Area Under The ROC Curve Research Articles

The Perioperative NonaGEnaRIan And cenTenarian suRgICal (GERIATRIC) Risk Stratification Tool

Objective: To develop age-appropriate nonaGEnaRIan And cenTenarian suRgICal (GERIATRIC) risk tool for classifying patients who may or may not develop postoperative complications or die within their index hospital admission. Background: There are no validated perioperative risk stratification tools for use in nonagenarian and centenarian patients—people aged 90 to 99 years and >100 years. Methods: In this retrospective observational study, nonagenarians and centenarians undergoing any surgical procedure were profiled. Surgery severity was stratified, and the incidence and grade of postoperative complications were recorded. Multivariable logistic regression analysis was performed on a training cohort, followed by calibration on a validation cohort, followed by performance evaluation on a testing cohort. The discriminative accuracy was compared to that of the age-adjusted Charlson Comorbidity Index for each outcome. The primary outcome was the ability of the risk stratification tool to effectively classify patients into those who may or may not experience a postoperative complications or mortality within their index hospital stay. Results: A total of 3085 patients were enrolled. The GERIATRIC risk tool had good discriminative accuracy for any postoperative complication [area under the receiver operating characteristic curves (AUROC), 0.857; 95% CI = 0.824–0.890] and any severe postoperative complication (AUROC, 0.833; 95% CI = 0.793–0.874), and fair discriminative accuracy for in-hospital mortality (AUROC, 0.780; 95% CI = 0.668–0.893). Conclusions: Compared to the age-adjusted Charlson Comorbidity Index, The GERIATRIC risk tool was accurate in classifying patients into those who may or may not experience severe complications or die during their index admission. The tool can be used to assist perioperative clinicians with shared decision-making and short-term prognostication.

Comparative Analysis of ML Models with Selection Methods for Early Predictive Analytics of Sepsis in ICU

Objectives: This study aims to make early prediction of Sepsis using ML algorithms and provide comparative analysis of different feature selection techniques. Methods: In this study, the Physionet website dataset has been used. Sepsis categorization is done on the basis of Sepsis-3 criteria. The dataset used is highly imbalanced, with only 2% of the data belonging to Sepsis. The SmoteTomek technique is used to handle an imbalanced dataset. Various filter, embedded, and wrapper feature selection techniques, like tree-based feature selection technique, recursive feature elimination (RFE), information gain, Bhattacharya distance, lasso, etc., have been applied for the top-performing classification ML models. These selection techniques were applied to RandomForest, K Nearest Neighbors (KNN), and Decision Tree models. We compared the impact of these selection techniques on the aforesaid machine learning models. Findings: After applying the RFE technique, the Area Under the Receiver Operating Characteristic curve (AUROC) score of the RandomForest model has slightly increased from 0.996 to 0.9974. KNN model with a tree-based feature selection technique showed the highest sensitivity of 0.934. which is slightly higher than the sensitivity of 0.922, which was without applying any feature selection technique. Along with the AUROC score, the highest performance, in terms of specificity (0.9976), accuracy (0.9959), and f-measure score (0.9062), is achieved when RFE is applied to the RandomForest model. The best selection algorithm for decision trees and KNN is the tree-based selection technique. RFE is the best selection technique for RandomForest. Novelty: In this research, the AUROC score is slightly increased to 0.9974, which has not been achieved yet. Instead of 40, the number of features chosen is 20. This research also provides a comparison of different feature selection techniques like tree-based feature selection, information gain, Bhattacharya, and RFE. It also analyses their impact on the performance of models, which has not been done yet with the same set of selection techniques. Keywords: Sepsis, Machine Learning, Feature selection, Early prediction, Predictive Analytics

Radiomics-based predictive model for preoperative risk classification of gastrointestinal stromal tumors using multiparametric magnetic resonance imaging: aretrospective study.

