Criteria for considering studies for this review

Search methods for identification of studies

Electronic searches
We will use electronic search strategies, to identify relevant trials (as specified under 'types of studies'), and reviews or meta‐analyses (for identification of additional eligible trials).
The following databases will be searched:

• MEDLINE (up to present) (http://www.ncbi.nlm.nih.gov/PubMed/);

• The Cochrane Library (recent issue);

• NHS Economic evaluation database (NHS EED) (up to present) (www.cochranelibrary.com);

• EMBASE (up to present).

We will also search databases of ongoing trials:
Current Controlled Trials (www.controlled‐trials.com);
The National Research Register (http://www.nrr.nhs.uk/search.htm).

Search strategies were adapted from the Cochrane Handbook (Robinson 2002; Clarke 2003) and the Collaborative Metabolic and Endocrine Disorders Review Group. For a detailed search strategy see under 'Additional tables' (Table 1).

Additional searches
The reference lists of the identified reviews and included studies will be checked for additional eligible publications.

Additional key words of relevance may be identyfied during any of the electronic or other searches. If this is the case, electronic search strategies will be modified to incorporate these terms.

Data collection and analysis

Trial selection
Two observers (LW, GN) will independently inspect the titles and abstracts of the references identified to evaluate their potential eligibility. Trials will be included if all of the following criteria are met in the study:
1. The population of the participants consists of type 2 diabetic patients, not using insulin.
2. The design is a randomised controlled trial.
3. The intervention of the trial is self‐monitoring of blood glucose compared with usual care without self/monitoring of blood glucose (SMBG). Trials studying the comparison between blood glucose and urine glucose monitoring will also be included.
4. The outcomes are measured using at least one of the following measurements: HbA1c, fasting blood glucose, quality of life and well‐being, patient satisfaction.

There will be no language restriction. If the title does not provide enough information to decide whether or not to include the trial in the selection, the abstract will be read. The full article will be retrieved for clarification, if the abstract does not provide enough information.
Studies will be eliminated if both reviewers agree that the study does not meet the criteria for including studies in the review. A third party will resolve differences in opinion (JD). Interrater agreement for study selection will be measured using the kappa statistics Cohen 1960).
In case of duplicate publications we will check the consistency of the publications and include the publication with the longest follow‐up period.

Methodological quality assessment
The methodological quality of the relevant trials will be assessed independently by two reviewers (LW, EB) by means of a score list. We will use the Maastricht‐Amsterdam score (Tulder van 2003) list for randomised and controlled clinical trials that includes all criteria of the lists by Jadad et al. (Jadad 1996) and Verhagen et al. (Verhagen 1998)
The original list consisted of 19 items. According to van Tulder et al. (Tulder van 2003) we will only apply the 11 items pertaining to internal validity:

1) Was the method of randomisation adequate? *+
2) Was the treatment allocation concealed? *+
3) Were the groups similar at baseline regarding the most important prognostic indicators? *+
4) Was the patient blinded to the intervention? *+
5) Was the care provider blinded to the intervention? *+
6) Was the outcome assessor blinded to the intervention? *+
7) Were co‐interventions avoided or comparable?+
8) Was the compliance acceptable in both groups?+
9) Was the withdrawal/drop‐out rate described and acceptable?+
10) Was the timing of the outcome assessment in all groups similar?+
11) Did the analysis include an intention‐to‐treat analysis? *+
* Item from the Delphi list (Verhagen 1998)
+ Item recommended by van Tulder (Tulder van 2003)
All 11 items are included in the complete Maastricht‐Amsterdam list which is a combination of the items of Jadad 1996 and Verhagen 1998.

Each item has a rating scale of 'yes', 'no' or 'don't know'. If bias is unlikely, the item will be rated positive. If bias is likely, the item will be rated negative or if information concerning the item is not available, it will be rated with 'don't know'.
Studies fulfilling 6 or more of the 11 quality criteria will be considered to be of 'high quality'. All studies scoring less than 6 of the criteria will be rated as 'low quality'. The methodological quality of each study will be taken into account when estimating the overall effect by exploring the quality of the studies in a sensitivity analysis. Studies will not be excluded on the basis of the methodological criteria.
The initial level of agreement on the 11 quality criteria between the two reviewers will be reported as Cohen's kappa (Cohen 1960). Discrepancies on item level will be discussed and if consensus cannot be reached, a third reviewer (GN) will take a final decision.
A pilot, using two trials excluded from the review, will precede the quality assessment of the RCTs in order to assess the feasibility of using this scale. All items will be clearly defined and both researchers should, by scoring articles with the list, obtain consensus on each item in the list.

Data extraction
Data concerning details of study population, intervention and outcomes will be extracted using items from the Cochrane Metabolic and Endocrine Disorders Editorial base generic data extraction form considered relevant for our review.

