ED, heal thyself

Identifying low acuity Emergency Department visits with a machine learning approach: The low acuity visit algorithms (LAVA)

OBJECTIVE

To improve the performance of International Classification of Disease (ICD) code rule-based algorithms for identifying low acuity Emergency Department (ED) visits by using machine learning methods and additional covariates.

DATA SOURCES

We used secondary data on ED visits from the National Hospital Ambulatory Medical Survey (NHAMCS), from 2016 to 2020.

STUDY DESIGN

We established baseline performance metrics with seven published algorithms consisting of International Classification of Disease, Tenth Revision codes used to identify low acuity ED visits. We then trained logistic regression, random forest, and gradient boosting (XGBoost) models to predict low acuity ED visits. Each model was trained on five different covariate sets of demographic and clinical data. Model performance was compared using a separate validation dataset. The primary performance metric was the probability that a visit identified by an algorithm as low acuity did not experience significant testing, treatment, or disposition (positive predictive value, PPV). Subgroup analyses assessed model performance across age, sex, and race/ethnicity.

DATA COLLECTION

We used 2016-2019 NHAMCS data as the training set and 2020 NHAMCS data for validation.

PRINCIPAL FINDINGS

The training and validation data consisted of 53,074 and 9542 observations, respectively. Among seven rule-based algorithms, the highest-performing had a PPV of 0.35 (95% CI [0.33, 0.36]). All model-based algorithms outperformed existing algorithms, with the least effective-random forest using only age and sex-improving PPV by 26% (up to 0.44; 95% CI [0.40, 0.48]). Logistic regression and XGBoost trained on all variables improved PPV by 83% (to 0.64; 95% CI [0.62, 0.66]). Multivariable models also demonstrated higher PPV across all three demographic subgroups.

CONCLUSIONS

Machine learning models substantially outperform existing algorithms based on ICD codes in predicting low acuity ED visits. Variations in model performance across demographic groups highlight the need for further research to ensure their applicability and fairness across diverse populations.

A Comparison Between Computer-Assisted Self-Triage by Patients and Triage Performed by Nurses in the Emergency Department

Background and objective

Emergency departments (EDs) often find the number of arriving patients exceeding their capacity and find it difficult to triage them in a timely manner. The potential risk to the safety of patients awaiting assessment by a triage professional has led some hospitals to consider implementing patient self-triage, such as using kiosks. Published studies about patient self-triage are scarce and information about patients’ ability to accurately assess the acuity of their condition or predict their need to be hospitalized is limited. In this study, we aimed to compare computer-assisted patient self-triage scores versus the scores assigned by the dedicated ED triage nurse (TN).

Methods

This pilot study enrolled patients presenting to a tertiary care hospital ED without ambulance transport. They were asked a short series of simple questions based on an algorithm, which then generated a triage score. Patients were asked whether they were likely to be admitted to the hospital. Patients then entered the usual ED system of triage. The algorithm-generated triage score was then compared with the Canadian Triage and Acuity Scale (CTAS) score assigned by the TN. Whether the patients actually required hospital admission was determined by checking their medical records.

Results

Among the 492 patients enrolled, agreement of triage scores was observed in 27%. Acuity was overestimated by 65% of patients. Underestimation of acuity occurred in 8%. Among patients predicting hospitalization, 17% were admitted, but the odds ratio (OR) for admission was 3.4. Half of the patients with cardiorespiratory complaints were correct in predicting the need for hospitalization.

Conclusion 

The use of a short questionnaire by patients to self-triage showed limited accuracy, but sensitivity was high for some serious medical conditions. The prediction of hospitalization was more accurate with regard to cardiorespiratory complaints.

A determination of emergency department pre-triage times in patients not arriving by ambulance compared to widely used guideline recommendations

OBJECTIVES

Emergency department (ED) lengths of stay are measured from the time of patient registration or triage. The time that patients wait in line prior to registration and triage has not been well described. We sought to characterize pre-triage wait times and compare them to recommended physician response times, as per the Canadian Triage and Acuity Scale (CTAS).

METHODS

This observational study documented the time that consenting patients entered the ED and the time that they were formally registered and triaged. Participants' CTAS scores were collected from the electronic record. Patients arriving to the ED by ambulance were excluded.

RESULTS

A total of 536 participants were timed over 13 separate intervals. Of these, 11 left without being triaged. Participants who scored either CTAS 1 or 2 (n=53) waited a median time of 3.1 (interquartile range [IQR]: 0.43, 11.1) minutes. Patients triaged as CTAS 3 (n=187) waited a median of 11.4 (IQR: 1.6, 24.9) minutes, CTAS 4 (n=139) a median of 16.6 (IQR: 6.0, 29.7) minutes, and CTAS 5 (n=146) a median of 17.5 (IQR: 6.8, 37.3) minutes. Of patients subsequently categorized as CTAS 1 or 2, 20.8% waited longer than the recommended time-to-physician of 15 minutes to be triaged.

CONCLUSIONS

All urban EDs closely follow patients' wait times, often stratified according to triage category, which are assumed to be time-stamped upon a patient's arrival in the ED. We note that pre-triage times exceed the CTAS recommended time-to-physician in a possibly significant proportion of patients. EDs should consider documenting times to treatment from the moment of patient arrival rather than registration.

The Patient Protection and Affordable Care Act's Effect on Emergency Medicine: A Synthesis of the Data

This review synthesizes the existing literature to provide evidence-based predictions for the future of emergency care in the United States as a result of the Patient Protection and Affordable Care Act, with a focus on emergency department (ED) visit volume, acuity, and reimbursement. Patient behavior will likely be quite different for patients gaining Medicaid than for those gaining private insurance through the Marketplaces. Despite the threat of the individual mandate, not all uninsured patients will enroll, and those who choose to enroll will likely be a different population from those who remain uninsured. New Medicaid enrollees will be a sicker population and will likely increase their number of ED visits substantially. Their acuity will be higher at first but will then revert to the traditionally high number of low-acuity visits made by Medicaid patients. Most patients enrolling through the Marketplace are choosing high-deductible health plans, and they will initially avoid the ED because of high out-of-pocket costs but may present later and sicker after self-rationing their care. Most patients gaining health coverage through the Affordable Care Act will be shifting from uninsured to either Medicaid or private insurance, both of which reimburse more than self-pay, so ED collections should increase. Because of the differences between Medicaid and Marketplace plans, there will be a difference in ED volume, acuity, and financial outcomes, depending on states' current demographics, whether states expand Medicaid, and how aggressively states advertise new options for coverage in Medicaid or state health insurance Marketplaces.