Differences between the protocol and the current update

For differences between other published versions, please see the 'Differences between protocol and review' sections within the original publications.

Objectives: In the protocol, we had planned to assess the effects of treatments, but in this and the previous updates we broadened this to include management strategies because waiting for natural resolution is a recognised option for dealing with molluscum contagiosum. We amended the text from that which was in the protocol to make our objectives more clear.

Types of studies: In the protocol, we said that "studies should compare one or more treatments with another, with placebo, or with no treatment (waiting for natural response)"; we removed this sentence in this and previous updates because it refers to comparisons rather than studies.

Types of interventions: We had planned to include randomised trials of all treatments for molluscum contagiosum, but narrowed this to include only treatments aimed at eradicating molluscum contagiosum lesions, and excluded studies on other aspects of the treatment of molluscum contagiosum, for example on reducing pain in the studies that assessed the effect of using an analgesic EMLA (eutectic mixture of local anaesthetics) cream before the actual intervention took place. This was because the analgesic was not used to eradicate the molluscum lesions.

Primary outcomes: We decided that our original choice for 'short‐term clinical cure' of one month was not realistic, and therefore changed it to three months. We have also clarified our primary outcome to make it more manageable: short‐term clinical cure (up to three months after treatment). (Please see Overall completeness and applicability of evidence for a more detailed description of why we felt the original choice was not realistic.) We also deleted the term 'elevated' in the description, as we felt it was unnecessary and could possibly cause confusion, the implication being that there are elevated and non‐elevated forms of the lesion.

Where included studies used the term 'complete clearance' or 'free of lesions' or 'cured or > 90% cleared', we classed these as our primary outcome 'short‐term clinical cure (up to three months after start of treatment)' or our secondary outcome 'medium‐ and long‐term cure (after three months and up to six months, and after six months, respectively)'. Where studies have referred to 'partial clearance', we took this to mean our secondary outcome 'improvement'.

Secondary outcomes: We did not initially specify the outcome 'disease‐related quality of life' in the protocol, but added it afterwards as we considered it to be a relevant additional measure.

We also added 'short‐, medium‐, and long‐term improvement (including cure, intervals as above)' as a secondary outcome as we considered it to be important. For this outcome we combined 'improvement' and 'cure' (even though cure alone was a seperate outcome) because 'improvement' would be hard to interpret without also including those who were cured. For example: suppose in group A, 30% were cured and another 20% improved. In group B, 40% were cured and 10% improved. Comparing improvement rates between A and B (20% versus 10%) is misleading, whereas combining cure and improvement (50% versus 50%) is not.

Electronic searches: We expanded the number of trial registries that we planned to search when we became aware of the existence of these registries and in line with current Cochrane Skin practices. For similar reasons, we added Google as an additional electronic search strategy.

Selection of studies: If a randomised controlled trial included a variety of skin diseases, of which one was molluscum contagiosum, the number of molluscum participants needed to be at least five in the active treatment and placebo groups in order to reduce the role of extremely small studies. We added this criterion after the protocol was approved when we found a study that included 10 molluscum participants with a 9:1 distribution over the two treatment groups (Caballero 1996). The criterion also applied to Manchanda 1997a.

Selection of studies: If the setting of the study was not explicitly mentioned in the text, we assumed it to be carried out at the affiliation of the first author. Also, if the full text of a study was not available, we considered published abstracts for this update, as we have done this for previous versions of the review.

Assessment of risk of bias in included studies: In this update, we assessed each study using Cochrane's 'Risk of bias' tool (Higgins 2011), as this is now required. Also items (5), (6) and (7), which differed from the original protocol or were absent, were added or amended for the 2009 update as recommended. In previous versions of this review, items (3) and (4) were combined. For the 2016 update we further clarified how we decided what constituted an 'adequate' assessment and therefore low risk of bias.

Measures of treatment effect: Following the recent Cochrane Skin Group recommendations, we decided post hoc to re‐analyse results from individual studies with borderline significance and with low numbers of events (fewer than 10 in total) or a total sample size of less than 30, using Fisher’s exact test. The resulting P value was leading in interpreting the results.

Data synthesis: We had planned to express dichotomous results as odds ratios, but changed this to risk ratios and as a number needed to treat where appropriate because these are easier for most readers to understand. We decided to report numbers needed to treat only for comparisons with more than one study and only in the case of statistically significant differences, the latter because numbers needed to treat for differences that are statistically not significant produce large and uncertain confidence intervals.

When the same comparison between two interventions was made in more than one study, and studies appeared to have been executed in similar groups and settings, we planned to use statistical tests for homogeneity between studies. In those studies where the available data were sufficiently homogenous and where a pooled estimate of the treatment effect made sense, we planned to conduct a meta‐analysis. However, we could not implement these plans in most cases due to lack of data.

Assessment of reporting biases: Subsequent to the protocol, we aimed to assess reporting bias by comparing the published trial publications with the study protocol, but no protocols were available.

Unit of analysis issues: In our methods we planned to use special analytic techniques for paired (split‐body) designs; however, we were unable to do this as the paired data were not available to us.

Dealing with missing data: Although this was not specified in the protocol, we considered participants who dropped out or were lost to follow‐up as treatment failures.

Unit of analysis issues/Assessment of heterogeneity/Sensitivity analysis: We had planned analyses not documented in the protocol, including the use of sensitivity analyses to examine the effects of excluding studies with lower reported methodological quality, as well as how to analyse cross‐over trials and within‐participant designed trials. However, we did not undertake these analyses because of the small number of studies for each comparison.

Sensitivity analysis: We planned to use sensitivity analyses to examine the effects of excluding studies with high risk of bias. However, we did not undertake these analyses because of the small number of studies for each comparison.

Summary of findings: We developed 'Summary of findings' tables subsequent to our protocol. We have produced one for this update.

Quality of evidence: We used GRADE to assess the quality of evidence for each primary outcome and key secondary outcomes.