Natural Language Processing in a Clinical Decision Support System for the Identification of Venous Thromboembolism:...

Area of Science:

• Medical Informatics
• Clinical Decision Support Systems
• Health Data Analytics

Background:

• The efficacy of Natural Language Processing (NLP) in detecting venous thromboembolism (VTE) from electronic health records (EHRs) across diverse clinical settings remains unclear.
• Accurate VTE detection is crucial for patient outcomes and effective healthcare management.

Purpose of the Study:

• To validate an NLP algorithm integrated into the DeVTEcare clinical decision support system for identifying VTEs within EHRs.
• To assess the performance of the NLP algorithm in various hospital departments and patient risk groups.

Main Methods:

• A cohort of 30,152 adult inpatients from a single hospital center was analyzed.
• The NLP tool's performance was evaluated using sensitivity, specificity, likelihood ratios, AUC, and F1-scores, with manual record review as the gold standard.
• Subgroup analyses were conducted across hospital departments, VTE risk levels, age, sex, and season.

Main Results:

• The NLP algorithm achieved high performance in VTE detection, with an overall sensitivity of 89.9% and specificity of 99.8%.
• The algorithm demonstrated superior performance in surgery departments and low-risk VTE departments, with F1-scores up to 0.95 and 0.97, respectively.
• Consistent high performance (>87% sensitivity/specificity, >89% AUC/F1-score) was observed across age, sex, and seasonal subgroups, with better results in patients ≤65 years.

Conclusions:

• The DeVTEcare NLP algorithm effectively identifies VTE in EHRs across various clinical environments, particularly in surgical units and among younger patients.
• This validated NLP tool can provide accurate in-hospital VTE prevalence data and support enhanced, risk-stratified VTE integrated care strategies.
• Further research can leverage this algorithm for improved VTE surveillance and management.