Criteria for considering studies for this review
Types of studies
We will include randomised controlled trials (RCTs) and quasi-RCTs, published and unpublished, in any language. Quasi-RCTs are those in which allocation of participants to groups is not fully random and, for example, hospital numbers or date of birth may be used.
Types of participants
We will include people of any age with type 1 or type 2 diabetes mellitus who have been diagnosed with polyneuropathy. Polyneuropathy will be defined in accordance with the Consensus Statement of the American Diabetes Association and the American Academy of Neurology (ADA consensus statement 1988). Participants in the trials must have met at least three of the following five criteria.
Typical symptoms of diabetic polyneuropathy.
Typical signs of diabetic polyneuropathy.
Electrophysiological results consistent with sensory neuropathy.
Quantitative sensory testing (QST) confirming neuropathy.
Quantitative autonomic testing (QAT) confirming neuropathy.
We have not used more recent criteria for the diagnosis of diabetic neuropathy (Tesfaye 2010), as studies are likely to have been conducted many years earlier. Although we plan to assess studies using the earlier criteria given above, some studies may have been published before even these were in use. Studies that do predate the publication of these criteria may still be eligible based on fulfillment of criteria one to three. If participant inclusion criteria are not fulfilled we will decide by discussion whether participants adequately met diagnostic criteria for type 1 or type 2 diabetes at the time the study was performed and will make a decision about inclusion accordingly, and document this in the results.
Types of interventions
We will consider the following interventions for inclusion in the review.
Gangliosides versus placebo.
Gangliosides versus no treatment.
Gangliosides versus another treatment.
Gangliosides will be of any formulation, mixed or single, at any dose, and for any duration. If different mixtures of gangliosides are used, we will report both aggregated results and results of each mixture. Diabetes is a long-term condition, and if the disease is not cured, its complications are likely to require long-term treatment; a single-dose study is unlikely to be of any use. No available data support the use of gangliosides over a prolonged time course. In animal studies, some evidence supports shorter durations of treatment, albeit in a rat hippocampus model (Skup 1987). In studies of which the authors are aware, length of treatment has varied between three and six months. Therefore, we have specified a minimum duration of three months for the intervention to be applied.
Gangliosides may be given in conjunction with other treatment such as insulin, or dietary restriction, as long as this is the same for both groups.
Types of outcome measures
No outcomes for diabetic neuropathy intervention studies have been universally accepted; no gold standard is used worldwide. The outcomes chosen reflect what we consider to be the most appropriate direct and surrogate markers of change in the status of diabetic neuropathy.
Change from baseline in impairment as assessed by a validated scale such as the Michigan Neuropathy Screening Instrument (Feldman 1994) or the Neuropathy Impairment Score (Dyck 2005) at 12 months. We deliberately have not specified a particular scale, as it is unlikely that all relevant studies would have utilised it. In addition, although we now prefer dichotomised measures of outcomes, for instance, 30% or 50% reduction in pain, it is unlikely that the included studies will have used such measures, and for this review, we will use changes in mean values as our primary analysis.
Diabetic peripheral neuropathy is a long-term condition. Changes in impairment and disability are unlikely to occur rapidly, although changes in pain may be relatively acute and may impact on other functions. Therefore we will assess secondary outcomes at three months and 12 months, scaling the outcomes when necessary and appropriate.
Change from baseline in a composite impairment score (as above) at three months.
Change in mean sensory impairment, as measured by the sensory component of a validated clinical scale or a validated sensory impairment scale, at one year. When multiple outcome measures have been used, data may be dichotomised. Participants will be classified into improved, stable or deteriorated, according to reported changes in the original data. Change will be defined as change by greater than or equal to the minimum clinically important difference (MCID) for the scoring system used (if published), or 10% of the mean baseline value if the MCID is not available.
Change in a validated patient-assessed global symptoms score at three months and 12 months.
Disability. A change in function as assessed by a recognised and validated disability measure at three months and 12 months. The Overall Neuropathy Limitations Scale (ONLS) (Graham 2006) has been used for other neuropathy types. The Rasch-built Overall Disability Scale (R-ODS) score (van Nes 2011) is under validation, and other similar scores may become available.
