Comparative efficacy of growth factor therapy in healing diabetes‐related foot ulcers: A network meta‐analysis of randomized controlled trials

Abstract Introduction This study examined the relative efficacy of growth factor therapies in healing diabetes‐related foot ulcers (DFU). Methods PubMed and Cochrane databases were searched for randomized controlled trials testing growth factor therapies for treating DFU. The primary outcome was complete wound closure. Results were reported as relative risk (RR) ± 95% credible intervals (CrI). The risk of bias was assessed using Cochrane's RoB‐2 tool. Results A total of 31 RCTs involving 2174 participants were included. Only 13 of the trials (n = 924) reported on the aetiology of the ulcers (85.4% neuropathic and 14.6% ischaemic). Epidermal growth factor (RR 3.83; 95% CrI 1.81, 9.10), plasma‐rich protein (PRP) (RR 3.36; 95% CrI 1.66, 8.03) and platelet‐derived growth factor (PDGF) (RR 2.47; 95% CrI 1.23, 5.17) significantly improved the likelihood of complete ulcer healing compared to control. Sub‐analyses suggested that PRP (3 trials ‐ RR 9.69; 95% CrI 1.37, 103.37) and PDGF (6 trials ‐ RR 2.22; 95% CrI 1.12, 5.19) significantly improved the likelihood of wound closure amongst trial mainly recruiting participants with neuropathic ulcers. Eleven trials had a low risk of bias, 9 had some concerns and 11 had a high risk of bias. Sub‐analysis of trials with a low risk of bias suggested that none of the growth factors significantly improved ulcer healing compared with control. Discussion This network meta‐analysis found low‐quality evidence that Epidermal growth factor, PRP and PDGF therapy improved DFU healing likelihood compared with control. Larger well‐designed trials are needed.


| INTRODUCTION
Diabetes-related foot ulcers (DFU) have been estimated to cause about four million years of living with disability. 1 Treatment of DFU costs approximately US$8659 per patient annually. 2 Approximately 10 billion US dollars are spent annually on the treatment of DFU. 3 About 85% of diabetes-related amputations are preceded by a DFU. 4 New treatments for DFU are urgently needed.
Growth factors play an important role in promoting wound healing and therefore have been tested as treatments for DFU. 5 Several meta-analyses have suggested that the administration of a number of different growth factors significantly improves DFU healing compared with control alone. [6][7][8][9][10] Despite these findings, The International Working Group of the Diabetic Foot has concluded that the available evidence does not support the use of growth factors in the treatment of DFU. 11,12 Prior meta-analyses have mainly focused on the comparison of two treatment strategies such as administration of a growth factor compared to control alone. Network meta-analysis (NMA) enables the comparison of multiple different treatments. One previous NMA comparing different growth factors suggested that the recombinant human epidermal growth factor (EGF) was the most beneficial growth factor for healing DFU. 13 However, seven clinical trials have been reported since the publication of this prior NMA. [14][15][16][17][18][19][20] Therefore, there is a need to perform an updated NMA to assess the evidence for different growth factors in treating DFU.
The aim of this study was to perform an up-to-date NMA of randomised clinical trials (RCT) evidence to test the relative efficacy of growth factor therapy in healing DFU in comparison to control.

| Search strategy
The systematic review and NMA was performed according to the Preferred Reporting Items for Systematic Review and Meta-Analysis with an extension for NMA statement 21 Table 1).

| Study selection
RCT testing growth factors for treating DFU were eligible for inclusion. Eligible growth factors included recombinant human platelet-derived growth factor (PDGF), autologous platelet-rich plasma (PRP), recombinant human EGF, transforming growth factor beta, recombinant human vascular endothelial growth factor (VEGF) and recombinant human fibroblast growth factor (FGF).
Control patients were defined as those receiving usual care, best or optimal medical treatment alone, or those receiving placebo. Trials were included only if the growth factor was compared with another type of growth factor or control. Trials which tested multiple different types of growth factors were included only if a control group for each intervention was available. For trials testing multiple doses of one growth factor compared to only one control group, the highest dose growth factor and control groups only were included. Trials were eligible for inclusion if the minimum data (i.e. number of patients with complete wound closure at the end of the study) were published or available from the corresponding author.
Trials were included only if all participants had diabetes and a lower limb ulcer. While it was expected that the majority of ulcers would be located in the foot, other lower limb sites were also acceptable. In this study, we defined the cut off to be at a minimum 75% of lower limb ulcers in the foot. When multiple publications arising from the same clinical trial were identified, data from the report with the longest follow-up were included. Trials published in languages other than English, non-randomized or crossover trials, observational studies and trials where complete wound closure data were not available were excluded. In addition, studies including participants with infected DFU or people with osteomyelitis were excluded. Eligibility was determined by two authors (ST and HJ), with discrepancies resolved by discussion with the senior author (JG).

