Development of ED Triage Tool for Acute Coronary Syndrome (ACS) – Assessing the Use of Logistic Regression for Model Development
| Autor: | Ellice Jane J. Tiu, 張純潔 |
|---|---|
| Rok vydání: | 2019 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 107 Acute Coronary Syndrome (ACS) patient management in the Emergency Department (ED) is known to be a challenging task as majority of patients do not present clear-cut evidence of this condition. This leads to difficulty in identifying patients who should be prioritized for thorough diagnoses in the Observation Unit (OU) and who can safely wait in the regular ED queue. This study aimed to validate and use Logistic Regression as the method for developing a set of triaging criteria which can identify patients with considerable or negligible risk of ACS upon presentation. At the same time, evaluate the effectiveness of having a separate triage model for patients arriving via ambulance (119) or self-arrival patients. Validation for Logistic Regression was done through the implementation of the ACS Triage Tool, previously developed by Tsai et al. (2018), on a new dataset and compared its consistency with the previously published result. Since the ACS Triage Tool previously utilized patient data which was not truly representative of the ED population, a new model using a more valid dataset was developed using Logistic Regression along with physician’s decision for triaging as basis for prediction. Validation of this newly developed model was done through comparing it with other Logistic Regression models formulated with a larger training dataset as well as prediction based on ACS discharge diagnosis. Individual Logistic Regression models for patients arriving via ambulance and self-transportation were also developed and compared with the general model. The ACS Triage Tool proved to have consistent results with its published results indicating that Logistic Regression is a reliable method for model development. The new Logistic Regression model which used a more inclusive set of patient data was comprised of 6 significant predictors and expressed as: Odds Ratio = - 4.566+ 1.056#westeur024#Age + 0.778#westeur024#Male + 2.24#westeur024#Chest Discomfort + 1.365#westeur024#Shock + 0.805#westeur024#Proximal Radiation Pain -1.308#westeur024#Arrhythmia with threshold value of 0.13 for Probability (ACS suspicion). This yielded a performance of 95% sensitivity, 13% specificity, and Area Under Curve of 65% for triaging ACS cases which were superior than other 6 models tested (Chest Pain Strategy, Triage Flowchart, Zarich’s Strategy, HBI Checklist and Modified HBI Checklist 1&2). Furthermore, the logistic regression model proved to be sufficient as developed models with increased training dataset and ACS discharge diagnosis-based model did not result to significant improvements (p>0.05). Meanwhile, a dedicated triage model for patients arriving via a specific mode-of-arrival (self-arrival and ambulance/119) is projected to have good potential in performance improvement. Current results have to be proceeded with caution due to the small amount of data available on-hand for patients arriving via ambulance (119). Therefore, further collection of data for 119 patients for model development is seen to be beneficial. This is foreseen to improve the reliability of the model and develop a dedicated 119 triage model with enhanced triaging capabilities. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|