# Developing an algorithm to identify individuals with psychosis in secondary care in England: application using the Mental Health Services Data Set

## TL;DR

Researchers created algorithms to identify psychosis cases in England's mental health data, aligning with national estimates.

## Contribution

Development of 12 algorithms using MHSDS data elements to identify psychosis in secondary care.

## Key findings

- The algorithms identified between 99,204 and 138,516 psychosis cases in 2017–2018.
- Most algorithms identified a similar core group of patients despite varying data elements.

## Abstract

There is currently no definitive method for identifying individuals with psychosis in secondary care on a population-level using administrative healthcare data from England.

To develop various algorithms to identify individuals with psychosis in the Mental Health Services Data Set (MHSDS), guided by national estimates of the prevalence of psychosis.

Using a combination of data elements in the MHSDS for financial years 2017–2018 and 2018–2019 (mental health cluster (a way to describe and classify a group of individuals with similar characteristics), Health of the Nation Outcome Scale (HoNOS) scores, reason for referral, primary diagnosis, first-episode psychosis flag, early intervention in psychosis team flag), we developed 12 unique algorithms to detect individuals with psychosis seen in secondary care. The resulting numbers were then compared with national estimates of the prevalence of psychosis to ascertain whether they were reasonable or not.

The 12 algorithms produced 99 204–138 516 and 107 545–134 954 cases of psychosis for financial years 2017–2018 and 2018–2019, respectively, in line with national prevalence estimates. The numbers of cases of psychosis identified by the different algorithms differed according to the type and number (3–6) of data elements used. Most algorithms identified the same core of patients.

The MHSDS can be used to identify individuals with psychosis in secondary care in England. Users can employ several algorithms to do so, depending on the objective of their analysis and their preference regarding the data elements employed. These algorithms could be used for surveillance, research and/or policy purposes.