Development and validation of a complexity score to rank hospitalized patients at risk for preventable adverse drug events

Abstract

The C-score was developed and validated for the identification of hospitalized patients at highest risk for pADEs. Aggregation of individual prediction models into a single score reduced its predictive power for most pADEs, compared with the individual risk models, but concentrated in the highest C-score decile a patient group more than two thirds of whom experienced at least 1 pADE.

Clinical usefulness of prediction tools to identify adult hospitalized patients at risk of drug‐related problems: A systematic review of clinical prediction models and risk assessment tools

“…The summary of the characteristics of all included studies is presented in Table 1. Of the 21 studies, eight involved risk assessment tools 38,40,42,43,56–59 and 13 clinical prediction models 15,32–37,39,41,60–63 . Among the risk assessment tool studies, four described the development of new tools, 38,40,42,43 three detailed validation process 56,57,59 and one assessed the impact of using the tools on clinical outcomes 58 .…”

“…The studies focused on a wide range of outcome measures: six studies focused on DRPs or MRPs, 15,41–43,57,62 five on ADEs, 35–38,56 four on ADRs, 32–34,63 five on MEs 39,40,58–60 and one on medication harm 61 . Such prediction tools were developed for use in identifying inpatients to receive targeted pharmaceutical care, such as medication review, 36,40,42,60 closely monitoring drug use 15,61 and medication therapy management (MTM) 37 …”

“…Discrimination was more often reported than calibration, in the form of AUROC or c‐statistic. The AUROC was reported in 10 of 13 studies 15,32–36,60–63 and c‐statistic reported in the remainder 37,39,41 . The AUROC ranged from of 0.63 35 to 0.70 61 (poor to fair discrimination) in internal validation and 0.62 32 to 0.78 15 (poor to acceptable discrimination) in external validations.…”

“…The AUROC ranged from of 0.63 35 to 0.70 61 (poor to fair discrimination) in internal validation and 0.62 32 to 0.78 15 (poor to acceptable discrimination) in external validations. The c‐statistic was only reported in internal validation, ranging from 0.657 41 to 0.838 37 (poor to good performance). Calibration was presented, with only four studies reporting the Hosmer–Lemeshow test 15,61 and a calibration plot, 39,41 which gave nearly perfect calibration.…”

“…27,28 Currently, many prediction tools have been developed to aid patient screening and prioritization in specific areas, such as fall risk assessments, 29 suicide risk assessment tools 30 and the Framingham risk score. 31 In pharmaceutical practice, prediction tools to identify patients at risk of adverse medication-related outcomes have been developed, for a variety of outcome measures such as ADRs, [32][33][34] ADEs, [35][36][37][38] MEs 39,40 and DRPs. 15,[41][42][43] There are two types of these tools based on the method of development.…”