Aldose reductase inhibitors for the treatment of diabetic polyneuropathy

Abstract

Background

Polyneuropathy, a common complication of diabetes mellitus, causes pain and sensory and motor deficits in the limbs, and is also an important independent predictor of foot ulceration. Inhibiting the metabolism of glucose by the polyol pathway using aldose reductase inhibitors is a potential mechanism to slow or reverse the neuropathy's progression.

Objectives

To assess the effects of aldose reductase inhibitors on the progression of symptoms, signs or functional disability in diabetic polyneuropathy.

Search methods

We searched the Cochrane Neuromuscular Disease Group Trials Register, MEDLINE (from January 1966 to May 2007), EMBASE (from January 1980 to May 2007) and LILACS (from 1982 to May 2007). We reviewed bibliographies of randomized trials identified, and contacted authors and experts in the field.

Selection criteria

We included randomized controlled trials comparing an aldose reductase inhibitor with control, and lasting at least six months. 
 The primary outcome measure was change in neurological function, measured in various ways, including strength testing, sensory examination, and composite scores of neurological examination. Secondary outcome measures were nerve conduction studies, neuropathic symptoms, quality of life, occurrence of foot ulcers and adverse effects.

Data collection and analysis

Trials included in the review were selected and assessed independently by at least two of us. Methodological criteria and study results were recorded on data extraction forms.

Main results

Thirty‐two randomized controlled trials meeting the inclusion criteria were identified. Many had significant methodological flaws. Change in neurological function, our primary outcome measure, was assessed in 29 trials, but sufficient data for meta‐analysis were only available in 13 studies, involving 879 treated participants and 909 controls. There was no overall significant difference between the treated and control groups (SMD ‐0.25, 95% CI ‐0.56 to 0.05), although one subgroup analysis (four trials using tolrestat) favored treatment. A benefit for neuropathic symptoms was suggested by a group of trials using a dichotomized endpoint (improvement or not), but this was contradicted by another group of trials which measured symptoms on a continuous scale. There was no overall benefit on nerve conduction parameters (27 studies) or foot ulceration (one study). Quality of life was not assessed in any of the studies. While most adverse events were infrequent and minor, three compounds had dose limiting adverse events that lead to their withdrawal from human use: severe hypersensitivity reactions with sorbinil, elevation of creatinine with zenarestat, and alteration of liver function with tolrestat.

Authors' conclusions

We found no statistically significant difference between aldose reductase inhibitors and placebo in the treatment of diabetic polyneuropathy. Any future clinical trials of aldose reductase inhibitors should be restricted to compounds proven to have substantial biological or preclinical advantages over previously tested agents.

Plain language summary

Aldose reductase inhibitors for the treatment of diabetic polyneuropathy

Polyneuropathy is a common complication of diabetes mellitus that causes pain and sensory and motor deficits in the arms and legs. It can also lead to foot ulcers and amputation. Aldose reductase inhibitors are a class of medications that block the breakdown of glucose by a specific metabolic pathway called the polyol pathway, and may potentially slow or reverse progression of neuropathy. The authors reviewed the results of all randomized trials that compared an aldose reductase inhibitor with a control and lasted at least six months. Many of the 32 randomized controlled trials identified had significant methodological flaws. The trials used a variety of measures to look for a benefit of treatment with aldose reductase inhibitors. The authors elected to focus primarily on changes in muscle strength and sensation. These were chosen because they are thought be the best indicator of the severity of polyneuropathy, and they have been used in a previous landmark study of the effects of intensive blood sugar control on diabetic neuropathy, as well as in studies of treatments in other types of polyneuropathy. Muscle strength or sensation were assessed in 29 trials, but sufficient data for analysis was only available in 13 studies, involving 879 treated participants and 906 controls. There was no overall significant difference between the treated and control groups. For one drug, tolrestat, there was possibly some benefit, but concerns about liver toxicity have lead to withdrawal of tolrestat from use in humans. A few trials did report that symptoms of neuropathy improved for the treated group, but this was contradicted by most other trials. No benefit was detected on electromyography (EMG) parameters (27 studies) or foot ulceration (one study). Quality of life was not assessed in any of the studies. Adverse effects were infrequent and were mostly minor, except for severe allergic reactions with sorbinil, impaired kidney function with zenarestat, and alteration of liver function with tolrestat. The authors concluded that there was no significant benefit of treatment with aldose reductase inhibitors for diabetic polyneuropathy.

Background

Diabetes mellitus is a metabolic disorder resulting from a defect in insulin secretion, insulin action, or both. A consequence of this is chronic hyperglycemia (elevated levels of blood glucose) with disturbances of carbohydrate, fat and protein metabolism. Long‐term complications of diabetes mellitus include retinopathy, nephropathy, neuropathy and increased risk of cardiovascular disease. For a detailed overview of diabetes mellitus, please see under 'Additional information' in the information on the Metabolic and Endocrine Disorders Group in the Cochrane Library (see 'About the Cochrane Collaboration', 'Collaborative Review Groups'). For an explanation of methodological terms, see the main Glossary in The Cochrane Library.

Peripheral neuropathy is a common complication of diabetes mellitus. Estimates of the prevalence of neuropathy in people with diabetes range from 10 to 50%, depending on factors such as how neuropathy is defined, methods of detection, and the population studied. A recent review estimates that neuropathy is present in about 30% of people attending hospital diabetes clinics, 20% of people with diabetes in primary care settings and 10% of the entire population of patients with diabetes (Shaw 2003).

The symptoms and signs of diabetic polyneuropathy mainly involve the feet and hands and are approximately symmetrical. Sensory loss and paresthesiae (often painful) are predominant, while muscle weakness is mild in most people. In addition to disability related to pain and the sensory and motor deficits, polyneuropathy is an important independent predictor of foot ulceration in people with diabetes. The annual incidence of foot ulcers in all people with diabetes is about 1%, but in those with established sensory loss from neuropathy, the incidence is seven times higher (Abbott 1998). Non‐healing foot ulcers precede the large majority of amputations in people with diabetes, and diabetic polyneuropathy is associated with a 2‐to‐25‐fold increase in the risk of amputation (Mayfield 1998).

Longitudinal studies of people with diabetic polyneuropathy have demonstrated that their clinical and electrophysiological abnormalities worsen with time, although slowly. For example, in the population‐based Rochester Diabetic Neuropathy Study, participants were assessed with the Neuropathy Impairment Score (NIS), in which the minimum detectable meaningful change is two points. In this patient cohort, those who were followed for at least two years worsened at a rate of only 0.85 NIS points per year (Dyck 1997). Thus, it is likely that it would require many months or several years of observation to demonstrate that a treatment slows or prevents worsening in diabetic neuropathy. Strategies to treat diabetic neuropathy include optimal glycemic control, management of associated risk factors, symptomatic therapies and treatments based on pathogenic mechanisms (Ziegler 2006).

The pathogenesis of diabetic polyneuropathy has been the subject of much study and discussion, and there is not universal agreement about its cause. Electrophysiological and nerve biopsy data suggest that the major pathological process is axonal degeneration, though there is evidence of secondary demyelination as well. Potential mechanisms of pathogenesis include an ischemic process secondary to microangiopathy, analogous to diabetic retinopathy and nephropathy (Dyck 1989), glycosylation of structural proteins consequent to chronic hyperglycemia, or injury by reactive oxygen species generated by altered glucose metabolism (Leinninger 2006). These mechanisms are not mutually exclusive, and may coexist and interact in many or most people with diabetic polyneuropathy.

