METHODS RESEARCH

Abstract

Research ObjectivePrimary care networks (PCNs) are a new organizational hierarchy with wide‐ranging responsibilities introduced in the NHS Long Term Plan. The vision is that they represent “natural” communities of primary care practices (PCPs) with boundaries that make sense to practices, other health care providers, and local communities. Our study aims to identify natural communities of PCPs based on patient registration patterns using network analysis methods and unsupervised clustering to create catchments for these communities.Study DesignWe used a series of novel methods for unsupervised graph clustering. A cosine similarity matrix was constructed representing similarities between each PCP to each other, based on registration of patients in each Lower Super Output Area (LSOA)—a geographic division similar to census block groups. Unsupervised graph partitioning using Markov multiscale community detection was conducted to identify communities of PCPs. Catchment areas for each PCN were assigned based on the majority attendance from an LSOA.Population StudiedPatients resident in and attending PCPs in London identified from Hospital Episode Statistics from 2017 to 2018.Principal Findings3,428,322 unique patients attended 1,334 GPs in 4,835 LSOAs in London. Our model grouped 1,291 PCPs (96·8%) and 4,721 LSOAs (97·6%), into 165 mutually exclusive PCNs. The median PCN list size was 53,490, with a lower quartile of 38,079 patients and an upper quartile of 72,982 patients. A median of 70·1% of patients attended a GP within their allocated PCN, ranging from 44·6% to 91·4%.ConclusionsWith PCNs expected to take a role in population health management and with community providers expected to reconfigure around them, it is vital we recognize how PCNs represent their communities. We find that stable, representative collaborations between PCPs may be identified based on the similarity of the geographic locations of their registered patients. Our method may be used by policy makers to understand the populations and geography shared between networks.Implications for Policy or PracticeCollaboration between primary care providers offers the opportunity to improve the coverage and integration of local health care services. In the absence of data‐driven methods, these processes may proceed through interpersonal relationships between practitioners, which may not be representative of the local delivery of health care. This study uses Markov multiscale community detection, a data‐driven, unsupervised clustering method to identify “naturally occurring” communities of GP practices, which collectively form 165 PCNs across London. In doing so, this technique produces PCNs which are most representative of the spatial communities of patients for whom PCNs provide care. This approach may be used by policy makers and local commissioners across England to identify naturally occurring relationships between PCPs which are informed by existing patterns of patient registration. As the need for integration between community health care services grows, this method enables data‐driven, unsupervised understanding of the relationships that exist between providers of primary care and the communities they serve.