The aim of this study was to develop and assess aradiomics model utilizing multiparametric magnetic resonance imaging (MRI) for the prediction of preoperative risk assessment in gastrointestinal stromal tumors (GISTs). An analysis was performed retrospectively on agroup of 121 patients who received ahistological diagnosis of GIST. They were then divided into two sets, with85 in the training set and 36 in the validation set through random partitioning. Radiomics features from five MRI sequences, totaling 600 per patient, were extracted and subjected to feature selection utilizing arandom forest algorithm. The discriminatory efficacy of the models was evaluated through receiver operating characteristic (ROC) and precision-recall (P-R) curve analyses. Model calibration was assessed via calibration curves. Subgroup analysis was performed on GISTs with apathological maximum diameter equal to or less than 5 cm. Furtherly, Kaplan-Meier (K-M) curves and log-rank tests were used to compare the differences in survival status among different groups. Cox regression analysis was employed to identify independent prognostic factors and to construct aprognostic prediction model. The clinical model (ModelC) displayed limited predictive efficacy in the context of GIST. Conversely, aradiomics model (ModelR) incorporating five parameters exhibited robust discriminative capabilities across both the training and validation sets, yielding area under the ROC curve (AUC) values of 0.893 (95% confidence interval [CI]: 0.807-0.949) and 0.855 (95% CI: 0.732-0.978), respectively. The F1max scores derived from the P‑R curves were 0.741 and 0.842 for the training and validation sets, respectively. Noteworthy was the exclusion of the two-dimensional tumor diameter and tumor location when constructing ahybrid model (ModelCR) that amalgamated radiomics and clinical features. ModelR demonstrated asubstantially enhanced discriminative capacity in the training set compared with ModelC (p < 0.005). The net reclassification improvement (NRI) corroborated the superior performance of ModelR over ModelC, thereby enhancing diagnostic accuracy and clinical applicability. Patients in the high-risk group had significantly worse recurrence-free survival (RFS, p < 0.001) and overall survival (OS, p = 0.004), and the radiomics signature is an independent risk factor for RFS. The extended model incorporating the radiomics signature outperformed the baseline model in terms of risk assessment accuracy (p < 0.001). Our investigation underscores the value of integrating radiomics analysis in conjunction with machine learning algorithms for prognostic risk stratification in GIST, presenting promising implications for informing clinical decision-making processes as well as optimizing management strategies.

Computed tomography enterography-based deep learning radiomics to predict stratified healing in patients with Crohn's disease: a multicenter study.

This study developed a deep learning radiomics (DLR) model utilizing baseline computed tomography enterography (CTE) to non-invasively predict stratified healing in Crohn's disease (CD) patients following infliximab (IFX) treatment. The study included 246 CD patients diagnosed at three hospitals. From the first two hospitals, 202 patients were randomly divided into a training cohort (n = 141) and a testing cohort (n = 61) in a 7:3 ratio. The remaining 44 patients from the third hospital served as the validation cohort. Radiomics and deep learning features were extracted from both the active lesion wall and mesenteric adipose tissue. The most valuable features were selected using univariate analysis and least absolute shrinkage and selection operator (LASSO) regression. Multivariate logistic regression was then employed to construct the radiomics, deep learning, and DLR models. Model performance was evaluated using receiver operating characteristic (ROC) curves. The DLR model achieved an area under the ROC curve (AUC) of 0.948 (95% CI: 0.916-0.980), 0.889 (95% CI: 0.803-0.975), and 0.938 (95% CI: 0.868-1.000) in the training, testing, and validation cohorts, respectively in predicting mucosal healing (MH). Furthermore, the diagnostic performance of DLR model in predicting transmural healing (TH) was 0.856 (95% CI: 0.776-0.935). We have developed a DLR model based on the radiomics and deep learning features of baseline CTE to predict stratified healing (MH and TH) in CD patients following IFX treatment with high accuracies in both testing and external cohorts. The deep learning radiomics model developed in our study, along with the nomogram, can intuitively, accurately, and non-invasively predict stratified healing at baseline CT enterography. Early prediction of mucosal and transmural healing in Crohn's Disease patients is beneficial for treatment planning. This model demonstrated excellent performance in predicting mucosal healing and had a diagnostic performance in predicting transmural healing of 0.856. CT enterography images of active lesion walls and mesenteric adipose tissue exhibit an association with stratified healing in Crohn's disease patients.

An interpretable (explainable) model based on machine learning and SHAP interpretation technique for mapping wind erosion hazard.

