Cochrane Evidence Synthesis and Methods最新文献

This tutorial focuses on cluster-randomized controlled trials (cluster-RCTs). We will explain what cluster-RCTs are, why they might be used, and how to include data from cluster-RCTs in systematic reviews.

What is a cluster-randomized controlled trial?

In most RCTs, individuals are randomly assigned to intervention groups. In a cluster-RCT, groups of individuals (e.g., schools, communities, or clinics) are randomized to intervention groups.

Why use a cluster-randomized controlled trial design?

Table 1 outlines reasons that a cluster-RCT design might be used by researchers, and provides an example for each of these reasons.

How do I perform risk of bias assessments for cluster-randomized controlled trials?

An adaptation of the Risk of Bias 2 tool [1] outlines issues that should be considered when assessing the risk of bias of cluster-RCTs. Detailed guidance on the use of the adapted tool is also available [2].

What is a “unit-of analysis” error?

Individuals from the same cluster are likely to respond in a similar way to each other, and therefore observations made for these individuals cannot be assumed to be independent. It is important that this dependency is accounted for when analyzing data from a cluster-RCT.

If the effects of clustering are ignored, and the analysis is conducted as if individuals were randomized, a “unit-of-analysis error” occurs [3], as the unit of analysis (the individual) is different to the unit of randomization (the cluster). When a “unit-of-analysis error” occurs, confidence intervals for the effect estimate will be artificially narrow and associated p-values will be artificially small. The trial will also have too much weight in any meta-analysis. Incorrect conclusions may therefore be drawn from the results of the cluster-RCT itself, and any meta-analyses that include the cluster-RCT.

How do I include data from a cluster-randomized controlled trial in a systematic review?

When a cluster-RCT is included in a systematic review (with or without a meta-analysis), it is important that the effect estimate and its corresponding confidence interval are adjusted for the clustered nature of the data.

The ideal approach is to extract cluster-adjusted effect estimates and a measure of uncertainty (i.e., confidence interval or standard error) that have been calculated by the trial authors using statistical methods such as multilevel models or generalized estimating equations. These effect estimates and the measure of uncertainty may be included in meta-analyses that use the generic inverse variance method.

Another acceptable approach is to conduct the analysis at the cluster level. The data set for analysis would include a summary measurement for each cluster, and the sample size is the number of clusters. These data can then be treated as if they were from an RCT that ra

In this article, we look at risk ratios (RR), odds ratios (OR), and the risk difference (RD); what they are, how to interpret them, and when they should be used.

The RR, the OR, and the RD are used to compare the occurrence of an event in two groups for dichotomous outcomes. For example, a recent Cochrane review compared oral misoprostol to placebo for induction of labour and reported whether the women in each group went on to have a caesarian section, or not [1].

The RR provides a measure of how much higher or lower the risk of the event happening in the intervention group is, compared to the risk of the same event happening in the control group. Risk might sound like a complicated term but it is actually just the probability of an event happening.

Example: Trial A is interested in whether patients were readmitted to hospital within 30 days of an operation (procedure X) and compared this with patients who had a different operation (procedure Y). One hundred out of 200 patients who had procedure X were readmitted within 30 days. The risk (or probability) of this event occurring is 100 divided by 200, or 0.5. One hundred and fifty patients out of 200 patients who had procedure Y were readmitted within 30 days. The risk (or probability) of this event occurring is 150 divided by 200 or 0.75.

This means that if patients have procedure X, their risk of being readmitted is reduced by 33% compared to their risk if they had procedure Y. We can also express this as a probability—the probability of them being readmitted is reduced by 33%.

On the other hand, if 140 out of 200 patients who had procedure X (risk: 140/200 = 0.7) and 120 of 200 patients who had procedure Y (risk: 120/200 = 0.6) were readmitted within 30 days, then the RR equals 0.7/0.6 = 1.17. This means that if patients have procedure X, their risk of being readmitted is increased by 17% compared to their risk if they have procedure Y. When the risk in the intervention group is the same as the risk in the control group, the RR will be 1, and we can say that the risks are the same for both groups.

The OR provides a measure of how much higher or lower the odds of the event happening in the intervention group are compared to the odds of the same event happening in the control group. Odds may also sound like a complicated term, but it merely refers to the probability of something happening compared to the probability of it not happening. If the odds of a horse winning the Kentucky Derby are 7 to 2 against, this means that over nine races it would be predicted to win twice and lose seven times.

Example: Let's go back to our hypothetical example of procedure X versus procedure Y. If 100 out of 200 patients who had procedure X were readmitted within 30 days, the odds of readmission would be 100/100 or 1. In betting terminology, this would be “evens”: the chances of being readmitted and not being readmitted are the same. If

Evidence reviews are important for informing decision-making and primary research, but they can be time-consuming and costly. With the advent of artificial intelligence, including machine learning, there is an opportunity to accelerate the review process at many stages, with study screening identified as a prime candidate for assistance. Despite the availability of a large number of tools promising to assist with study screening, these are not consistently used in practice and there is skepticism about their application. Single-arm evaluations suggest the potential for tools to reduce screening burden. However, their integration into practice may need further investigation through evaluations of outcomes such as overall resource use and impact on review findings and recommendations. Because the literature lacks comparative studies, it is not currently possible to determine their relative accuracy. In this commentary, we outline the published research and discuss options for incorporating tools into the review workflow, considering the needs and requirements of different types of review.