Personalized diagnosis in suspected myocardial infarction

• DOI: 10.1007/s00392-023-02206-3
• Publication Date: 2023-05-02T16:01:51Z
• Authors: Johannes Tobias N...; Raphael Twerenbold; Francisco Ojeda; Sally J. Aldous; Brandon R. Allen; Fred S. Apple; Hugo Babel; Robert H. Christe...; Louise Cullen; Eleonora Di Carlu...; Dimitrios Doudesis; Ulf Ekelund; Evangelos Giannitsis; Jaimi Greenslade; Kenji Inoue; Tomas Jernberg; Peter Kavsak; Till Keller; Kuan Ken Lee; Bertil Lindahl; Thiess Lorenz; Simon A. Mahler; Nicholas L. Mills; Arash Mokhtari; William Parsonage; John W. Pickering; Christopher J. Pe...; Christoph Reich; A. Mark Richards; Yader Sandoval; Martin P. Than; Betül Toprak; Richard W. Troughton; Andrew Worster; Tanja Zeller; Andreas Ziegler; Stefan Blankenberg; ; Emily Brownlee; Kai M. Eggers; Gavin Fincher; Norbert Frey; Niranjan Gaikwad; Vinay Gangathimmaiah; Emma Hall; Paul M. Haller; Christian Hamilto...; Rebecca Hancock; Andrew Hobbins-King; Gerben Keijzers; Maryam Khorramsha...; Georgios Koliopanos; Jonas Lehmacher; Lina Ljung; Troy Madsen; Ehsan Mahmoodi; Ellyse McCormick; Bryn Mumma; Richard Nowak; Vanessa Blazquez; Siegfried Perez; Vazhma Qaderi; Isuru Ranasinghe; Alina Schock; Nils A. Sörensen; Andrew Staib; Laura Stephensen; Michael Weaver; R. Gentry Wilkerson; Anna Zournazi
• Keywords: Discriminative model

Abstract Background In suspected myocardial infarction (MI), guidelines recommend using high-sensitivity cardiac troponin (hs-cTn)-based approaches. These require fixed assay-specific thresholds and timepoints, without directly integrating clinical information. Using machine-learning techniques including hs-cTn routine variables, we aimed to build a digital tool estimate the individual probability of MI, allowing for numerous assays. Methods 2,575 patients presenting emergency department with two ensembles models single or serial concentrations six different assays were derived MI (ARTEMIS model). Discriminative performance was assessed area under receiver operating characteristic curve (AUC) logLoss. Model validated in an external cohort 1688 tested global generalizability 13 international cohorts 23,411 patients. Results Eleven routinely available variables age, sex, cardiovascular risk factors, electrocardiography, included ARTEMIS models. validation generalization cohorts, excellent discriminative confirmed, superior only. For measurement model, AUC ranged from 0.92 0.98. Good calibration observed. measurement, model allowed direct rule-out very high similar safety but up tripled efficiency compared guideline-recommended strategy. Conclusion We developed diagnostic accurately which allow variable use flexible timing resampling. Their application may provide rapid, safe efficient personalized patient care. Trial Registration numbers Data following used this project: BACC ( www.clinicaltrials.gov ; NCT02355457), stenoCardia NCT03227159), ADAPT-BSN www.australianclinicaltrials.gov.au ACTRN12611001069943), IMPACT , ACTRN12611000206921), ADAPT-RCT www.anzctr.org.au ANZCTR12610000766011), EDACS-RCT ANZCTR12613000745741); DROP-ACS https://www.umin.ac.jp UMIN000030668); High-STEACS NCT01852123), LUND NCT05484544), RAPID-CPU NCT03111862), ROMI NCT01994577), SAMIE https://anzctr.org.au ACTRN12621000053820), SEIGE SAFETY NCT04772157), STOP-CP NCT02984436), UTROPIA NCT02060760). Graphical