Chronic hyperglycemia is hypothesized to exert its effects on peripheral nerves by both extracellular and intracellular mechanisms (Oates 2002). The extracellular route comprises various types of glycation reaction, including direct glycosylation of structural proteins, as well as chemical rearrangements resulting in the production of advanced glycosylation end products (AGE). The intracellular route results from increased flux of glucose through the polyol pathway (Cameron 1997). The polyol pathway is a sequence of intracellular chemical reactions, and is a mechanism by which cells can produce fructose from glucose. The main interest in the pathway has been its potential role in the development of complications, particularly neuropathy, in people with diabetes. The polyol pathway has two steps, each energy‐dependent and enzymatically catalyzed. In the first step, which is catalyzed by aldose reductase, glucose is converted to sorbitol, and nicotinamide adenine dinucleotide phosphate is oxidized (i.e. NADPH is converted to NADP). The second step, catalyzed by sorbitol dehydrogenase, converts sorbitol to fructose, and nicotinamide adenine dinucleotide is reduced (i.e. NAD is reduced to NADH). Chronic hyperglycemia results in increased activity of the polyol pathway, with several potentially deleterious consequences: (1) a decrease in cellular NADPH levels, resulting in decreased concentrations of glutathione (a free radical scavenger) and nitric oxide (a vasodilator); (2) increased cellular sorbitol levels, leading to decreased levels of myo‐inositol (necessary for Na‐K ATPase function); and (3) increased fructose levels, which enhance formation of AGEs (fructose is ten‐fold more active than glucose in glycosylation reactions).

Suppressing the metabolism of glucose via the polyol pathway by inhibiting aldose reductase is a potential way to prevent the putative deleterious consequences noted above. A great deal of research effort in the last two decades has been expended to test the effects of drugs which inhibit aldose reductase on human diabetic polyneuropathy. However, a meta‐analysis of controlled trials of aldose reductase inhibitors published between 1981 and 1993 concluded that it was not possible to make a clear statement about the efficacy of these compounds on diabetic neuropathy (Nicolucci 1996). After this meta‐analysis was first published as a Cochrane Review in The Cochrane Library, several concerns were raised about its methodology (inclusion of short duration trials, use of a surrogate measure of neuropathy severity (nerve conduction studies) as the primary outcome measure). As the concerns were not addressed and the Review was not updated, it was subsequently withdrawn from the Cochrane Library.

Several factors appear to have contributed to the uncertain status of aldose reductase inhibitors, including problems of trial design and types of outcome measures employed, in addition to serious adverse effects for some compounds. For example, tolrestat was approved for clinical use in several countries, but was subsequently withdrawn from most markets worldwide in 1996 following reports of fatal hepatic necrosis. Two other compounds did not progress beyond clinical trials because of serious adverse effects: sorbinil (lymphadenopathy, rash, fever, pancytopenia) and zenerastat (impaired renal function). One aldose reductase inhibitor, epalrestat, is approved for clinical use in Japan, but none of the other aldose reductase inhibitors are being used outside clinical trials. 
 
 Treatments that could prevent, improve, or even slow the progression of diabetic polyneuropathy would result in important improvements in the quality of life for many people with diabetes mellitus. Despite a wealth of experimental data that implicate the polyol pathway in the pathogenesis of diabetic polyneuropathy, and over 20 years of clinical trials of aldose reductase inhibitors, the clinical efficacy of these compounds is unclear. The goal of this review was to assess whether there is now sufficient evidence to support the use of aldose reductase inhibitors in clinical practice.

Objectives

The objective of this review was to assess the effects of aldose reductase inhibitors on the progression of symptoms, signs or functional disability in diabetic polyneuropathy.

Methods

Criteria for considering studies for this review

Types of studies

We included randomized controlled trials, regardless of publication status, language, or period of patient inclusion. Given the slow rate of progression of diabetic polyneuropathy, it is unlikely that any short‐duration intervention will produce a detectable effect. Therefore, only trials with treatment lasting at least six months were included.

Types of participants

Participants were male or female, aged 18 years or older. Criteria for the diagnosis of diabetes mellitus must have been clearly stated, using the standard criteria valid at the time of the beginning of the trial. Likewise, criteria for the diagnosis of diabetic polyneuropathy must have been clearly stated, including appropriate investigations to exclude other causes of polyneuropathy.

Types of interventions

Aldose reductase inhibitors (sorbinil, tolrestat, ponalrestat, zopolrestat, zenarestat, epalrestat, ranirestat, and fidarestat) given orally were compared with placebo or with no treatment. Treatment duration in any included study must have been at least six months.

Types of outcome measures

Primary outcomes

The primary outcome measure was defined as change in neurological status after a treatment period of at least six months, as measured by an evaluation of neurological function. This type of outcome measure is clinically relevant and was used to detect alterations in the progression of diabetic neuropathy in a landmark study of the effects of intensive glycemic control on diabetic neuropathy (DCCTRG 1993). Evaluation of neurological function may be measured in various ways, including strength testing, sensory examination (quantitative and non‐quantitative) and composite scores of neurological examination. In order to compare studies which used different scales to measure neurological function we treated a standardized unit change in one neurological scale as equivalent to a standardized unit change in another neurological scale, and compared proportional changes in these scales with and without use of an aldose reductase inhibitor. Unit change was defined as a standard deviation for that measurement scale as implemented in the RevMan standardized mean difference facility.

Secondary outcomes

Secondary outcome measures after a treatment period of at least six months included:

1. nerve conduction studies (changes in peroneal and tibial motor conduction velocity, peroneal compound muscle action potential amplitude, sural sensory nerve action potential amplitude);

2. neuropathic symptoms (measured by change in symptom scales including pain scales);

3. quality of life (measured by QOL scales);

4. occurrence of foot ulcers;

5. adverse effects (for example: liver enzyme elevation, organ failure, rash, hospitalisations, death, etc.).

Search methods for identification of studies

Electronic searches

We searched the Cochrane Neuromuscular Disease Group Trials Register using the search terms 'aldose reductase inhibitors' or 'aldehyde reductase inhibitors' or 'alrestatin' or 'sorbinil' or 'eparestat' or 'statil' or 'tolrestat' or 'ponalrestat' or 'fidalrestat' or 'zenarestat' or 'zopolrestat' and 'diabetic neuropathies' or 'neuropathy' or 'peripheral nervous system diseases'. We also adapted this strategy to search the Cochrane Metabolic and Endocrine Disorders Group Trials Register. We searched MEDLINE (from 1966 to May 2007) using the strategy in Additional Table 2 and adapted this strategy to search EMBASE (from 1980 to May 2007) and LILACS (from 1982 to May 2007). We reviewed the bibliographies of the randomised trials identified, contacted the authors and known experts in the field and approached pharmaceutical companies to identify additional published or unpublished data. We also searched electronic registers of ongoing trials such as the Current Controlled Trials register (http://www.controlled‐trials.com/).

Search strategies for MEDLINE, EMBASE and LILACS can be found in Appendix 1; Appendix 2 and Appendix 3 respectively.

Data collection and analysis

Selection of studies

An initial inclusion/exclusion form was used to assess whether a study should be included in the review. The titles and abstracts of every record retrieved by the searching process were scanned. The citations were divided into three equal groups alphabetically by first author, and each citation was assessed for inclusion independently by two review authors (authors A‐G were assessed by CC and TJB, H‐P by CC and FGAM, and Q‐Z by TJB and FGAM). Discrepancies between assessments were resolved by discussion and, when necessary, in consultation with the third review author. If disagreements could not be resolved at this stage, the full article was collected and assessed. If a trial was excluded at any time after this point, a record of both the article and exclusion reason was kept.

Data extraction and management

The methodological details and data from publications or unpublished study reports were extracted independently by two reviewers using a data extraction form designed for this review (we followed the same division of studies according to first author as described above in Selection of studies). The data extraction form included:

1. details of the study quality as outlined above,

2. details of the study design, including use of placebo, duration of treatment and duration of follow‐up,

3. details of the intervention, including specific ARI and dose;

4. details of the inclusion and exclusion criteria of the participants, numbers of participants, number of withdrawals and reasons for withdrawals;

5. baseline and end of study patient outcome parameters, including adverse effects. If there were insufficient data in the report or publication to complete the data extraction form then the original authors were contacted to supply additional information.

