Machine learning models for predicting early hemorrhage progression in traumatic brain injury.

This study explores the progression of intracerebral hemorrhage (ICH) in patients with mild to moderate traumatic brain injury (TBI). It aims to predict the risk of ICH progression using initial CT scans and identify clinical factors associated with this progression. A retrospective analysis of TBI patients between January 2010 and December 2021 was performed, focusing on initial CT evaluations and demographic, comorbid, and medical history data. ICH was categorized into intraparenchymal hemorrhage (IPH), petechial hemorrhage (PH), and subarachnoid hemorrhage (SAH). Within our study cohort, we identified a 22.2% progression rate of ICH among 650 TBI patients. The Random Forest algorithm identified variables such as petechial hemorrhage (PH) and countercoup injury as significant predictors of ICH progression. The XGBoost algorithm, incorporating key variables identified through SHAP values, demonstrated robust performance, achieving an AUC of 0.9. Additionally, an individual risk assessment diagram, utilizing significant SHAP values, visually represented the impact of each variable on the risk of ICH progression, providing personalized risk profiles. This approach, highlighted by an AUC of 0.913, underscores the model's precision in predicting ICH progression, marking a significant step towards enhancing TBI patient management through early identification of ICH progression risks.

Guidelines for Manufacturing and Application of Organoids: Lung.

The objective of standard guideline for utilization of human lung organoids is to provide the basic guidelines required for the manufacture, culture, and quality control of the lung organoids for use in non-clinical efficacy and inhalation toxicity assessments of the respiratory system. As a first step towards the utilization of human lung organoids, the current guideline provides basic, minimal standards that can promote development of alternative testing methods, and can be referenced not only for research, clinical, or commercial uses, but also by experts and researchers at regulatory institutions when assessing safety and efficacy.

Dynamic insights into infection risk over time in two-stage implant-based breast reconstruction: a retrospective cohort study.

BACKGROUND: Infections following postmastectomy implant-based breast reconstruction (IBR) can compromise surgical outcomes and lead to significant morbidity. This study aimed to discern the timing of infections in two-stage IBR and associated risk factors. METHOD: A review of electronic health records was conducted on 1,096 breasts in 1,058 patients undergoing two-stage IBR at ** National University Hospital (2015-2020). Infections following the first-stage tissue expander (TE) insertion and second-stage TE exchange were analyzed separately, considering associated risk factors. RESULTS: Over a median follow-up of 53.5 months, infections occurred in 2.9% (32/1096) after the first stage and 4.1% (44/1070) after the second stage. Infections following the first-stage procedure exhibited a bimodal distribution across time, while those after the second-stage procedure showed a unimodal pattern. When analyzing risk factors for infection after the first-stage procedure, axillary lymph node dissection (ALND) was associated with early (≤7 weeks) infection, while both ALND and obesity were independent predictors of late (>7 weeks) infection. For infections following the second-stage procedure, obesity, postmastectomy radiotherapy, a history of expander infection, ALND, and the use of textured implants were identified as independent risk factors. Postmastectomy radiotherapy was related to non-salvaged outcomes after infection following both stages. CONCLUSION: Infections following first and second-stage IBR exhibit distinct timelines reflecting different pathophysiology. Understanding these timelines and associated risk factors will inform patient selection for IBR and aid in tailored postoperative surveillance planning. These findings contribute to refining patient suitability for IBR and optimizing personalized postoperative care strategies.

Quantitative and qualitative mutational impact of ionizing radiation on normal cells.

The comprehensive genomic impact of ionizing radiation (IR), a carcinogen, on healthy somatic cells remains unclear. Using large-scale whole-genome sequencing (WGS) of clones expanded from irradiated murine and human single cells, we revealed that IR induces a characteristic spectrum of short insertions or deletions (indels) and structural variations (SVs), including balanced inversions, translocations, composite SVs (deletion-insertion, deletion-inversion, and deletion-translocation composites), and complex genomic rearrangements (CGRs), including chromoplexy, chromothripsis, and SV by breakage-fusion-bridge cycles. Our findings suggest that 1 Gy IR exposure causes an average of 2.33 mutational events per Gb genome, comprising 2.15 indels, 0.17 SVs, and 0.01 CGRs, despite a high level of inter-cellular stochasticity. The mutational burden was dependent on total irradiation dose, regardless of dose rate or cell type. The findings were further validated in IR-induced secondary cancers and single cells without clonalization. Overall, our study highlights a comprehensive and clear picture of IR effects on normal mammalian genomes.

