Proceedings of the 2013 annual meeting of the Netherlands Epidemiology Society
Volume 1 Issue S1 Abstract 5
Maarten J.G. Leening, Erasmus Medical Center, Rotterdam, the Netherlands
Moniek M. Vedder, Erasmus Medical Center, Rotterdam, the Netherlands
Ewout W. Steyerberg, Erasmus Medical Center, Rotterdam, the Netherlands
In 2008, a measure to evaluate improvements in risk predictions, the net reclassification improvement (NRI), has been introduced. However, despite its immense popularity, misconceptions persist in its interpretation. We aimed to assess the interpretation of the NRI and the reporting of methods and results in high-impact medical journals.
Using the Thomson Reuters Web-of-Knowledge, we selected all articles citing one of five methodological papers on NRI, published in NEJM, Lancet, JAMA, and Ann Intern Med. Two evaluators independently extracted data from all publications.
A total of 57 publications (53 original articles) were reviewed, of which the majority (75%) related to cardiovascular research. Of the 36 studies presenting NRI results, 34 (94%) pertained to category-based NRI. Of these, 31 (91%) publications detailed the risk categorization used, which corresponded to clinically meaningful thresholds in only 6 (18%) instances. The event and nonevent components of the NRI were reported in 23 (64%) of the
36 studies that presented NRI results. The overall NRI was expressed as a percentage in 22 (61%) instances and erroneously interpreted as the percentage of the study population correctly reclassified in 8 (22%) studies. Many other incorrect interpretations of the NRI were encountered in this review.
In high-impact medical journals, NRI is frequently misinterpreted and the reporting of methods for the NRI computation is often incomplete. The category-based version of the NRI should not be used unless clinically relevant risk thresholds can be well motivated. More emphasis should be given to the NRI components, since their sum is difficult to interpret.
Published: 06 Jun, 2013