Development of a hospital electronic record frailty index (HerFI): an enhanced care alert score to identify older patients likely to require enhanced care on discharge from hospital

This thesis explores the different methods that are currently used to define frailty and the development of a new frailty index using routinely collected hospital data. The increasing ageing population means that older people account for the majority of the UK healthcare usage and spend therefore if the need for intervention can be quantified, adverse outcomes could be prevented. Data were extracted from the local systems at the Queen Elizabeth Hospital Birmingham for patients over 65 who were admitted as an emergency. A combination of 31 routinely collected test results were extracted and used to calculate a frailty score called FI-QEHB, by taking the sum of deficits divided by the total number of measurements for each patient. Machine Learning techniques were then used to firstly perform multiple imputation on missing data and then Classification and Regression Tree Analysis to determine the most important variables that predict mortality. This technique reduced the number of variables required to calculate a frailty score down to 6 from 31, and the area under the receiving operating characteristic (ROC) curve was used to assess the performance when the frailty score was added into a multivariable logistic regression model to predict emergency readmissions, mortality and whether a patient was discharged to a care home