Prediction of complications in diabetes mellitus using machine learning models with transplanted topic model features.

Purpose: This study aims to predict the progression of Diabetes Mellitus (DM) from the clinical notes through machine learning based on latent Dirichlet allocation (LDA) topic modeling. Particularly, 174,427 clinical notes of DM patients were collected from the electronic medical record (EMR) system of the Seoul National University Hospital outpatient clinic. Method: We developed a model to predict the development of DM complications. Topics developed by the topic model were exploited as the key feature of our machine-learning model. The proposed model generalized a correlation between topic structures and complications. Results: The model provided acceptable predictive performance for all four types of complications (diabetic retinopathy, diabetic nephropathy, nonalcoholic fatty liver disease, and cerebrovascular accident). Upon employing extreme gradient boosting (XGBoost), we obtained the F1 scores of the predictions for each complication type as 0.844, 0.921, 0.831, and 0.762. Conclusion: This study shows that a machine learning project based on topic modeling can effectively predict the progress of a disease. Furthermore, a unique way of topic model transplanting, which matches the dimension of the topic structures of the two data sets, is presented. Supplementary Information: The online version contains supplementary material available at 10.1007/s13534-023-00322-7.

Current Status and Key Issues of Data Management in Tertiary Hospitals: A Case Study of Seoul National University Hospital.

OBJECTIVES: In the era of the Fourth Industrial Revolution, where an ecosystem is being developed to enhance the quality of healthcare services by applying information and communication technologies, systematic and sustainable data management is essential for medical institutions. In this study, we assessed the data management status and emerging concerns of three medical institutions, while also examining future directions for seamless data management. METHODS: To evaluate the data management status, we examined data types, capacities, infrastructure, backup methods, and related organizations. We also discussed challenges, such as resource and infrastructure issues, problems related to government regulations, and considerations for future data management. RESULTS: Hospitals are grappling with the increasing data storage space and a shortage of management personnel due to costs and project termination, which necessitates countermeasures and support. Data management regulations on the destruction or maintenance of medical records are needed, and institutional consideration for secondary utilization such as long-term treatment or research is required. Government-level guidelines for facilitating hospital data sharing and mobile patient services should be developed. Additionally, hospital executives at the organizational level need to make efforts to facilitate the clinical validation of artificial intelligence software. CONCLUSIONS: This analysis of the current status and emerging issues of data management reveals potential solutions and sets the stage for future organizational and policy directions. If medical big data is systematically managed, accumulated over time, and strategically monetized, it has the potential to create new value.

Selective UMLS knowledge infusion for biomedical question answering.

One of the artificial intelligence applications in the biomedical field is knowledge-intensive question-answering. As domain expertise is particularly crucial in this field, we propose a method for efficiently infusing biomedical knowledge into pretrained language models, ultimately targeting biomedical question-answering. Transferring all semantics of a large knowledge graph into the entire model requires too many parameters, increasing computational cost and time. We investigate an efficient approach that leverages adapters to inject Unified Medical Language System knowledge into pretrained language models, and we question the need to use all semantics in the knowledge graph. This study focuses on strategies of partitioning knowledge graph and either discarding or merging some for more efficient pretraining. According to the results of three biomedical question answering finetuning datasets, the adapters pretrained on semantically partitioned group showed more efficient performance in terms of evaluation metrics, required parameters, and time. The results also show that discarding groups with fewer concepts is a better direction for small datasets, and merging these groups is better for large dataset. Furthermore, the metric results show a slight improvement, demonstrating that the adapter methodology is rather insensitive to the group formulation.

Cryobiopsy: A Breakthrough Strategy for Clinical Utilization of Lung Cancer Organoids.