Assessment of risk of bias in included studies

Only randomized clinical trials of at least six months duration were selected for review. Each study was reviewed by two of the review authors to evaluate characteristics of methodological quality (we followed the same division of studies according to first author as described above in Selection of studies). The characteristics assessed were defined prior to review and included methods of randomization, adequacy of allocation concealment, methods of patient and observer blinding, completeness of follow‐up (> 80% of randomized participants) and use of intention‐to‐treat analysis. Each characteristic was graded as adequate, unclear, inadequate or not done. Differences in grading by the review authors were discussed and resolved by consensus by all three authors. When the quality of a characteristic was unclear, a letter of inquiry was sent to the lead author of the study for clarification. This classification of methodological quality was used as the basis of a sensitivity analysis. Additionally, we explored the influence of individual quality criteria in a sensitivity analysis.

Measures of treatment effect

The primary outcome measure, change in neurological examination, was measured in several ways (see Results). Several secondary outcome measures, some of which were surrogate measures of nerve function, were also analyzed. For continuous data measured in a similar way, we calculated the weighted mean differences (WMD) with 95% confidence intervals. Where the outcome measure or data measurement differed between studies we calculated the standardized mean differences (SMD).

It was anticipated that adverse effects for the different aldose reductase inhibitors would vary and might be compound‐specific. Therefore, to capture variable types of adverse effects we recorded all adverse effects reported in the analyzed studies and tabulated these separately for each of the included compounds. There were insufficient data to calculate relative risk and number needed to harm for specific agents.

Assessment of reporting biases

Funnel plots were used to assess for evidence of small study bias (Egger 1997).

Subgroup analysis and investigation of heterogeneity

Heterogeneity was explored by statistical testing (to look for variability in effect sizes between studies), qualitative review of the confidence intervals for the results of each study and careful review of the design of each study. We considered heterogeneity to be present if statistically significant variation in effect sizes above that consistent with the within‐studies variation existed between the studies. The I‐squared statistic measures the variability in studies not attributable to chance. I‐squared values from 75% to 100% were regarded as representing high levels of heterogeneity. We also considered heterogeneity to be present if the confidence intervals for the results of each study did not overlap, or if, in our opinion, the studies differed greatly in their design (type of compound studied, length of study, and quality of study). Outcome variables were analyzed using a random‐effects model if heterogeneity was detected, and with a fixed‐effects model if there was not significant heterogeneity. When heterogeneity was present and appeared to be attributable to one or more studies, analyses were repeated omitting these studies.

As this review evaluated several aldose reductase inhibitors, we performed subgroup analysis of the different compounds for benefit and harm.

Sensitivity analysis

Sensitivity analyses were performed on allocation concealment and other aspects of study quality that we assessed.

Results

Description of studies

Results of the search

We conducted an initial search in 2003, followed by updates in August 2006 and May 2007, yielding a total of 877 citations. Following the citation assessment process described in the Methods section above, we reviewed 47 full papers in detail. Thirty‐two of these described RCTs of at least six months duration (see Characteristics of included studies). Fourteen other studies were excluded for a variety of reasons, most often because they proved to be non‐randomized comparisons or had treatment periods less than six months (see Characteristics of excluded studies). One study (Nakajima 2005) is awaiting assessment.

Included studies

Study designs

The included studies were all randomized parallel group designs, except for one crossover study (Martyn 1987). Most studies lasted six to 12 months (range 24 weeks to three years), and most involved comparison of an ARI with placebo. In four studies (Baba 2006; Fabiani 1995; Hotta 2006; Nakayama 2001), control subjects received only usual care without a placebo.

Participants

All studies recruited participants 18 years and older. The 32 studies involved a total of 4970 participants, of whom 2922 received an ARI. Type 2 diabetics were preponderant, although diagnostic criteria for diabetes were not explicitly stated in most studies. Criteria for the diagnosis of diabetic polyneuropathy, and the severity of the neuropathy, varied considerably from study to study. In most studies there was a reasonable attempt to exclude other causes of neuropathy, although the possibility of inherited neuropathy was considered only in a minority of studies. Overall, the intervention and control participants were comparable at baseline, except in two small studies in which neuropathy at baseline seemed to be more severe in the control groups (Guy 1988; Jennings 1990), and one with a relative excess of female participants in the ARI group (Salmon 1991).

Interventions

Participants received either an ARI or placebo as oral tablets daily for the duration of the study, except in four studies (Baba 2006; Fabiani 1995; Hotta 2006; Nakayama 2001) which did not use a placebo in the control group. Seven different ARIs were used: epalrestat (386 participants: Baba 2006; Hotta 2006; Nakayama 2001), fidarestat (139 participants: Hotta 2001), ponalrestat (Statil) (416 participants: Dietrich 1990; Faes 1993; Florkowski 1991; Krentz 1992; Salmon 1991; Stracke 1994; Sundkvist 1992; Ziegler 1991), sorbinil (362 participants: Christensen 1987; Fagius 1985; Guy 1988; Jennings 1990; Jespersen 1986; Martyn 1987; O'Hare 1988; Sima 1988; SRG 1993), tolrestat (453 participants: Boulton 1989; Boulton 1990; Daniele 1995; Fabiani 1995; Giugliano 1993; Giugliano 1995; Macleod 1992; van Gerven 1992), zenarestat (1114 participants: Brown 2004; Greene 1999), and zopolrestat (54 participants: Johnson 2004).

Outcomes

Our primary outcome measure, change in neurological examination, was an endpoint in 29 studies. Most often this involved only the sensory examination, using several quantitative sensory testing techniques (e.g. hand‐held vibrometer). More comprehensive neurological examination scales were used in some (Baba 2006; Brown 2004; Fagius 1985; Giugliano 1995; Guy 1988; Hotta 2006; Martyn 1987; Sima 1988; SRG 1993; van Gerven 1992). Among our secondary outcome measures, nerve conduction studies were used in 27 studies, neurological symptoms in 21, and the development of foot ulcers in one (O'Hare 1988). Quality of life measures were not used in any of the included studies.

Risk of bias in included studies

Additional Table 1 summarizes the methodological grading of included studies. Thirteen of 32 studies had one or more methodological characteristics that were graded as inadequate. The majority of studies had characteristics that were unclear.

1. Summary of Methodological Quality of Included Studies.

Study	Randomization	Allocation	Blinding	Follow‐up	Intention‐to‐treat
Baba 2006	B	B	C	A	B
Boulton 1989	B	B	B	B	B
Boulton 1990	B	B	B	C	B
Brown 2004	B	B	B	C	A
Christensen 1987	B	B	B	A	B
Daniele 1995	B	B	B	A	A
Dietrich 1990	B	B	B	A	B
Faes 1993	B	B	B	A	A
Fabiani 1995	B	B	C	A	A
Fagius 1985	A	A	A	A	C
Florkowski 1991	A	A	A	A	A
Guigliano 1993	B	B	A	A	A
Guigliano 1995	B	B	A	A	A
Greene 1999	A	A	B	C	B
Guy 1988	A	A	A	A	A
Hotta 2001	A	A	A	C	C
Hotta 2006	A	A	C	C	C
Jennings 1990	B	B	B	C	C
Jespersen 1986	B	B	B	A	A
Johnson 2004	B	B	B	A	A
Krentz 1992	B	B	B	A	A
Macleod 1992	A	A	A	A	B
Martyn 1987	B	B	B	A	B
Nakayama 2001	B	B	C	A	A
O'Hare 1987	B	B	B	A	B
Salmon 1991	A	A	A	A	C
Sima 1988	B	B	B	A	B
SRG 1993	A	B	A	A	B
Stracke 1994	B	B	B	A	B
Sundkvist 1992	A	B	B	A	A
van Gerven 1992	B	B	B	B	B
Ziegler 1991	B	B	B	C	B
Abbreviations: A = adequate, B = unclear, C = inadequaute or not done

Effects of interventions

Most studies presented outcome data numerically in tables. In some cases outcome data were presented in graphs only, and numeric data for meta‐analysis were derived from the graphs.

