In a crowded emergency department (ED), several crucial decisions must be made to provide effective patient care [1,2]. Prehospital decision-making, such as transport location and timing decisions, are crucial for continuity of care from prehospital to hospital levels [3-5]. Appropriate triage is required for selective and targeted care of critically ill patients [6]. To improve patient outcomes, a prompt investigation and intervention decision should be achieved [7,8]. The final disposition choice must consider patient safety [9]. Management of the overall process and the allocation of resources are also essential for ED operation [10]. Overall, the decision-making points should be considered based on patient state and ED situation.
Clinical decision support (CDS) based on artificial intelligence (AI) has recently been introduced into numerous medical fields, including emergency care. Increased data sources such as wearable devices and electronic medical records, higher computing power, and increased storage locations such as online cloud databases have contributed to the expansion of AI technology. As a result, several AI-based algorithms have been introduced for use in the ED journey.
AI-CDS can be categorized according to the decision phase of the ED journey. From the standpoint of patient care and based on a previous study, the ED journey can be classified into phases of prehospital phase, triage, investigation, intervention, and final disposition [11]. Each stage has its own decision objective and may be repeated based on the situation in the ED and the patient condition (Fig. 1).
During the prehospital period, decisions are based on the experience of the medical staff at the scene and the protocols of the local emergency medical service. However, it is difficult to make decisions with limited resources and patient information. With AI-CDS, efforts have been made to improve the accuracy of initial diagnoses based on prehospital information and data. Dispatching could also be aided by AI-CDS; for example, when the conversation between callers and bystanders shows certain context or words for critical conditions, systems would suggest higher-level providers to be dispatched. With prehospital data, a machine-learning-based prediction of hospital admission was attempted in a previous study to inform prehospital providers of patient prognosis after ED care [12]. Natural language processing techniques were implemented in prehospital paramedic reports for diagnosis of stroke [13]. A previous study predicted early sepsis using prehospital data [14].
Nations and institutions have their own triage system for proper early care of a critical patient and effective resource distribution [7,15]. Underestimation or overestimation of triage can result in resource loss or deterioration of critically ill patients due to delayed care [6]. For triage, several AI-CDS system or novel AI-based methods have been introduced. There have been attempts to predict outcomes such as mortality or admission, including development of certain clinical conditions such as sepsis [16-18]. However, few studies demonstrate prospective AI-based triage research or implementation [19,20]. Attempts should be sought to match ED triage with ED resource allocation, including staff, beds, and investigatory efforts.
Prescription and interpretation of diagnostic tests are included in the investigation phase. The intervention phase consists of interventions and medication to improve patient conditions. Investigations and interventions that are time consuming and equipment restrictive should be limited. AI can assist in allocation of investigatory and treatment options or to plan expert consultations.
Investigation and intervention tools can be inputs or outputs of an AI algorithm. Some research has attempted to predict the need for computed tomography in the evaluation of disease [21]. Selected interventions can be a result of a prediction [22]. The majority of investigational studies are concerned with interpretation of diagnostic tests in relation to final diagnosis and outcome [23-25].
In an ED, the decision for further management should be made carefully and appropriately. A patient-specific disposition, such as intensive care unit admission, is a frequently targeted outcome [25-27]. Predicting the readmission or return of a discharged patient is also essential [28,29]. Before making a final determination, these algorithms can help physicians reconsider the conditions of their patients.
Several barriers exist regarding the nature of AI. The first issue involves the data. Due to the nature of AI-based algorithms, data quality is as important as data quantity. Using low-quality data as input renders the algorithm less effective. However, unlike that from the intensive care unit, ED data collection is challenging due to rapid patient turnover and the complex environment. Additionally, patient information is easily lost along the path from injury to ward admission. Data collection should be encouraged, and data quality must be maintained.
It is well known that AI-based algorithms have a “black box” nature [30,31]. Therefore, it is difficult to determine the decision’s cause or the decision-making process. To overcome the uncertainty of AI, explainable AI has recently been introduced, such as modified scoring systems or showing feature importance [32,33]. It is anticipated that explainable artificial intelligence-based predictive models will be applicable to emergency medicine as they are implemented throughout the medical field [34].
Recent review articles revealed that only a small number of AI-based models are utilized in actual practice, despite the development of numerous AI-based models for diverse decision phases [35-37]. The health care field is strictly managed by regulations, which is considered as the most important factor for delayed implementation of AI-CDS. For example, a randomized controlled trial is an important strategy to analyze the regulatory process, which is often difficult or impractical for AI-CDS. With these reasons, we do not frequently witness AI-CDS in practice.
Several AI-CDSs have been developed to cope with ED problems and can be categorized by the decision point in an ED: prehospital, triage, investigation, intervention, and disposition phases. Recently, only few of AI-CDS had been implemented and used in actual practice. These obstacles should be evaluated from the aspects of data, algorithm, and application. AI implementation is a practical decision that involves continuous and harmonized efforts by the ED, information technology, and data science.
CONFLICT OF INTEREST
No potential conflict of interest relevant to this article was reported.
FUNDING
None.
