ABSTRACT
BACKGROUND: In the event of a mass casualty incident (MCI), the situation-related shortage of medical resources does not end when the patients are transported from the scene of the incident. Consequently, an initial triage is required in the receiving hospitals. In the first step, the aim of this study was to create a reference patient vignette set with defined triage categories. This allowed a computer-aided evaluation of the diagnostic quality of triage algorithms for MCI situations in the second step. METHODS: A total of 250 case vignettes validated in practice were entered into a multistage evaluation process by initially 6 and later 36 triage experts. This algorithm-independent expert evaluation of all vignettes-served as the gold standard for analyzing the diagnostic quality of the following triage algorithms: Manchester triage system (MTS module MCI), emergency severity index (ESI), Berlin triage algorithm (BER), the prehospital algorithms PRIOR and mSTaRT, and two project algorithms from a cooperation between the Federal Office of Civil Protection and Disaster Assistance (BBK) and the Hashemite Kingdom of Jordan-intrahospital Jordanian-German project algorithm (JorD) and prehospital triage algorithm (PETRA). Each patient vignette underwent computerized triage through all specified algorithms to obtain comparative test quality outcomes. RESULTS: Of the original 250 vignettes, a triage reference database of 210 patient vignettes was validated independently of the algorithms. These formed the gold standard for comparison of the triage algorithms analyzed. Sensitivities for intrahospital detection of patients in triage category T1 ranged from 1.0 (BER, JorD, PRIOR) to 0.57 (MCI module MTS). Specificities ranged from 0.99 (MTS and PETRA) to 0.67 (PRIOR). Considering Youden's index, BER (0.89) and JorD (0.88) had the best overall performance for detecting patients in triage category T1. Overtriage was most likely with PRIOR, and undertriage with the MCI module of MTS. Up to a decision for category T1, the algorithms require the following numbers of steps given as the median and interquartile range (IQR): ESI 1 (1-2), JorD 1 (1-4), PRIOR 3 (2-4), BER 3 (2-6), mSTaRT 3 (3-5), MTS 4 (4-5) and PETRA 6 (6-8). For the T2 and T3 categories the number of steps until a decision and the test quality of the algorithms are positively interrelated. CONCLUSION: In the present study, transferability of preclinical algorithm-based primary triage results to clinical algorithm-based secondary triage results was demonstrated. The highest diagnostic quality for secondary triage was provided by the Berlin triage algorithm, followed by the Jordanian-German project algorithm for hospitals, which, however, also require the most algorithm steps until a decision.
Subject(s)
Mass Casualty Incidents , Triage , Humans , Triage/methods , Berlin , Algorithms , Computer SimulationABSTRACT
The Covid-19 pandemic has pushed many hospitals to their capacity limits. Therefore, a triage of patients has been discussed controversially primarily through an ethical perspective. The term triage contains many aspects such as urgency of treatment, severity of the disease and pre-existing conditions, access to critical care, or the classification of patients regarding subsequent clinical pathways starting from the emergency department. The determination of the pathways is important not only for patient care, but also for capacity planning in hospitals. We examine the performance of a human-made triage algorithm for clinical pathways which is considered a guideline for emergency departments in Germany based on a large multicenter dataset with over 4,000 European Covid-19 patients from the LEOSS registry. We find an accuracy of 28 percent and approximately 15 percent sensitivity for the ward class. The results serve as a benchmark for our extensions including an additional category of palliative care as a new label, analytics, AI, XAI, and interactive techniques. We find significant potential of analytics and AI in Covid-19 triage regarding accuracy, sensitivity, and other performance metrics whilst our interactive human-AI algorithm shows superior performance with approximately 73 percent accuracy and up to 76 percent sensitivity. The results are independent of the data preparation process regarding the imputation of missing values or grouping of comorbidities. In addition, we find that the consideration of an additional label palliative care does not improve the results.
