Complementary and alternative medicine for treatment of atopic eczema in children under 14 years old: a systematic review and meta-analysis of randomized controlled trials

We conducted systematic literature searches in 12 electronic databases, including 4 Chinese databases (China National Knowledge Infrastructure (CNKI), Wanfang Database, Chinese Scientific Journal Database (VIP), and SinoMed), and 8 English databases: PubMed, EMBASE via OVID, AMED (Allied and Complementary Medicine Database) via OVID, CINAHL (Cumulative Index to Nursing and Allied Health Literature) via EBSCO, PsychoInfo, CAM-QUEST, the GREAT database (the Global Resource for Eczema Trials: www.​Greatdatabase.​org.​uk), and the Cochrane Library from their inception date until May 2018. The filters were English and Chinese language (Additional file 1). We also searched in the grey literature such as conference proceedings and dissertations in CNKI and Wanfang for unpublished trials and trial protocols. References of all included studies were hand searched for additional eligible studies.

Two authors (CL Lu and SB Liang) independently examined the full text to identify the eligible trials. Four authors in pair (CL Lu, XH Liu, X Wang, and X Bai) extracted data independently from the included studies according to a predesigned data sheet. Any disagreement was resolved by discussion with a third author (JP Liu). Following items were extracted: publication year, study type, funding, inclusion/exclusion criteria, diagnostic criteria, study methodology, demographic characteristics of the participants, details of intervention and controls, outcome measures methods, adverse events, and results.

Two authors (CL Lu and XH Liu) used the risk of bias tool [16] to assess the methodological quality of the included trials. Seven items including random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting and other bias such as pharmaceutical funding, were used to be judged as “low risk”, “high risk”, or “unclear risk”. Any disagreements were resolved by discussion with a third author (JP Liu).

We used RevMan 5.3 software for data analysis. For continuous data, we used mean difference (MD) with 95% confidence intervals (CI), and for dichotomous data we used relative risk (RR) with 95% CI. We performed meta-analyses for trials if the study design, participants, interventions, control, and outcome measures were similar. Bulk data were synthesized quantitatively by descriptive counting. Other data not suitable for pooling analysis were synthesized qualitatively.

We used I-square (I²) to test the statistical heterogeneity as recommended by the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). We considered I² statistic value greater than 50% as a suggestion that there might be substantial heterogeneity [16]. We used random effects model for data pooling with significant heterogeneity (I² ≥ 50%), otherwise a fixed effect model was applied. If the data were available, we did subgroup analyses for subcategories of CAM modalities.

A sensitivity analysis was conducted to explore the influence of the type of randomized trials (parallel or cross-over randomized) and the quality of trials (high or low) if the data were available. A funnel plot was generated to explore possible publication bias if more than ten trials were included in a meta-analysis.

Our searches identified 4807 citations. After reviewing the titles and abstracts, 3034 citations were excluded due to duplication, reviews, and non RCTs. After scanning the full texts to identify the participants who were over 14 years, we excluded 1648 publications. Among 125 publications that were eligible, three publications were excluded [17‐19] due to inappropriate allocation of participants. We excluded 98 trials for the intervention of TCM in separate systematic review. Finally, there were 24 trials [20‐43] with a total of 2233 children (< 14 years) included in this review (Fig. 1). Eleven trials [33‐43] were published in English, and 13 trials [20‐32] were in Chinese. We did not identify any unpublished study. Twenty-two trials [20‐41] had two arms with parallel groups, one trial [42] had three arms, and one trial [43] had five arms.

The details of the 24 trials are presented in Table 1. The sample size of these studies ranged from 15 to 298 participants. The age ranged from 2 months to 13 years. We defined the conventional therapy with more than two modalities (e.g. topical and systemic anti-allergic, and immunomodulatory therapy) as “usual care” in 24 trials [9]. Every trial had more than two modalities of conventional therapy except placebo. Therefore, from the component level, CAM modalities of 22 trials [21, 23‐43] in the intervention group could be classified as probiotics, diet, biofilm, and borage oil (undershirts coated with oil) (more in Table 2). Moreover, two trials [20, 22] had their main modality referring to swimming and biofilm while accompanying other modalities. So, we considered for three different comparisons in the two-arm trials from system level: CAM versus usual care [20‐22], CAM plus usual care versus usual care [23‐32], and CAM (probiotics) versus placebo [33‐41]. For the two trials with three or more arms [42, 43], probiotics was compared with other formula of probiotics, placebo, usual care, or observation with no intervention.

The trials only reporting that the study was “randomized” were defined as “unclear” risk of bias, while the trials describing a specific method of randomized sequences generation, allocation concealment, and blinding as “low” risk of bias. Other bias accessed the funding scheme. Trials supported by non-commercial funding were defined as “low” risk of bias, trials funded by pharmaceutical companies were classified as “high” risk of bias, no information as “unclear” risk of bias. Eight trials [22, 24, 29, 31, 33, 34, 37, 41] reported the random allocation by random number table or computer generated-list. Only five trials [33, 34, 37, 38, 40, 41] reported the allocation concealment by using computer-generated random numbers or randomization software, which can conceal the allocation automatically. Ten trials [33, 34, 36‐43] reported double blinding. Eight trials [33‐37, 39, 41, 43] reported the drop out in both intervention group and control group, and only four trials [33, 39, 41, 43] used intention-to-treat (ITT) to analyze for all outcome [33, 39, 41] and primary outcome [43], and the other four trials [34‐37] analyzed data by per-protocol (PP) and reported the data of available participants. Besides, two trials [39, 41] analyzed by both ITT and PP. We considered one trial [34] as high risk of incomplete data for loss to follow up without ITT analysis because of 13 withdrawals in intervention group and 16 withdrawals in control group among the 100 participants in the trial. Four trials [33, 39, 41, 43] were considered as low risk of bias and the others as unclear. Eleven trials [21, 25, 29, 33, 35‐37, 40‐43] mentioned non-commercial funding. Three trials [33, 39, 43] reported that probiotic manufacturers produced the drugs used. We considered these three trials [33, 39, 43] as high risk of bias for conflict of interest (Fig. 2). Sample size calculation was reported in five trials [33, 35, 37, 41, 43] according to disease prevalence [33], symptoms and signs reduction in treatment group by 30% [34, 41], 34% [37] and in placebo group by 15% [34], 17% [37], 10% [41], and symptoms and signs scale of SCORAD for a standard deviation increments of 7.65 [43]. The other trials did not report any detail of sample size calculation. We considered the other bias of power calculation as unclear.

