Criteria for considering studies for this review
Types of studies
We considered all randomised and blinded trials of the effects of orally-administered calcium channel blockers compared with placebo, or other drug therapy, in people with primary Raynaud's phenomenon, using any method of randomisation. We included trials not analysed on an intention-to-treat basis, provided all randomised participants were accounted for. We included cross-over trials only if they incorporated (a) a placebo or no-treatment 'run-in' of at least one week (or provided baseline data for quantification of the severity of the condition), and (b) a 'wash-out' period of at least one week between treatment arms for all participants. We identified but did not include in the meta-analyses studies in which primary and secondary cases could not be resolved. We also excluded open studies and single-dose studies. There was no restriction on language, with publications in languages other than English translated and assessed for eligibility.
Types of participants
Adults, over 18 years of age, with clinical features of primary Raynaud's phenomenon. Trials including a mixture of primary and secondary participants were included if primary participants could be identified and analysed separately.
Types of interventions
Types of outcome measures
Physiological measurements (including digital temperature and blood flow response to hand cooling)
Adverse reactions (e.g. flushing, headache, tachycardia and ankle swelling), and withdrawal of medication.
Search methods for identification of studies
There were no language or publication status restrictions.
The Cochrane Peripheral Vascular Diseases Group Trials Search Co-ordinator (TSC) searched the Specialised Register (last searched February 2013) and the Cochrane Central Register of Controlled Trials (CENTRAL) (2013, Issue 1), part of The Cochrane Library, (www.thecochranelibrary.com). See Appendix 1 for details of the search strategy used to search CENTRAL. The Specialised Register is maintained by the TSC and is constructed from weekly electronic searches of MEDLINE, EMBASE, CINAHL, AMED, and through handsearching of relevant journals. The full list of the databases, journals and conference proceedings which have been searched, as well as the search strategies used, are described in the Specialised Register section of the Cochrane Peripheral Vascular Diseases Group module in The Cochrane Library (www.thecochranelibrary.com).
The TSC searched the following trial databases (February 2013) for details of ongoing and unpublished studies using the term 'Raynaud';
World Health Organization International Clinical Trials Registry (apps.who.int/trialsearch/)
Current Controlled Trials (www.controlled-trials.com/)
Nederlands Trials Register (www.trialregister.nl/trialreg/admin/rctsearch.asp)
Searching other resources
In February 2013, all trials listed under 'channel blockers' on www.trialscentral.org were checked by the authors for reference to Raynaud's phenomenon with no relevant results found.
Reference lists of relevant studies and reviews were also checked for potentially relevant studies.
Data collection and analysis
Selection of studies
We obtained full-text articles of all the references identified and where necessary translated them. Three review authors (HE, MA and AH) independently reviewed the articles and resolved disagreements by discussion and consensus. One review author (HE) contacted six trial authors or co-authors for additional information on articles where data from subgroups could not be identified. We obtained email addresses by searching the relevant article, the author's or co-author's most recent reference in PubMed, and 'Google'. We received replies from four.
Data extraction and management
Three review authors (HE, MA and AH) independently reviewed each study identified as being eligible for inclusion, and extracted data from included studies using the Cochrane Peripheral Vascular Diseases Group 'Data Extraction Table'. This includes method of allocation, degree of blinding, power calculations, exclusions post-randomisation, losses to follow-up, source of funding, number of participants, age and sex of participants, inclusion and exclusion criteria, treatment, control group, duration of study, and outcome measures. We resolved disagreements by discussion and consensus. Where a trial was described in multiple publications, we extracted data from the most complete report.
Assessment of risk of bias in included studies
Two review authors (MA, AH) independently assessed trial quality as being at 'low', 'high' or 'unclear' risk of bias across the following areas, with reference to the criteria listed in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011):
1. Adequate sequence generation.
2. Allocation concealment.
3. Blinding of participants and personnel, and of outcome assessment.
4. Incomplete outcome data addressed.
5. Free of selective reporting.
6. Free of any other bias: i.e. if no other potential sources of bias were identified.
We resolved disagreement about whether or not a trial fulfilled certain quality criteria by consensus, in discussion with the other review authors (HE, JW). All quality criteria ratings and supporting information are listed in the 'Risk of bias' tables (See Characteristics of included studies).
