Is Nerve Electrophysiology a Robust Primary Endpoint in Clinical Trials of Treatments for Diabetic Peripheral Neuropathy?

There is currently no FDA-approved disease-modifying therapy for diabetic peripheral neuropathy (DPN). Nerve conduction velocity (NCV) is an established primary endpoint of disease-modifying therapies in DPN and clinical trials have been powered with an assumed decline of 0.5 m/s/year. This paper sought to establish the time-dependent change in NCV associated with a placebo, compared to that observed in the active intervention group. A literature search identified twenty-one double-blind, randomised controlled trials in DPN of ≥1 year duration conducted between 1971 and 2021. We evaluated changes in neurophysiology, with a focus on peroneal motor and sural sensory NCV and amplitude in the placebo and treatment groups. There was significant variability in the change and direction of change (reduction/increase) in NCV in the placebo arm, as well as variability influenced by the anatomical site of neurophysiological measurement within a given clinical trial. A critical re-evaluation of efficacy trials should consider placebo-adjusted effects and present the placebo-subtracted change in NCV rather than assume a universal annual decline of 0.5 m/s/year. Importantly, endpoints such as corneal confocal microscopy (CCM) have demonstrated early nerve repair, whilst symptoms and NCV have not changed, and should thus be considered as a viable alternative.


Introduction
The prevalence of diabetic peripheral neuropathy (DPN) is estimated to be~50% but may be higher depending on the duration and type of diabetes [1]. Around 30% of people with diabetes experience symptoms which may be positive (tingling, burning) and/or negative (numbness) [2,3], affecting the lower limb [4]. Although the optimization of glycaemic control and cardiovascular risk factors is recommended to limit the development or progression of diabetic neuropathy [5][6][7], regrettably, there are no currently approved (US Food and Drug Administration (FDA), or European Medicines Agency (EMA)) diseasemodifying treatments for DPN [8].
The efficacy of several pathogenetic treatments, including aldose reductase inhibitors, antioxidants, and protein kinase C inhibitors, has been assessed in clinical trials over the past four decades [9]. All have failed in phase III clinical trials with none approved as treatments for DPN by the either the FDA or EMA [10,11]. Numerous reasons have been cited for these failures, including a lack of drug efficacy, trial duration and inappropriate endpoints. Composite clinical scores and quantitative sensory testing rely on patient responses and are deemed to be subjective [12] and prone to high variability. The results of neurophysiology-based assessments continue to be endorsed as a primary endpoint in clinical trials of DPN [13][14][15], despite a high inter-observer variability [16]. Most clinical trials are powered assuming a nerve conduction velocity (NCV) decline of 0.5 m/s/year in the placebo group [17,18]. However, the overall improvement in glycaemic control (with safer, more effective glucose lowering therapies) and a wider use of ACE inhibitors and statins over the last two decade will have modified the natural history of DPN. Thus, the application of this (exaggerated) rate of decline could lead to false interpretations in treatment effects and rejection of the approval of effective treatments for DPN. To this point, Dyck et al. [19] showed an enhanced placebo effect on symptoms and signs and much slower NCV reduction when assessing the natural history of DPN in the placebo arms of two large clinical trials and the Rochester DPN epidemiological study [19,20].
To objectively address the issue of the variable rate of decline in NCV, we have evaluated the change in NCV in the placebo and treatment arms of trials of pathogenetic treatments for DPN with a duration ≥12 months.

Methods
Electronic literature searches of the MEDLINE database were performed for trials of pathogenetic treatments in diabetic neuropathy. The searches were restricted to the English language from January 1971 to July 2021.
All search results were combined using Endnote X7.8 and duplicates removed. Reference lists of the primary and secondary literature were browsed to identify additional studies and hand searching of reference lists were also performed.
Peroneal motor NCV and sural sensory NCV and amplitude were extracted to an electronic spreadsheet and the change analysed in the placebo and treatment arms.