Funnel plots did not suggest small study bias, although the relatively low number of studies made interpretation of funnel plots difficult.

Primary outcome measure

Change in neurological examination

Change in the neurological examination was an outcome measure in 29 studies. This most frequently involved the measurement of vibration perception threshold. Other measurements included the use of an "overall assessment score" (Sima 1988), measurement of thermal threshold (Ziegler 1991) or cool thermal sensation (Brown 2004), measurement of light touch and pain sensation (Baba 2006), and measurement of tendon reflexes or muscle power (Fabiani 1995).

There were no data or insufficient data for analysis in 15 studies involving approximately 770 treated participants (Bertelsmann 1991; Boulton 1989; Dietrich 1990 (exact number of participants unclear); Fagius 1985; Greene 1989; Greene 1999; Jennings 1990; Jespersen 1986; Johnson 2004; Krentz 1992; Martyn 1987; O'Hare 1988; SRG 1993; Stracke 1994; van Gerven 1992). Of these studies, one reported a statistically significant difference between the treated and control groups (van Gerven 1992), and several others reported no significant difference (Fagius 1985; Greene 1999; Jespersen 1986; Johnson 2004; Martyn 1987). One study (Baba 2006) reported change in neurological signs only as a dichotomous outcome (2/62 treated participants versus 9/72 untreated participants developed signs of diabetic neuropathy, a statistically significant difference), so that these data could not be included in the meta‐analysis based on continuous data.

Thirteen studies, involving 879 treated participants and 909 controls, reported sufficient data for meta‐analysis (Brown 2004; Fabiani 1995; Florkowski 1991; Giugliano 1993; Giugliano 1995; Guy 1988; Hotta 2006; Macleod 1992; Nakayama 2001; Salmon 1991; Sundkvist 1992; Ziegler 1991). As the heterogeneity of these studies was significant, we used a random effects approach. There was no significant difference between the treated and control groups (SMD ‐0.25, 95% CI ‐0.56 to 0.05) (see Analysis 01.01). One study (Brown 2004) reported results for both vibration sensation and cool thermal sensation. Analysis using either of these outcomes produced almost identical results; Analysis 01.01 used the vibration data. Another study reported a statistically significant improvement favoring treatment for elements of the neurological exam, including tendon reflexes and muscle power that were rated on a three point scale (Fabiani 1995). However, raw data could not be obtained so only the data for vibration perception threshold from this study were included in the meta‐analysis.

Analysis for each ARI separately also showed no difference between treated and control groups, with the exception of tolrestat, where there was a statistically significant benefit (WMD ‐4.15, 95% CI ‐7.43 to ‐0.87) based on 175 treated participants and 163 controls (Fabiani 1995; Giugliano 1993; Giugliano 1995; Macleod 1992) (see Analysis 01.02). If the tolrestat study that was not placebo‐controlled (Fabiani 1995) is excluded from the analysis, the WMD favours tolrestat, but the 95% CI spans zero (see Analysis 01.03). In Analysis 01.01, one study (Giugliano 1995) appears to be an outlier. This trial, which used tolrestat, enrolled participants with a shorter duration of diabetes mellitus than most trials. A subgroup analysis including trials with duration of diabetes less than six years suggested a small statistically significant benefit (SMD ‐1.33, 95% CI ‐2.39 to ‐0.27), but the number of treated participants is only 79 (see Analysis 01.04).

Several sensitivity analyses did not change the results (excluding (1) studies using ponalrestat (which was subsequently found to not enter peripheral nerve and therefore would not be expected to show any benefit (Hamada 2004)), (2) poor quality studies (those with any inadequate or more than 2 unclear quality items ‐ see Additional Table 1), and (3) studies of less than one year duration).

Secondary outcome measures

Nerve conduction studies

Nerve conduction studies (NCS) were performed as an outcome measure in 27 studies. The nerves evaluated were variable. The NCS measures that were assessed for meta‐analysis were changes in peroneal motor conduction velocity, peroneal compound muscle action potential amplitude, and sural sensory nerve action potential amplitude. Data were not gathered for these specific variables in some studies (Baba 2006; Hotta 2006). Tibial conduction velocity data was reported in some studies that did not have peroneal conduction velocity data (Fagius 1985; Florkowski 1991; Hotta 2001; Nakayama 2001). An analysis of the combined peroneal and tibial conduction velocity data was performed. Continuous data were available for the majority of studies that reported data for these NCS variables. Some studies stated whether there was a statistically significant difference between treatment and control groups, but reported either no data or insufficient data for meta‐analysis (Guy 1988 ‐ not significant (ns), Dietrich 1990 ‐ ns, Jespersen 1986 ‐ ns, Greene 1989 ‐ significant in favour of treatment, Boulton 1989 ‐ significant in favour of control). Several other studies were excluded from the pooled analysis because incomplete data were reported: motor nerve studied was not stated (van Gerven 1992), ranges of values without standard deviations (Jennings 1990), least squares means without standard deviations (Krentz 1992), only percentage of participants improved, unchanged and worsened (Daniele 1995). 
 
 Lower extremity motor conduction velocity data were reported in 17 studies, involving 1074 treated participants and 992 controls. A total of 13 studies, involving 938 treated participants and 845 controls, reported sufficient peroneal motor nerve conduction velocity data for meta‐analysis. As the heterogeneity of these studies was significant, we used a random‐effects approach. Standardized mean differences were used, as one study (Boulton 1989) reported only percentage change from onset. Treatment was associated with a slightly worse SMD in change in peroneal motor conduction velocity (‐0.02 m/s) but the 95% CI was ‐0.29 to 0.25 (Analysis 02.01). Including data from the four studies which reported tibial motor nerve conduction velocity (Fagius 1985; Florkowski 1991; Hotta 2001; Nakayama 2001) produced a SMD marginally favoring treatment, but the 95% CI spanned zero (SMD 0.09 m/s, 95% CI ‐0.13 to 0.31) (see Analysis 02.02).

Peroneal nerve compound muscle action potential amplitude data were available from 3 studies involving 298 treated participants and 209 controls (Greene 1999; SRG 1993; Sundkvist 1992). The difference between groups was not significant (WMD 0.03 mV, 95% CI ‐0.37 to 0.42) (see Analysis 02.03).

Sural sensory nerve action potential amplitude data was available from 6 studies, involving 570 treated participants and 496 controls (Brown 2004; Greene 1999; Martyn 1987; Salmon 1991; Sima 1988; Sundkvist 1992). The difference between groups was not significant (WMD ‐0.10 µV, 95% CI ‐0.48 to 0.27) (see Analysis 02.04).

Neuropathic symptoms

Many of the included studies that used neuropathic symptoms as an outcome measure simply reported that no significant effect was seen, or otherwise did not provide any data amenable to meta‐analysis. When presented, data were either continuous (e.g. a symptom score measured on a visual‐analog scale) or dichotomous (e.g. symptoms present or absent, worsened or stable/improved). There was no significant difference between ARI and control subjects in the studies using quantitative pain scales, (WMD 0.31, 95% CI ‐0.32 to 0.93) more improvement with control than ARIs (Boulton 1990; Florkowski 1991; Hotta 2001; Macleod 1992; O'Hare 1988; Ziegler 1991) (see Analysis 03.03). By contrast, in studies which evaluated symptoms as a dichotomous variable (Baba 2006; Fabiani 1995; Giugliano 1993; Giugliano 1995; Guy 1988; Hotta 2006), participants treated with an ARI were significantly less likely to have unchanged or worse symptoms than were controls (relative risk 0.58, 95% CI 0.46 to 0.74) (see Analysis 03.01). Two of these studies were open‐label (Baba 2006;Hotta 2006), and one was not placebo‐controlled (Fabiani 1995); if these are removed from the analysis, the relative risk is still less than one (0.35, 95% CI 0.17 to 0.75) (see Analysis 03.02).