Soil erosion by wind poses a significant threat to various regions across the globe, such as drylands in the Middle East and Iran. Wind erosion hazard maps can assist in identifying the regions of highest wind erosion risk and are a valuable tool for the mitigation of its destructive consequences. This study aims to map wind erosion hazards by developing an interpretable (explainable) model based on machine learning (ML) and Shapley additive exPlanation (SHAP) interpretation techniques. Four ML models, namely random forest (RF), support vector machine (SVM), extreme gradient boosting (XGB), and quadratic discriminant analysis (QDA) were used. Thirteen features associated with wind erosion were mapped spatially and then subjected to a multivariate adaptive regression spline (MARS) feature selection algorithm, and then, tolerance coefficient (TC) and variance inflation factor (VIF) statistical tests were used to explore multicollinearity among the variables. MARS analysis shows that eight features consisting of elevation (or DEM), soil bulk density, precipitation, aspect, slope, soil sand content, vegetation cover (or NDVI), and lithology were the most effective for wind erosion, while no collinearity existed among these variables. The ML models were used for ranking the effective features, and the research introduces the application of an interpretable ML model for the interpretation of predictive model's output. The ranking of effective features by RF-as the most typical ML model-revealed that elevation and soil bulk density were the two most important features. According to the area under the receiver operating characteristic curve (AUROC) (with a value > 90%) and precision-recall (PR) (with a value > 90%) curves, all four ML models performed with great accuracy. According to the PR curve, the SVM model performed slightly better than others, and its results revealed that 20.9%, 23%, and 16.6% of the total area in Hormozgan Province is characterized by moderate, high, and very high hazard classes to wind erosion, respectively. SHAP revealed that soil sand content and elevation are the most important variables contributing to the predictive model output. Overall, our research is one of the pioneering applications of interpretable ML models in mapping wind erosion hazards in Southern Iran. We recommend that future research should address the aspect of interpretability in order to better understand predictive model outputs.

Performance of Polygenic Risk Scores for Primary Open-Angle Glaucoma in Populations of African Descent.

Primary open-angle glaucoma (POAG) polygenic risk scores (PRSs) continue to be evaluated in primarily European-ancestry populations despite higher prevalence and worse outcomes in African-ancestry populations. To evaluate how established POAG PRSs perform in African-ancestry samples from the Genetics in Glaucoma Patients of African Descent (GIGA), Genetics of Glaucoma in Individuals of African Descent (GGLAD), and Million Veteran Program (MVP) datasets and compare these with European-ancestry samples. This was a multicenter, cross-sectional study of POAG cases and controls from Tanzania, South Africa, Nigeria, Ghana, and the US. Included were individuals of African descent from South Africa and Tanzania from the GIGA dataset; individuals of African descent from Ghana, Nigeria, and the US from the GGLAD dataset; and individuals of African or European descent from the US in the MVP dataset. Data were analyzed from January 2022 to July 2023. Three PRSs derived from large meta-analyses of European and Asian populations, namely Gharahkhani et al (Gharahkhani PRS), Han et al (Han PRS), and Craig et al (Craig PRS). Odds ratios (ORs) for POAG risk stratification comparing the highest and lowest quintiles; area under the receiver operating characteristic curve (AUROC), and liability coefficient of determination (R2) for the addition of PRS to a baseline of age, sex, and first 5 principal components. A total of 11 673 cases and 66 432 controls were included in this study across 7 ancestral groups. Mean (SD) age of the total participants was 76.9 (8.7) years, with 74 304 males (95.1%). The following were included in each dataset: GIGA (663 cases, 476 controls), GGLAD (1471 cases, 1482 controls), and MVP (9559 cases, 64 474 controls). Increases in ORs were found for the highest POAG risk quintile ranging from an OR of 1.68 (95% CI, 1.17-2.43) in Ghanaians to 7.05 (95% CI, 2.73-19.6) in the South African multiple ancestry group (which derives from at least 5 distinct ancestral groups: Khoisan, Bantus, Europeans, Indians, and Southeast Asians) with the Gharahkhani PRS. The Han PRS showed OR increases for the highest POAG risk quintile ranging from 2.27 (95% CI, 1.49-3.47) in African American individuals in the GGLAD dataset to 7.24 (95% CI, 6.47-8.12) in Europeans. The Craig PRS predicted OR increases in the highest quintile for all groups ranging from 1.51 (95% CI, 1.05-2.18) in Ghanaians to 6.31 (95% CI, 5.67-7.04) in Europeans. However, AUROC and R2 increases above baseline were lower for all African-ancestry compared with European-ancestry groups in the 3 tested PRSs. In this cross-sectional study, despite some improvements in OR-based risk stratification using the Gharahkhani PRSs, Han PRSs, and Craig PRSs, consistently lower improvements in AUROC and R2 for African-ancestry compared with European-ancestry groups highlight the need for risk prediction models tailored to diverse populations.

In vivo evaluation of complex polyps with endoscopic optical coherence tomography and deep learning during routine colonoscopy: a feasibility study.

