Care Home Quality Changes

One of CQC's four strategic priorities is to become more intelligence driven; a key component of this is to use intelligence to prioritise which adult social care services should be inspected.

As part of this goal, CQC together with partner organisations have estimated machine learning models to predict ratings of residential and nursing care homes using a variety of quantitative data.

Data for approximately 16,000 care homes were available for this work, including: • Overall ratings data from CQC's inspections; most care homes have received only one inspection rating.

Ratings were aggregated into two classes for the purpose of this work. • Monthly time series data on the number of notifications received by CQC on events that care home providers are required to report (e.g. deaths, serious injuries etc.) • Data on bed occupancy, staffing and funding sources of residents.

These are collected at irregular time periods, typically but not always within a three-month period prior to inspection. • Static characteristics of care homes (type of care home, bed capacity, service user characteristics, regional location etc.) • Socio-economic characteristics of care homes' local areas • Certifications and results of inspections by third parties (e.g.

Foods Standards Agency) The following classifiers were tested: logistic regression, linear discriminant analysis, support vector classification, random forest, oblique random forest.

We were able to achieve a 60% true positive rate with a 5% false positive rate.