case studies

Recalibration of Predictive Models

The predictive models contained within the Johns Hopkins ACG System were originally derived from United States (US) health care data. In 2006, work at Imperial College and University College London demonstrated that the performance of these models using National Health Service (NHS) data was as robust as the performance in the US.

It is considered good practice to recalibrate predictive models using local data to align them more closely with the way in which data is recorded within that country, how activity is costed and the different ways in which care is provided. For this reason, the ACG System team undertook a major recalibration of the primary predictive models within the ACG System in 2013. The exercise was repeated in 2016.

This short report provides details of the third recalibration of the predictive models within the ACG System. This recalibration was carried out by the Johns Hopkins ACG System Team in collaboration with NHS South, Central & West Commissioning Support Unit (SCW) during the summer of 2020.