Purpose: In this study we developed a novel prediction model to identify which patients will need to undergo non-contrast head CT scan at time of admission to the emergency department (ED).
Materials and Methods: We collected data of all adult ED visits in our institution during five consecutive years (1/2013 – 12/2017). Retrieved variables included: demographics, mode of arrival to the ED, comorbidities, home medications history, structured and unstructured chief complaints, vital signs, pain scale, emergency severity index (ESI), ED wing assignment, documentation of previous ED visits, hospitalizations and previous non-contrast head CTs. The outcome evaluated was current visit non-contrast head CT usage.
A machine learning gradient boosting model (CatBoost) was trained on data from the years 2013 - 2016 and validated on data from 2017. Area under the curves (AUC) was used as metrics. Single variables AUCs were also determined. Youden’s index evaluated optimal sensitivity and specificity of the models.
Results: The final cohort included 561,933 ED visits. The non-contrast head CT usage rate was 11.8%. Each ED visit was coded into an input vector of 171 variables. Single variable analysis showed that chief complaint had the best single predictive analysis (AUC=0.87), followed by age, ED wing assignment and mode of arrival to the ED. Chief complaint of trauma and headache showed the highest frequency of performed head CTs (number of CT scans) whereas suspected cerebrovascular accident (CVA) and neurological related complaints had the highest CT usage rate (number of CT scans/number of patients with chief complaint). The best model showed an AUC of 0.93 (95% CI: 0.931 - 0.936) for predicting non-contrast head CT usage at triage level. The model had a sensitivity of 87.7% and specificity of 86.1% for non-contrast head CT utilization.
Conclusion: The developed model can identify patients that need to undergo head CT scan in the triage level, and by that help to promote early diagnosis and treatment. As a diagnostic tool in the decision-making process, it can improve efficiency in healthcare facilities by flagging patients and optimizing workflow.