Standard-of-care (SoC) imaging for assessing colorectal polyps during colonoscopy, based on white-light colonoscopy (WLC) and narrow-band imaging (NBI), does not have sufficient accuracy to assess the invasion depth of complex polyps non-invasively during colonoscopy. We aimed to evaluate the feasibility of a custom endoscopic optical coherence tomography (OCT) probe for assessing colorectal polyps during routine colonoscopy. Patients referred for endoscopic treatment of large colorectal polyps were enrolled in this pilot clinical study, which used a side-viewing OCT catheter developed for use with an adult colonoscope. OCT images of polyps were captured during colonoscopy immediately before SoC treatment. A deep learning model was trained to differentiate benign from deeply invasive lesions for real-time diagnosis. 35 polyps from 32 patients were included. OCT imaging added on average 3:40min (range 1:54-8:20) to the total procedure time. No complications due to OCT were observed. OCT revealed distinct subsurface tissue structures that correlated with histological findings, including tubular adenoma (n = 20), tubulovillous adenoma (n = 10), sessile serrated polyps (n = 3), and invasive cancer (n = 2). The deep learning model achieved an area under the receiver operating characteristic curve (AUROC) of 0.984 (95%CI 0.972-0.996) and Cohen's kappa of 0.845 (95%CI 0.774-0.915) when compared to gold standard histopathology. OCT is feasible and safe for polyp assessment during routine colonoscopy. When combined with deep learning, OCT offers clinicians increase confidence in identifying deeply invasive cancers, potentially improving clinical decision-making. Compared to previous studies, ours offers a nuanced comparison between not just benign and malignant lesions, but across multiple histological subtypes of polyps.

Usefulness of Dual-Energy CT for Differentiating Hemorrhage From Iodine Extravasation in Meningiomas After Preoperative Embolization

Objective Discriminating between hemorrhage and iodine extravasation can pose challenges in conventional computed tomography (CCT) images following preoperative embolization for meningioma. This study aimed to assess the efficacy of dual-energy computed tomography (DECT) in differentiating hemorrhage from iodine extravasation after preoperative embolization for meningioma. Methods Twenty-one consecutive meningioma patients who underwent CCT before and DECT immediately after preoperative embolization were included in this study. Two independent observers conducted qualitative assessments on CCT and virtual noncontrast (VNC) images and iodine maps (IMs) to differentiate between hemorrhage and iodine extravasation. One observer recorded CT values of hemorrhage and iodine extravasation on CCT and VNC images. The ratio of maximum attenuation to minimum attenuation on VNC images was defined as the VNC ratio. Statistical analysis included Kappa (κ) statistics, unpaired t tests, and receiver operating characteristic (ROC) analysis. Results Interobserver agreement for qualitative assessment was fair (κ = 0.231) for CCT alone and good (κ = 0.723) for CCT plus VNC imaging and IM. The addition of VNC imaging and IM to CCT improved differential confidence in 16 (76%) and 18 (86%) cases of the two observers, respectively, increasing the area under the receiver operating characteristic curve (AUROC) from 0.868 to 0.895 and 0.658 to 0.947, respectively. At a cutoff value of 1.527, the VNC ratio was significantly higher for hemorrhage than iodine extravasation (P &lt; 0.05), with the highest diagnostic performance (AUROC, 1). Conclusions DECT with VNC imaging and IM is useful for differentiating hemorrhage from iodine extravasation in meningiomas with preoperative embolization.

Performance of a Deep Learning System and Performance of Optometrists for the Detection of Glaucomatous Optic Neuropathy Using Colour Retinal Photographs

Background/Objectives: Glaucoma is the leading cause of irreversible blindness, with a significant proportion of cases remaining undiagnosed globally. The interpretation of optic disc and retinal nerve fibre layer images poses challenges for optometrists and ophthalmologists, often leading to misdiagnosis. AI has the potential to improve diagnosis. This study aims to validate an AI system (a convolutional neural network based on the Inception-v3 architecture) for detecting glaucomatous optic neuropathy (GON) using colour fundus photographs from a UK population and to compare its performance against Australian optometrists. Methods: A retrospective external validation study was conducted, comparing AI’s performance with that of 11 AHPRA-registered optometrists in Australia on colour retinal photographs, evaluated against a reference (gold) standard established by a panel of glaucoma specialists. Statistical analyses were performed using sensitivity, specificity, and area under the receiver operating characteristic curve (AUROC). Results: For referable GON, the sensitivity of the AI (33.3% [95%CI: 32.4–34.3) was significantly lower than that of optometrists (65.1% [95%CI: 64.1–66.0]), p &lt; 0.0001, although with significantly higher specificity (AI: 97.4% [95%CI: 97.0–97.7]; optometrists: 85.5% [95%CI: 84.8–86.2], p &lt; 0.0001). The optometrists demonstrated significantly higher AUROC (0.753 [95%CI: 0.744–0.762]) compared to AI (0.654 [95%CI: 0.645–0.662], p &lt; 0.0001). Conclusion: The AI system exhibited lower performance than optometrists in detecting referable glaucoma. Our findings suggest that while AI can serve as a screening tool, both AI and optometrists have suboptimal performance for the nuanced diagnosis of glaucoma using fundus photographs alone. Enhanced training with diverse populations for AI is essential for improving GON detection and addressing the significant challenge of undiagnosed cases.