Inclusion and Exclusion Criteria
Studies were included in the final evaluation based on an a priori inclusion criteria. For inclusion, studies were required: (1) to be a double-blind, randomised controlled trials, (2) to be a minimum of one-year in duration, and (3) to have neurophysiology included as an endpoint (4) in diabetic peripheral neuropathy.
Studies were excluded if they: (1) were not a human study, or (2) not in the English language.
A.J.N. screened all articles and selected those that satisfied the inclusion criteria for full article evaluation. This was then subsequently reviewed by D.Y.A. All titles and abstracts of articles were screened to remove irrelevant studies, and the remaining shortlisted articles were screened for eligibility in depth. The full texts of relevant articles were retrieved, screened, and selected using the inclusion and exclusion criteria ( Figure 1). The original authors were not contacted for additional data. Any concerns regarding inclusion or exclusion were decided by the senior author (U.A). Before data extraction, U.A. assessed all articles to confirm their eligibility in the study. Our study partly utilises a systematic review methodology, although it was primarily designed as a standard review of the literature. A meta-analysis/summary data synthesis was not undertaken due to an a priori decision that the quality of data reporting from included studies was likely to be poor. The study or study protocol was therefore not registered with a systematic review database such as PROSPERO.
Diagnostics 2022, 12, x FOR PEER REVIEW 3 of 17 A.J.N. screened all articles and selected those that satisfied the inclusion criteria for full article evaluation. This was then subsequently reviewed by D.Y.A. All titles and abstracts of articles were screened to remove irrelevant studies, and the remaining shortlisted articles were screened for eligibility in depth. The full texts of relevant articles were retrieved, screened, and selected using the inclusion and exclusion criteria ( Figure  1). The original authors were not contacted for additional data. Any concerns regarding inclusion or exclusion were decided by the senior author (U.A). Before data extraction, U.A. assessed all articles to confirm their eligibility in the study. Our study partly utilises a systematic review methodology, although it was primarily designed as a standard review of the literature. A meta-analysis/summary data synthesis was not undertaken due to an a priori decision that the quality of data reporting from included studies was likely to be poor. The study or study protocol was therefore not registered with a systematic review database such as PROSPERO.

Results
Twenty-one studies published between 1988 and 2016 met the a priori inclusion criteria (Table 1), with an average duration of 67.9 ± 37.3 weeks (median 52 weeks; IQR 52-64.5 weeks) (placebo: n = 2395, treatment: n = 3541). The available study characteristics and demographic data are presented for each study in Table 2. The populations recruited had a heterogeneity of diabetes sub-type (type 1 and type 2 diabetes), with more partic- ipants with type 2 diabetes being enrolled in later years. A range of putative treatments were studied: aldose reductase inhibitors (ARI), protein kinase C beta (PKC-β) inhibitors, amino acid/Carnitine acetyltransferase, nerve growth factor-rhNGF, neurotrophic peptide/ACTH 4-9 analogue, antioxidants, angiotensin-converting-enzyme (ACE) inhibitors, γ-Linolenic acid (GLA) and C-peptide.
Trials evaluated between 1980 and 2009 with available NCS data (n = 19), both peroneal motor NCV and sural sensory NCV exhibited an increase in the placebo arm (+0.17 ± 0.77 m/s/year and +0.09 ± 1.31 m/s/year, respectively), while sural amplitude showed a marginal reduction (−0.11 ± 0.48 uV/year). These results are consistent with those from the most recent trial undertaken [21] which demonstrated an increase in peroneal motor NCV and sural sensory NCV (+0.3 ± 0.2 m/s/year and +1.2 ± 0.29 m/s/year) with a slight reduction in sural amplitude (−0.2 ± 0.2 mV/year) in the placebo group [21]. A further trial demonstrated a marginal reduction in peroneal motor NCV at 4 years (−0.06 ± 4.07 m/s) [22].