Pain. We will assess change in mean pain scores from baseline, or the proportion of participants showing improvement of 30% or more, using any validated assessment scale, for example, a visual analogue scale (VAS) (Wewers 1990), the Gracely Pain Scale (Gracely 1988) or Likert scales (Likert 1932), at three months and 12 months.
Electrophysiological parameters in the following order of preference: (1) motor conduction velocity (MCV) in the peroneal nerve, median nerve and ulnar nerve; (2) distal compound muscle action potential amplitude (CMAP) in the peroneal nerve, median nerve and ulnar nerve; and (3) sensory nerve action potential amplitude (SNAP) in the sural nerve, at three months and at 12 months.
Adverse events. We will include all adverse events and will analyse them in three groups: any adverse events, adverse events leading to cessation of the interventions and serious adverse events leading to hospitalisation or prolonged admission, serious illness or disability or death. If this information is not available from the trials themselves, we will endeavour to obtain it from non-randomised sources and will include this information in the Discussion.
Search methods for identification of studies
We will search the Cochrane Neuromuscular Disease Group Specialized Register, The Cochrane Central Register of Controlled Trials (CENTRAL) (current issue), MEDLINE (1966 to present) and EMBASE (1980 to present).
We will use the following terms:
gangliosides, sygen, cronassial, diabetes, diabetic, diabetes mellitus, neuropathy, polyneuropathy, peripheral neuropathy, diabetic neuropathy.
The search strategies are presented in the appendices: CENTRAL—Appendix 1, MEDLINE—Appendix 2 and EMBASE—Appendix 3.
We will also conduct a search of the US National Institutes of Health Clinical Trials Registry, ClinicalTrials.gov (www.ClinicalTrials.gov), and the World Health Organization (WHO) International Clinical Trials Registry Portal (ICTRP) (http://apps.who.int/trialsearch/) for ongoing trials.
Searching other resources
All references in the identified trials will be forward and backward tracked (for references and cited articles) and study authors contacted to identify additional published or unpublished data.
Data collection and analysis
Selection of studies
Two review authors (SM and AGS) will independently check titles and abstracts of studies identified from the searches and will discard studies not fulfilling the inclusion criteria. We will obtain the full text of all potentially relevant studies, and two of the review authors (SM and AGS) will read them and make a consensus decision about inclusion.
Data extraction and management
Two review authors (SM and AGS) will independently extract relevant data from included papers using specially prepared data extraction sheets. They will compare extractions and resolve disputes by discussion. A third review author (MPTL) will settle unresolved disputes.
We will enter details of relevant included and excluded studies into the reference tables. SM will enter extracted data into the Cochrane statistical software Review Manager 5 (RevMan) (RevMan 2008) data tables and will analyse them as below. AGS will check the data.
We will record the selection process in sufficient detail to complete a PRISMA flow diagram and a 'Characteristics of excluded studies' table.
Assessment of risk of bias in included studies
Two review authors (SM and AGS) will independently assess risk of bias using the methods described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). We will consider the attributes of explicit diagnostic criteria, sequence generation, allocation concealment, blinding, completeness of follow-up, freedom from selective reporting and other sources of bias. 'Other sources of bias' will include single-centre and solo investigator trials and trials carried out by the pharmacological company producing the product. We will grade these items as being at low risk of bias, high risk of bias or unclear risk of bias and will describe the evidence on which our conclusions are based in a 'Risk of bias' table. If the assessments differ, we will seek agreement by consensus, if necessary in consultation with a third review author (MPTL).
Measures of treatment effect
As most outcomes are continuous data, the main measures of effect will be differences in means. In some cases, such as the primary outcome, when we anticipate the use of different scales in different studies, we will use the standardised mean difference (SMD).
With certain outcomes, especially the Patient Global Impression of Change, and neuropathic pain symptoms, for which Likert scales may be used, the data can be analysed as dichotomous or continuous. We will decide how Likert scale data will be analysed based on the number of options in the scale and the suitability of combining options into two main categories. If the data are treated as dichotomous, we will use risk ratio (RR) as the measure of treatment effect. If the number of choices is high, and it would be inappropriate to treat the data as dichotomous, we will treat them as continuous, and the mean difference will be the appropriate measure.
