Types of studies

Randomised controlled trials (RCTs) that randomised people or limbs to either the test intervention or a control intervention.

We did not find a reason to exclude right‐left comparative studies. Nevertheless, pretreatment of the skin of a body part investigated in a cross‐over design may influence the results as opposed to no pretreatment. To prevent the concerning additional bias, we excluded cross‐over designs. We suggested that simultaneous application of the intervention of interest on multiple body parts of each participant may distract the scoring of the Psoriasis Area and Severity Index (PASI) response and other subjective assessments. We are positive that right‐left comparative study should not include comparing different body parts. Different body parts may be affected differently and the assessment may introduce bias. We did not find a reason to exclude cluster‐randomised trials. Throughout the review, we used the term 'people or limbs' to accommodate people who were randomised, as well as randomised legs, arms, and elbows.

We planned to include the following designs.

• Simple parallel group design according to the Cochrane Handbook for Systematic Reviews of Interventions, section 9.3.1, quote: "participants are individually randomised to one of two intervention groups, and a single measurement for each outcome from each participant is collected and analyzed" (Higgins 2011).

• Multiple observations for the same outcome, specifically repeated measurements. If the data were to be included in a meta‐analysis, we planned to select a single time point and to analyse only data at this time point for studies in which it was presented.

• Multiple observations for the same outcome, specifically, multiple body parts receive different interventions according to the Handbook section 9.3.8, quote: "These trials have similarities to cross‐over trials: whereas in cross‐over trials individuals receive multiple treatments at different times, in these trials they receive multiple treatments at different sites." (Higgins 2011). In the present review, we considered paired comparisons within participants, such as comparisons of right versus left arm, leg, elbow, or knee. We planned to separately analyse the within‐participant data and the between‐participant data.

• Cluster‐randomised trials according to the Cochrane Handbook section 9.3.2, quote: "groups of participants, such as schools or families are randomized to different interventions. [...] Participants within any cluster often tend to respond in a similar manner, and thus their data can no longer be assumed to be independent of one another" (Higgins 2011).

We did not include the following study designs.

• Cross‐over design according to the Handbook section 9.3.3, quote: "all participants receive all interventions in sequence: they are randomized to an ordering of interventions, and participants act as their own control.". We expected a high risk of carryover according to the Handbook section 16.4.2, quote: "Carryover is the situation in which the effects of an intervention given in one period persist into a subsequent period, thus interfering with the effects of different subsequent intervention" (Higgins 2011).

• Multiple observations for the same outcome, specifically, multiple body parts receive the same intervention according to the Handbook section 9.3.7, quote: "people are randomized, but multiple parts (or sites) of the body receive the same intervention, a separate outcome judgement being made for each body part, and the number of body parts is used as the denominator in the analysis. [...] This is similar to the situation in cluster‐randomized trials, except that participants are the 'clusters" (Higgins 2011).

• Comparison of different body parts within participants, such as comparing arm versus leg.

Types of participants

Adults (i.e. 18 years of age or older) of any ethnic background or gender who have been diagnosed with chronic plaque type psoriasis by a dermatologist. We excluded people with pustular psoriasis, guttate psoriasis, or inverse psoriasis.

Types of interventions

The aim of this review was to assess the efficacy of salt bath added to ultraviolet B (UVB) light. We did not seek to assess efficacy of salt bath only.

Types of outcome measures

We did not use the outcomes listed here as criteria for including studies; they are the outcomes of interest within studies identified for inclusion.

Electronic searches

The Cochrane Skin Information Specialist searched the following databases up to 4 June 2019:

• the Cochrane Skin Group Specialised Register using the search strategy in Appendix 1;

• the Cochrane Central Register of Controlled Trials (CENTRAL) 2019, Issue 6 in the Cochrane Library using the strategy in Appendix 2;

• MEDLINE via Ovid (from 1946) using the strategy in Appendix 3;

• Embase via Ovid (from 1974) using the strategy in Appendix 4; and

• LILACS (Latin American and Caribbean Health Science Information database, from 1982) using the strategy in Appendix 5.

Searching other resources

Selection of studies

While preparing this systematic review, we endorsed the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta‐Analyses) statement, adhered to its principles and conformed to its checklist (Moher 2009). We downloaded all titles and abstracts retrieved by electronic searching to an Excel spreadsheet (Microsoft Corp 2011), and removed any duplicates. Two review authors (FP, SP) independently examined any remaining references. We included a study selection flow chart in the review (Figure 1).

Figure 1

---

---

Study flow diagram.