Quality of life

This was not an outcome for any of the included studies.

Occurrence of foot ulcers

This outcome was assessed in only one included study (O'Hare 1988). This was a 12‐month double‐blind randomized trial in which 21 participants treated with sorbinil were compared with 10 treated with placebo. During the trial four sorbinil participants developed new episodes of ulceration, compared with one in the placebo group. The authors concluded that there was no evidence supporting a role for sorbinil in preventing foot ulcers.

Adverse effects

Adverse effects were reported with varying degrees of detail, with almost all studies providing some details regarding adverse effects. Insufficient data were available to perform statistical comparisons between groups. Tolrestat was associated with elevation of liver enzymes in four of the seven studies (Boulton 1989; Boulton 1990; Macleod 1992; van Gerven 1992). Dizziness and reduced blood pressure were also more frequent in the treated group in one of these studies (Boulton 1990). Ponalrestat, in general, was not associated with adverse effects different from placebo, except for one study, which reported decreases in hemoglobin, red blood cell count and lymphocyte count (Krentz 1992). These were not felt to be clinically significant. Sorbinil was associated with rash in seven of nine studies (Christensen 1987; Fagius 1985; Guy 1988; Jennings 1990; Martyn 1987; O'Hare 1988; SRG 1993). The rash was often reported as severe and associated with malaise and sometimes lymphadenopathy; in some instances the hypersensitivity reaction was labeled Stevens Johnson syndrome. In the SRG 1993 study, hypersensitivity reactions were reported in 7.4% of treated participants and 1% of placebo participants (P = 0.003). Two studies also reported elevation of liver enzymes in sorbinil treated participants (Martyn 1987; O'Hare 1988). The single study using zopolrestat reported one patient with increased liver enzymes (Johnson 2004). The single study using fidarestat reported no difference in adverse events or laboratory abnormalities between groups (Hotta 2001). In one of the three epalrestat studies increased liver enzymes were reported in 2% of participants, and nausea and diarrhea in 3% (Hotta 2006). The development of zenarestat was halted due to increases in creatinine and this was reported by both studies using zenarestat (Brown 2004; Greene 1999). One zenarestat study also reported decreased hemoglobin and three participants with non‐Hodgkins lymphoma in the treated group (Greene 1999). The authors postulated that the lymphoproliferative disorders may have preceded the use of zenarestat.

Discussion

There is now a substantial body of published studies of clinical trials of ARIs in diabetic polyneuropathy. Even though we excluded all trials of less than six months duration, we still identified 32 RCTs, published between 1985 and 2006, involving almost 5000 participants. Unfortunately, there are important issues of methodological quality involving most of these studies; in a substantial number, we had incomplete or inadequate information about concealment of treatment allocation, follow‐up of randomized participants, and methods of randomization and blinding.

Unlike previous meta‐analyses of ARIs in diabetic neuropathy, which focused on surrogate measures of neuropathy (primarily nerve conduction data), we elected to evaluate the effect of ARIs on the neurological examination, arguing that this is a clinically meaningful and important endpoint. Furthermore, neurological examination scales and related functional outcomes are generally accepted as the standard metric in therapeutic trials in other neuromuscular diseases, and their use facilitates comparison of the magnitudes of treatment effects between different diseases.

Our meta‐analysis showed no statistically significant overall benefit of ARIs on neurological examination findings in diabetic polyneuropathy. A subgroup analysis of studies using tolrestat suggested benefit, although the number of treated participants was small, and the benefit did not persist when only placebo‐controlled trials were included. We cannot entirely exclude the possibility that tolrestat may have benefit. However, larger trials of tolrestat were conducted and apparently did not demonstrate benefit. These data were not published and we could not obtain them. Subsequent development of tolrestat was halted because of concerns about liver toxicity. Various other subgroup analyses, such as those excluding studies involving ponalrestat (which proved not to enter peripheral nerve), poorer quality studies, or studies lasting less than one year, showed no benefit of ARIs on neurological examination findings. 
 
 Among our secondary outcome measures, only one (neuropathic symptoms measured as a dichotomous outcome), based on six studies, reached statistical significance. However, two of these studies were non‐blinded, and one was not placebo‐controlled, which are likely to introduce an important bias when evaluating symptoms. When these studies were removed from the analysis, the overall relative risk still suggested benefit, but the 95% confidence interval approached one. This rather tenuous suggestion of benefit was contradicted by the lack of benefit in studies evaluating symptoms with continuous scales, and the 12 studies that simply stated that there was no benefit, without reporting data. Taken together, we think the evidence that ARIs may benefit neuropathic symptoms is unconvincing. Also noteworthy among our secondary outcome measures was the overall lack of clear improvement in electrophysiological measures, particularly motor nerve conduction velocity. Although several individual studies found that ARIs produced modest improvements in conduction velocity, the pooled results showed no benefit. Some ARIs were associated with significant adverse effects, which appeared to be compound specific. Severe hypersensitivity reactions (sorbinil), hepatic toxicity (tolrestat) and renal toxicity (zenarestat) resulted in the withdrawal of these compounds from further development. Other ARIs have not been impacted by significant adverse effect issues.

Some limitations of our review process must be noted. Many of the studies were conducted more than a decade ago, making it particularly difficult to clarify details of methodology. We received replies only from about half the authors we attempted to contact, and often learned that study protocols or data were no longer available. In addition to concerns about variable methodological quality and outcome measurement in the included studies noted above, there may be an important non‐publication bias in ARI studies. This is suggested by the number of studies reported only in abstract or other incomplete forms, such that data for meta‐analysis are either insufficient or absent. A further issue is that most of the ARI trials have been industry sponsored, and data may exist which are proprietary and thus unavailable to us.

Unlike many neuromuscular disorders, in which conclusions about the value of treatments are limited by a lack of RCTs, for ARIs in diabetic polyneuropathy there is a large body of RCT data to assess. Although the available RCTs are imperfect, we conclude that the accumulated evidence has not demonstrated that ARIs are effective in human diabetic neuropathy, despite attractive theoretical arguments for ARIs and promising experimental data in animals. If any further clinical trials of ARIs are undertaken, we feel it is essential that (1) any compounds tested should be proven to have substantial advantages over previously tested compounds in biological properties or preclinical results, and (2) any trial should be of at least one year duration, use clear methods of randomization, allocation concealment, and blinding, have good follow‐up of subjects, and employ measures based on clinical outcomes such as change in neurological examination findings, symptoms, or functional disability.

Authors' conclusions

Implications for practice.

There is no overall evidence that aldose reductase inhibitors can slow or halt the natural progression of diabetic polyneuropathy.

Implications for research.

If further clinical trials of aldose reductase inhibitors are undertaken, they should only involve compounds with substantial biological or preclinical advantages over the agents that have been tested to date. Any future trials should last at least one year, use rigorous methodology, and employ clinical endpoints as outcome measures.

What's new

Date	Event	Description
28 October 2008	Amended	Converted to new review format.

History

Protocol first published: Issue 1, 2004
 Review first published: Issue 4, 2007

Date	Event	Description
22 August 2007	New citation required and conclusions have changed	Substantive amendment

Notes

This review replaces the ' Aldose reductase inhibitors for the prevention and treatment of diabetic peripheral neuropathy' review by M Airey, C Bennett, A Nicolucci, R Williams which was withdrawn from The Cochrane Database for Systematic Reviews pending this update.

