Predictive validity of tools used to assess the risk of unplanned admissions: a rapid review of the evidence
PATON Fiona, WILSON Paul, WRIGHT Kath
A synthesis of evidence assessing the predictive ability of tools used to identify frail elderly and people living with multiple long-term chronic health conditions who are at risk of future unplanned hospital admissions. There are now a large number of models available that can be used to predict the risk of unplanned hospital admissions and this study aims to provide a summary of their comparative performance. Overall, the models identified in this review show reasonable concordance in terms of their predictive performance (based on c-statistics). Models reporting other performance indications showed that at different thresholds, as sensitivity increased, specificity would decrease. As the algorithms become more complex or incorporate longer term horizons specificity increased but the ability of the models to identify future high cost individuals reduced. It should also be noted that whilst the reported c-statistics are broadly similar, the underlying populations, data sources and coding may differ.