We excluded those studies that clearly did not meet the inclusion criteria and obtained copies of the full text of potentially relevant references. Two review authors (FP, SP) independently assessed the eligibility of retrieved papers. We resolved any disagreements by discussion between the two review authors; no third party arbitration was necessary. We documented reasons for exclusion in the Characteristics of excluded studies tables. If we identified multiple reports of one study we used the most‐up‐to‐date full‐text results. We checked the multiple reports for possible duplicate data, addressed the issue and did not include duplicate data in the analysis.

Data extraction and management

For each included study, two review authors (FP, SP) extracted study characteristics and outcomes independently onto a data extraction form. The data included: information on study design, participant characteristics (such as inclusion criteria, age, disease severity, comorbidity, previous treatment, number enrolled in each arm), interventions (such as type of irradiation, type of bath, dose applied, duration of therapy, control treatment), risk of bias, follow‐up duration, outcome measures, and deviations from the study protocol. We pilot tested The Cochrane Editorial Resources Committee's data collection form for intervention reviews (for RCTs only) available from the Cochrane Skin Group website: www.skin.cochrane.org/resources, then used this for data extraction. We resolved differences between review authors by discussion.

Assessment of risk of bias in included studies

Two review authors (FP, SP) independently appraised the risk of bias in the included studies. We resolved differences between review authors by discussion. We used the items listed within Cochrane's tool for assessing risk of bias (Higgins 2011):

• random sequence generation (selection bias);

• allocation concealment (selection bias);

• blinding of participants and personnel (performance bias);

• blinding of outcome assessment (detection bias);

• incomplete outcome data such as missing data (attrition bias);

• selective reporting such as not reporting pre‐specified outcomes (reporting bias); and

• other sources of bias such as bias related to the specific study design (other bias).

In general, a judgement of "low risk" of bias was given if plausible bias is unlikely to seriously alter the results, for example, if the participants and investigators enrolling those participants could not foresee the assignment. A judgement of "high risk" of bias was given if plausible bias seriously weakens confidence in the results, for example, if the participants or investigators enrolling those participants could possibly foresee the assignments. A judgement of "unclear" risk of bias was given if plausible bias raises some doubt about the results, for example, the method of concealment is not described or not described in sufficient detail to allow a definite judgement.

To draw conclusions about the overall risk of bias for an outcome, it was necessary to summarise the results of the 'Risk of bias' assessments performed for each individual study. We assessed the risk of bias at study level, not by outcome. We aligned the summary assessment to the Cochrane 'Possible approach for summary assessments of the risk of bias for each important outcome (across domains) within and across studies', as shown in Table 8.7a in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). We assessed whether the risk of bias for a single outcome was low, unclear, or high, then we interpreted whether the plausible bias was unlikely to seriously alter the results, raises some doubts about the results, or seriously weakened the confidence in the results. We assessed within a study whether there was a low risk of bias for all key domains, an unclear risk of bias for one or more domains, or a high risk of bias for one or more key domains. We assessed across studies whether most information was from studies at low risk of bias or at high or unclear risk of bias, and we assessed across studies whether the proportion of information from studies at high risk of bias was sufficient to affect the interpretation of results.

Measures of treatment effect

For time‐to‐event data such as survival, we planned to extract the hazard ratio (HR) and its standard error or confidence interval (CI) from trial reports; if these were not reported, we planned to estimate the logHR and its standard error using the methods of Parmar 1998 and using the tool provided by Tierney 2007.

For dichotomous data, such as PASI‐75 and treatment‐related adverse events requiring withdrawal, we estimated the proportion of events among the assessed participants. We pooled the data and estimated a risk ratio (RR) and a P value for overall effect by using the Mantel‐Haenszel statistic and a fixed‐effect model. If the P value was not reported by the study, then we estimated the P value by using the web‐based Easy Fisher Exact Test Calculator (Social Science Statistics 2018). In case of rare events, we planned to use Peto odds ratios instead, but this was not applied.

For continuous outcomes (e.g. PASI), we planned to extract the final value or change from baseline and corresponding standard deviation (SD) of the outcome of interest and the number of people or limbs assessed at endpoint in each treatment arm at the end of follow‐up. This was in order to estimate the mean difference (MD) or standardised mean difference (SMD) between treatment arms. We planned to analyse and present continuous data using the MD if all results were measured on the same scale (e.g. PASI). If this was not the case (e.g. fear, or quality of life), we planned to use the SMD. Eligible continuous outcomes were not reported.

