Safety and Outcomes of Different Surgical Techniques for Cubital Tunnel Decompression

Key Points Question For adults with primary cubital tunnel syndrome, which operation is associated with the best chance of symptomatic cure and lowest risk of complications? Findings This network meta-analysis included 30 studies comparing 8 different operations in 2894 limbs. It found that 87% of patients improve with surgery and that open in situ decompression (with or without a medial epicondylectomy) was associated with the greatest response to treatment and lowest complication risk. Meaning The findings of this study suggest that for adults with primary cubital tunnel syndrome, the most beneficial operation appears to be open in situ decompression.


Introduction
Cubital tunnel syndrome is the second most common compressive neuropathy, affecting up to 6% of the population 1 or 36 per 100 000 person-years. 2 Surgical decompression of the cubital tunnel is the most effective treatment. [3][4][5] Consequently, approximately 15 000 people across the UK 6  During the last decade, at least 15 systematic reviews and pairwise meta-analyses have failed to resolve uncertainty about the efficacy and safety of these different operations for primary cubital tunnel syndrome, [10][11][12][13][14][15][16][17][18][19][20][21][22] which is manifested in persistent variation in practice. 23 This uncertainty is important to resolve because as many as 30% of patients do not improve after surgery 24 and many are subject to revision surgery, which is rarely curative. 25 Network meta-analysis is a technique for comparing multiple treatments simultaneously by combining direct evidence from clinical studies and indirect evidence from within a network. This advanced form of meta-analysis has several distinct advantages over standard (pairwise) metaanalyses, including better precision and power, 26,27 the ability to compare interventions that have not been directly compared before (ie, in a real-life head-to-head study), and the capacity to rank competing treatments to inform clinical decisions. 28 Therefore, network meta-analysis has the potential to address some of the remaining uncertainties about the efficacy and safety associated with different operations for cubital tunnel syndrome. In this study, we aimed to rank the safety and outcomes of different techniques for adults with primary cubital tunnel decompression.

Methods
This review was registered on the PROSPERO database (CRD42019127892); it was designed and conducted in accordance with the Cochrane Handbook for Systematic Reviews of Interventions, version 5.1.0 29 and was written in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guideline 30 and the PRISMA Network Meta-analysis extension statement. 31

Study Selection
We included experimental and observational studies directly comparing the outcomes of at least 2 surgical treatments for adults (aged >16 years) with primary cubital tunnel syndrome. We excluded case reports, and when comparative studies had a subgroup with 1 participant, the single-participant subgroup was excluded. The intervention had to be 1 of the following open, minimally invasive, or endoscopic techniques: in situ decompression; in situ decompression with medial epicondylectomy; anterior subcutaneous, anterior subfascial, intramuscular, or submuscular transposition; or any combination thereof. The comparator could be sham surgery or any of the earlier mentioned techniques.

Outcomes
The primary outcome was response to treatment. In the absence of a core outcome set, symptomatic improvement was measured with a variety of well-known tools, such as the McGowan, Bishop, Dellon, Yasutaka, and Wilson-Krout classifications. All tools assess similar parameters and broadly agree in cubital tunnel syndrome. 32 They cannot be approximated to a scale, but changes after surgery (for better or worse) can be dichotomized into responders and nonresponders. We used the outcome measures in the original study to classify patients as responders or nonresponders. For example, if a patient's McGowan score improved after surgery, they were classified as a responder.
Conversely, if a patient's McGowan score did not change or worsened after surgery, then they were defined as a nonresponder (treatment failure). When multiple outcome measures were reported, the patient-reported assessment was used because this is the most meaningful approach for patients. 33 There was no minimum or maximum severity (clinical or electrodiagnostic) required for inclusion.
The secondary outcomes included short-term surgical site complications that warranted any form of medical or surgical intervention, including bleeding, infection, and wound dehiscence.
Numbness around the surgical site was not considered a complication unless it was caused by the division of a named cutaneous nerve and treated by microsurgical neurorrhaphy. Reoperation was defined as repeated surgery for any reason (eg, evacuation of a hematoma, debridement of an infected or necrotic wound, revisional surgery for recurrence) and recurrence of symptoms (as defined by the original study) after a period of symptomatic relief, whether or not additional treatment was required.

Data Extraction
Three review authors (R.G.W., T.T.G., and R.F.) extracted details of the study design, demographic characteristics, and statistics of interest. Where data were missing or unclear, the author(s) were contacted. The authors of 1 article 37 provided data on request. In 1 study, 36 2 groups with single participants were discarded. In 1 study, 38 we were unable to disaggregate the outcomes for intramuscular transposition and submuscular transposition, requiring the exclusion of these groups.
The unit of analysis was the unit reported in the study; bilateral procedures are rarely performed simultaneously, so we considered that bilateral procedures (if not otherwise stated) were performed at times sufficiently separated to be considered independent.