| Data extraction
Study characteristics, participant risk factors, wound characteristics and primary outcome data were extracted on a customised spreadsheet independently by three authors (ST, HJ and UA). Any inconsistencies were resolved through discussion and confirmed with the senior researcher (JG). The primary outcome was complete wound closure or healing. The following additional data were extracted: Age, sex, body mass index, smoking, Hypertension (HTN), diabetes, Ankle brachial pressure index (ABPI), toe pressure, transcutaneous oxygen pressure, medications, sample size, duration of treatment and duration of follow-up. In addition, ulcer duration, grade, size and depth were also collected. Additional information from three trials was requested from the corresponding authors 22-24 of whom one author responded with the requested information. 24

| Risk of bias assessment
Two authors (HJ and UA) independently assessed the risk of bias of included trials using the RoB-2 tool, which assessed key aspects of the reporting including random sequence allocation, allocation concealment, randomisation between intervention groups, deviations 2 of 14 -THANIGAIMANI ET AL. from the intended interventions, missing outcome data, the suitability of outcome measurement, blinding of participants and outcome assessors and pre-specified analysis plan. 25 The trials were assessed as either at low risk of bias, some concerns (probably low risk of bias), or high risk of bias based on these aspects as per the RoB2 tool. Any inconsistencies were resolved through discussion between the authors until a consensus was reached. No studies were excluded on the basis of risk of bias.

| Data analysis
The Bayesian random-effects NMA was performed using the R statistical package "BUGSnet", which uses the arm based model to assess the geometry of the treatment network and provides effect size estimates for multiple comparisons. 26 The package develops a random effects model with Bayesian hierarchy using Markov Chain Monte Carlo (MCMC) simulation. 27 MCMC simulations were run using three chains with different initial values for 100,000 iterations.
The model assumes consistency between trials and follows a noninformative uniform distribution and a weakly informative prior distribution with a variance scaling factor of 2.5. The priors were calculated using the package without any user input using justifications made previously, similar to those used in other NMA packages such as 'GeMTC'. 28 The convergence of the resulting model was assessed using the league plots. The fit of the model was assessed by producing leverage plots and a better fit model was selected based on the lowest posterior mean of residual deviance (D res ). D res is the magnitude of the difference between observed data and that predicted by the model. In a well fit model, D res should be closer to the number of data points (one data point per arm). Inconsistencies within the network model were explored using leverage plots that compare the posterior mean deviance of each data point between the consistency and the inconsistency models. The results of the NMA were reported as relative risk (RR) � 95% credible intervals (CrI) using the league tables, odds ratio relative to control using forest plots and surface under the cumulative ranking curve (SUCRA) plots.
A minimum of 5 trials with a similar method of measurement and reporting was required for any meta-analysis. Sub-analyses were performed focused on studies that included participants with largely neuropathic DFUs (85% or above) and trials deemed to be at low risk of bias. A p-value of ≤0.05 was considered statistically significant.
None of the included trials with mixed aetiology DFU reported complete wound healing of neuropathic and ischaemic ulcers separately. The control provided to control participants in each trial varied as tabulated in Table 2.

| Network models
The network plot suggested that the comparison of the primary outcome at different time points was feasible as model convergence  Although there were no significant differences in outcomes between the different growth factor therapies, a SUCRA plot suggested that EGF treatment had the best outcome followed by PRP and PDGF ( Figure 4). F I G U R E 1 Preferred reporting items for systematic reviews and meta-analyses flow diagram. A total of 1013 publications were screened and after the exclusion of irrelevant studies, 31 publications were included.