Protein Kinase C Beta (PKC-β) Inhibitor-Ruboxistaurin (RBX)
PKCβ is activated by hyperglycaemia and disordered fatty acid metabolism, and it is involved in the pathogenesis of endothelial damage in diabetes. Ruboxistaurin (RBX), a specific inhibitor of PKCβ, showed favourable results in animal models and in shortduration clinical trials of diabetic neuropathy [23,24].
Vinik et al. showed that neither 32 mg/d nor 64 mg/d of RBX altered the vibration detection threshold (VDT) or Neuropathy Total Symptom Score (NTSS-6) compared to placebo and electrophysiological data were not presented [27]. In patients with significant symptomatic DPN at baseline (NTSS-6 total score > 6; n = 83), there was a significant improvement in the NTSS-6 total score over 12 months in the RBX 64 mg/d treatment arm compared to placebo (p = 0.015). A subgroup of participants with clinically significant symptoms (NTSS-6 total score > 6) and less severe DPN (sural sensory nerve action potential >0.5 uV) demonstrated a significant change in the NTSS-6 total score with RBX 64 mg compared to placebo (−5.25 vs. −1.61 points; p = 0.006) [27].

Sorbinil
The Sorbinil retinopathy trial was conducted in patients with type 1 diabetes (n = 497) in the mid 1980s and found that approximately 30% of patients had worsening clinical measures of symmetric distal neuropathy in both the Sorbinil-treated and placebo arms [29]. Only 192 of 497 patients had NCS and there was no significant difference in the change in NCV or amplitude in the two groups [29]. There was also a trend for NCV to decline in the Sorbinil group, although the peroneal motor NCV improved in the Sorbinil group (+0.5 ± 4.4 m/s) and declined in the placebo group (−0.9 ± 3.5) (p = 0.02). The median motor NCV (−0.8 ± 4.1), sensory forearm NCV (−1.7 ± 7.5), sensory distal NCV (1.1 ± 7.0) and peroneal motor NCV (−0.9 ± 3.5) declined in the placebo arm [29]. A regression model demonstrated that higher baseline nerve conduction velocity and amplitudes were associated with a greater decrease over time [29].
In a two-year study of patients with severe symptomatic DPN (n = 21) [30], there was no significant change between the Sorbinil or placebo groups for 12 neurophysiological parameters [30]. However, the study had an attrition rate of 33%, due to adverse reactions and only 14 patients completed the trial (n = 8; Sorbinil, n = 6; placebo). Similarly, in a double-blind trial of Sorbinil 250 mg daily over 12 months in patients with DPN, there was no change in symptoms or neurophysiology compared to placebo and, paradoxically, neuropathic diabetic foot ulcers developed in 19% of patients in the Sorbinil group (n = 21) compared to 10% in the placebo group (n = 10) [31].

Ranirestat
Bril et al. investigated the effect of multiple doses of Ranirestat (10, 20 or 40 mg/day) and placebo over 1 year in patients with mild-to-moderate DPN [32]. There was no difference in sensory NCS vs. placebo, but there was an improvement in the summed motor NCV of the peroneal, tibial, and median nerves in the higher doses at weeks 12, 24, and 36 (p ≤ 0.05), and the peroneal NCV at 36 and 52 weeks (p ≤ 0.05) for intermediate dosing. At week 52, the summed sensory NCV (bilateral sural and proximal median sensory) increased in both the placebo (+2.0 m/s) and Ranirestat (+3.2-3.8 m/s) group, with no significant difference between groups.

Zenerastat
In a 52-week clinical trial, Zenerastat was associated with an increase in sural NCV, composite rank score (p = 0.004) and density of sural nerve myelinated fibres <5 microns [33]. Peroneal, median, and sural NCV all reduced by >0.25 m/s in the placebo group and increased by +1-1.5 m/s in the 600 mg twice-daily treatment group, with intermediate trends in the 150 and 300 mg twice-daily groups. Median NCV declined in all groups, attributed to carpal tunnel syndrome, although the decline tended to be greater in the placebo group (−1.41 ± 0.90 m/s) than in the 150, 300, and 600 mg twice-daily groups (−0.37 ± 0.61, −0.30 ± 0.67, and −0.28 ± 0.60 m/s, respectively, p = 0.085). Peroneal NCV tended to decrease, and the total myelinated fibre density in the largest fascicle also tended to decrease in the placebo and 150 mg twice-daily treatment groups with an increase in the 300 and 600 mg twice-daily groups (p = 0.105; p = 0.077 respectively).