One major challenge associated with lung cancer organoids (LCOs) is their predominant derivation from surgical specimens of patients with early-stage lung cancer. However, patients with advanced lung cancer, who are in need of chemotherapy, often cannot undergo surgery. Therefore, there is an urgent need to successfully generate LCOs from biopsy specimens. Conventional lung biopsy techniques, such as transthoracic needle biopsy and forceps biopsy, only yield small amounts of lung tissue, resulting in a low success rate for culturing LCOs from biopsy samples. Furthermore, potential complications, like bleeding and pneumothorax, make it difficult to obtain sufficient tissue. Another critical issue is the overgrowth of normal lung cells in later passages of LCO culture, and the optimal culture conditions for LCOs are yet to be determined. To address these limitations, we attempted to create LCOs from cryobiopsy specimens obtained from patients with lung cancer (n = 113). Overall, the initial success rate of establishing LCOs from cryobiopsy samples was 40.7% (n = 46). Transbronchial cryobiopsy enables the retrieval of significantly larger amounts of lung tissue than bronchoscopic forceps biopsy. Additionally, cryobiopsy can be employed for peripheral lesions, and it is aided via radial endobronchial ultrasonography. This study significantly improved the success rate of LCO culture and demonstrated that the LCOs retained characteristics that resembled the primary tumors. Single-cell RNA sequencing confirmed high cancer cell purity in early passages of LCOs derived from patients with advanced lung cancer. Furthermore, the three-dimensional structure and intracellular components of LCOs were characterized using three-dimensional holotomography. Finally, drug screening was performed using a specialized micropillar culture system with cryobiopsy-derived LCOs. LCOs derived from cryobiopsy specimens offer a promising solution to the critical limitations of conventional LCOs. Cryobiopsy can be applied to patients with lung cancer at all stages, including those with peripheral lesions, and can provide sufficient cells for LCO generation. Therefore, we anticipate that cryobiopsy will serve as a breakthrough strategy for the clinical application of LCOs in all stages of lung cancer.

PAIP 2020: Microsatellite instability prediction in colorectal cancer.

Microsatellite instability (MSI) refers to alterations in the length of simple repetitive genomic sequences. MSI status serves as a prognostic and predictive factor in colorectal cancer. The MSI-high status is a good prognostic factor in stage II/III cancer, and predicts a lack of benefit to adjuvant fluorouracil chemotherapy in stage II cancer but a good response to immunotherapy in stage IV cancer. Therefore, determining MSI status in patients with colorectal cancer is important for identifying the appropriate treatment protocol. In the Pathology Artificial Intelligence Platform (PAIP) 2020 challenge, artificial intelligence researchers were invited to predict MSI status based on colorectal cancer slide images. Participants were required to perform two tasks. The primary task was to classify a given slide image as belonging to either the MSI-high or the microsatellite-stable group. The second task was tumor area segmentation to avoid ties with the main task. A total of 210 of the 495 participants enrolled in the challenge downloaded the images, and 23 teams submitted their final results. Seven teams from the top 10 participants agreed to disclose their algorithms, most of which were convolutional neural network-based deep learning models, such as EfficientNet and UNet. The top-ranked system achieved the highest F1 score (0.9231). This paper summarizes the various methods used in the PAIP 2020 challenge. This paper supports the effectiveness of digital pathology for identifying the relationship between colorectal cancer and the MSI characteristics.

A Deep Learning Model Using Chest Radiographs for Prediction of 30-Day Mortality in Patients With Community-Acquired Pneumonia: Development and External Validation.

BACKGROUND. Chest radiography is an essential tool for diagnosing community-acquired pneumonia (CAP), but it has an uncertain prognostic role in the care of patients with CAP. OBJECTIVE. The purpose of this study was to develop a deep learning (DL) model to predict 30-day mortality from diagnosis among patients with CAP by use of chest radiographs to validate the performance model in patients from different time periods and institutions. METHODS. In this retrospective study, a DL model was developed from data on 7105 patients from one institution from March 2013 to December 2019 (3:1:1 allocation to training, validation, and internal test sets) to predict the risk of all-cause mortality within 30 days after CAP diagnosis by use of patients' initial chest radiographs. The DL model was evaluated in a cohort of patients diagnosed with CAP during emergency department visits at the same institution from January 2020 to March 2020 (temporal test cohort [n = 947]) and in two additional cohorts from different institutions (external test cohort A [n = 467], January 2020 to December 2020; external test cohort B [n = 381], March 2019 to October 2021). AUCs were compared between the DL model and an established risk prediction tool based on the presence of confusion, blood urea nitrogen level, respiratory rate, blood pressure, and age 65 years or older (CURB-65 score). The combination of CURB-65 score and DL model was evaluated with a logistic regression model. RESULTS. The AUC for predicting 30-day mortality was significantly larger (p < .001) for the DL model than for CURB-65 score in the temporal test set (0.77 vs 0.67). The larger AUC for the DL model than for CURB-65 score was not significant (p > .05) in external test cohort A (0.80 vs 0.73) or external test cohort B (0.80 vs 0.72). In the three cohorts, the DL model, in comparison with CURB-65 score, had higher (p < .001) specificity (range, 61-69% vs 44-58%) at the sensitivity of CURB-65 score. The combination of DL model and CURB-65 score, in comparison with CURB-65 score, yielded a significant increase in AUC in the temporal test cohort (0.77, p < .001) and external test cohort B (0.80, p = .04) and a nonsignificant increase in AUC in external test cohort A (0.80, p = .16). CONCLUSION. A DL-based model consisting of initial chest radiographs was predictive of 30-day mortality among patients with CAP with improved performance over CURB-65 score. CLINICAL IMPACT. The DL-based model may guide clinical decision-making in the care of patients with CAP.

