ABSTRACT
Clinical prediction models are often based solely on the use of structured data in electronic health records, e.g. vital parameters and laboratory results, effectively ignoring potentially valuable information recorded in other modalities, such as free-text clinical notes. Here, we report on the development of a multimodal model that combines structured and unstructured data. In particular, we study how best to make use of a clinical language model in a multimodal setup for predicting 30-day all-cause mortality upon hospital admission in patients with COVID-19. We evaluate three strategies for incorporating a domain-specific clinical BERT model in multimodal prediction systems: (i) without fine-tuning, (ii) with unimodal fine-tuning, and (iii) with multimodal fine-tuning. The best-performing model leverages multimodal fine-tuning, in which the clinical BERT model is updated based also on the structured data. This multimodal mortality prediction model is shown to outperform unimodal models that are based on using either only structured data or only unstructured data. The experimental results indicate that clinical prediction models can be improved by including data in other modalities and that multimodal fine-tuning of a clinical language model is an effective strategy for incorporating information from clinical notes in multimodal prediction systems. © 2022 IEEE.
ABSTRACT
BACKGROUND: As most automated surveillance (AS) methods to detect healthcare-associated infections (HAIs) have been developed and implemented in research settings, information about the feasibility of large-scale implementation is scarce. AIM: To describe key aspects of the design of AS systems and implementation in European institutions and hospitals. METHODS: An online survey was distributed via e-mail in February/March 2019 among (i) PRAISE (Providing a Roadmap for Automated Infection Surveillance in Europe) network members; (ii) corresponding authors of peer-reviewed European publications on existing AS systems; and (iii) the mailing list of national infection prevention and control focal points of the European Centre for Disease Prevention and Control. Three AS systems from the survey were selected, based on quintessential features, for in-depth review focusing on implementation in practice. FINDINGS: Through the survey and the review of three selected AS systems, notable differences regarding the methods, algorithms, data sources, and targeted HAIs were identified. The majority of AS systems used a classification algorithm for semi-automated surveillance and targeted HAIs were mostly surgical site infections, urinary tract infections, sepsis, or other bloodstream infections. AS systems yielded a reduction of workload for hospital staff. Principal barriers of implementation were strict data security regulations as well as creating and maintaining an information technology infrastructure. CONCLUSION: AS in Europe is characterized by heterogeneity in methods and surveillance targets. To allow for comparisons and encourage homogenization, future publications on AS systems should provide detailed information on source data, methods, and the state of implementation.