Prediction scores in gastrointestinal bleeding: a systematic review and quantitative appraisal

Background and study aims: Several algorithms predicting outcomes in acute gastrointestinal bleeding have been developed over the past three decades. These algorithms differ substantially and therefore the aim of the current study was to conduct a systematic review to compare their predictive performance and methodological quality in gastrointestinal bleeding. Methods: A PubMed literature search was performed up to 1 July 2011. All studies reporting prediction scores in gastrointestinal bleeding were included. Studies were analyzed for predictive performance, and a quality appraisal of these rules was performed for which a score range of 0 (lowest) to 29 (highest) was used. Results: A total of 372 studies were identified, of which 16 were eligible for inclusion. The studies evaluated different outcomes: mortality (n=5), rebleeding (n=2), intervention required (n=2), or a combination (n=7). The predictive performance of the identified prediction scores varied between an area under the curve of 0.71-0.92 (if given). The mean overall quality rating was 17 (SD 4.0, range 9-25). Major methodological shortcomings were the absence of validation and absence of impact analyses. Eight of 16 scores (50%) were determined "easy to use," and five scores (31%) reported some type of action based on the results. Conclusion: Substantial heterogeneity in outcomes and results was seen in the 16 identified prediction scores. Moreover, the methodological quality was suboptimal in most studies. However, we suggest that clinicians should use the "best available" scores according to performance and quality, which are the Blatchford score to assess the need for intervention, and the scores of Villanueva et al. for poor outcome, Guglielmi et al. for rebleeding, and Chiu et al. for mortality risk.