Comparative and expression analyses of AP2/ERF genes reveal copy number expansion and potential functions of ERF genes in Solanaceae.

BACKGROUND: The AP2/ERF gene family is a superfamily of transcription factors that are important in the response of plants to abiotic stress and development. However, comprehensive research of the AP2/ERF genes in the Solanaceae family is lacking. RESULTS: Here, we updated the annotation of AP2/ERF genes in the genomes of eight Solanaceae species, as well as Arabidopsis thaliana and Oryza sativa. We identified 2,195 AP2/ERF genes, of which 368 (17%) were newly identified. Based on phylogenetic analyses, we observed expansion of the copy number of these genes, especially those belonging to specific Ethylene-Responsive Factor (ERF) subgroups of the Solanaceae. From the results of chromosomal location and synteny analyses, we identified that the AP2/ERF genes of the pepper (Capsicum annuum), the tomato (Solanum lycopersicum), and the potato (Solanum tuberosum) belonging to ERF subgroups form a tandem array and most of them are species-specific without orthologs in other species, which has led to differentiation of AP2/ERF gene repertory among Solanaceae. We suggest that these genes mainly emerged through recent gene duplication after the divergence of these species. Transcriptome analyses showed that the genes have a putative function in the response of the pepper and tomato to abiotic stress, especially those in ERF subgroups. CONCLUSIONS: Our findings will provide comprehensive information on AP2/ERF genes and insights into the structural, evolutionary, and functional understanding of the role of these genes in the Solanaceae.

A database of 5305 healthy Korean individuals reveals genetic and clinical implications for an East Asian population.

Despite substantial advances in disease genetics, studies to date have largely focused on individuals of European descent. This limits further discoveries of novel functional genetic variants in other ethnic groups. To alleviate the paucity of East Asian population genome resources, we established the Korean Variant Archive 2 (KOVA 2), which is composed of 1896 whole-genome sequences and 3409 whole-exome sequences from healthy individuals of Korean ethnicity. This is the largest genome database from the ethnic Korean population to date, surpassing the 1909 Korean individuals deposited in gnomAD. The variants in KOVA 2 displayed all the known genetic features of those from previous genome databases, and we compiled data from Korean-specific runs of homozygosity, positively selected intervals, and structural variants. In doing so, we found loci, such as the loci of ADH1A/1B and UHRF1BP1, that are strongly selected in the Korean population relative to other East Asian populations. Our analysis of allele ages revealed a correlation between variant functionality and evolutionary age. The data can be browsed and downloaded from a public website ( https://www.kobic.re.kr/kova/ ). We anticipate that KOVA 2 will serve as a valuable resource for genetic studies involving East Asian populations.

Two different domain architectures generate structural and functional diversity among bZIP genes in the Solanaceae family.

The bZIP gene family is one of the largest transcription factor families and has important roles in plant growth, development, and stress responses. However, bZIP genes in the Solanaceae family have not been extensively investigated. Here, we conducted genome-wide re-annotation in nine Solanaceae species and Arabidopsis thaliana. We annotated 935 bZIP genes, including 107 (11%) that were newly identified. Structural analyses of bZIP genes in the Solanaceae family revealed that the bZIP domain displayed two types of architectures depending on the presence of an additional domain, suggesting that these architectures generate diversified structures and functions. Motif analyses indicated that the two types of bZIP genes had distinct sequences adjacent to the bZIP domain. Phylogenetic analyses suggested that the two types of bZIP genes distinctly evolved and ultimately adapted in different lineages. Transcriptome analyses in pepper (Capsicum annuum) and tomato (Solanum lycopersicum) revealed putative functional diversity between the two types of bZIP genes in response to various abiotic stresses. This study extensively updated bZIP gene family annotations and provided novel evolutionary and functional evidence for the role of bZIP genes in Solanaceae plants. Our findings provide evolutionary and functional characteristics of bZIP genes for a better understanding of their roles in Solanaceae plants.

