Outcome choice and potential loss of valuable information - an example from a Cochrane Eyes and Vision systematic review

Article type
Authors
Clearfield E1, Money S1, Saldanha I1, Chuck R2, Lindsley K1
1Cochrane Eyes and Vision US Satellite, Johns Hopkins Bloomberg School of Public Health
2Department of Ophthalmology and Visual Sciences, Albert Einstein College of Medicine
Abstract
Background: Outcomes selected for systematic reviews (SRs) should address clinical uncertainties to help make treatment decisions. However, when reviewers select outcomes by specifying the outcome’s 5 elements (domain, measurement, method of aggregation, metric, and time point) these may not match outcomes reported in trials included in SRs.

Objectives: To examine from a selected Cochrane SR outcomes reported from eligible trials not meeting the review-outcome definition.

Methods: We selected a Cochrane SR comparing classes of medications given after cataract surgery, which had no trials contributing to meta-analyses of the primary outcome. Clinician authors of the SR defined the primary outcome as the proportion of participants with grade >1 on the Standardisation of Uveitis Nomenclature (SUN) scale at 1-week follow-up (dichotomous). The SUN scale ranges from 0-4 and indicates the amount of cells and flare in the anterior chamber of the eye; higher grades indicate worse inflammation. Cell and flare can be measured by a slit-lamp or a cell and flare meter, and are recorded as the number of cells, amount of flare, or a combination. We compared the number of studies providing inflammation data per the SR outcome definition with the number of studies providing inflammation data using other outcome definitions.

Results: Of 48 studies included in the SR, none reported dichotomous inflammation data. Eighteen studies reported inflammation as a continuous outcome; however, there was variation in outcome elements. Replacing the review outcomes with mean inflammation at 1-week follow-up, we were able to include data from 7 studies (n=484 participants) in meta-analysis. Extending follow-up to 1-month postoperatively would have added data from 4 more studies.

Conclusions: The choice to use a dichotomous rather than continuous outcome for inflammation scores resulted in a potential missed opportunity to use available data from trials. Additionally, dichotomization of continuous outcomes at arbitrary cut-points runs the risk of losing valuable information. Our results underscore the importance for core outcomes sets to address all 5 outcome elements.