AI to Improve Data From Danish Cardiac Arrest Registry

Overview

Out-of-hospital cardiac arrest is a leading cause of death worldwide and patient outcome vary substantially throughout regions suggesting further evaluation and potential for improvement.When focussing on subgroups of OHCA, data in certain areas remains scarce and the need of revised guidelines is evident. Furthermore, enhanced knowledge on these varieties of OHCA's apply to substantial number of patients, also among vulnerable populations. The Danish Emergency Medical System introduced a nationwide registry of electronic medical reports in 2016. This report system allows electronic searches and thereby the opportunity to identify subgroups of OHCA's. Thus, this novel reporting enables the evaluation of new characteristics of cardiac arrests of non-cardiac origin, in cases where an automated external defibrillator (AED) is retrieved but did not recommend defibrillation and finally in OHCA related to foreign body obstruction. With the advantages of artificial intelligence, this project will enhance and strengthen data from the Danish Cardiac Arrest Registry. It may substitute the manual validation of the around 9000 cases per year in Denmark. Further, it proposes improvement of quality and development of observational health research.

OHCA of non-cardiac origin is a heterogenous subgroup within a range of causes including trauma, electrocution, asphyxia (external), drowning and drug overdose. This classification of presumed etiology accounts for up to 40% of all OHCA's, but varying rates have been reported also advocating for standardised reporting and coherence to the Utstein reporting template.Compared to OHCA's with presumed cardiac etiology, the group of non-cardiac etiology has been associated with a worse prognosis.(7,9) Increased effort has been put into improvement in both prehospital and in-hospital treatment during the last two decades, but morbidity and mortality remains high with reported survival rates of around 10%. The Utstein recommendations have, since the establishment in 1991 undergone revisions in 2004 and 2015.Through the progress and revisions of the Utstein reporting template, registries have emerged worldwide; systematic reporting is, however still lacking. Besides that, most data within the field of OHCA is on cases with presumed cardiac cause. With aid of machine learning, the hypothesis is that advances text searches may lead to improvement of quality of data from the Danish registry of OHCA. With this improvement this project might contribute to the handling and strategies concerning defined subgroups within OHCA. Furthermore, we speculate that this novel data from the EMS medical reports provide new and central data reversible on causes, which presumably are linked to enhanced survival of OHCA of non-cardiac origin. Thus, this study aims to: * Investigate descriptives of traumatic out-of-hospital cardiac arrest, drug overdose, electrocution and external asphyxia leading to cardiac arrest * Assess outcomes of cardiac arrests of non-cardiac origin * Evaluate inter- and intrasystem comparisons Data on OHCA's with attempted resuscitation in Denmark have been collected in the electronic based Danish Medical Service reporting system since 2016. The registry covers detailed data including the EMS report. The data consists of executive entries and advanced text searches of prehospital charts in conjunction, augmenting the identification and collecting all OHCA's in Denmark. All cases have been through an elaborate validation process of which all identified events were read through manually. This was conducted by an external verification team to corroborate high quality of data throughout the approximately 5400 cases of OHCA in Denmark annually. Within this practice of verification, supplementary sources of data have been linked to each individual case of OHCA; information of certain interest was survival, localization, initiation of bystander CPR and actions from EMS personnel. Identification of subgroups Within this diverse entity of OHCA's further investigation of subgroups are required. Through advanced text searches using a bag-of-words model within the prehospital medical record, cases of interest among the predefined subgroups will be identified; thus, is text-string searches developed for every subgroup. These predefined subgroups will be defined as trauma, drowning, electrocution, drug overdose and asphyxia (external) according to the Utstein template. The identified cases will be coupled to the national registry after external manual validation. Variables included * Age: Age will be defined as the subject age at the time of the event. * Sex: Sex will be stated as either male, female or undetermined derived from personal identification numbers. Gender will be characterized based on EMS-charts, in those cases without an identification number. * Initial rhythm: The initial rhythm will be defined as the first rhythm observed by EMS personnel, and categorized as either shockable, non-shockable (asystole), non-shockable (other) or undetermined. * Etiology of cardiac arrest: Presumed causes will be stated as either presumed cardiac, other medical cause, trauma, drug overdose, drowning, electrocution or asphyxia (external). * Location of Incidence: Location will be characterized as either residential area, outdoors and nature, private home or other. * Observation of occurrence: Cardiac arrests will be defined as either unwitnessed, bystander witnessed or EMS-witnessed. * Cardio-pulmonary-resuscitation: CPR will include bystander-initiated CPR and EMS treatment with CPR. * Defibrillation and use of automatic external defibrillators: Defibrillation will include defibrillation by bystanders and/or EMS personnel. Use of AED's includes using the device also in the case it did not deliver shocks. * EMS-response time: This will be defined as the time between a dispatcher receiving the emergency call and the arrival of the first EMS-personnel. * Hospitalization: Hospitalization will be considered as either; transported to hospital or declared dead by EMS-personnel. * Return-of-spontaneous-circulation (ROSC): ROSC will be classified as the achievement of ROSC anytime between recognition of the event and termination (defined as either hospital admission or declaration of death by EMS-personnel). * State at hospital admission: State of the case upon arrival at the hospital will be defined as either ROSC or ongoing CPR. * Survival: Survival will be defined as ROSC at the time of hospital admission, additionally 30-day survival will be included, this derived from data from the National Patient Registry. * Airway management: This includes the airway manoeuvres performed on each case. It is defined by listing the use of airway adjuncts such as nasopharyngeal airway, oropharyngeal airway as well as endo-tracheal intubation, supraglottic airway device and isolated bag-valve-mask ventilation. * Use of Medication: This includes a dichotomous (yes/no) variable based upon the potential administration of medication (ie epinephrine) for each case. Analysis and Data Presentation All data will be pseudo-anonymized, and all analyses will be accomplished on an aggregated nationwide level. Data is collected using the STROBE (STrengthening the Reporting of OBservational studies in Epidemiology) statement. The objective of this project is to clarify etiology, descriptive data and prehospital interventions of different subgroups of OHCA's. Descriptive statistics includes the above-mentioned variables labelled with absolute numbers and percentages. Comparative analyses will be carried out using non-parametric testing to examine subgroups. Forward logistic regression analysis will be performed for multivariate analysis. Within this multivariate logistic regression analysis both known and unknown variables will be processed. Odds ratio for survival will be calculated stratified by etiology, localization, bystander and EMS-actions. Risk Management Strategy and Ethical Considerations Since it is registry-based research, there is no patient contact and of such no harm possible or further ethical considerations of concern. GDPR will be followed according to danish law and the studies will be registered at the Danish Data Protection Agency, capital region of Denmark. According to Danish legislation, register-based research does not require ethical committee approval nor patient consent. Perspectives This project provides novel information regarding these subgroups of OHCA; The way of using artificial intelligence within text mining allows high quality use of data in order to strengthen the results. Thus, the descriptive statistics provides relevant data based on a reviewed, high-quality database. Furthermore, throughout analyses, a better understanding of the preceding circumstances, etiology and prehospital assessment might contribute to improve handling these types of arrests. This, eventually through use of campaigns, education and training targeted laypersons and healthcare personnel, respectly. Publication The results are targeted for publication in an international peer reviewed journal. Signe Amalie Wolthers is primary investigator of this project and will be listed as first author. Participation as coauthors will be decided according to the Vancouver criteria or acknowledged for providing access to data. All Danish regional EMS regions will receive this manuscript prior to publication for eventual comments.