Harmonizing occupational classifications in the UK health sector: challenges and opportunities for SOC 2020 and ISCO-08 integration

Purpose Occupational classification (OC) systems like SOC 2020 and ISCO-08 are vital for healthcare workforce planning; however, differences between these standards reduce comparability. This study examines the technical feasibility of harmonizing UK health OC through natural language processing (NLP) models. Design/methodology/approach First, a hierarchical analysis of OC systems was developed. Then, NLP models (i.e. Bag of Words, TF-IDF, SBERT and an ensemble) were proposed for semantic mapping and benchmarking of OC systems in the United Kingdom. A manual validation study with two domain-aware annotators assessed mapping accuracy. Findings We found that the SBERT model matches occupations more accurately than other models. Indeed, 1,308 matches from one job title to many job titles pairings were identified. Moreover, we benchmarked four UK health systems and showed that using SBERT, OC systems can achieve an average similarity and standard deviation of 63.21% (17.13%) for Scotland-Wales, 58.80% (16.49%) for Scotland-England and 61.28% (15.71%) for Scotland-Northern Ireland. Manual validation yielded precision scores of 0.73–0.98 and F1 scores of 0.84–0.99 across datasets. Using NLP simplifies the classification process, reduces administrative tasks and improves consistency with international standards. However, agreement levels varied across comparisons (76–90%), indicating that expert oversight remains necessary. Originality/value This study shows the technical feasibility of using NLP models for healthcare OC harmonization and offers a practical and scalable solution for improving OC alignment. This research applies NLP models jointly with benchmarking processes prioritizing practical adoption over methodological novelty, focusing on the UK healthcare sector.