Tolrestat
Santiago et al. [39] randomly assigned patients treated with Tolrestat for an average of 4.2 years to either placebo or continued Tolrestat therapy for 52 weeks. There was a decrease (−0.9 ± 0.2 m/s) in summed NCV (median, ulnar, peroneal, and sural nerve) in the placebo group, whereas in patients who continued on Tolrestat, MNCV did not differ significantly from baseline with changes ranging from −0.6 ± 0.4 to +0.7 ± 0.5 m/s; p ≤ 0.05). After 3 months, patients were given the choice to switch to placebo, and they showed a significant decrease in MNCV (−1.6 ± 0.6 m/s), whilst those on placebo who switched to Tolrestat showed a significant increase in NCV (p < 0.05), toe sensation and pain (p ≤ 0.005). Tolrestat was approved in 1988 and marketed in several countries but was discontinued in 1996 due to severe liver toxicity and death.

Recombinant Human Nerve Growth Factor (rhNGF)
Nerve growth factor (NGF) promotes the survival of small-fibre sensory and sympathetic neurons. A multicenter phase-three study in 1019 patients with type 1 or type 2 diabetes and DPN assessed the effects of recombinant NGF (rhNGF) 0.1 mcg/kg (n = 504) or placebo (n = 515) subcutaneously three times per week over 48 weeks [41]. The global symptom assessment (p = 0.03), severity of pain in the legs (p = 0.05) and six-month symptoms in the feet and legs (p = 0.003) demonstrated an improvement with rhNGF. However, sural nerve amplitude marginally decreased in the placebo (−0.1 ± 1.7 µV) and rhNGF (−0.1 ± 1.9 µV) groups, whilst the results of the median NCV and sural NCV were not reported for the study.

Angiotensin-Converting-Enzyme (ACE) Inhibitor-Trandolapril
ACE inhibitors reduce oxidative stress, improve endothelial dysfunction, and help to delay the progression of retinopathy and nephropathy. Trandolapril was compared to placebo in 41 normotensive patients with T1DM or T2DM and mild neuropathy over 12 months [45]. Peroneal MNCV increased in the Trandolapril group (38.3 ± 4.1 to 39.4 ± 4.1, p = 0.03) compared to a decrease in the placebo group (37.4 ± 4.8 to 36.6 ± 5.4, p = 0.03), and the sural nerve amplitude increased in the Trandolapril group (8.3 ± 6.4 to 10.0 ± 7.8, p = 0.04) and decreased in the placebo group (6.1 ± 6.3 to 5.6 ± 5.8). The peroneal M-wave amplitude (p = 0.03) and F-wave latency (p = 0.03) increased in the Trandolopril group with no change in vibration-perception threshold, cardiac autonomic function, neuropathy symptom or deficit score [45].

Fatty Acids: γ-Linolenic Acid (GLA)
GLA is a vital component of the microcirculation and neurons, and those with diabetes have a reduced ability to convert dietary linoleic acid to GLA, which may result in DPN [46]. In a clinical trial comparing GLA to placebo over 1 year [47], the placebo group demonstrated a significant decline in peroneal motor NCV (−1.86 ± 0.99 m/s), median sensory NCV (−2.14 ± 1.02 m/s) and sural sensory nerve amplitude (−0.96 ± 0.65 µV).

