Evaluation of ambient AI-based clinical documentation in physiotherapy: A case study of Western Uusimaa Wellbeing Services County

Abstract

Ambient AI- assisted documentation shows significant potential for reducing documentation workload in healthcare. Empirical evidence on AI-assisted documentation in a physiotherapy setting remains limited, particularly outside English-language contexts. This case study evaluates the implementation of a Large Language Model (LLM)-based ambient AI scribe in public outpatient physiotherapy services at Western Uusimaa Wellbeing Services County in Finland. A longitudinal single-case design was applied using multiple data sources, including automatically logged timestamp data, user ratings, bi-weekly questionnaires and a self-reported control group comparison. Documentation time was analysed using clinician-level medians and non-parametric statistical comparisons. Sensitivity analyses were performed to address potential artefacts in the timestamp data. Median documentation time was lower in the AI scribe group compared to the control group, with reductions ranging from 27 % to 39 % depending on timestamp processing assumptions. Considerable between-clinician variability was observed, suggesting that time savings depend on workflow adaptation and documentation practices. User rating and error rate improvements were particularly related to an update to the speech recognition model. Low to somewhat positive effects were reported on therapeutic alliance and job satisfaction, though survey response rates were limited. The findings suggest that ambient AI-assisted documentation is viable in public outpatient physiotherapy in a Nordic language setting and may reduce documentation burden. The results contribute empirical evidence from a physiotherapy setting and highlight the importance of workflow integration to realize the full potential of AI-assisted documentation.