Prognostic Scores: One Fits All?

Ye et al.1 assessed the utility of CHA2DS2-VASc score in estimating the risk of death, ischaemic and haemorrhagic events in patients with heart failure with reduced ejection fraction (HFrEF) and sinus rhythm. In addition to the classical parameters of discrimination and calibration, which are usually considered to judge the reliability of a score, we believe that each of the crucial characteristics should be systematically appraised before a score is routinely incorporated into clinical practice. As previously published by our group, when dissecting the general structure of a score, the following pivotal components must be systematically considered: determination cohort, validation cohort, performance, and the applicability in clinical practice.2 The CHA2DS2-VASc score was derived on a cohort of patients with documented paroxysmal or persistent atrial fibrillation, and thus truly different in terms of age and mean ejection fraction from the general population of patients with heart failure. Second, validation is a crucial step in the development of a score. It is meant to grant its applicability to different subsets of patients. The general validity of a score is best achieved by challenging it on a independent, community-based cohort.3 In this study, CHA2DS2-VASc score was validated on the Warfarin Versus Aspirin in Reduced Ejection Fraction (WARCEF) trial population. Although the validation cohort dataset should be independent, the characteristics of the patients included in the two subsets (i.e. determination and validation cohorts) should be homogeneous. In the WARCEF trial, patients were treated, by protocol, with either aspirin or warfarin, and therefore do not reflect the characteristics of patients with HFrEF managed in the community. This discrepancy might explain the poor performance of this score, as the generalized use of either an antiplatelet or anticoagulant agent may lead to a so called ‘zero inflated’ model. In other words, the quantitative disproportion of the pool of patients without events may limit the efficiency of risk estimation. Finally, we must appreciate that CHA2DS2-VASc can be considered a parsimonious score, including a limited number of clinical variables easily retrievable. However, we must also bear in mind that an AUC in the range of 0.6 means that only six patients out of ten are correctly classified by the score, which is close to the play of chance, represented by a value of 0.5. This observation raises a great concern in the use of such a score for clinical purposes. In summary, although prognostic scores should be incorporated into daily medical practice as routine diagnostic test, caution should be exercised in avoiding their use as a ‘magic’ number.