Subject(s)
COVID-19 , Triage , Humans , Triage/methods , Critical Pathways , Pandemics , Algorithms , Emergency Service, Hospital , Artificial IntelligenceABSTRACT
BACKGROUND: In the event of a mass casualty incident (MCI), the situation-related shortage of medical resources does not end when the patients are transported from the scene of the incident. Consequently, an initial triage is required in the receiving hospitals. In the first step, the aim of this study was to create a reference patient vignette set with defined triage categories. This allowed a computer-aided evaluation of the diagnostic quality of triage algorithms for MCI situations in the second step. METHODS: A total of 250 case vignettes validated in practice were entered into a multistage evaluation process by initially 6 and later 36 triage experts. This algorithm-independent expert evaluation of all vignettes-served as the gold standard for analyzing the diagnostic quality of the following triage algorithms: Manchester triage system (MTS module MCI), emergency severity index (ESI), Berlin triage algorithm (BER), the prehospital algorithms PRIOR and mSTaRT, and two project algorithms from a cooperation between the Federal Office of Civil Protection and Disaster Assistance (BBK) and the Hashemite Kingdom of Jordan-intrahospital Jordanian-German project algorithm (JorD) and prehospital triage algorithm (PETRA). Each patient vignette underwent computerized triage through all specified algorithms to obtain comparative test quality outcomes. RESULTS: Of the original 250 vignettes, a triage reference database of 210 patient vignettes was validated independently of the algorithms. These formed the gold standard for comparison of the triage algorithms analyzed. Sensitivities for intrahospital detection of patients in triage category T1 ranged from 1.0 (BER, JorD, PRIOR) to 0.57 (MCI module MTS). Specificities ranged from 0.99 (MTS and PETRA) to 0.67 (PRIOR). Considering Youden's index, BER (0.89) and JorD (0.88) had the best overall performance for detecting patients in triage category T1. Overtriage was most likely with PRIOR, and undertriage with the MCI module of MTS. Up to a decision for category T1, the algorithms require the following numbers of steps given as the median and interquartile range (IQR): ESI 1 (1-2), JorD 1 (1-4), PRIOR 3 (2-4), BER 3 (2-6), mSTaRT 3 (3-5), MTS 4 (4-5) and PETRA 6 (6-8). For the T2 and T3 categories the number of steps until a decision and the test quality of the algorithms are positively interrelated. CONCLUSION: In the present study, transferability of preclinical algorithm-based primary triage results to clinical algorithm-based secondary triage results was demonstrated. The highest diagnostic quality for secondary triage was provided by the Berlin triage algorithm, followed by the Jordanian-German project algorithm for hospitals, which, however, also require the most algorithm steps until a decision.
Subject(s)
Mass Casualty Incidents , Triage , Humans , Triage/methods , Berlin , Algorithms , Computer SimulationABSTRACT
BACKGROUND: Major incidents are characterized by a lack of resources compared to an overwhelming number of casualties, requiring a prioritization of medical treatment. Triage algorithms are an essential tool for prioritizing the urgency of treatment for patients, but the evidence to support one over another is very limited. We determined the influence of blood pressure limits on the diagnostic value of triage algorithms, considering if pulse should be palpated centrally or peripherally. METHODS: We used a database representing 500 consecutive HEMS patients. Each patient was allocated a triage category (T1/red, T2/yellow, T3/green) by a group of experienced doctors in disaster medicine, independent of any algorithm. mSTaRT, ASAV, Field Triage Score (FTS), Care Flight (CF), "Model Bavaria" and two Norwegian algorithms (Nor and TAS), all containing the question "Pulse palpable?", were translated into Excel commands, calculating the triage category for each patient automatically. We used 5 blood pressure limits ranging from 130 to 60 mmHg to determine palpable pulse. The resulting triage categories were analyzed with respect to sensitivity, specificity and Youden Index (J) separately for trauma and non-trauma patients, and for all patients combined. RESULTS: For the entire population of patients within all triage algorithms the Youden Index (J) was highest for T1 (J between 0,14 and 0,62). Combining trauma and non-trauma patients, the highest J was obtained by ASAV (J = 0,62 at 60 mmHg). ASAV scored the highest within trauma patients (J = 0,87 at 60 mmHg), whereas Model Bavaria (J = 0,54 at 80 mmHg) reached highest amongst non-trauma patients. FTS performed worst for all patients (J = 0,14 at 60 mmHg), showing a lower score for trauma patients (J = 0,0 at 60 mmHg). Change of blood pressure limits resulted in different diagnostic values of all algorithms. DISCUSSION: We demonstrate that differing blood pressure limits have a remarkable impact on diagnostic values of triage algorithms. Further research is needed to determine the lowest blood pressure value that is possible to palpate at a peripheral artery compared to a central artery. CONCLUSION: As a consequence, it might be important in which location pulses are palpated according to the algorithm at hand during triage of patients.