Unit of analysis issues
We anticipate that most studies will be parallel-group randomised trials, with randomisation occurring at the individual level.
If some studies are cross-over trials, we will use generic inverse variance (GIV) to analyse the results once differences in effects and standard errors have been estimated.
Dealing with missing data
We will attempt to minimise studies missing from our review by using a comprehensive search strategy.
In the case of any missing data, we will in the first instance attempt to contact the study authors to acquire this information.
As most outcome measures are continuous data, we may be faced with the problem of missing standard deviations (SDs). In this scenario, provided that most of the included studies have provided SDs, we will impute the missing SDs, using the average value from other trials measuring the same outcome in this review.
For other missing data, the steps taken to account for this will depend on whether the data are considered to be missing at random, or not at random. In the former instance, analysis will ignore the missing data, whereas in the latter, imputed values will need to be substituted and analysed with the rest of the data set.
In the case of individual dropouts, we will undertake analysis on an intention-to-treat basis.
Assessment of heterogeneity
We will assess studies for differences in methodology and participant characteristics. We anticipate significant heterogeneity in participant selection and the methodology of included studies. It may be appropriate to carry out subgroup analysis, although this may be limited by the amount of detail provided in the individual studies.
We will use a fixed-effect model in the first instance. We will assess statistical heterogeneity using the I2 test. If a high degree of heterogeneity is noted (I2 > 50%), we will inspect the studies and forest plots for sources of heterogeneity. If none is found, we will utilise the random-effects model.
Assessment of reporting biases
To reduce the risk of reporting biases, we will perform a comprehensive search, looking for both published and unpublished data from several different sources, including trial registries.
If sufficient studies are included, we will construct a funnel plot, which initially will be visually inspected for symmetry. If the meta-analysis includes more than 10 studies, we will perform tests to look for funnel asymmetry (e.g. the Egger 1997 test), which may show evidence of reporting bias. We recognise that funnel plots and their analysis may yield misleading results, and we will discuss this if we believe it to be relevant.
We will pool results from the included studies in meta-analyses using RevMan 5. We will pool these data based on measured outcomes with a fixed-effect model using the RevMan 5 SMD facility or a random-effects model if heterogeneity requires it. Results based on dichotomous outcomes will use the RevMan 5 facility for pooling RRs based on the Mantel-Haenszel method.
'Summary of findings' table
We will create a 'Summary of findings' table using the following outcomes.
Change from baseline in impairment at one year as assessed by a validated scale such as the Michigan Neuropathy Screening Instrument (Feldman 1994) or the Neuropathy Impairment Score (Dyck 2005).
Adverse events before one year.
Change in mean sensory impairment at one year, as measured by the sensory component of a validated clinical scale or a validated sensory impairment scale.
Change in a validated patient-assessed global symptom score at one year.
Disability: a change in function as assessed by a validated scale of disability at one year.
Pain: a change in mean pain score at one year.
We will use the five GRADE considerations (study limitations, consistency of effect, imprecision, indirectness and publication bias) to assess the quality of a body of evidence (studies that contribute data to the prespecified outcomes). We will use methods and recommendations described in Section 8.5 and Chapter 12 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011) using GRADEpro software. We will justify all decisions to downgrade or upgrade the quality of studies using footnotes, and we will make comments to aid readers' understanding of the review when necessary.
Subgroup analysis and investigation of heterogeneity
It is unlikely that the trials will provide sufficient data for a subgroup analysis to be performed. However, if available, we will analyse data in subgroups as follows.
Less than five years.
Longer than five years.
Neuropathy severity judged to be mobility limiting (needing a walking aid to mobilise, or worse) or not, as assessed by any clinical scale.
Type 1 and type 2 diabetes.
When appropriate, we will repeat meta-analysis with changes to important variables to assess the effects of those variables on the overall effect of the intervention.
We will endeavour to perform the following sensitivity analyses.
Trials with more than 50 participants, and trials with fewer than 50 participants.
Trials with a treatment duration of three months, and trials with a treatment duration of longer than three months.
Trials with a perfect methodological score, and trials with a less than perfect methodological score.
Fixed-effect model and random-effects model.