Identifying General Practice workforce patterns in England: a machine learning to support Evidence-Based health and social care policy

Background
General practices in England continue to face increasing workload pressures, recruitment and retention challenges, and inequalities in healthcare provision. Analysing national workforce composition and workload patterns is critical for informing health and social care policy decisions, and resource planning and allocation. This study applies machine learning methods to national workforce data to identify distinct clusters of general practices based on skill mix and workload. The aim is to generate an evidence base to inform strategic primary care planning.
Methods
Publicly available NHS workforce, demographic, and geographic datasets were used to conduct a cross-sectional secondary analysis of 6,202 general practices across England. After excluding practices with missing workforce data, 5,846 practices were included. The Skill Mix Index was calculated using Shannon Entropy across five direct care roles (GPs, nurses, pharmacists, physician associates, paramedics), while the Workload Index measured patient demand relative to workforce capacity. Both indicators were standardised before applying K-means clustering. A three-cluster solution was identified, followed by ANOVA and chi-square tests to examine differences in socio-demographic, geographical, and workforce characteristics. The geographical distribution of these clusters across England is illustrated in Figure 1.
Results
Three workforce clusters emerged. Cluster A (n=1,628) had the highest skill mix (Mean=0.60) and a low–moderate workload (Mean=0.30), representing large, well-resourced practices with broad multidisciplinary skill sets. Cluster B (n=2,916) had the lowest skill mix (Mean=0.35) but lowest workload (Mean=0.28), characterised by mid-sized practices with high proportions of GP FTE and the largest rural share. Cluster C (n=1,302) showed a low–moderate skill mix (Mean=0.38) combined with the highest workload burden (Mean=0.69), predominantly located in highly urban areas with high patient to GP ratios and the lowest appointment rates.
Conclusions
This study provides useful insights to develop primary care tailored policies. Cluster A practices are comparatively resilient and well-resourced, with broad, multidisciplinary teams that serve as a benchmark for effective workforce configuration. Cluster B shows relative stability but signals the need for continuity of care and GP retention strategies, particularly in mixed rural settings. Cluster C practices are under substantial burden and require prioritised policy action, including investment in workforce expansion, multidisciplinary role deployment, enhanced appointment capacity, and recruitment incentives. These insights can inform the NHS Long Term Workforce Plan by guiding decisions on staffing skill mix, improving access in high-demand urban areas, and strengthening resilience in mid-sized practices. The machine-learning approach offers a scalable, repeatable national framework for monitoring workforce patterns, enabling early identification of emerging pressures and supporting more dynamic, data-driven resource allocation.