| Effect of growth factor therapies on ulcer healing
T A B L E 1 Baseline characteristics of participants in the included studies.
None of the other growth factor therapies significantly improved the likelihood of ulcer healing in comparison to control (Supplementary Figures 4 and 5). Leverage plots suggested that the NMA model in the sub-analysis was well fit (Supplementary Figure 6).  T A B L E 3 Risk of bias assessed using the revised Cochrane risk-of-bias tool for randomized trials (RoB 2). ischaemic ulcers were identified in the included trials, it was not possible to perform a sub-analysis focused on ischaemic ulcers. This is a major limitation of the included RCTs as ischaemic ulcers make up the majority of chronic non-healing DFUs and are much more likely than neuropathic ulcers to be complicated by major amputations, mortality and other complications. 54 Furthermore, a subanalysis restricted to high-quality trials found that none of the growth factor therapies significantly improved the likelihood of ulcer healing.

Randomisation process
Our results are in agreement with a smaller previous NMA, which similarly reported that PRP, EGF and PDGF significantly improved the likelihood of ulcer healing compared to control. 13 Both NMAs suggest that there was no significant difference in the efficacy of the three growth factor therapies tested. Similar to the previous study which reported that EGF was likely to have the best efficacy, our results were in agreement with EGF having the highest probability of being the best treatment for DFU. 13 Our results are more updated in comparison to the previous NMA, which included four trials investigating PRP in 113 participants, whereas the current NMA included 10 trials with 368 participants treated with PRP. 13 Improved wound closure using PRP 10,55 or EGF 56 compared to control has been reported in previous conventional meta-analyses of RCTs as well.
A number of limitations of this NMA and the included RCTs should be noted. The follow up of included participants was relatively short (<24 weeks) and no information on ulcer recurrence was reported. All trials included a small sample size with heterogeneity in terms of ulcer size and duration. This is important as larger and longer term ulcers are less likely to heal 57 and more likely to be complicated by amputation 58 and death. 59 Amputation and mortality data were rarely reported and none of the studies reported information about medications. Furthermore, there was no information reported on modifiable risk factor control, which is an important determinant of outcome. 60,61 One trial did not report the number of participants who had to be derived from the reported percentage of F I G U R E 2 Network plot of all comparisons in the included trials. EGF, Epidermal growth factor; FGF, Fibroblast growth factor; PDGF, Platelet-derived growth factor; PRP, plasma-rich protein, and VEGF, vascular endothelial growth factor.

F I G U R E 3
Forest plot showing odds ratio of difference in wound closure of growth factors relative to control. EGF, Epidermal growth factor; FGF, Fibroblast growth factor; PDGF, Platelet-derived growth factor; PRP, plasma-rich protein, and VEGF, vascular endothelial growth factor. THANIGAIMANI ET AL.
-11 of 14 healed ulcers in relation to the total number of ulcers treated. 23 One trial crossed over the patients between control and treatment groups after 2 weeks based on responders. 46 Most importantly, it is rather surprising that none of the trials reported the outcomes for neuropathic and ischaemic DFU independently. Ischaemic ulcers are much less likely to heal and more likely to precipitate major amputation. 54 For all these reasons, the beneficial effects of growth factors reported in this NMA may not be generalisable to routine clinical practice. 62 Furthermore, a sub-analysis of trials deemed to be of low risk of bias suggested that none of the included growth factor therapies were effective in increasing the likelihood of ulcer healing compared to control. Larger well-designed high-quality trials are needed for better evaluation of the value of growth factors over control for treating non-healing ulcers. These should include costeffective analyses.
In conclusion, this study suggests that EGF, PRP and PDGF treatments are more effective in healing DFU than control, but these findings could not be replicated in a sub-analysis restricted to highquality trials or those including ischaemic ulcers.

AUTHOR CONTRIBUTIONS
Shivshankar Thanigaimani was involved in developing the search terms, screening the studies, data extraction, analysis, writing the initial draft and finalising the manuscript. Harry Jin and Usama Ahmad were involved in data extraction and manuscript editing.
Jonathan Golledge was involved in conceiving the study, developing the search terms, manuscript writing, critical review of the manuscript and funding acquisition.

F I G U R E 4
Surface under the cumulative ranking curve (SUCRA) plot showing the probability of the best treatment strategy for attaining wound closure in patients with diabetes-related foot ulcer within the included trials. EGF, Epidermal growth factor; FGF, Fibroblast growth factor; PDGF, Platelet-derived growth factor; PRP, plasma-rich protein, and VEGF, vascular endothelial growth factor.