New tool outdoes existing triage scores at estimating mortality risk in ED
The novel Score for Emergency Risk Prediction (SERP) performs better than other point-based clinical scores at triaging patients according to mortality risk in the emergency department (ED), according to a study from Singapore.
“SERP scores were more accurate in identifying patients who died during short- or long-term care than other point-based clinical tools (ie, Patient Acuity Category Scale [PACS], National Early Warning Score [NEWS], Modified Early Warning Score [MEWS], Cardiac Risk Assessment Triage [CART], Rapid Acute Physiology Score [RAPS], and Rapid Emergency Medicine Score [REMS]),” said a team of researchers from Duke–National University of Singapore Medical School and Singapore General Hospital.
They pointed out that the score could be used to support clinical acumen during the consultation, when a clinician intends to admit a patient and weighs up the level of service that might be appropriate for that individual.
In the study, the SERP scoring models were developed using a machine learning–based clinical score generation algorithm, known as the AutoScore, which combined machine learning and logistic regression, as well as automated the development of parsimonious sparse-score risk models for predefined outcomes. The scores were based on the primary outcomes of 2-, 7-, and 30-day mortality.
The investigators used data from three cohorts. These included a training cohort of 224,666 patients admitted to ED (mean age 63.60 years, 50.5 percent female) to generate tentative models using the AutoScore framework, a validation cohort of 56,167 patients (mean age 63.58 years, 50.6 percent female) to evaluate the candidate models, and a testing cohort of 42,676 patients (mean age 64.85 years, 50.5 percent female) to calculate the performance metrics of the final SERP model.
The mortality rates in the training cohort were 0.8 percent at 2 days, 2.2 percent at 7 days, and 5.9 percent at 30 days. AutoScore identified 26 candidate variables for use in developing the SERP scores. Five of the said variables were selected for all the three SERP scores (2d, 7d, and 30d), including age, heart rate, respiration rate, diastolic blood pressure, and systolic blood pressure.
On the area under the curve (AUC) in the receiver operating characteristic analysis, the SERP score showed promising discriminatory ability in estimating all mortality-related outcomes. The SERP-30d had the best performance for predicting short-term and long-term mortality, with AUCs of 0.821 (95 percent CI, 0.796–0.847) for 2-day mortality, 0.826 (95 percent CI, 0.811–0.841) for 7-day mortality, 0.823 (95 percent CI, 0.814–0.832) for 30-day mortality, and 0.810 (95 percent CI, 0.799–0.821) for inpatient mortality. Furthermore, the score outperformed several benchmark scores. [JAMA Netw Open 2021;4:e2118467]
“Several possible reasons exist for SERP-30d to excel,” according to the investigators. “The 2-day mortality rate in our cohort was as low as 0.8 percent. Thus, SERP-2d was developed based on highly imbalanced data, for which the abundance of samples from the majority class (survival group) could overwhelm the minority class in predictive modeling. As a comparison, the 30-day mortality contained all 2- or 7-day mortality cases and was more prevalent at a rate of 5.9 percent, making the SERP-30d score more reliable and accurate.”
The findings, they added, also reconfirmed the value of 30-day mortality as an essential indicator for the ED. The investigators were positive that the SERP scores could offer an objective measure to estimate a patient’s mortality risk during ED triage. [Ann Emerg Med 2020;76:291-300; Eur J Intern Med 2020;77:36-43; J Am Geriatr Soc 2020;68:1755-1762]
“Although physicians can generally ascertain the severity of a patient’s acute condition and the threat to life, their decisions are often subjective and depend on an individual’s experience and knowledge… Like the Emergency Severity Index, some triage scores may achieve better performance in risk estimation but require some subjective variables,” they said. [Acad Emerg Med 2000;7:236-242]
Meanwhile, the SERP scores consist only of objective variables, the investigators said. As such, the scores can be easily computed by trained medical assistants or integrated into an existing hospital electronic hospital record, without the need for professional medical personnel.
“Therefore, one can rapidly estimate a patient’s risk of death without adversely affecting ED workloads, which is important in the fast-paced ED environment and other heterogenous emergency care systems run by generalists rather than emergency medicine specialists,” they added.