SARS-CoV-2 spike N-terminal domain modulates TMPRSS2-dependent viral entry and fusogenicity.

The severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) spike N-terminal domain (NTD) remains poorly characterized despite enrichment of mutations in this region across variants of concern (VOCs). Here, we examine the contribution of the NTD to infection and cell-cell fusion by constructing chimeric spikes bearing B.1.617 lineage (Delta and Kappa variants) NTDs and generating spike pseudotyped lentivirus. We find that the Delta NTD on a Kappa or wild-type (WT) background increases S1/S2 cleavage efficiency and virus entry, specifically in lung cells and airway organoids, through use of TMPRSS2. Delta exhibits increased cell-cell fusogenicity that could be conferred to WT and Kappa spikes by Delta NTD transfer. However, chimeras of Omicron BA.1 and BA.2 spikes with a Delta NTD do not show more efficient TMPRSS2 use or fusogenicity. We conclude that the NTD allosterically modulates S1/S2 cleavage and spike-mediated functions in a spike context-dependent manner, and allosteric interactions may be lost when combining regions from more distantly related VOCs.

Accuracy of Cloud-Based Speech Recognition Open Application Programming Interface for Medical Terms of Korean.

BACKGROUND: There are limited data on the accuracy of cloud-based speech recognition (SR) open application programming interfaces (APIs) for medical terminology. This study aimed to evaluate the medical term recognition accuracy of current available cloud-based SR open APIs in Korean. METHODS: We analyzed the SR accuracy of currently available cloud-based SR open APIs using real doctor-patient conversation recordings collected from an outpatient clinic at a large tertiary medical center in Korea. For each original and SR transcription, we analyzed the accuracy rate of each cloud-based SR open API (i.e., the number of medical terms in the SR transcription per number of medical terms in the original transcription). RESULTS: A total of 112 doctor-patient conversation recordings were converted with three cloud-based SR open APIs (Naver Clova SR from Naver Corporation; Google Speech-to-Text from Alphabet Inc.; and Amazon Transcribe from Amazon), and each transcription was compared. Naver Clova SR (75.1%) showed the highest accuracy with the recognition of medical terms compared to the other open APIs (Google Speech-to-Text, 50.9%, P < 0.001; Amazon Transcribe, 57.9%, P < 0.001), and Amazon Transcribe demonstrated higher recognition accuracy compared to Google Speech-to-Text (P < 0.001). In the sub-analysis, Naver Clova SR showed the highest accuracy in all areas according to word classes, but the accuracy of words longer than five characters showed no statistical differences (Naver Clova SR, 52.6%; Google Speech-to-Text, 56.3%; Amazon Transcribe, 36.6%). CONCLUSION: Among three current cloud-based SR open APIs, Naver Clova SR which manufactured by Korean company showed highest accuracy of medical terms in Korean, compared to Google Speech-to-Text and Amazon Transcribe. Although limitations are existing in the recognition of medical terminology, there is a lot of rooms for improvement of this promising technology by combining strengths of each SR engines.

Airway secretory cell fate conversion via YAP-mTORC1-dependent essential amino acid metabolism.

Tissue homeostasis requires lineage fidelity of stem cells. Dysregulation of cell fate specification and differentiation leads to various diseases, yet the cellular and molecular mechanisms governing these processes remain elusive. We demonstrate that YAP/TAZ activation reprograms airway secretory cells, which subsequently lose their cellular identity and acquire squamous alveolar type 1 (AT1) fate in the lung. This cell fate conversion is mediated via distinctive transitional cell states of damage-associated transient progenitors (DATPs), recently shown to emerge during injury repair in mouse and human lungs. We further describe a YAP/TAZ signaling cascade to be integral for the fate conversion of secretory cells into AT1 fate, by modulating mTORC1/ATF4-mediated amino acid metabolism in vivo. Importantly, we observed aberrant activation of the YAP/TAZ-mTORC1-ATF4 axis in the altered airway epithelium of bronchiolitis obliterans syndrome, including substantial emergence of DATPs and AT1 cells with severe pulmonary fibrosis. Genetic and pharmacologic inhibition of mTORC1 activity suppresses lineage alteration and subepithelial fibrosis driven by YAP/TAZ activation, proposing a potential therapeutic target for human fibrotic lung diseases.

De novo phasing resolves haplotype sequences in complex plant genomes.