Abstract B063: DiMMER: A robust computational pipeline for differential methylation marker evaluation in R of cell-free DNA fragments

Abstract Background: Decoding the origins of cell-free DNA (cfDNA) released from dying cells in a liquid biopsy sample offers the potential to provide insight into the dynamic, organism-wide changes reflective of health and disease, making cfDNA an ideal target for serial, minimally invasive monitoring of disease-related changes. To this end, cell-type specific DNA methylation patterns offer a promising target to facilitate tissue of origin analysis, yet limited methods exist to identify differentially methylated regions that distinguish cell-types. Methods: We develop a robust differentially methylated marker region identification pipeline, DiMMER, that leverages cfDNA fragment level information of neighboring, co-regulated CpG sites on methylome-wide sequencing reads. Compared to prior methods that utilize simple heuristics in one-vs-all average methylation rate comparisons, we implement an individual pairwise statistical test procedure across all cell-types or groups under consideration. We assess the cell-specific nature of identified differentially methylated marker regions by generating in-silico mixtures from known cell-type of origin at each region, and calculating the area under the receiver operating characteristic curve (AUROC). We utilize our differentially methylated marker finding pipeline to identify melanocyte specific methylation marker regions and assess their functional role through annotations. We show these cell-specific regions are conserved in melanoma cell-lines and are distinct from the aberrant changes in a cancer context. Results: We identified the most cell-type specific differentially methylated marker regions across 24 distinct cell-type groups using our pipeline. Compared to a simpler one-vs-all heuristic, we demonstrate improved cell-type specificity evidenced by a higher AUROC on average. Using our pipeline we identified melanocyte-specific marker regions, which were commonly found in intronic and promoter regions of genes related to melanocyte development and differentiation. When comparing methylation patterns between melanocytes and melanoma cell-lines, we demonstrate melanocyte specific methylation marks are conserved through the transformation to malignant melanoma. Conclusion: As methylome-wide sequencing methods continue to rapidly develop and more cell-specific methylation data is generated, there is unmet need for more statistically robust differentially methylated marker finding tools. Here we present one such tool, DiMMER, and demonstrate the potential utility of identifying such regions to be used to assess organ-specific load as a measure of residual disease, as opposed to traditional circulating tumor DNA quanitification. Citation Format: Arthur P McDeed, Sidharth S Jain, Megan E McNamara, Amber Alley, Anton Wellstein, Jaeil Ahn. DiMMER: A robust computational pipeline for differential methylation marker evaluation in R of cell-free DNA fragments [abstract]. In: Proceedings of the AACR Special Conference: Liquid Biopsy: From Discovery to Clinical Implementation; 2024 Nov 13-16; San Diego, CA. Philadelphia (PA): AACR; Clin Cancer Res 2024;30(21_Suppl):Abstract nr B063.

A two-stage deep-learning model for determination of the contact of mandibular third molars with the mandibular canal on panoramic radiographs.