Risk of Bias Assessment
The risk of methodological bias was assessed by 3 authors (R.G.W., T.T.G., and R.F.) independently, using the Cochrane Risk of Bias tool 39 (for randomized trials) or ROBINS-I tool 40 (for observational studies). Assessments were displayed graphically with RevMan version 5 (Cochrane Collaboration) and the Confidence in Network Meta-Analysis 41 tool. Disagreements were resolved by discussion.

Assessing the Transitivity Assumption in Network Meta-analysis
An important concept of network meta-analysis is that all patients in a network should be equally eligible (in principle) to receive any of the treatments, a phenomenon that is typically termed jointly randomizable. This means that all patients in our networks should in principle be eligible to undergo any of the decompressive operations described. We assessed the validity of the transitivity assumption underlying the network meta-analysis conceptually by considering whether participants in the identified studies were jointly randomizable (ie, could in principle receive any of the treatments in the network) and whether the distribution of effect modifiers was similar across nodes. 27 In this case, an effect modifier is a factor that changes the effectiveness of surgery. For example, it is noted that older patients benefit less from decompressive surgery than younger patients; therefore, age might modify the effectiveness of surgery. We tested the distribution of commonly espoused effect modifiers (ie, age, sex, and symptom severity) to ensure that they were balanced and thus our estimates were not confounded.

Statistical Analysis
We produced a network plot to summarize the treatments followed by a series of frequentist, random-effects, network meta-analyses, using the netmeta package in R version 3 (R Project for Statistical Computing) 42 and assuming a single heterogeneity parameter. To assess the agreement between randomized and nonrandomized evidence, we first performed separate network metaanalyses and compared the results. 43 Because no important discrepancies were observed, we performed a joint analysis that included both study types (so-called naive network meta-analysis).
Interventions were ranked by their P scores 44 with the netrank function; P scores are assumed to take a value between 0 and 1, with a higher score indicating a better treatment. 44 With the netleague package, network meta-analysis results are summarized in league tables and treatments ordered by their P score. Forest plots of relative risks and 95% CIs were generated with open in situ decompression as the reference treatment. Heterogeneity was quantified through the standard deviation of random effects (τ, assumed common for all comparisons in the network). Inconsistency was assessed with both global and local methods with the netsplit package 45,46 and displayed via heat plots 45,47 with the netheat command. We produced forest plots to show the relative risk and 95% CIs for the outcomes of interest, with open in situ decompression as the reference operation. To assess possible small study effects for the primary outcome, we produced a comparison adjusted funnel plot 48 in Stata version 15 (StataCorp) 48 with the netfunnel package.
Next, we performed a series of designed-adjusted analyses, 43 whereby data from randomized studies were combined with data from nonrandomized studies after down-weighting of the effect of the latter. These analyses involved a variance-inflation factor 43 (ie, an extra parameter used to increase the variance of nonrandomized studies), thus reducing their effect on the pooled network meta-analysis estimate. We used the following variance inflation factors: w = 1 (corresponding to the naive network meta-analysis [ie, including all studies at face value]), 0.8, 0.6, 0.4, 0.2, and 0 [ie, 0 excluded nonrandomized studies from the analysis]). Randomized clinical trials were not downweighted in these designed-adjusted analyses. We produced forest plots with the results of all treatments vs the reference (open in situ decompression) for all analyses to show how gradually allowing nonrandomized evidence to inform the estimates of relative treatment effects. In our designed-adjusted analyses, we did not adjust the point estimates from nonrandomized studies 43 because we could not be confident of the direction and magnitude of potential bias in the treatment effects.
Given that the secondary outcomes were rare, we used sensitivity fixed-effects Mantel-Haenszel network meta-analyses 49 (using the netmetabin package), which synthesize odds ratios; however, for rare events, odds and risks are almost identical. Inconsistency in these networks was assessed with the netsplit package and SIDDE approach. 49 The RStudio version 1.3 (R Project for Statistical Computing) package metaprop 50 was used to estimate the pooled prevalence of outcomes, using Hartung-Knapp-Sidik-Jonkman random effects and the Freeman-Tukey double arcsine transformation to stabilize the variances of proportions close to 0 or 1.
Recent publications have highlighted problems with null hypothesis testing, 51,52 particularly in network meta-analysis. 53 Therefore, we did not use the concept of statistical significance when presenting or discussing results from network meta-analyses but instead focused on the clinical interpretation in relation to the corresponding point estimates and their respective confidence intervals.

Study Selection
After review of 68 full texts, 38 articles were excluded with reasons (eFigure 1 in the Supplement), and 30 studies 36-38,54-80 describing 8 operations were included.

Risk of Bias Within Studies
The average risk of bias contributions for each comparison within the network are shown in eFigure 2 in the Supplement. The assessments of the risk of methodological bias for randomized clinical trials and nonrandomized studies are shown in eFigure 3 and eFigure 4 in the Supplement, respectively.

Assessment of Transitivity
After grouping the studies by treatment comparison and inspecting the distribution of possible effect modifiers, there were no significant differences between the demographic characteristics or preoperative McGowan grades for all treatments (eTable 2 in the Supplement). Therefore, they were judged to be sufficiently similar to be jointly synthesized in a network meta-analysis.

