Below is an overview of “Machine Learning Techniques for Predicting Patient-Level Emergency Department Workload,” published in the January 2023 issue. emergency medicine by Joseph et al.
Labor Relative Value Units (wRVU) are incorporated into various pay scales as a measure of time and effort spent on patient care. Therefore, being able to predict with high accuracy the number of her wRVUs that a patient will generate during triage would have many operational and clinical implications, such as reducing the burden on individual physicians by more evenly distributing their workload. you get the advantage. In this study, we tested whether deep learning techniques can accurately predict a patient’s her wRVU using data typically collected during triage. Participants were adults who visited the ER of an urban university between 1 July 2016 and 1 March 2020.
Structured (age, gender, vital signs, urgency severity index scores, language, race, standardized chief complaints) and unstructured data (free text chief complaints) were used for anonymized triage information. , wRVU served as the outcome measure. . The researchers considered five models, including mean wRVU per chief complaint, linear regression, neural networks, gradient-boosted trees for structured data, and neural networks for unstructured text data. Mean absolute error was used as a metric to rank model quality. We analyzed 204,064 visits between 1 January 2016 and 28 February 2020. Age, gender and race significantly influenced wRVU, with a median wRVU of 3.80 (interquartile range 2.56 to 4.21).
The model error decreased as the model complexity increased. Predictions using chief complaints showed a mean error of 2.17 wRVU per visit (95% CI 2.07-2.27). The linear regression model showed an error of 1.00 wRVU (95% CI 0.97-1.04), the gradient boosted tree showed an error of 0.85 wRVU (95% CI 0.84-0.86), and the neural network with structured data showed an error of 0.86. showed. wRVU (95% CI 0.85-0.87), neural networks with unstructured data showed an error of 0.78 wRVU (95% CI 0.76-0.80). However, deep learning techniques show potential to overcome the limitations of chief complaints as predictors of the time required for patient assessment. These algorithms could have many useful applications, such as reducing bias in the triage process, quantifying crowding and resource mobilization, and balancing emergency physicians and compensation..
sauce: sciencedirect.com/science/article/abs/pii/S0736467922005686