Discussion
This systematic review highlights a potential explanation for the universal failure of clinical trials assessing disease-modifying therapies in DPN to identify meaningful outcomes [48]. Our analysis identifies a major flaw in the application of NCS-based assessment as a tool and the use of NCV as a primary endpoint in clinical trials. We find no evidence to support the assumption that NCV would decrease by 0.5 m/s/year in the placebo group, which raises concern about the validity of the statistical power calculations used in the previously reported clinical trials. Indeed, NCV shows considerable variability in its magnitude and direction of change when assessing placebo-associated change. Confounding factors influencing this variability could not be addressed due to limited demographic data, missing datapoints and inconsistency in the way that endpoints were reported in these studies. Furthermore, the clinical heterogeneity of the patients incorporating differing ethnicities and mixed aetiologies of diabetes (T1D and T2D) may have further confounded outcomes. Components of the metabolic syndrome (hypertension, dyslipidaemia, obesity, etc.), inflammation and different stringencies of glycaemic control will all affect nerve conduction parameters and contribute to DPN in those with T2D [49,50].
All 21 studies utilised NCS as a primary or co-primary study endpoint for DPN. However, small fibres constitute 70-90% of peripheral nerve fibres, conveying pain and thermal sensation and regulating sweat, tissue blood flow, inflammation, and wound healing [51]. Indeed, studies have demonstrated that significant small-fibre abnormalities may exist despite normal NCV in subjects with diabetes [51]. Shabeeb et al. concluded that the best method for quantitative evaluation and diagnosis of DPN was electrophysiology-based [52]. However, the implementation of a valid diagnostic test such as nerve electrophysiology does not necessarily translate to a robust (primary) endpoint in clinical trials in assessing the impact of an intervention.
A variety of other assessment tools may be considered. For example, sural nerve biopsy studies have previously demonstrated that unmyelinated nerve fibre damage precedes myelinated nerve fibre damage in DPN [53]. However, nerve biopsy is an invasive procedure requiring a specialised laboratory and expertise for quantification, which restricts its use in clinical trials of DPN [54]. In contrast, skin biopsy is a minimally invasive procedure which allows quantification of intra-epidermal nerve fibres [55,56] and the intra-and inter-observer variability for the evaluation of intra-epidermal nerve fibre density shows good agreement [56,57], and an international investigator consortium collated a normative database of intra-epidermal nerve fibre density [58]. Skin biopsy is advocated alongside an assessment of typical symptoms and sensory evaluation for the diagnosis of small-fibre neuropathy [30] and is also recommended as an endpoint in clinical trials [32]. Unfortunately, the availability of facilities to undertake skin biopsy and assess intra-epidermal nerve fibres is limited. Another technique that shows promise is corneal confocal microscopy (CCM)-a non-invasive, reproducible test which detects small nerve fibre loss in diabetic neuropathy [59,60], reliably [61] and with high sensitivity and specificity [62,63] and is comparable to intra-epidermal nerve fibre density [63,64]. CCM also predicts incident DPN [65] and detects nerve regeneration in people with DPN [66]. It has all the attributes of an ideal endpoint to identify early neuropathy, define at-risk individuals and monitor the progression or improvement of diabetic neuropathy [67]. An increasing number of interventional studies have shown that CCM can identify early nerve fibre regeneration following simultaneous pancreas and kidney transplantation [68], bariatric surgery [69], and GLP-1 [70] or insulin [71], as well as Cibinetide [72,73] and Omega-3 [74,75], which importantly precedes any improvement in symptoms, signs, and NCS. Importantly, endpoints such as CCM have demonstrated early nerve repair, whilst symptoms and NCV have not altered in interventional trials of pancreatic/kidney transplantation and Omega-3, and should thus be considered as a viable and reliable alternative [71,74,75]. Further studies are warranted to evaluate the efficacy of CCM to predict clinically meaningful improvements in people with DPN undergoing disease modifying interventions.
We acknowledge publication bias and English language bias as a limitation of this study. The clinical trials were small, generally poorly conducted, and had significant heterogeneity in their reporting and nerve conduction study protocols, a major confounding factor.

Conclusions
Nerve conduction studies have shown marked variability in the direction and magnitude of change of NCV in the placebo arm of multiple clinical trials of disease-modifying therapies for DPN; these findings are at odds with the assumed constant rate of annual decline suggested. These findings challenge the continued application of NCS-related measures as a primary endpoint in clinical trials of disease-modifying therapies for DPN, as endorsed by the FDA. This measure overlooks a growing body of evidence that suggests that small nerve fibre regeneration should be assessed in clinical trials of DPN, which may identify early nerve repair and demonstrate therapeutic efficacy of pathogenetic therapies for DPN, which, with other techniques, may erroneously be discounted. With this is mind, measures using corneal confocal microscopy, a rapid non-invasive and reiterative technique which can quantify small nerve fibre repair (thus fulfilling all FDA criteria), could act as a primary endpoint in clinical trials of DPN [76].