This study aimed to assess the accuracy of a two-stage deep learning (DL) model for (1)detecting mandibular third molars (MTMs) and the mandibular canal (MC), and (2)classifying the anatomical relationship between these structures (contact/no contact) on panoramic radiographs. MTMs and MCs were labeled on panoramic radiographs by a calibrated examiner using bounding boxes. Each bounding box contained MTM and MC on one side. The relationship of MTMs with the MC was assessed on CBCT scans by two independent examiners without the knowledge of the condition of MTM and MC on the corresponding panoramic image, and dichotomized as contact/no contact. Data were split into training, validation, and testing sets with a ratio of 80:10:10. Faster R-CNN was used for detecting MTMs and MCs and ResNeXt for classifying their relationship. AP50 and AP75 were used as outcomes for detecting MTMs and MCs, and accuracy, precision, recall, F1-score, and the area-under-the-receiver-operating-characteristics curve (AUROC) were used to assess classification performance. The training and validation of the models were conducted using the Python programming language with the PyTorch framework. Three hundred eighty-seven panoramic radiographs were evaluated. MTMs were present bilaterally on 232 and unilaterally on 155 radiographs. In total, 619 images were collected which included MTMs and MCs. AP50 and AP75 indicating accuracy for detecting MTMs and MCs were 0.99 and 0.90 respectively. Classification accuracy, recall, specificity, F1-score, precision, and AUROC values were 0.85, 0.85, 0.93, 0.84, 0.86, and 0.91, respectively. DL can detect MTMs and MCs and accurately assess their anatomical relationship on panoramic radiographs.

Abstract 4119992: 12-lead electrocardiograms predict adverse cardiovascular outcomes of emergency department patients

Introduction: Chest pain is a common cause for presentation to the emergency department (ED). Identifying patients at increased risk of adverse cardiovascular outcomes may help guide testing. Aims: We sought to determine if deep learning enabled 12-lead electrocardiogram (ECG) analysis may improve prediction of adverse cardiovascular outcomes of ED patients presenting with chest pain compared to conventional approaches. Methods: A pretrained convolutional neural network was applied to derive numerical representations of the index ECG waveforms from patients presenting to the Brigham and Women’s Hospital (BWH) ED with chest pain. We trained a neural network model (‘CP-AI’) to predict 30-day major adverse cardiovascular events (MACE) defined as myocardial infarction (MI), pulmonary embolism (PE), aortic dissection (AD), or mortality using age, sex, cardiac biomarkers (troponin and d-dimer), and the numerical ECG representations as inputs ( Figure 1 ). In a holdout sample of Massachusetts General Hospital (MGH) ED patients, we compared the performance of the CP-AI model to alternative models fit on (1) demographics (‘Age and Sex’) (2) demographics + cardiac biomarkers (‘Biomarker Model’), and (3) demographics + numerical ECG representations (‘ECG Model’). Results: The BWH derivation sample included 18,811 individuals (51% female, mean age 65 ± 16 years). The MGH validation sample included 14,476 individuals (45% female, mean age 65 ± 15 years). The CP-AI model significantly outperformed the comparison models for prediction of 30-day MACE with an area under the receiver operating characteristic curve (AUROC) of 0.82 [95% CI 0.81-0.83] ( Figure 1 ). CP-AI also significantly outperformed the comparison models for classification of the 30-day MACE components including MI (AUROC 0.91 [95% CI 0.90-0.92]), PE (AUROC 0.74 [95% CI 0.72-0.76]), and mortality (AUROC 0.86 [95% CI 0.85-0.87]), but not aortic dissection (0.71 [95% CI 0.68-0.75]). Conclusions: The integration of deep learning ECG analysis with conventional assessment improved prediction of adverse cardiovascular of ED patients presenting with chest pain.

Abstract 4137770: Development of a User-Friendly Self-Screening Tool for Assessing Metabolic Syndrome Risk in young adults from economically challenged regions

Background: Metabolic syndrome is a cluster of conditions that increase the risk of heart disease and diabetes. Early identification and management are crucial, particularly in economically challenged regions where access to healthcare may be limited. Research Questions/Hypothesis: User-friendly self-report data accurately predict metabolic outcomes. Aims: To develop and validate nomograms for individualized estimation of metabolic syndrome risk. Methods: Data from 521 college students (60.1% aged 17-20 years; 68.7% female; 28.0% white) were collected in 2022/2023 from two Brazilian cities. These cities are located in the country's poorest states, with Gini indices of 0.56 and 0.43. The potential predictors include demographic and economic variables, school-related factors, behaviors, and body weight. Based on predictors for abdominal obesity identified through multilevel logistic regression, we created a nomogram model. We performed the Hosmer-Lemeshow test to assess model calibration and used a bootstrapping approach (B = 150) for internal validation. To evaluate external validity, we assessed metabolic syndrome in a subset of 375 students. The area under the receiver operating characteristic curve (AUROC), with a threshold of 0.70, was used to evaluate the model's discrimination accuracy. Results: We identified 114 (23.0%) college students who were abdominally obese. We found ten variables associated with the primary outcome: age, biological sex, physical education facilities, enrollment in sports competition (during elementary school); grade retention, preferred subject, physical education classes per week; enrollment in sports training (during secondary school); adherence of 24-hour movement behaviors and body weight. The proposed nomogram showed acceptable performance in the AUROC (0.94 [95% CI: 0.92-0.96). The calibration assessment indicated reasonable consistency of our model (p &gt; 0.05). In the internal validation, we observed a decreased predictive capability (AUROC = 0.86). Conclusion: The 24h-MESYN risk score offers an effective self-screening tool for college students from diverse racial and ethnic backgrounds in economically challenged regions to assess their risk of developing metabolic syndrome.