REFERENCESJonesPGMountainDForeroRReview article: emergency department crowding measures associations with quality of care: a systematic review202133592600PinesJMHiltonJAWeberEJInternational perspectives on emergency department crowding201118135870KangKKimTRoYSKimYJSongKJShinSDPrehospital endotracheal intubation and survival after out-of-hospital cardiac arrest: results from the Korean nationwide registry20163412832SekiTTamuraTSuzukiMSOS-KANTO 2012 Study GroupOutcome prediction of out-of-hospital cardiac arrest with presumed cardiac aetiology using an advanced machine learning technique201914112835EisnerZJDelaneyPGWidderPPrehospital care for traumatic spinal cord injury by first responders in 8 sub-Saharan African countries and 6 other low- and middle-income countries: a scoping review20211133946MoonSHShimJLParkKSParkCSTriage accuracy and causes of mistriage using the Korean Triage and Acuity Scale201914e0216972IsersonKVMoskopJCTriage in medicine, part I: concept, history, and types20074927581OredssonSJonssonHRognesJA systematic review of triage-related interventions to improve patient flow in emergency departments20111943ChenWLinthicumBArgonNTThe effects of emergency department crowding on triage and hospital admission decisions2020387749ChangHYuJYYoonSYImpact of COVID-19 pandemic on the overall diagnostic and therapeutic process for patients of emergency department and those with acute cerebrovascular disease202093842GrantKMcParlandAMehtaSAckeryADArtificial intelligence in emergency medicine: surmountable barriers with revolutionary potential2020757216ShirakawaTSonooTOguraKInstitution-specific machine learning models for prehospital assessment to predict hospital admission: prediction model development study20208e20324MayampurathAParnianpourZRichardsCTImproving prehospital stroke diagnosis using natural language processing of paramedic reports20215226769DesaiMDTootooniMSBobayKLCan prehospital data improve early identification of sepsis in emergency department? An integrative review of machine learning approaches202213189202BullardMJMusgraveEWarrenDRevisions to the Canadian Emergency Department Triage and Acuity Scale (CTAS) guidelines 2016201719S1827KwonJMLeeYLeeYLeeSParkHParkJValidation of deeplearning-based triage and acuity score using a large national dataset201813e0205836YunHChoiJParkJHPrediction of critical care outcome for adult patients presenting to emergency department using initial triage information: an XGBoost algorithm analysis20219e30770YuJYChangHJungWPredicting mid-term survival of patients during emergency department triage for resuscitation decision2022111MilesJTurnerJJacquesRWilliamsJMasonSUsing machine-learning risk prediction models to triage the acuity of undifferentiated patients entering the emergency care system: a systematic review2020416Sanchez-SalmeronRGomez-UrquizaJLAlbendin-GarciaLMachine learning methods applied to triage in emergency services: a systematic review202260101109KonikoffTGorenIYalonMMachine learning for selecting patients with Crohn’s disease for abdominopelvic computed tomography in the emergency department202153155964ChangHYuJYYoonSKimTChaWCMachine learningbased suggestion for critical interventions in the management of potentially severe conditioned patients in emergency department triage20221210537MatsuoKAiharaHNakaiTMorishitaATohmaYKohmuraEMachine Learning to predict in-hospital morbidity and mortality after traumatic brain injury20203720210HwangEJNamJGLimWHDeep learning for chest radiograph diagnosis in the emergency department201929357380KirubarajanATaherAKhanSMasoodSArtificial intelligence in emergency medicine: a scoping review202011691702BarfodCLauritzenMMDankerJKAbnormal vital signs are strong predictors for intensive care unit admission and inhospital mortality in adults triaged in the emergency department: a prospective cohort study20122028LeeYKwonJMLeeYParkHChoHParkJDeep learning in the medical domain: predicting cardiac arrest using deep learning20183311720MisicVVGabelEHoferIRajaramKMahajanAMachine learning prediction of postoperative emergency department hospital readmission202013296880HongWSHaimovichADTaylorRAPredicting 72-hour and 9-day return to the emergency department using machine learning2019234652PapernotNMcDanielPGoodfellowIJhaSCelikZBSwamiAPractical black-box attacks against machine learningIn: Proceedings of the 2017 ACM on Asia Conference on Computer and Communications Security2017 Apr 2-6Abu Dhabi, United Arab EmiratesNew YorkAssociation for Computing Machinery201750619RudinCStop explaining black box machine learning models for high stakes decisions and use interpretable models instead2019120615XieFChakrabortyBOngMEGoldsteinBALiuNAutoScore: a machine learning-based automatic clinical score generator and its application to mortality prediction using electronic health records20208e21798MaoQJayMHoffmanJLMulticentre validation of a sepsis prediction algorithm using only vital sign data in the emergency department, general ward and ICU20188e017833PawarUO’SheaDReaSO’ReillyRIncorporating Explainable Artificial Intelligence (XAI) to aid the understanding of machine learning in the healthcare domainLongoLRizzoLHunterEPakrashiAAachenCEUR Workshop Proceedings202016980MoulikSKKotterNFishmanEKApplications of artificial intelligence in the emergency department2020273558TangKJAngCKConstantinidesTRajinikanthVAcharyaURCheongKHArtificial intelligence and machine learning in emergency medicine20214115672MuellerBKinoshitaTPeeblesAGraberMALeeSArtificial intelligence and machine learning in emergency medicine: a narrative review20229e740Figure
Artificial intelligence-based clinical decision support points during emergency department (ED) phases. There are four points at which emergency physicians must make important decisions, where artificial intelligence can produce substantial effects.