The search for a more specific algorithm/score to be used by physicians to predict acute liver failure (ALF) outcomes in established drug-induced liver injury (DILI) cases is a topic of major relevance to improve patient clinical care. We read with interest the article by Lo Re et al, 1 who proposed a new prognostic model for early prediction of ALF, encompassing platelet count and total bilirubin with a sensitivity of 0.91 and a specificity of 0.76. These results were obtained from a retrospective cohort of 15,353 Kaiser Permanente Northern California members presumed to have DILI.These figures show that of 2383 patients classified at high risk of ALF according to the proposed prognostic model, only 20 (0.8%) patients did in fact develop either intrinsic or idiosyncratic ALF (true cases), and in the remaining 2363 (99.2%) patients the score actually failed in predicting ALF outcome. In a clinical setting, such intensive monitoring is hardly justifiable in a population in which the score wrongly classifies the bulk of the patients as having a high risk of ALF.On the other hand, Lo Re et al 1 stated that the ability (sensitivity) of their score to predict ALF was higher than that reported by the Spanish DILI group. 2 However, sensitivity is not the only test operating characteristic that needs to be considered. A test is very sensitive if it determines a high proportion of positives that are identified correctly as such. However, its utility is questionable if the test is unable to discriminate false positives (ie, low positive predictive value). Indeed, the Spanish DILI algorithm identified 24 (14.7%) cases of ALF of 163 predicted, as compared with the 0.8% identified using Lo Re et al 1 scores. We wonder whether Lo Re et al 1 have applied the Spanish DILI prognostic algorithm to their DILI cohort and how it performed. If not, perhaps it could be worth performing this analysis.