External validation of a modified cardiovascular sequential organ failure assessment score in patients with suspected infection using the MIMIC-IV database.

We developed a modified cardiovascular (CV) Sequential Organ Failure Assessment (SOFA) score using an emergency department-based cohort data, incorporating norepinephrine equivalent dose and lactate to represent current clinical practice patterns for vasopressor utilization and the diagnostic significance of lactate, respectively. In this study, we sought to validate this modified CV-SOFA score in intensive care unit patients with suspected infection using the Marketplace for Medical Information in Intensive Care (MIMIC)-IV database. This was a retrospective study that utilized data from the MIMIC-IV database. Modified CV/total SOFA score and original CV/total SOFA score were compared for predicting in-hospital mortality. Area under the receiver operating characteristic curve (AUROC) and the calibration curve were employed to evaluate discrimination and calibration, respectively. A total of 29,618 ICU patients with suspected infections was analyzed. The in-hospital mortality rate was 12.4% (n = 3,675). Modified CV-SOFA score (AUROC 0.667; 95% confidence interval [CI] 0.657-0.677 vs. 0.663; 95% CI 0.654-0.673; p = 0.283) and modified total SOFA score (0.784 [95% CI 0.776-0.793] vs. 0.785 [95% CI 0.777-0.793], p = 0.490) did not differ significantly from the original CV-SOFA score and original total SOFA score, respectively. The calibration curve of the original CV-SOFA score was inferior to that of the modified CV-SOFA score. The modified CV- and total SOFA scores were better calibrated than the original CV- and total SOFA scores, but their discriminative performance was not significantly different. Further studies of the modified CV-SOFA score in different settings and populations are required to assess the generalizability of this score.

Comprehensive analysis of psychological symptoms and quality of life in early patients with IBD: a multicenter study from China.

To investigate the prevalence and risk factors of psychological symptoms and quality of life (QoL) in early patients with inflammatory bowel disease (IBD). From September 2021 to May 2022, a unified questionnaire was developed to collect clinical data from early patients with IBD from 42 tertiary care hospitals. The influencing factors of psychological symptoms and poor QoL are screened by logistic regression analysis for constructing model in predicting poor QoL. The consistency index, receiver operating characteristic (ROC) curve, area under the ROC curve (AUC), net reclassification improvement (NRI), integrated discrimination improvement (IDI), calibration curve, and decision curve analysis (DCA) were used to evaluate the performance of the model. A total of 939 early patients with IBD were surveyed, Among them, 20.3% exhibited anxiety, 21.7% had depression, 57.3% experienced sleep disturbance, and 41.9% reported poor QoL. The factors influencing psychological symptoms varied between ulcerative colitis (UC) and Crohn's disease (CD) patients. The QoL was primarily affected by disease activity, income level and depression. The AUC value of the model in the training group was 0.781 (95% CI: 0.748-0.814). The calibration diagram of the model closely matched the ideal curve. Compared to other prediction models, our model showed superior predictive capability, with NRI and IDI values of 0.324 (95%CI:0.196-0.4513) and 0.026 (95%CI:0.014-0.038), respectively. DCA indicated that the nomogram model could provide clinical benefits. Early patients with IBD exhibit a high prevalence of psychological symptoms and poor QoL. The nomogram prediction model we constructed demonstrates high accuracy and performance in predicting QoL in early patients with IBD.