We planned to extract the proportion of participants who reached the 50% reduction of the respective psoriatic lesion score compared to the baseline value at a fixed time point to allow pooling of data. If this was not possible, we planned to extract the time necessary to reach 50% reduction compared to baseline value. To calculate a mean difference, the number of people at risk and the probability of an event at a given time is required. Where studies did not provide that information, we chose to describe the individual results and not to pool the data. If data were given in a chart, we deduced the numbers from the chart.

Where possible, all data that we extracted were those relevant to an intention‐to‐treat (ITT) analysis in which all data were analysed in groups to which they were assigned. We stated if this was not possible. We noted the time points at which outcomes were collected and reported. We reported the 95% CIs for all analyses.

Unit of analysis issues

We included between‐participant data as well as within‐participant data. In general, within‐participant studies apply the intervention to a body part such as a limb and the comparator to a different body part such as the opposite limb (e.g. right arm versus left arm). Those data are distinct. Consequently, we separately analysed and reported information on between‐participant data and within‐participant data. PASI‐75 is the primary efficacy outcome for both data types.

If studies reported multiple intervention groups, we designated the intervention group and each different comparator group.

Dealing with missing data

We conformed to Cochrane's principal options for dealing with missing data (Higgins 2011). If data were missing, or only imputed data were reported, we planned to contact trial authors to request data on the outcomes of the people or limbs of the study. When relevant data regarding study selection, data extraction, and 'Risk of bias' assessment were missing, we planned to contact study authors to retrieve the missing data. If data remained missing after contact with authors, we analysed only the available data and addressed the potential impact on the findings in the Discussion.

Assessment of heterogeneity

We planned to assess heterogeneity (composed of dissimilar parts) between studies by visual inspection of forest plots; by estimation of the percentage heterogeneity between trials that could not be ascribed to sampling variation (I² statistic) (Higgins 2003); and if possible, by subgroup analyses. If there was evidence of substantial heterogeneity, we planned to investigate and report the possible reasons for this. An I² statistic greater than 50% was considered to indicate substantial heterogeneity, demonstrating a considerable variation in results. In this case, we planned to present the graphical display of a forest plot, but we did not plan to report an average value for the intervention effect.

Assessment of reporting biases

We planned to assess study protocols to see if the planned outcomes have been reported, but did not identify any. We planned to assess reporting bias, such as publication bias, by constructing a funnel plot if there were a sufficient number of included studies for a particular outcome (that is, at least 10 studies included in a meta‐analysis).

Data synthesis

We analysed the data using the Review Manager 5.3 software (Review Manager 2014); one review author entered the data (FP) and another review author checked it (SP). If sufficient clinically similar studies were available, we deemed it feasible to pool data. We planned to consider clinical homogeneity and statistical heterogeneity before pooling the data. If the I² statistic value was greater than 50%, we did not combine studies in order to report an average value for the intervention effect. We planned to use random‐effects models with inverse variance weighting for all analyses (DerSimonian 1986).

Concerning the dichotomous events of PASI‐75 and treatment‐related adverse events requiring withdrawal, we pooled the data and estimated a risk ratio and a P value for overall effect by using the Mantel‐Haenszel statistic and a fixed‐effect model.

Where possible, all data that we extracted were those relevant to an intention‐to‐treat (ITT) analysis in which all data were analysed in groups to which they were assigned. We stated if this was possible. We noted the time points at which outcomes were collected and reported. We reported the 95% confidence intervals (CIs) for all analyses.

We planned to use ordinal data from people who answered the Dermatology Life Quality Index (DLQI) questionnaire (Table 2) (Melilli 2006) or the visual analogue scale (VAS) on pruritus (itching), as well as from people who achieved a remission and people who experienced secondary malignancy. However, these data were not reported.

In the protocol, we addressed a problem that might arise if we included only a single study in an analysis of dichotomous data. In this instance, the confidence intervals around risk ratios calculated in Review Manager 2014 are unreliable. Where results were estimated for individual studies with low numbers of outcomes (less than 10 in total), or where the total sample size was less than 30 people, or limbs and a risk ratio was used, we planned to report the proportion of outcomes in each treatment group together with a P value from a Fisher's exact test. But this problem did not arise.

The analysis should be in intention‐to‐treat and all missing data were considered as failure.

Results on outcomes that were not predefined were also reported in Effects of interventions as additional information, though, the results were not relevant for the conclusion of the present review.

Subgroup analysis and investigation of heterogeneity

We did not undertake any subgroup analyses.

Sensitivity analysis

We planned to conduct a sensitivity analysis removing studies at high or uncertain risk of bias. However, we did not undertake any sensitivity analyses due to the limited number of studies included in this review.