A Bayesian Network Meta-Analysis and Systematic Review of Guidance Techniques in Botulinum Toxin Injections and Their Hierarchy in the Treatment of Limb Spasticity

Abstract

Accurate targeting of overactive muscles is fundamental for successful botulinum neurotoxin (BoNT) injections in the treatment of spasticity. The necessity of instrumented guidance and the superiority of one or more guidance techniques are ambiguous. Here, we sought to investigate if guided BoNT injections lead to a better clinical outcome in adults with limb spasticity compared to non-guided injections. We also aimed to elucidate the hierarchy of common guidance techniques including electromyography, electrostimulation, manual needle placement and ultrasound. To this end, we conducted a Bayesian network meta-analysis and systematic review with 245 patients using the MetaInsight software, R and the Cochrane Review Manager. Our study provided, for the first time, quantitative evidence supporting the superiority of guided BoNT injections over the non-guided ones. The hierarchy comprised ultrasound on the first level, electrostimulation on the second, electromyography on the third and manual needle placement on the last level. The difference between ultrasound and electrostimulation was minor and, thus, appropriate contextualization is essential for decision making. Taken together, guided BoNT injections based on ultrasound and electrostimulation performed by experienced practitioners lead to a better clinical outcome within the first month post-injection in adults with limb spasticity. In the present study, ultrasound performed slightly better, but large-scale trials should shed more light on which modality is superior.

Keywords: anatomical localization; botulinum neurotoxin; electromyography; electrostimulation; injections; limb spasticity; ultrasound.

Figure 1

Flow diagram for study selection.

Figure 2

(a) Rhombus-shaped network plot with four nodes corresponding to one guidance technique. The size of each node depends on the number of patients assigned to this technique. (b) Weighted network graph showing the number of studies in each edge. EMG: electromyography, ES: electrical stimulation, MNP: manual needle placement, US: ultrasound.

Figure 3

(a) The hierarchy of guidance techniques as shown in the Bayesian network meta-analysis. (b) Forest plot depicting the effect estimate with the corresponding confidence intervals for each guidance technique when compared to the non-guided approach. The guidance techniques are ranked from the worst at the top to the best at the bottom of the plot. (c) Ranking table showing the probability of each technique being ranked as first, second, third and fourth as well as the SUCRA values. Higher SUCRA values indicate a higher probability to be ranked as the best technique. EMG: electromyography, ES: electrical stimulation, MNP: manual needle placement, SUCRA: surface under the cumulative ranking curve, US: ultrasound, 95% CrI: credible interval.

Figure 4

Main ranking results with two different plot types: (a) The Litmus Rank-O-Gram shows the cumulative probability for each guidance technique to be ranked as first, second, third and fourth alongside with the SUCRA values. (b) Radial SUCRA plot displaying SUCRA values for each treatment radially with a network diagram of evidence overlaid. EMG: electromyography, ES: electrical stimulation, MNP: manual needle placement, SUCRA: surface under the cumulative ranking curve, US: ultrasound.

Figure 5

(a) The NMA/UME residual deviance plot displays the contribution of each data point (study arm) to the residual deviance for the NMA consistency and the UME inconsistency models. If the data points are located on the line of equality, there is no inconsistency, because the model fit does not improve if the UME inconsistency model is applied. Points below the equality line have a better fit for the UME inconsistency model. On the contrary, points above the equality line have a better fit for the NMA consistency model. Hence, data points above or on the equality line have a smaller residual deviance from the NMA consistency model and there is no proof of inconsistency. In our case, all study arms were above or on the equality line, meaning that there was not inconsistency. (b) The stem plot visualizes the posterior residual deviance of each study arm. The shorter the stem, the smaller the residual deviance and thus, the better the model fit. In total, there are 15 study arms and in all cases posterior residual deviance is lower than 1, pointing to a good model fit. (c) The leverage plot is used to evaluate the influence of each data point to the model fit and DIC. The average leverage is depicted on the y-axis and the residual deviance is on the x-axis. There are parabolas characterized by a number, represented by c. Points that lie on each parabola or in-between contribute an amount of c to the model estimation. Points lying outside the curve with c = 3 contribute to a poor model fit. In our study, all study arms lie below the parabola with c = 1.5 and contribute to a good model fitness. (d) Gelman convergence assessment plots for the parameters d.EMG.ES, d.EMG.US and d.EMG.MNP and d.sd. Four Markov chains were used to compute the scale reduction factor. In each case, there were 25,000 iterations and convergence was achieved at 5000 iterations approximately. Variance between and within the chains was almost identical, in other words there was a satisfactory convergence of the Markov chains. The chain steps were stable over the time apart from the d.sd parameter in which they exhibited small divergence. Abbreviations: EMG: electromyography, ES: electrical stimulation, MNP: manual needle placement, NMA: network meta-analysis, sd: standard deviation, UME: unrelated mean effect, US: ultrasound.

Figure 6

Heatmap showing the results of between-group comparisons in each study across all utilized assessment scales. Superiority of a guidance technique is labeled as red. Orange codifies the second-best technique in three-arm trials. Yellow designates a technique which is the worst in threarm or two-arm studies. When both techniques are colored with yellow, they are equal and there was no difference in their efficacy. Techniques that were not evaluated in this study are marked with green. It becomes apparent that US has more superiority “hits” across different spasticity scales compared to the other guidance techniques but not in non-spasticity scales. Abbreviations: AS: Ashworth Scale, DAS: Disability Assessment Scale, EMG: electromyography, EQ-5D: Quality-of-Life Scale, ES: electrical stimulation, MAS: Modified Ashworth Scale, MNP: manual needle placement, PROM: ankle passive dorsiflexion range of motion, RMS-EMG: root mean square of surface electromyographic activity during the Ashworth maneuver, TCA: Tardieu catch angle, TSA: Tardieu spasticity angle, US: ultrasound, 10 mWT: 10 m walking test. * In this study the original Ashworth scale was used. Since the summary measure was the difference between the means pre- and post-injection and not the mean estimates per se, the difference by one level between the original and the modified scale was counterbalanced. ** TCA (Tardieu catch angle) is the same as TSA (Tardieu spasticity angle) [52,53,54,55,56,57].

Figure 7

Frequentist network meta-analysis results: (a) Forest plot showing the output of comparisons between each guidance technique and non-guided injections. Ultrasound and electrostimulation appear to be equivalent. (b) League table showing the results of pairwise comparisons as mean difference and the corresponding credible intervals. The four techniques are ranked from the top to the bottom of the diagonal (from the best to the worst). Of note, ultrasound is again depicted at the top of the diagonal, although the statistical estimates within the frequentist framework point to an equality of ultrasound and electrical stimulation. Abbreviations: EMG: electromyography, ES: electrical stimulation, MNP: manual needle placement, US: ultrasound.