Performance assessment as an application of causal inference

N.T. Longford

Abstract

Institutions in health care, education and other public services are under constant pressure to perform to high standards and with efficiency. Assessment of their performance is often problematic because it either ignores important background variables of their patients, students or clients (the case-mix), or adjusts for them in a way that is not equitable or transparent. We apply a method motivated by the potential outcomes framework, in which we consider the hypothetical scenario of an institution's caseload being dispersed for treatment throughout the domain of assessment (the country's institutions). The target of estimation is the difference of the means of the outcomes in the realised and hypothetical settings. The method is applied to estimating the prevalence of bronchopulmonary dysplasia (BPD) in extremely preterm-born infants in the neonatal units and their networks in Great Britain in 2017. The prevalence of BPD is an audit measure in the UK National Neonatal Audit Programme.

Submitted

October 2018.