The number needed to treat in pairwise and network meta-analysis and its graphical representation

ID: 

263

Session: 

Poster session 2

Date: 

Monday 17 September 2018 - 12:30 to 14:00

All authors in correct order:

Veroniki AA1, Bender R2, Glasziou P3, Straus S4, Tricco A4
1 Department of Primary Education, School of Education, University of Ioannina, Greece
2 Department of Medical Biometry, Institute for Quality and Efficiency in Health Care (IQWiG), Germany
3 Centre for Research on Evidence-Based Practice, Bond University, Australia
4 Li Ka Shing Knowledge Institute, St. Michael’s Hospital, Canada
Presenting author and contact person

Presenting author:

Areti Angeliki Veroniki

Contact person:

Areti Angeliki Veroniki
Abstract text
Background:
The number needed to treat (NNT) is an absolute measure of effect used to communicate the effectiveness or safety of an intervention. The NNT is frequently used in the pairwise meta-analytical literature and several attempts have been made to optimise its graphical representation. However, the need to compare multiple treatments for a clinical condition to make informed healthcare decisions requires the use of more complex statistical approaches, such as network meta-analysis (NMA). In spite of the widespread use of NNTs, they have not widely been used in NMA.

Objectives:
To suggest ways to calculate a NNT in pairwise meta-analysis and NMA using dichotomous data, and to present graphical approaches of NNTs from NMA to facilitate interpretation of results.

Methods:
A barrier to using NNT in (network) meta-analysis when an odds ratio (OR) or risk ratio (RR) is used, is the determination of a single control event rate (CER). We discuss five potential approaches to calculate the CER in meta-analysis, and present six graphical approaches for NNT for single or multiple outcomes assessed in a NMA. We illustrate the graphical approaches using a published systematic review and NMA on the comparative effectiveness and safety of cognitive enhancers for treating Alzheimer’s dementia.

Results:
The NNT calculation using a relative effect measure, such as OR and RR, requires a CER value, but different CERs may result in different NNTs. It is advised to calculate NNTs along with confidence intervals to include uncertainty around CER estimates. NNT results can be presented in several graphs. We suggest the NNT graphical representation in a barplot, Cates plot or forest plot for a single outcome, and in a bubble plot, scatterplot or rank-heat plot for at least two outcomes. Each plot is associated with different properties and can be used for different needs. For example, if uncertainty around NNT should be considered in decision making, then a barplot or a forest plot can be used. If multiple outcomes need to be considered, then a rank-heat plot can be used.

Conclusions:
Various graphs with different properties can be used to depict NNT. However, caution is needed when different considerations in NMA are made, including effect size and CER assumption across multiple comparisons, as these may impact NNT results and healthcare decision making.

Relevance to patients and consumers: 

The number needed to treat (NNT) provides insight into the clinical relevance of an effect size because it is defined as the average number of patients who need to be treated to prevent one extra person from having a bad outcome compared to a control or standard care of treatment. The NNT is a clinically friendly measure, and is commonly used in research. However, patients, healthcare providers, and policy-makers are faced with a multitude of intervention options and the need to compare several treatments for a clinical condition are required to make informed health-care decisions.