Menu Back to Poster-Presentations-Details

T-29: Term Indexing Technology to Support the Feasibility Assessment of a Network Meta-Analysis to Support Benefit-Risk Assessment

Poster Presenter

      Scott Snyder

      • Senior Safety Data Scientist
      • AbbVie, Inc.
        United States


Improve feasibility assessment and outcome selection for a network meta-analysis (NMA) across two therapeutic classes to support benefit-risk assessment through digital term collection (indexing) and synonym (ontology) management enabled by a semi-automated, technology based process.


Outcome terms reported across a study set were indexed, and synonyms were unified (bound) using an ontology management tool (late binding). A frequency report of available outcomes guided the selection of NMA endpoints. NMA was performed to quantitatively inform the benefit-risk assessment.


Twenty-one published randomized controlled trials across two therapeutic classes were selected via a literature search and included for digital indexing of all outcome terms. The terms without associated quantitative data were collected as reported, and term binding was performed by an ontology specialist to unify synonyms using a semi-automated ontology management tool. The outcome terms were categorized as adverse events, clinical endpoints, laboratory values/diagnostic, mortality, and withdrawal/drug discontinuation, and tagged as binary, continuous, count or association measure. The resulting frequency report of 516 bound outcome terms was used to make an assessment of reported safety and efficacy outcomes across the body of literature. Outcomes of interest were then selected based on frequency and severity for full extraction and inclusion into the NMA via the user-facing technology interface that guided outcome selection. NMA was performed across both safety and efficacy outcomes to characterize the benefit-risk profile in aggregate between treatments with two different mechanisms of action. Severe safety outcomes with a low frequency were described within the benefit-risk assessment independently of the NMA, as aggregate analysis of outliers was not scientifically appropriate. This frequency based approach to NMA enabled comprehensive benefit-risk assessment between two therapeutic classes with increased efficiency.


This case study highlights several benefits of using term indexing technology to guide outcome selection for NMA for benefit-risk assessment. Traditionally, outcomes for NMAs are selected based on prior knowledge, judgments, or even guesses. In the absence of visibility into all available data, unanticipated outcomes, notably safety, may be overlooked. The traditional approach therefore has several limitations: a) lack of a priori visibility into available outcome data may result in poor outcome selection, particularly with safety outcomes that are not yet fully understood; b) the diverse ontology across studies may result in omissions (missed synonyms); and c) term binding decisions are made at the time of individual outcome extraction (“early binding”), thus unanticipated terminology may be lost or mischaracterized. The frequency reporting approach described addresses many of the above limitations by allowing visibility across the entire landscape of reported outcomes. This supplements the traditional outcome selection processes and enables researchers to make informed decisions during outcome selection. It is important to note that this approach does not address the major limitation of NMA, reliance on statistical inference to draw conclusions. Additionally, the application of technology helps reviewers manage ontology (as opposed to binding prior to the collection of all terms) with both transparency and flexibility to make changes when scientifically warranted. Scientific integrity is maintained through the frequency based approach as the researcher has access to the frequency of studies that contain an outcome while blinding them to the quantitative magnitude of the outcome. This mitigates selection or binding bias that may unduly influence results. Rare, severe, unanticipated events that would otherwise be excluded from NMA can be identified and characterized to complement the results of the NMA providing a comprehensive summary of benefits and risks.