Design and Retrospective Cohort Validation of an Algorithm for Surgical Site Infection Surveillance after Hip Replacement Surgery Using Natural Language Processing and Extreme Gradient Boosting.
Flores-Balado, A.; Castresana Mendez, C.; Herrero Gonzalez, A.; Meson Gutierrez, R.; Arcos, J.; Martin-Rios, M. D.; Surgical Site Infection Surveillance Group,
Show abstract
BackgroundSurgical site infection (SSI) surveillance is a labor-intensive endeavor. We present the design and retrospective cohort validation of a multivariable algorithm to screen for SSI in patients undergoing hip replacement surgery, permitting real-time case detection, and reducing the number of clinical records to be reviewed manually. MethodsWe designed a multivariable algorithm using extreme gradient-boosting (XGBoost) to screen for SSI in patients undergoing hip replacement surgery. The development and validation cohort included all healthcare episodes dating from the initial hospital admission for surgery until 90 days after joint replacement (n=19661) between January 2014 and December 2021 from four hospitals in Madrid, Spain. A NLP pipeline was implemented to obtain text variables from free-text fields of the EHR. Clinical, prescription, microbiology, and laboratory variables were also collected. Episodes were split randomly into training (70%) and testing (30%) datasets. Hyperparameters were adjusted to penalize false negatives. FindingsThe presence of positive microbiological cultures, the text variable "infection", and the prescription of clindamycin were strong markers of SSI. Statistical analysis of the final model indicated a high sensitivity (99{middle dot}18%) and specificity (91{middle dot}01%) with an F1-score of 0{middle dot}32, AUC of 0{middle dot}989 and accuracy of 91{middle dot}27%. The model correctly classified 5129 episodes as negative for SSI, with only one case of SSI escaping detection (NPV 99{middle dot}98%). InterpretationWe conclude that the combination of NLP and extreme gradient boosting is a sensitive tool for SSI surveillance in hip replacement surgery, permitting real-time, semi-automatic surveillance. In practice, our results translate as an 88{middle dot}95% reduction in the total volume of clinical records to be reviewed manually. To the best of our knowledge, this is the first time that an algorithm incorporating data from multiple sources using NLP and extreme gradient boosting has been developed for orthopedic SSI surveillance. FundingThis study did not receive any funding.
Matching journals
The top 12 journals account for 50% of the predicted probability mass.
Similar papers in this journal
- Predicting bloodstream infection outcome using machine learning 93%
- Using explainable machine learning to identify patients at risk of reattendance at discharge from emergency departments 92%
- AKI Risk Score (AKI-RiSc): Developing an Interpretable Clinical Score for Early Identification of Acute Kidney Injury for Patients Presenting to the Emergency Department 92%
Similar papers in this journal
- A Risk Prediction and Clinical Guidance System for Evaluating Patients with Recurrent Infections. 94%
- Clinical prediction rule for SARS-CoV-2 infection from 116 U.S. emergency departments 92%
- A comparison of machine learning models versus clinical evaluation for mortality prediction in patients with sepsis 92%
Similar papers in this journal
- Development and validation of automated computer aided-risk score for predicting the risk of in-hospital mortality using first electronically recorded blood test results and vital signs for COVID-19 hospital admissions: a retrospective development and validation study 92%
- Development and validation of multivariable prediction models for adverse COVID-19 outcomes in IBD patients 92%
- Validity of intraoperative imageless navigation (Naviswiss™) for component positioning accuracy in primary total hip arthroplasty: Protocol for a prospective observational cohort study in a single-surgeon practice 92%
Similar papers in this journal
- Machine Learning-based Clinical Decision Support for Infection Risk Prediction 92%
- Development and Validation of an Interpretable 3-day Intensive Care Unit Readmission Prediction Model Using Explainable Boosting Machines 91%
- A Predictive Nomogram for In-ICU Deterioration of Stage 1 Pressure Injuries: A Retrospective Study 90%
Similar papers in this journal
- Identifying Cases of Shoulder Injury Related to Vaccine Administration (SIRVA) Using Natural Language Processing 91%
- Initial Experience in Predicting the Risk of Hospitalization of 496 Outpatients with COVID-19 Using a Telemedicine Risk Assessment Tool 91%
- Comorbidities might be a risk factor for the incidence of COVID-19: Evidence from a web-based survey of 780,961 participants 90%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.