Our literature search yielded 783 studies, of which 6 fulfilled our inclusion criteria [6, 8–12]. From them, only 1 study presented calibration analysis data (Hosmer–Lemeshow test) for their predictive model [10], whereas 4 reported discriminative analysis data (AUC: area under the curve) [6, 8, 10, 12]. The Hosmer–Lemeshow test is a statistical tool for checking goodness-of-fit of a logistic regression model by determining whether the observed event rates match expected event rates in subgroups of the model population. In one of these studies [8], a bootstrap procedure was performed on 200 subsamples, and the 95% confidence interval (CI) of the receiver operating characteristic was based on the bootstrap distribution. None of the studies mentioned Brier scoring, which is a measure of the accuracy of probabilistic predictions. In some of these studies, frailty measures were compared with established risk stratification scores [6, 10, 11]. A flow chart of our search results according to PRISMA guidelines is shown in Fig. 1. Overall, 9 different frailty assessment tools were identified (Table 1). The predictive effect sizes of the association of these tools with morbidity and mortality outcomes were reported as odds ratios (OR) or hazard ratios (HR) and 95% CI comparing frail versus non-frail cohorts (Table 2). Frailty has been used either as a continuous [6, 9] or as a categorical [8, 10–12] variable. Outcomes of interest for this review were 30-day, in-hospital and 1-year mortality rates; mid-term survival; composite outcome (mortality, major morbidity and complications); and prolonged length of stay or time to institutional discharge. The objectives of these studies varied from just assessing the relationship of frailty to complications [12] to evaluating the predictive ability of frailty [8, 9] or even to comparing the predictive ability of frailty to that of existing risk scoring systems [6, 10, 11].