One of the biggest challenges in evidence-based practice today is the vast and ever-increasing volume of information, studies, emerging theories, and ideas. High-quality systematic reviews and meta-analyses are especially useful to summarize knowledge relevant to clinical questions. Therefore, the latest meta-analysis by Huang et al1 regarding the efficacy of dressings for preventing pressure ulcers (PrUs) is a welcome contribution to inform decision-making in this emerging area of clinical interest. Because a meta-analysis is an influential addition to the body of knowledge, this work merits some comments:
* In total, Huang et al1 identified, appraised, and summarized 35 studies. Of these, 22 studies were conducted in China and published in Chinese. Other significant and recent systematic reviews and meta-analyses on this clinical topic2,3 have failed to identify and include these 22 relevant studies, indicating substantial language and publication bias. The general impact of language and publication bias is well known,4 but this recent contribution provides a striking case. Obviously, this is an issue not only of language, but also of a question of database selection. MEDLINE and EMBASE are generally regarded as the most applicable databases to the disciplines of medicine and health research worldwide, but researchers and reviewers must be aware that there is a substantial amount of relevant research that is not indexed in either database. This case also illustrates the advantage of multilingual research teams and/or translation services in capturing pertinent research.
* The statistical pooling of results across studies provides a powerful way to estimate overall treatment effects. A prerequisite for this process is comparability of study designs, populations, interventions, and outcome measures. For instance, equivalent follow-up period is of utmost importance when comparing and pooling the findings from PrU prevention trials. Although initial pressure injuries may develop at one point in time (eg, during surgery or during patient transfers), the end result-a PrU-may take several days to become visible.5,6 According to Table 1, the follow-up period was unclear in 19 studies included in the Huang et al1 meta-analysis, and in 3 studies, it was 24 hours or less. Given that the time between development of a PrU and visible alterations to the skin varies from several hours to 3 to 5 days,7-9 a 24-hour follow-up period is clearly insufficient to adequately evaluate a preventive strategy.
* A major, yet unsolved, challenge in PrU trials is the use of appropriate endpoints. The dichotomous outcome (yes/no) of PrU formation depends in part on the definition of a PrU that is used, and the PrU classification system that is applied. Challenges in establishing validity and reliability when using PrU classification systems are well documented.10 One common approach to enhance study validity is to exclude Category/Stage I PrUs when analyzing the outcome data. Whether Category/Stage I PrUs were included or excluded in the outcome measures of the studies included in the meta-analysis by Huang et al1 is unclear. Furthermore, it is now accepted that there are 2 main PrU categories: superficial and deep.2,6,10 Both types represent different pathophysiological pathways (ie, "top down" and "bottom up") and should be considered differently in clinical trials,11,12 through stratification of patients based on PrU severity12 and use of different outcome measures13 (eg, wound closure is not an appropriate outcome measure for a Category/Stage 1 PrU). Pooling findings for different PrU severities in a meta-analysis may produce inaccurate conclusions regarding efficacy, particularly in respect to the effectiveness of a dressing for preventing or healing deep and/or chronic PrUs. The authors further complicate the endpoints for the review by first stating that proxy outcome measures (eg, interface pressure) will be excluded from the review,1 but later reporting on ability of dressings to reduce skin friction and shear, both of which are also proxy outcome measures.
* It is widely accepted that pooling of study results is appropriate and useful only when the primary studies are free from bias. Therefore, only well-conducted randomized controlled trials (RCTs) should be used for pooling treatment effects of intervention studies. Although the results of quasi-experimental studies or other study designs might also be pooled, review authors should assume greater heterogeneity and risk of bias, and the pooling of different study designs into one single effect estimate should be avoided.4
* In determining the quality of studies included in their meta-analysis, Huang et al1 used the Cochrane risk of bias tool, which is an established methodological appraisal tool for RCTs.4 However, this tool is not appropriate for the appraisal of quasi-experiments or other study designs. For instance, the included study by Meaume et al14 was an observational study with no experimental intervention. Appraising randomization, allocation concealment, and blinding is inappropriate for observational study designs.
* The authors used the I2 statistic to guide selection of a fixed effect versus a random-effects model. Irrespective of the applied test statistic, it is important to remember that the fixed-effects model assumes that there is one true effect size that is shared by all the included studies. Studies need to have equivalent characteristics and study effects.15 We doubt that this was the case in all pooled studies, especially considering the comparator in most included studies was the "standard care protocol,"1 which is unclear terminology with widely varied meaning, particularly when comparing studies conducted in international settings.
In conclusion, the contribution by Huang et al1 prompts reflection on the following 2 major points:
* Reviewers, researchers, and clinicians perusing the literature must be aware that there may be substantial research that is not included in the major databases MEDLINE and EMBASE. Search strategies in future systematic reviews should consider other databases and aim for reflection of the global body of knowledge.
* Because clinicians often seek out evidence summaries before considering the evidence presented in single studies, it is essential that meta-analyses are conducted using appropriate methodology. Reviewers should address the methodological problems we have outlined here in order to accurately reflect the evidence and provide clinicians with appropriate guidance.
-Jan Kottner, PhD
Charite-Universitatsmedizin Berlin, Berlin, Germany
-Emily Haesler, BN, PGradDipAdvNsg
Australian National University Medical School, Canberra, Australia; Department of Nursing and Midwifery, La Trobe University, Melbourne, Australia
References