Appendices

Appendix 1. Ovid MEDLINE Search Strategy

1. randomized controlled trial.pt. 
 2. controlled clinical trial.pt. 
 3. randomized controlled trials/ 
 4. random allocation/ 
 5. double‐blind method/ 
 6. single‐blind method/ 
 7. or/1‐6 
 8. animals/ not humans/ 
 9. 7 not 8 
 10. clinical trial.pt. 
 11. exp clinical trials/ 
 12. (clin$ adj25 trial$).ti,ab. 
 13. ((singl$ or doubl$ or tripl$ or trebl$) adj25 (blind$ or mask$)).ti,ab. 
 14. placebos/ 
 15. placebo$.ti,ab. 
 16. random$.ti,ab. 
 17. research design/ 
 18. or/10‐17 
 19. 18 not 8 
 20. 19 not 9 
 21. comparative study/ 
 22. exp evaluation studies/ 
 23. follow up studies/ 
 24. prospective studies/ 
 25. (control$ or prospectiv$ or volunteer$).ti,ab. 
 26. or/21‐25 
 27. 26 not 8 
 28. 27 not (9 or 20) 
 29. 9 or 20 or 28 
 30. exp Diabetes Mellitus/ 
 31. diabet$.mp. 
 32. exp diabetes mellitus, non‐insulin‐dependent/ 
 33. or/30‐32 
 34. neuropath$.mp. 
 35. exp Peripheral Nervous System Diseases/ 
 36. polyneuropath$.mp. 
 37. or/34‐36 
 38. exp diabetic neuropathies/ 
 39. diabetic neuropath$.mp. 
 40. diabetic polyneuropath$.mp. 
 41. or/38‐40 
 42. exp Aldehyde Reductase/ 
 43. (aldose reductase or aldehyde reductase).mp. 
 44. alrestatin.mp. 
 45. sorbinil.mp. 
 46. epalrestat.mp. 
 47. statil.mp. 
 48. tolrestat.mp. 
 49. ponalrestat.mp. 
 50. fidarestat.mp. 
 51. zenarestat.mp. [mp=title, original title, abstract, name of substance word, subject heading word] 
 52. zopolrestat.mp. [mp=title, original title, abstract, name of substance word, subject heading word] 
 53. or/42‐52 
 54. 33 or 37 or 41 
 55. 53 and 54 
 56. 29 and 55

Appendix 2. Ovid EMBASE Search Strategy

1. Randomized Controlled Trial/ 
 2. Clinical Trial/ 
 3. Multicenter Study/ 
 4. Controlled Study/ 
 5. Crossover Procedure/ 
 6. Double Blind Procedure/ 
 7. Single Blind Procedure/ 
 8. exp RANDOMIZATION/ 
 9. Major Clinical Study/ 
 10. PLACEBO/ 
 11. Meta Analysis/ 
 12. phase 2 clinical trial/ or phase 3 clinical trial/ or phase 4 clinical trial/ 
 13. (clin$ adj25 trial$).tw. 
 14. ((singl$ or doubl$ or tripl$ or trebl$) adj25 (blind$ or mask$)).tw. 
 15. placebo$.tw. 
 16. random$.tw. 
 17. control$.tw. 
 18. (meta?analys$ or systematic review$).tw. 
 19. (cross?over or factorial or sham? or dummy).tw. 
 20. ABAB design$.tw. 
 21. or/1‐20 
 22. human/ 
 23. nonhuman/ 
 24. 22 or 23 
 25. 21 not 24 
 26. 21 and 22 
 27. 25 or 26 
 28. exp diabetes mellitus/ 
 29. diabet$.mp. [mp=title, abstract, subject headings, heading word, drug trade name, original title, device manufacturer, drug manufacturer name] 
 30. 28 or 29 
 31. exp NEUROPATHY/ 
 32. NEUROPATH$.mp. [mp=title, abstract, subject headings, heading word, drug trade name, original title, device manufacturer, drug manufacturer name] 
 33. POLYNEUROPATH$.mp. [mp=title, abstract, subject headings, heading word, drug trade name, original title, device manufacturer, drug manufacturer name] 
 34. 31 or 32 or 33 
 35. aldehyde reductase/ 
 36. exp aldose reductase inhibitor/ 
 37. (aldose reductase or aldehyde reductase).mp. [mp=title, abstract, subject headings, heading word, drug trade name, original title, device manufacturer, drug manufacturer name] 
 38. alrestatin.mp. [mp=title, abstract, subject headings, heading word, drug trade name, original title, device manufacturer, drug manufacturer name] 
 39. sorbinil.mp. [mp=title, abstract, subject headings, heading word, drug trade name, original title, device manufacturer, drug manufacturer name] 
 40. epalrestat.mp. [mp=title, abstract, subject headings, heading word, drug trade name, original title, device manufacturer, drug manufacturer name] 
 41. statil.mp. [mp=title, abstract, subject headings, heading word, drug trade name, original title, device manufacturer, drug manufacturer name] 
 42. tolrestat.mp. [mp=title, abstract, subject headings, heading word, drug trade name, original title, device manufacturer, drug manufacturer name] 
 43. ponalrestat.mp. [mp=title, abstract, subject headings, heading word, drug trade name, original title, device manufacturer, drug manufacturer name] 
 44. fidarestat.mp. [mp=title, abstract, subject headings, heading word, drug trade name, original title, device manufacturer, drug manufacturer name] 
 45. or/35‐44 
 46. 30 and 34 
 47. 45 and 46 
 48. 27 and 47 
 49. 48

Appendix 3. LILACS Search Strategy

((Pt randomized controlled trial OR Pt controlled clinical trial OR Mh randomized controlled trials OR Mh random allocation OR Mh double‐blind method OR Mh single‐blind method) AND NOT (Ct animal AND NOT (Ct human and Ct animal)) 
 
 OR (Pt clinical trial OR Ex E05.318.760.535$ OR (Tw clin$ AND (Tw trial$ OR Tw ensa$ OR Tw estud$ OR Tw experim$ OR Tw investiga$)) OR ((Tw singl$ OR Tw simple$ OR Tw doubl$ OR Tw doble$ OR Tw duplo$ OR Tw trebl$ OR Tw trip$) AND (Tw blind$ OR Tw cego$ OR Tw ciego$ OR Tw mask$ OR Tw mascar$)) OR Mh placebos OR Tw placebo$ OR (Tw random$ OR Tw randon$ OR Tw casual$ OR Tw acaso$ OR Tw azar OR Tw aleator$) OR Mh research design) AND NOT (Ct animal AND NOT (Ct human and Ct animal)) 
 
 OR (Ct comparative study OR Ex E05.337$ OR Mh follow‐up studies OR Mh prospective studies OR Tw control$ OR Tw prospectiv$ OR Tw volunt$ OR Tw volunteer$) AND NOT (Ct animal AND NOT (Ct human and Ct animal))) 
 
 AND ((Mh diabetes mellitus OR diabet$) 
 
 AND (Mh NEUROPATHY OR NEUROPATH$ OR POLYNEUROPATH$) 
 
 AND (Mh aldehyde reductase OR Mh aldose reductase inhibitor OR aldose reductase OR aldehyde reductase OR alrestatin OR sorbinil OR epalrestat OR statil OR tolrestat OR ponalrestat OR fidarestat))

Data and analyses

Comparison 1. Change in neurological examination.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Change in neurological examination	13	1788	Std. Mean Difference (IV, Random, 95% CI)	‐0.25 [‐0.56, 0.05]
2 Change in neurological examination, tolrestat alone	4	338	Mean Difference (IV, Random, 95% CI)	‐4.15 [‐7.43, ‐0.87]
3 Change in neurological examination, tolrestat, placebo‐controlled studies only	3	272	Mean Difference (IV, Random, 95% CI)	‐3.82 [‐9.72, 2.09]
4 Change in neurological examination, diabetes duration less than 6 years	3	156	Std. Mean Difference (IV, Random, 95% CI)	‐1.33 [‐2.39, ‐0.27]

1.1. Analysis.

Comparison 1 Change in neurological examination, Outcome 1 Change in neurological examination.

1.2. Analysis.

Comparison 1 Change in neurological examination, Outcome 2 Change in neurological examination, tolrestat alone.

1.3. Analysis.

Comparison 1 Change in neurological examination, Outcome 3 Change in neurological examination, tolrestat, placebo‐controlled studies only.