Genome phasing is a recently developed assembly method that separates heterozygous eukaryotic genomic regions and builds haplotype-resolved assemblies. Because differences between haplotypes are ignored in most published de novo genomes, assemblies are available as consensus genomes consisting of haplotype mixtures, thus increasing the need for genome phasing. Here, we review the operating principles and characteristics of several freely available and widely used phasing tools (TrioCanu, FALCON-Phase, and ALLHiC). An examination of downstream analyses using haplotype-resolved genome assemblies in plants indicated significant differences among haplotypes regarding chromosomal rearrangements, sequence insertions, and expression of specific alleles that contribute to the acquisition of the biological characteristics of plant species. Finally, we suggest directions to solve addressing limitations of current genome-phasing methods. This review provides insights into the current progress, limitations, and future directions of de novo genome phasing, which will enable researchers to easily access and utilize genome-phasing in studies involving highly heterozygous complex plant genomes.

Altered TMPRSS2 usage by SARS-CoV-2 Omicron impacts infectivity and fusogenicity.

The SARS-CoV-2 Omicron BA.1 variant emerged in 20211 and has multiple mutations in its spike protein2. Here we show that the spike protein of Omicron has a higher affinity for ACE2 compared with Delta, and a marked change in its antigenicity increases Omicron's evasion of therapeutic monoclonal and vaccine-elicited polyclonal neutralizing antibodies after two doses. mRNA vaccination as a third vaccine dose rescues and broadens neutralization. Importantly, the antiviral drugs remdesivir and molnupiravir retain efficacy against Omicron BA.1. Replication was similar for Omicron and Delta virus isolates in human nasal epithelial cultures. However, in lung cells and gut cells, Omicron demonstrated lower replication. Omicron spike protein was less efficiently cleaved compared with Delta. The differences in replication were mapped to the entry efficiency of the virus on the basis of spike-pseudotyped virus assays. The defect in entry of Omicron pseudotyped virus to specific cell types effectively correlated with higher cellular RNA expression of TMPRSS2, and deletion of TMPRSS2 affected Delta entry to a greater extent than Omicron. Furthermore, drug inhibitors targeting specific entry pathways3 demonstrated that the Omicron spike inefficiently uses the cellular protease TMPRSS2, which promotes cell entry through plasma membrane fusion, with greater dependency on cell entry through the endocytic pathway. Consistent with suboptimal S1/S2 cleavage and inability to use TMPRSS2, syncytium formation by the Omicron spike was substantially impaired compared with the Delta spike. The less efficient spike cleavage of Omicron at S1/S2 is associated with a shift in cellular tropism away from TMPRSS2-expressing cells, with implications for altered pathogenesis.

Use of Time-to-Event Analysis to Develop On-Scene Return of Spontaneous Circulation Prediction for Out-of-Hospital Cardiac Arrest Patients.

STUDY OBJECTIVE: We aimed to train and validate the time to on-scene return of spontaneous circulation prediction models using time-to-event analysis among out-of-hospital cardiac arrest patients. METHODS: Using a Korean population-based out-of-hospital cardiac arrest registry, we selected a total of 105,215 adults with presumed cardiac etiologies between 2013 and 2018. Patients from 2013 to 2017 and from 2018 were analyzed for training and test, respectively. We developed 4 time-to-event analyzing models (Cox proportional hazard [Cox], random survival forest, extreme gradient boosting survival, and DeepHit) and 4 classification models (logistic regression, random forest, extreme gradient boosting, and feedforward neural network). Patient characteristics and Utstein elements collected at the scene were used as predictors. Discrimination and calibration were evaluated by Harrell's C-index and integrated Brier score. RESULTS: Among the 105,215 patients (mean age 70 years and 64% men), 86,314 and 18,901 patients belonged to the training and test sets, respectively. On-scene return of spontaneous circulation was achieved in 5,240 (6.1%) patients in the former set and 1,709 (9.0%) patients in the latter. The proportion of emergency medical services (EMS) management was higher and scene time interval longer in the latter. Median time from EMS scene arrival to on-scene return of spontaneous circulation was 8 minutes for both datasets. Classification models showed similar discrimination and poor calibration power compared to survival models; Cox showed high discrimination with the best calibration (C-index [95% confidence interval]: 0.873 [0.865 to 0.882]; integrated Brier score at 30 minutes: 0.060). CONCLUSION: Incorporating time-to-event analysis could lead to improved performance in prediction models and contribute to personalized field EMS resuscitation decisions.