The use of composite outcomes in neonatal trials: an analysis of Cochrane Reviews




Poster session 3


Tuesday 18 September 2018 - 12:30 to 14:00

All authors in correct order:

Lai N1, Yap AQY2, Ong HC3, Wai SX1, Yeoh JHH4, Koo CYY5, Lah WC5, Lim YS5, Roger S6, Colleen O6
1 School of Medicine, Faculty of Health and Medical Sciences, Taylor's University, Malaysia
2 Hospital Penang, Ministry of Health, Malaysia
3 Hospital Miri, Ministry of Health, Malaysia
4 School of Medicine, Faculty of Health and Medical Sciences, Taylor's University, Malaysia
5 School of Medicine, Faculty of Health and Medical Sciences Taylor's University, Malaysia
6 Cochrane Neonatal, USA
Presenting author and contact person

Presenting author:

Nai Ming Lai

Contact person:

Abstract text
Researchers combine multiple outcomes to form a composite outcome, either to increase the power of an analysis via increased event rates or due to the 'competing' nature of the components. However, the use of composite outcomes may mislead patients when the components differ substantially in their patient-importance, event rate and effect size.

We assessed how often composite outcomes were incorporated in neonatal trials and how comparable their components were.

We analysed randomised controlled trials (RCTs) in Cochrane Neonatal reviews published up to November 2017 and we extracted up to five outcomes per study. We selected all composite outcomes for further assessment. Two researchers independently determined the patient-importance of the components of a composite outcome, with discussion leading to a consensus. We compared the effect size and event rate between the components, and considered the components to have a substantially different event rate and/or effect size when their ratios were > 1.5.

We identified 2115 RCTs published between 1952 and November 2017 in 311 Cochrane Reviews. We extracted a total of 7621 outcomes and among them we identified 53 composite outcomes (0.70%) in 45 RCTs. There were two components in 49 composite outcomes and more than two components in four outcomes. The components in 34 composite outcomes (64.2%) had clearly different patient-importance, while the effect sizes and event rates differed substantially between the components in 24 (45.3%) and 34 (64.2%) outcomes respectively, with up to a 43-fold difference in the event rates between components.

A small proportion of neonatal RCTs included composite outcomes. However, when they were included, the majority had contrasting effect sizes and event rates between the components, with substantially different patient-importance between components in nearly two-thirds of the outcomes. When the evidence from such trials is used in communication with carers, health care providers should not only present the composite outcome as a whole, but also highlight the individual components sequentially, taking into account the relationship between the components.

Patient or healthcare consumer involvement:
Two authors (newly graduated medics) consulted their lay relatives on the patient-importance of the components during their judgement process.

Relevance to patients and consumers: 

Communicating evidence to patients requires careful interpretation of the evidence. Often the evidence presented may be misleading, and one example of potentially misleading presentation is when a composite outcome (an outcome that is made of two or more individual components combined) is reported in a trial. This is especially problematic when the components differ greatly in patient-importance and frequency of occurence. We studied the use of such composite outcomes in neonatal trials, specifically, how often they were used, how many had different patient-importance, frequency of occurence and treatment effects between the components.