Measures of treatment effect
The rate of attacks was the only outcome for which we deemed it appropriate to conduct meta-analysis. The results of the remaining outcomes were described separately for each study. For attack rate, the measure of treatment effect used was the difference in mean number of Raynaud's attacks in the treatment and placebo groups. Mean numbers of attacks were reported over different time periods across the studies and, in the case of RTS 2000, geometric means were reported, requiring log transformation in order to make the comparison. Accordingly, we used a standardised mean difference as the parameterisation of treatment effect, calculated by the difference in means divided by the pooled standard deviation (SD) for each study. We calculated the individual study estimates of treatment effect in R (R Development Core Team 2012).
Numbers of attacks were reported on different scales across trials. We calculated standardised mean differences for each trial, in order to provide commensurable estimates of treatment effect. Following data synthesis, we transformed the pooled estimate of treatment effect and its 95% confidence interval to an interpretable scale by multiplying by a typical SD, obtained by pooling the standard deviations of the trials in the analysis (Higgins 2011, section 12.6.4).
Unit of analysis issues
The individual participant was the unit of analysis. For cross-over trials, the calculation of standard errors accounted for the fact that observations were paired (Elbourne 2002).
Dealing with missing data
Five trials had missing data due to participant withdrawal. In three of these (Sarkozi 1986; Vayssairat 1991; Wollersheim 1991), participant-level data were not reported and no intention-to-treat analysis had been attempted. For these studies, we analysed the data as reported based on all participants retained in the study. As such, we assumed missing data to be missing completely at random. In one of the five studies (Ettinger 1984), participant-level data were reported, but the number of participants with primary Raynaud's phenomenon was too low to permit meaningful imputation of missing values. One study (RTS 2000) presented results having conducted multiple imputation to account for missing data. We used these results in the meta-analysis. We contacted two authors for missing data or if the reporting of data was unclear and, while both responded, no additional data were available.
Assessment of heterogeneity
We examined statistical heterogeneity using a Cochrane's Q test and the I² statistic (Higgins 2003).
Wherever there appeared to be at least moderate statistical heterogeneity (I² > 30%), we performed a random-effects analysis to test the robustness of the results derived from the fixed-effect approach. Assessment of clinical heterogeneity was based on consideration of the protocols of the studies included in the analysis and their individual estimates of effect size. Wherever we suspected clinical heterogeneity, we performed a sensitivity analysis by repeating the meta-analysis with the clinically heterogeneous trials excluded.
Assessment of reporting biases
We entered primary outcome data from all included studies into a funnel plot (trial effect against trial size) to investigate the possibility of publication bias, but the small number of studies (seven) meant that we could reach no definitive conclusion.
We performed statistical analysis according to the statistical guidelines provided for authors by the Cochrane Peripheral Vascular Diseases Group and from the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011).
We combined the results of five cross-over trials and two parallel-group trials in a meta-analysis. These compared the calcium channel blockers nifedipine and nicardipine with placebo for the outcome 'frequency of Raynaud's attacks'. We calculated estimates of treatment effect and standard errors for each of the studies.
Two studies did not report SDs for treatment or placebo groups, and did not contain sufficient information to calculate them. This did not, however, prevent the calculation of a standardised mean difference for either study. For one study (Wollersheim 1991), the mean difference in the number of attacks, and therefore the standardised difference, was zero. We made a conservative assumption that there was no correlation between participants' measurements, resulting in a larger standard error and a relatively low weighting for the study within the meta-analysis. The other study (RTS 2000) reported geometric means for attack rates and adjusted P values. By using the upper limit of the P value and log-transforming the geometric means, it was possible to calculate a standard error for the difference in means. From this, it was possible to calculate a standardised mean difference and a corresponding standard error.
We then pooled the estimates using the generic inverse variance method, whereby each study is weighted according to the precision of its estimate of treatment effect. Following synthesis, we back-transformed the pooled estimate to an interpretable scale, as described above.
We used a forest plot to show individual study estimates and 95% confidence intervals (CI), together with the pooled estimate. We conducted the meta-analysis using the Review Manager 5 software (RevMan 2012) provided by The Cochrane Collaboration.
Subgroup analysis and investigation of heterogeneity
We performed separate subgroup analyses for the nifedipine and nicardipine trials. To investigate the influence of clinically heterogeneous trials, we performed sensitivity analyses with the trials in question excluded.
We conducted sensitivity analyses to examine the robustness of the meta-analysis as follows:
If a study had a discordant estimate of treatment effect, we repeated the analysis with that study removed.
If a study had a weighting that seemed disproportionate to its sample size, we repeated the analysis with that study removed.
If an analysis displayed at least moderate statistical heterogeneity (e.g. I² > 30%) we repeated the analysis using the random-effects method.