1.4. Analysis.

Comparison 1 Change in neurological examination, Outcome 4 Change in neurological examination, diabetes duration less than 6 years.

Comparison 2. Nerve conduction studies.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Change in peroneal motor nerve conduction velocity (m/s)	13	1783	Std. Mean Difference (IV, Random, 95% CI)	‐0.02 [‐0.29, 0.25]
2 Change in peroneal or tibial nerve motor conduction velocity (m/s)	17	2066	Std. Mean Difference (IV, Random, 95% CI)	0.09 [‐0.13, 0.31]
3 Change in peroneal compound muscle action potential amplitude (mV)	3	507	Mean Difference (IV, Fixed, 95% CI)	0.03 [‐0.37, 0.42]
4 Change in sural sensory nerve action potential amplitude (microvolts)	6	1066	Mean Difference (IV, Fixed, 95% CI)	‐0.10 [‐0.48, 0.27]

2.1. Analysis.

Comparison 2 Nerve conduction studies, Outcome 1 Change in peroneal motor nerve conduction velocity (m/s).

2.2. Analysis.

Comparison 2 Nerve conduction studies, Outcome 2 Change in peroneal or tibial nerve motor conduction velocity (m/s).

2.3. Analysis.

Comparison 2 Nerve conduction studies, Outcome 3 Change in peroneal compound muscle action potential amplitude (mV).

2.4. Analysis.

Comparison 2 Nerve conduction studies, Outcome 4 Change in sural sensory nerve action potential amplitude (microvolts).

Comparison 3. Neuropathic symptoms.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Symptoms as dichotomous outcome (same or worse)	6	691	Risk Ratio (M‐H, Fixed, 95% CI)	0.58 [0.46, 0.74]
2 Symptoms as dichotomous outcome (same or worse), open‐label and non‐placebo‐controlled studies excluded	3	136	Risk Ratio (M‐H, Fixed, 95% CI)	0.35 [0.17, 0.75]
3 Change in pain as a continuous variable	5	497	Mean Difference (IV, Random, 95% CI)	0.31 [‐0.32, 0.93]
3.1 4‐point scale	2	389	Mean Difference (IV, Random, 95% CI)	0.17 [‐0.03, 0.37]
3.2 Visual analog scale	3	108	Mean Difference (IV, Random, 95% CI)	8.42 [2.82, 14.02]

3.1. Analysis.

Comparison 3 Neuropathic symptoms, Outcome 1 Symptoms as dichotomous outcome (same or worse).

3.2. Analysis.

Comparison 3 Neuropathic symptoms, Outcome 2 Symptoms as dichotomous outcome (same or worse), open‐label and non‐placebo‐controlled studies excluded.

3.3. Analysis.

Comparison 3 Neuropathic symptoms, Outcome 3 Change in pain as a continuous variable.

Characteristics of studies

Characteristics of included studies [ordered by study ID]

Baba 2006.

Methods	Randomized parallel group, 3 years.
Participants	76 treatment, 87 controls; comparable at baseline.
Interventions	Epalrestat 150 mg vs no treatment.
Outcomes	Neurological examination, electrophysiology symptoms.
Notes	Open‐label study (not blinded).
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

Boulton 1989.

Methods	Randomized parallel group, 6 months.
Participants	98 treatment, 92 placebo. Age range, baseline features not given.
Interventions	Tolrestat 200 mg daily vs placebo.
Outcomes	VPT, electrophysiology symptoms.
Notes	Abstract, with further details from Makin 1990.
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

Boulton 1990.

Methods	Randomized parallel group, 52 weeks.
Participants	112 treatment, 107 control subjects; comparable at baseline.
Interventions	Tolrestat 200 mg daily vs placebo.
Outcomes	Electrophysiology symptoms.
Notes	1. Trial also included groups given tolrestat 50, 100, and 100 mg BID, but data are presented only for 200 mg QD group, as no changes were seen with other groups. 2. Less than 80% of randomized participants included in analysis. 3. Unclear if intention‐to‐treat analysis used.
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

Brown 2004.

Methods	Randomized, parallel group, 12 months.
Participants	481 & 475 treatment subjects (2 doses), 472 placebo. Baseline comparable.
Interventions	Zenarestat 600 and 1200 mg vs placebo.
Outcomes	QST (CASE IV), clinical neuropathy scales, electrophysiology.
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

Christensen 1987.

Methods	Randomized parallel group, 52 weeks.
Participants	13 treatment, 7 control subjects. Comparable but baseline data limited.
Interventions	Sorbinil 250 mg vs placebo.
Outcomes	Electrophysiology.
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

Daniele 1995.

Methods	Randomized parallel group, 6 months.
Participants	39 treatment, 35 control subjects. Comparable but incomplete baseline data.
Interventions	Tolrestat 200 mg vs placebo.
Outcomes	Symptoms, electrophysiology.
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

Dietrich 1990.

Methods	Randomized parallel group, 56 weeks.
Participants	60 patients entered, 50 completed (numbers in treatment vs control not stated); baseline characteristics not given.
Interventions	Statil 600 mg vs placebo.
Outcomes	QST, electrophysiology symptoms.
Notes	Abstract only.
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

Fabiani 1995.

Methods	Randomized parallel group, 12 months.
Participants	33 treatment, 33 control patients: comparable at baseline.
Interventions	Tolrestat 200 mg vs "usual care".
Outcomes	neurological examination, QST, symptoms.
Notes	Main focus of study was esophageal motility.
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

Faes 1993.

Methods	Randomized parallel group, 28 weeks.
Participants	18 treatment, 16 control patients; comparable at baseline.
Interventions	Statil 600 mg vs placebo.
Outcomes	Symptoms.
Notes	1. Unclear whether participants or evaluators were blinded. 2. Main focus of study was autonomic function.
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

Fagius 1985.

Methods	Randomized parallel group, 24 weeks.
Participants	32 treatment, 32 control patients; comparable at baseline.
Interventions	Sorbinil 250 mg vs placebo.
Outcomes	Neurological examination, QST, electrophysiology symptoms.
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Low risk	A ‐ Adequate

Florkowski 1990 *.

Methods	Same data published in Florkowski 1991.
Participants
Interventions
Outcomes
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	D ‐ Not used

Florkowski 1991.

Methods	Randomized parallel group, 24 weeks.
Participants	37 treatment, 17 control; baseline characteristics comparable.
Interventions	Ponalrestat 300 or 600 mg vs placebo.
Outcomes	QST, electrophysiology symptoms.
Notes	Same data also published in Florkowski 1990.
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Low risk	A ‐ Adequate

Giugliano 1993.

Methods	Randomized parallel group, 12 months.
Participants	25 treatment, 20 control patients; comparable at baseline.
Interventions	Tolrestat 200 mg vs placebo.
Outcomes	QST, symptoms.
Notes	Autonomic function was study's main focus.
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

Giugliano 1995.

Methods	Randomized parallel group, 12 months.
Participants	30 treatment, 30 control patients; comparable at baseline.
Interventions	Tolrestat 200 mg vs placebo.
Outcomes	Neurological examination, QST, symptoms.
Notes	Main focus of study was autonomic function
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

Greene 1999.

Methods	Randomized parallel group, 52 weeks.
Participants	158 treatment (3 doses) vs 50 placebo patients; comparable at baseline.
Interventions	Zenarestat 150, 300, and 600 mg vs placebo.
Outcomes	QST, electrophysiology.
Notes	Less than 80% of randomized participants included in analysis.
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Low risk	A ‐ Adequate

Guy 1988.

Methods	Randomized parallel group, 52 weeks.
Participants	21 treatment vs 18 control patients; comparable at baseline except slight excess of foot ulcers in controls (13/18 vs 8/21).
Interventions	Sorbinil 250 mg vs placebo.
Outcomes	Neurological examination, QST, electrophysiology symptoms.
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Low risk	A ‐ Adequate

Hotta 2001.