The 24 trials tested five CAM interventions: combination of probiotics (72%, n = 18) [23‐25, 27‐43], biofilm (12%, n = 3) [22, 26, 32], diet (8%, n = 2) [21, 43], swimming (4%, n = 1) [20], and undershirts coated with borage oil (4%, n = 1) [38] (Table 2).

Twenty-two two-arms trials [20‐41] involved CAM modalities, including probiotics, swimming, diet, borage oil, and biofilm. Three different comparisons were summarized: CAM versus placebo, CAM versus usual care, and CAM plus usual care versus usual care alone. In addition to this, two multi-arm trials [42, 43] used different probiotic formulae in different groups to compare with placebo, observation with no intervention or diet. Table 3 showed the detailed results of the effect estimation.

Since the fact that each comparison did not include more than 10 trials, we were not able to conduct meaningful funnel plot analysis in order to identify the publication bias. Due to the same quality and the type of randomized trials, we could not conduct sensitivity analysis in this aspect. Besides, significant heterogeneity in two outcomes with two comparisons was more than 50% (I² ≥ 50%), so we conducted a subgroup meta-analysis or a meaningful sensitivity analysis.

The global improvement (≥95% improvement) and ≥ 50% improvement of symptoms and signs in probiotics compared with usual care in three trials with 566 participants [20‐22] showed I² as 91 and 92%. The interventions were very heterogeneous in the trials including swimming [20], diet [21], and biofilm [22]. One trial [20] investigated not only swimming but also Chinese herbal medicine lotion and tuina. We conducted a sensitivity analysis, which showed improvement from CAM both for global improvement ≥95% improvement) (RR 1.77, 1.36–2.31; n = 2) and ≥ 50% improvement of symptoms and signs (RR 1.33, 1.16–1.52; n = 2).

This review identified 24 RCTs involving 2233 children (< 14 years) with AE. The findings suggest that some of the CAM modalities used alone or in combination with usual care may relieve the symptoms and signs of AE with ≥95% and ≥ 50% improvement, such as itchiness, skin lesions, swelling, and papula, in addition to reduce relapse of eczema. Moreover, some CAM modalities (such as probiotics) showed significant effect compared with placebo. The evaluated modalities appear to be safe and tolerated for lower relapse rate in CAM modality group. In spite of unclear pathogenesis of AE, CAM modalities may reduce symptoms and signs, and relapse of AE compared to conventional therapies.

The majority of trials had unclear risk of bias in many domains such as allocation concealment, blinding, missing data, and sample size calculation. Due to the unclear risk of bias of included trials, we could not come to firm conclusions from the evidence of the included trials in this review.

By searching the Cochrane Library with “AE”, “AD”, and “CAM”, there are 19 Cochrane reviews and 11 other reviews published, and after scanning, 9 reviews [13, 44‐51] with CAM related to children (< 14 years) for treating or preventing. These reviews or protocols included both children and adults, and even pregnant women to prevent, cure and explore the pathogenesis of AE. We found only one protocol [44] similar to our review, but was withdrawn as the author gave up the title “Complementary and alternative medicine treatments for AE”. In addition, the previous studies did not exclude the allergic diseases (such as asthma, and intestinal diseases) to focus on AE. Our review included children (< 14 years) suffering only from AE. Moreover, most reviews investigated the specific treatment like probiotics, based on pathological mechanisms but ignoring the complex and unclear pathogenesis of AE. The findings of our review are based on the symptom relief of AE and we included more comprehensive trials involving CAM for children (< 14 years).

Although great effort was made to retrieve all trials, we still cannot confirm that we were able to cover all the evidence due to non identified unpublished data. Besides, selecting and extracting data may also lead to some bias. We included only children under the age of 14 as this is the maximum age as younger adolescents defined by WHO, which may exclude some studies due to unavailable data for their participants over the age of 14 years. In addition, due to the various treatment for AE with an integrated “whole system” of care approach, we considered the control group with usual care as the system effect. The intervention group with CAM for a specific modality as the component efficacy, which cannot be used to document or disprove the effectiveness of a “whole system” treatment approach [15]. Additionally, in terms of the statistical heterogeneity and the variability in the CAM modalities, we were not able to conduct a subgroup meta-analysis, meaningful sensitivity analysis and funnel plot analysis. These factors limit the conclusiveness and robustness of this systematic review.

In fact of the limitations of this review, future trials should be designed as multi-center, double-blind placebo controlled trials with sufficient power, and reported according to the CONSORT (Consolidated Standards for Reporting Trials) Statement [52]. In addition, trials should record the relapse with sufficient length of follow-up. Besides, it is important to provide the definite safe treatments to the patients. Thus, adverse events in each group should be reported respectively so that we could retrospect the reason of different modalities and be easy to estimate the safety of CAM modalities.