0.876 and the confidence interval spanning zero indicate no significant relationship between age and the likelihood of experiencing phantom pain.

The coefficient for sample size is also very close to zero, with a p-value of 0.160, suggesting no significant association between sample size and phantom pain incidence. The confidence interval includes zero, further supporting this lack of effect.

The coefficient is nearly zero, with a p-value of 0.772, indicating no significant effect of male gender on phantom pain incidence. The confidence interval also supports this non-significance.

The coefficient is negative, suggesting females might have a lower incidence of phantom pain compared to males. The p-value of 0.051 is close to the traditional threshold for significance (0.05), hinting at a potential effect, but it is not definitively significant. The confidence interval almost excludes zero, suggesting a possible trend worth further investigation. The summary of the meta-regression is available in Table 2 and The Q-Q plot of the meta-regression can be seen in below Figure 9 and Figure 10.

Figure 9. Q-Q for meta-regression of VAS Score.

Meta-regression analyzes the effect of Age (A), sample size (B), male gender (C), and female gender (D) on the VAS Score.

Figure 10. Q-Q Plot for meta-regression of no-phantom pain incidence.

Meta-regression analyzes the effect of Age (A), sample size (B), male Gender (C), and female gender (D) on the VAS Score.

Preview

In Targeted Muscle Reinnervation (TMR) treatment, severely damaged nerves are rerouted to new, redundant muscles that are no longer needed for their original function after an amputation. This procedure successfully reinnervates the muscles, allowing them to respond towards the transferred nerves neural signals transmission.¹³ Once reinnervated, these muscles contract and produce EMG signals in response to neural commands targeted directly towards the missing limb. For instance, if a user tries to plantar flexion, the transferred nerve stimulates the target muscle in the pedis to transmit EMG signals that control the prosthetic foot.^14,15

Neural control signals for movements such as knee and foot flexion are carried through nerves severed by amputation. Without muscle effectors, this vital neural information cannot be used to control prosthetics. The primary aim of Targeted Muscle Reinnervation (TMR) is to create control sites for four key prosthetic functions: dorsal flexion and plantar flexion of the foot, as well as flexion and extension of the knee joint. This is accomplished through innovative nerve transfers to establish new control sites and preserve existing ones for the appropriate prosthetic functions.¹⁶ Although prosthetists can typically address these issues with modifications to the prosthesis, consulting a plastic surgeon may offer a more durable solution for soft tissue problems.¹⁷

Several options are available for improving soft tissue conditions. Rearrangement of local tissue techniques, such as rotation flaps, advancement flaps, or Z-plasty, could free scar contractures. During TMR, proximally based adipofascial flaps are used to expose nerves and target muscles of the upper extremity or chest. This not only thins the tissue around the subcutaneous adipose, enhancing the EMG signal detected on the skin, but also helps place these flaps to isolate individual signals between target muscle segments. Meanwhile, direct lipectomy of circumferential lipofaction is able to reduce the amount of subcutaneous tissue in patients with excessive subcutaneous fat that dampens EMG signals and affects socket fit.¹⁷

Clinical outcome

VAS score

The Visual Analog Scale (VAS) is widely used to gauge pain intensity. In this meta-analysis, four studies involving 121 participants were reviewed to assess the impact of an intervention on VAS scores. As shown in the forest plot (Figure 9), the intervention group demonstrated a significant improvement in VAS scores, with a pooled mean difference (MD) of -1.74 [95% CI: -2.46 to -1.02, P<0.00001] compared to the control group. This negative mean difference indicates a notable reduction in pain intensity due to the intervention.

The significance of the results (P<0.00001) highlights the effectiveness of the intervention in reducing pain, as measured by the VAS. The confidence interval (CI) of -2.46 to -1.02 further reinforces the precision of this estimate, with all values suggesting a decrease in pain. The narrow CI boosts confidence in the intervention’s effectiveness.

Regarding heterogeneity, the funnel plot (Figure 10) indicates no significant heterogeneity among the studies. This suggests that the intervention’s effect on VAS scores is consistent across the studies, adding robustness to the findings. The uniformity across studies is important for generalizing the results, implying that the intervention may be broadly effective in various settings and populations with similar outcomes expected.

Phantom pain incidents

Phantom pain, the sensation of pain in a missing limb, is a major issue following amputation. This meta-analysis examined five studies involving 346 participants to evaluate the effect of an intervention on the occurrence of phantom pain. As illustrated in the forest plot (Figure 6), the intervention group had a non-statistically significant higher number of patients without phantom pain compared to the control group, with a pooled risk ratio (RR) of 1.58 [95% CI: 0.61 to 4.11, P=0.35].

The risk ratio of 1.58 suggests that those in the intervention group are 58% more likely to be free from phantom pain compared to the control group. However, this result is not statistically significant (P=0.35), and the wide confidence interval (0.61 to 4.11) indicates considerable variability in the effect estimate, which reduces the precision and reliability of the intervention’s impact on phantom pain.

The funnel plot (Figure 8) indicates significant heterogeneity among the studies. This variability suggests that the intervention’s effect on phantom pain may differ across studies due to differences in study design, participant characteristics, intervention types, or other factors. Such heterogeneity makes it challenging to interpret and generalize the findings consistently.

Clinical application and procedure

Included studies have highlighted various clinical applications of Targeted Muscular Reinnervation (TMR). TMR can be applied to patients with infections, trauma, or oncologic issues in limbs.¹⁸ Additionally, TMR has been shown to reduce the need for opioid consumption in amputees.¹⁹

TMR addresses chronic residual limb pain and phantom pain by transferring a transected proximal nerve stump to a muscle that is denervated surgically, improving possibility of reinnervation.⁷ This approach focuses on nerves that are relocated into specific muscle segments to produce distinct electromyographic (EMG) signals, enhancing control over myoelectric prostheses. The goal of TMR is to provide a substantial amount of healthy recovering axons to downstream motor muscle units, resulting in clear and detectable surface EMG signals. Large proximal recipient motor nerves are utilized to allow for formal neurorrhaphy, thus defining the denervation and reinnervation of targeted muscle segments.^20–22

Strength and Limitation

The use of TMR for scc pedis has not been mentioned in many publications and scientific articles, which can provide more information regarding the treatment of scc pedis with TMR techniques.

Registration

On August 5, 2024, this systematic review and meta-analysis was registered with the Open Science Framework (OSF) under the title, “Targeted Muscle Reinnervation (TMR) significantly reduces phantom limb pain and improves functional outcomes in below-knee amputees, including those with squamous cell carcinoma (SCC) of the pedis, by providing better control over prosthetic devices and enhancing overall quality of life” (https://doi.org/10.17605/OSF.IO/FG5B7).