Related Experiment Video
Updated: May 23, 2025
Implementation of a Real-Time Psychosis Risk Detection and Alerting System Based on Electronic Health Records using CogStack
Published on: May 15, 2020
Using Natural Language Processing in the LACE Index Scoring Tool to Predict Unplanned Trauma and Surgical
Umit Tokac1, Jennifer Chipps2, Petra Brysiewicz3
1College of Nursing, University of Missouri-Saint Louis, Saint Louis, Missouri, USA.
Background:
Unplanned and potentially avoidable readmission within 30 days post discharge is a major financial burden.
Aim:
To use text-based electronic patient records to calculate the Charlson Comorbidity Index (CCI) score using a natural language processing technique to establish the feasibility and usefulness of the text-based electronic patient records in identifying patients at risk for unplanned readmission.
Methods:
A retrospective review of electronic patient records for general and trauma surgery in a hospital in South Africa (2012-2022) was conducted using the LACE score. Validated sentiment analysis analyzed free text components of electronic patient records to compute the CCI score and to establish the feasibility and usefulness of the LACE score in identifying patients at risk for unplanned readmission.
Results:
Trauma surgery patients had a mean LACE score of 5.91 (SD = 2.41), with 8.44% scoring 10 or higher and a specificity and sensitivity of 91.63% and 13.81%, respectively. The general surgery patients had a mean LACE score of 7.75 (SD = 3.04), with 10.63% scoring 10 or higher and a specificity of 71.47% and a sensitivity of 44.80%, respectively. Logistic regression analysis revealed that LACE scores significantly predicted unplanned readmissions in both trauma (β = 0.11, p < 0.001; OR = 1.112, 95% CI [1.082, 1.143]) and general surgery (β = 0.15, p < 0.001; OR = 1.162, 95% CI [1.130, 1.162]) patients.
Conclusion:
The LACE score demonstrated the predictive value for readmission in trauma and general surgery patients. The LACE score was relatively effective in identifying patients who were less likely to be readmitted but showed limitations in identifying patients at higher risk of readmission. However, the successful use of natural language processing for data extraction of comorbidities shows promise on addressing the challenges around text-based medical records.