The adapted data extraction form will include the following items:
1. General information: authors, title, details of journal, year of publication, country, duplicate publication.
2. Trials characteristics: randomisation, concealment of allocation, intention‐to‐treat analysis.
3. Patients: inclusion and exclusion criteria, diagnostic criteria, age, gender, ethnicity, years of education, diabetes duration, weight, body mass index, fasting plasma glucose, HbA1c, co‐medications, other baseline characteristics.
4. Intervention(s): type of intervention in both groups, number of intervention events, number of participants in each group, duration of study and follow‐up, timing of the intervention.
5. Characteristics of methodological quality: power calculation, similarity of groups at baseline, blinding (patient, care provider and outcome assessor), number of withdrawals/drop‐outs, co‐interventions, compliance, and adverse effects.
6. Results: all available outcome measures reported in the trial.
7. Notes: any other available information reported in the study that can be important for the review.

A pilot test, again using two trials excluded from the review will precede the data extraction of the selected RCT's. This test is likely to identify data that are not needed or missing to optimse the data extraction sheet. Data extraction and data entry will be performed independently by two reviewers (LW, EB). Any discrepancies between reviewers will be resolved by discussion and if consensus cannot be reached, a third reviewer (GN) will take a final decision.

Data analysis
The results of each RCT will be plotted as point estimates with corresponding 95% confidence intervals. Statistical heterogeneity will be tested using the Z score and the Chi square statistic with significance being set at p < 0.10. Quantification of the effect of heterogeneity will be assessed by means of I squared, ranging from 0‐100% including its 95% confidence interval (Higgins 2002). I squared demonstrates the percentage of total variation across studies due to heterogeneity and will be used to judge the consistency of evidence. If the evidence of statistical heterogeneity is substantial, the potential sources of variation between the RCTs will be investigated. Regardless of any evidence of statistical heterogeneity, the influence of specific differences between the RCTs will be explored.
A quantitative analysis (statistical pooling) will be limited to clinical homogeneous studies for which the study populations, interventions and outcomes are considered to be similar by the reviewers by comparing the data extraction forms of the studies. If studies are clinically and statistically homogeneous, a meta‐analysis will be conducted using the fixed effect model. However, if significant heterogeneity is found, it is unreasonable to assume that there is one 'true' effect underlying the data, that is contant across different populations, and then a random effects model will be used. Separate meta‐analysis will be performed for the different outcome measures.
If studies are clinically and statistically heterogeneous or if data are lacking, a qualitative analysis (best‐evidence synthesis) will be performed using a rating system of levels of evidence to summarize the results of the studies in terms of strength of the scientific evidence. Findings will be considered consistent if more than one of the studies reports the same direction of the effect on the outcome measure. The rating system consists of five levels of evidence, based on the quality and the outcome of the studies (Tulder van 2003).
1. Strong evidence ‐ consistent findings among multiple high quality RCTs.
2. Moderate evidence ‐ consistent findings among multiple low quality RCTs and/or one high quality RCT.
3. Limited evidence ‐ one low quality RCT.
4. Conflicting evidence ‐ inconsistent findings among multiple RCTs.
5. No evidence ‐ no RCTs.
Possible sources of heterogeneity can also be assessed by subgroup and sensitivity analyses as described below.
The analyses will be carried out using the statistical module MetaView 4.2 in Review Manager 4.2 (Cochrane software).

Subgroup analysis
If there is a statistically significant effect for one or more of the main outcome measures and if the amount of data permits, we will aim to perform subgroup analyses to determine whether there are any systematic differences between groups of patients.

A subgroup analysis may be performed for:

• Glycated haemoglobin level and/or fasting plasma glucose level at baseline (subdividing into three groups of low, medium and high level‐based on data)

• Age groups (below 60 years, over 60 years);

• Gender;

• Duration of diabetes (based on data);

• Presence of complications (e.g. diabetic complications);

• Different comparison interventions;

• Type of treatment: oral hypoglycaemic agents, diet, exercise, no treatment;

• Duration of intervention (short, medium, long term‐based on data);

• Weight (normal (BMI: women less than 25, men less than 27), overweight (BMI: women 25‐30, men 27‐30) obese (BMI more than 30)).

Sensitivity analyses
We will perform a sensitivity analysis in order to explore the influence of the following factors on effect size:

• Repeating the analysis excluding particular studies on e.g. combined interventions;

• Repeating the analysis taking account of study quality, as specified above;

• Repeating the analysis excluding studies with a long follow‐up or with a large group of patients;

• Repeating the analysis excluding studies using the following filters: diagnostic criteria, language of publication, source of funding (industry versus other), country.

The robustness of the results will also be tested by repeating the analysis using different measures of effect size (risk difference, odds ratio etc.) and different statistic models (fixed and random effects models).

Small study bias
If the studies provide sufficient data, small study bias will be assessed by using a funnel plot, whereby effect estimates of the common outcome measure are plotted against sample size. The funnel plot will be examined visually. In the absence of bias, the plot will resemble a symmetrical inverted funnel (Sterne 2001).