Methods	Randomized parallel group, 52 weeks.
Participants	139 treatment vs 140 control patients; comparable at baseline.
Interventions	Fidarestat 1 mg vs placebo.
Outcomes	Electrophysiology symptoms.
Notes	1. Less than 80% of randomized subjects included in analysis. 2. Analysis not intention‐to‐treat.
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Low risk	A ‐ Adequate

Hotta 2006.

Methods	Randomized parallel group, 3 years.
Participants	295 treatment vs 308 control patients; comparable at baseline.
Interventions	Epalrestat 50 mg TID vs no treatment.
Outcomes	Neurological examination, QST, electrophysiology symptoms.
Notes	1. Open‐label study (not blinded) 2. Less than 80% of randomized subjects included in analysis. 3. Analysis not intention‐to‐treat.
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Low risk	A ‐ Adequate

Jennings 1990.

Methods	Randomized parallel group, 24 months.
Participants	13 treatment vs 7 control patients; more severe neuropathy and longer duration of diabetes in control group at baseline.
Interventions	Sorbinil 250 mg vs placebo.
Outcomes	QST, electrophysiology.
Notes	1. Less than 80% of randomized subjects included in analysis. 2. Analysis not intention‐to‐treat.
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

Jespersen 1986.

Methods	Randomized parallel group, 52 weeks.
Participants	12 treatment vs 7 control patients; baseline characteristics not stated.
Interventions	Sorbinil 250 mg vs placebo.
Outcomes	QST, electrophysiology.
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

Johnson 2004.

Methods	Randomized parallel group, 1 year.
Participants	54 treatment vs 27 control patients; comparable at baseline.
Interventions	Zopolrestat 500 and 1000 mg (pooled for analysis) vs placebo.
Outcomes	QST.
Notes	Study's main focus was cardiac function.
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

Krentz 1992.

Methods	Randomized parallel group, 52 weeks.
Participants	25 treatment vs 25 control patients; baseline characteristics not given.
Interventions	Ponalrestat 600 mg vs placebo.
Outcomes	QST, electropysiology symptoms.
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

Macleod 1992.

Methods	Randomized parallel group, 6 months.
Participants	98 treatment, 92 control patients; comparable at baseline
Interventions	Tolrestat 200 mg vs placebo.
Outcomes	QST, electrophysiology symptoms.
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Low risk	A ‐ Adequate

Martyn 1987.

Methods	Randomized crossover study, 24 weeks.
Participants	24 treatment, 24 control patients; comparable at baseline.
Interventions	Sorbinil 125 mg vs placebo.
Outcomes	Neurological examination, QST, electrophysiology
Notes	1. All study patients had prior exposure to sorbinil, and had not developed rash or other adverse effects. 2. Analysis not intention‐to‐treat.
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

Nakayama 2001.

Methods	Randomized parallel group, 24 weeks.
Participants	15 treatment, 15 control patients; comparable at baseline.
Interventions	Epalrestat 150 mg vs "usual care".
Outcomes	Electrophysiology.
Notes	Also used pupillary and cardiovascular reflex outcomes.
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

O'Hare 1988.

Methods	Randomized parallel group, 12 months.
Participants	21 treatment, 10 control patients, comparable at baseline.
Interventions	Sorbinil 250 mg vs placebo.
Outcomes	QST, electrophysiology symptoms.
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

Salmon 1991.

Methods	Randomized parallel group, 24 weeks.
Participants	21 treatment, 19 control patients; comparable at baseline except higher proportion of females in treatment group.
Interventions	Ponalrestat 600 mg vs placebo.
Outcomes	QST, electrophysiology.
Notes	Analysis not intention‐to‐treat.
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Low risk	A ‐ Adequate

Sima 1988.

Methods	Randomized parallel group, 1 year.
Participants	10 treatment, 6 control patients; unclear whether comparable at baseline.
Interventions	Sorbinil 250 mg vs placebo.
Outcomes	Neurological examination, QST, electrophysiology symptoms.
Notes	1. Subset of a larger multicenter trial. 2. Major focus of study was nerve biopsy evaluation.
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

SRG 1993.

Methods	Randomized parallel group, 21 months.
Participants	216 treatment vs 225 control patients; comparable at baseline.
Interventions	Sorbinil 250 mg vs placebo.
Outcomes	"Development of neuropathy" (symptoms & signs); electrophysiology.
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

Stracke 1994.

Methods	Randomized parallel group, 52 weeks.
Participants	30 treatment, 30 control patients; baseline data comparable but incomplete.
Interventions	Statil 600 mg vs placebo.
Outcomes	QST, electrophysiology symptoms.
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

Sundkvist 1992.

Methods	Randomized parallel group, 18 months.
Participants	216 treatment, 99 control patients; comparable at baseline.
Interventions	Ponalrestat 600 mg vs placebo.
Outcomes	QST, electrophysiology.
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

van Gerven 1992.

Methods	Randomized parallel group, 6 months.
Participants	18 treatment, 17 control patients, comparable at baseline.
Interventions	Tolrestat 200 mg vs placebo.
Outcomes	Neurological examination, QST, electrophysiology symptoms.
Notes	Same data also published as van Gerven 1993.
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

van Gerven 1993 *.

Methods	Same data published as van Gerven 1992
Participants
Interventions
Outcomes
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	D ‐ Not used

Ziegler 1991.

Methods	Randomized parallel group, 12 months.
Participants	39 treatment, 21 control patients; comparable at baseline.
Interventions	Ponalrestat 600 mg vs placebo.
Outcomes	QST, electrophysiology symptoms.
Notes	1. Less than 80% of randomized participants included in analysis. 2. Analysis not intention‐to‐treat.
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

Abbreviations: VPT = vibration perception threshold; QST = quantitative sensory testing; CASE IV = computer‐assisted sensory examination, model 4

Characteristics of excluded studies [ordered by study ID]

Study	Reason for exclusion
Bertelsmann 1991	Abstract only, with insufficient detail for analysis.
Bril 2006	Control group used for first 12 weeks only. Subsequent 48 weeks of study compared 2 doses of ranirestat, but without control group.
Canal 1984	Drug tested is not an ARI.
Fagius 1981	Treatment given for 12 weeks only.
Gianni 1992	Paper could not be retrieved for detailed review, but unlikely to be an eligible study.
Gonen 1991	Same study as Santiago 1993 ‐ an ARI withdrawal study, not an RCT.
Goto 1993	Only 12 weeks treatment.
Greene 1989	Abstract only, with insufficient detail for analysis
Guigliano 1991	Abstract only, with only autonomic function as outcome measures. Some or all of these patients may have been reported in Guigliano 1993 (see Included Stuides).
Jaspan 1983	Not an RCT.
Santiago 1993	Same study as Gonen 1991. An ARI withdrawal study, not an RCT.
Sima 1993	Short report, without sufficient detail for meta‐analysis.
Terranova 1993	Nonrandomized comparison, not an RCT.
Young 1983	Only 4 weeks treatment.

Contributions of authors

CC drafted the Background to the protocol; the inclusion criteria, outcomes and analysis sections were written primarily by TJB, with input from CC; FGAM devised the search strategy. Assessment of search strategy citations, evaluation of methodological quality and data extraction were shared by CC, TJB, and FGAM. Meta‐analyses were performed by FGAM (neurological examination), TJB (nerve conduction data) and CC (neuropathic symtoms). All reviewers contributed equally to the final version of the text of the review.

Declarations of interest

CC received consulting fees from Wyeth‐Ayerst, Inc in 1993‐4 for assisting with analysis of nerve biopsies from a clinical trial of tolrestat. 
 TJB has participated in clinical trials utilizing tolrestat sponsored by Wyeth‐Ayerst, zenarestat sponsored by Parke‐Davis, lidorestat sponsored by the Institute for Diabetes Discovery, Inc., and AS 3201 sponsored by Dainippon. TJB received a grant from ICI Pharma to study pre‐clinical effects of ponalrestat.

Edited (no change to conclusions)