JAMA Network Open | Surgery
Safety and Outcomes of Different Surgical Techniques for Cubital Tunnel Decompression nonrandomized evidence. This was further corroborated after testing for differences between the 2 estimates for each treatment comparison (P > .05 for all χ 2 tests). Thus, there was no evidence of incompatibility between randomized and nonrandomized evidence, so we proceeded with a joint (ie, naive) analysis. The randomized and nonrandomized studies contributing to the analyses are disaggregated in eFigure 7 and eFigure 8 in the Supplement.

Response to Treatment
The network was composed of 30 studies, 36  Detailed results are shown in Table 2. There was no measurable heterogeneity (τ 2 = 0) (eTable 4 in the Supplement) or inconsistency within the network (eFigure 11 in the Supplement).  Table 3. There was no measurable heterogeneity (τ 2 = 0) (eTable 7 in the Supplement) or inconsistency within the network (eFigure 14 in the Supplement).

Recurrence
Overall, 15  ranked as the best technique with the lowest risk of recurrence. Conversely, and with 78% probability, submuscular transposition was the most hazardous operation and was associated with the highest risk of recurrence (Table 4). There was no measurable heterogeneity (τ 2 = 0.93) (eTable 8 in the Supplement) or inconsistency within the network (eFigure 16 in the Supplement).
A sensitivity fixed-effects Mantel-Haenszel network meta-analysis yielded similar findings (eTable 9 in the Supplement). There was still no measurable heterogeneity (τ 2 = 0) (eTable 10 in the Supplement) or inconsistency within the network (eFigure 17 in the Supplement).

Small-Study Effects
An adjusted funnel plot showed no evidence of small-study effects. eFigure 18 in the Supplement presents the details.

Assessing Confidence in Results From the Analyses
There was moderate confidence in the mixed evidence but low confidence in the indirect evidence. eTable 11 in the Supplement presents the details.

Discussion
This systematic review and network meta-analysis found that open in situ decompression with or without medial epicondylectomy was associated with the greatest response to treatment and the lowest risk of complications, reoperation, and recurrence. Our network meta-analysis provides a central reference point for the global evidence on cubital tunnel syndrome surgery to help inform clinician practice, training, and international guidelines.
Our findings show that in situ decompression (whether by open, endoscopic, or minimally invasive means) was associated with lower risk of complications than any form of transposition procedure for primary cubital tunnel syndrome (Table 2); furthermore, the addition of an epicondylectomy was associated with an increased probability of symptomatic cure without increasing the risks of complications. The 95% CIs around these estimates are narrow, indicating a high degree of certainty, which is corroborated by the sensitivity analysis. Clearly, selecting an operation with the highest success rate and lowest complication risk is beneficial to patients. The reduced operative time and complexity of in situ decompression 77,81 are also beneficial to surgeons.
Furthermore, health care services stand to gain from our findings because in situ decompressions are 18% to 55% less expensive than transposition procedures, 81  neuropathy. 82 Thereafter, we echo calls 13,83 for comparative studies of operative vs nonoperative treatments. There is a paucity of data on nonoperative management, 84 and we have a responsibility to inform patients about the evidence for and against all management options.

Limitations
This study has limitations. The surveillance period used in most studies is arguably insufficient to capture all cases of reoperation and recurrence because relapse typically occurs between 6 and 21 months postoperatively. 25 Therefore, our estimates may underestimate the true prevalence of recurrence, which, compounded by biases of attrition and reporting, may misrepresent the true risk of recurrence for a given procedure. As such, we recommend cautious interpretation of these outcomes.
Ideally, the analyses of response to treatment would have included nonoperative treatments, although this might violate transitivity assumptions, given that some surgeons may not accept or offer nonoperative treatment to patients with moderate or severe cubital tunnel syndrome.
Bilateral surgery was described in 6 studies, 57,66,68,70,74,75 which raises concerns about the unit of analysis 85 and makes it impossible to judge how this may have affected our network metaanalyses. Despite this, it is likely that bilateral operations were performed at times sufficiently separated to be considered independent events, and all studies that reported bilateral operations used the same procedure on both limbs.
We transformed binomial data (to pool) with the Freeman-Tukey method because it stabilizes the variances of proportions close to 0 or 1; however, this method can yield unreliable estimates when back-transformed. Similarly, we used the DerSimonian-Laird method to synthesize binomial data, and this can induce biased estimates with falsely high precision; better methods exist but are not yet available. Therefore, caution is recommended when the pooled prevalence of outcomes is interpreted.

Conclusions
Overall, the results of this study suggest that the rate of cure for patients with cubital tunnel syndrome who receive surgery is high and complications are uncommon. According to the available evidence and notwithstanding some uncertainty regarding the estimates, open in situ decompression (with or without medial epicondylectomy) appeared to be the best procedure for patients with primary cubital tunnel syndrome. We suggest that future research focus on defining the disorder and generating core outcome measures before further (necessary) comparative studies are undertaken.