Abstract 4121733: Comprehensive Plasma Proteomics Profiling Identifies Circulating Biomarkers to Distinguish Hypertrophic Cardiomyopathy from Other Cardiomyopathies with Left Ventricular Hypertrophy

Introduction: Hypertrophic cardiomyopathy (HCM) is the most common genetic cardiomyopathy. It is sometimes challenging to distinguish HCM from other cardiomyopathies that cause left ventricular hypertrophy (LVH) such as hypertensive LVH, transthyretin cardiac amyloidosis (ATTR-CA), and aortic stenosis (AS). We aimed to identify a set of plasma protein biomarkers that distinguishes HCM from the other cardiomyopathies with LVH. Methods: In this multi-center case-control study, we conducted plasma proteomics profiling of 4,979 proteins in cases with HCM (n=879) and controls with hypertensive LVH (n=331), ATTR-CA (n=169), and AS (n=36). We identified clinical parameters that were significantly (univariable P&lt;0.05) different between HCM and hypertensive LVH. We then specified proteins that were significantly (multivariable P&lt;0.001) upregulated or downregulated in HCM compared to hypertensive LVH by performing multivariable logistic regression analysis on each protein adjusting for the identified clinical parameters (comparison #1). We performed the same analysis between HCM and ATTR-CA (comparison #2) as well as between HCM and AS (comparison #3). We identified proteins that were upregulated in HCM in all 3 comparisons. We also specified proteins downregulated in HCM throughout all 3 comparisons. Finally, using these proteins, we constructed a logistic regression model to distinguish HCM from all 3 control groups combined, and calculated an area under the receiver-operating-characteristics curve (AUROC). Results: Eight clinical parameters listed in Image 1 were significantly different between HCM and the controls and adjusted for in the multivariable logistic regression analyses. In the comparison between HCM and hypertensive LVH, 444 proteins were upregulated and 267 downregulated with multivariable P&lt;0.001 ( Image 1 ). There were 226 upregulated and 170 downregulated proteins comparing HCM with ATTR-CA, and 57 upregulated and 74 downregulated proteins when comparing HCM with AS ( Image 1 ). Among these proteins, 4 were upregulated in all 3 comparisons with multivariable P&lt;0.001 and 5 were downregulated ( Image 2 ). The logistic regression model with these 9 proteins had AUROC of 0.86 (95% confidence interval 0.84-0.88, Image 3 ). Conclusions: This study serves as the first to apply comprehensive proteomics profiling to identify circulating biomarkers that distinguish HCM from other cardiomyopathies with LVH independently from potential confounders.

Abstract 4136550: In-hospital cardiac arrest can be detected by Innovative smart healthcare syste

Background: Predicting in-hospital cardiac arrest (IHCA) is crucial for preventing patient fatalities. However, it remains challenging due to the irregularities in time series data across various features. Previous studies relied on traditional medical scoring systems, but their accuracy was limited as the models were overly simplistic. Recent research using machine learning methods has shown improved results. In our work, we employ deep learning and time series approaches, which are considered the most powerful models for addressing complex problems in recent years. Method/Research Design: The cohort study took place in a hospital in Taiwan from January 2016 to August 2022, involving 252,000 patients in general wards. Input data encompassed basic information, vital signs, Glasgow Coma Scale scores, medication, etc., to predict IHCA occurrences within 24 hours. Given the immediacy of IHCA events, we utilized left alignment to handle time series. Evaluation metrics included the area under the receiver operating characteristic curve (AUROC). Results: We compared three types of settings. The first was based on scoring systems, where the national early warning system (NEWS) achieved an AUROC of 0.756, and the modified early warning score (MEWS) reached 0.653. In the next setting, based on the data summary within a given time interval, the machine learning models XGBoost and deep learning model TabNet achieved AUROCs of 0.863 and 0.839, respectively. This indicates that machine learning outperforms traditional medical methods. The final setting involved time series analysis, where XGBoost and the Gate Recurrent Unit (GRU) network achieved 0.848 and 0.904, respectively, surpassing all previous settings. Conclusion: Our study demonstrates that learning-based methods outperform rule-based methods in predicting IHCA events. Machine learning algorithms offer more possibilities and have the potential to address complex problems effectively. Moreover, considering time series data rather than summarizing within fixed intervals preserves information and leads to better scores. Leveraging the time series deep learning method, GRU shows promising potential in significantly reducing the risk of IHCA events.

Abstract Su804: Use of Machine Learning Models to Predict Neurologically Intact Survival for Advanced Age Adults Following Out-of-hospital Cardiac Arrest

Background: We recently published a multivariable logistic regression model that successfully predicted neurologic outcomes in advanced age adults (≥ 65 years old) who achieve return of spontaneous circulation following out-of-hospital cardiac arrest (OHCA). Prior to externally validating this model, we sought to compare the predictive performance of our logistic regression model against common machine learning (ML) algorithms using the same dataset. Methods: We performed a retrospective observational analysis of the Cardiac Arrest Registry to Enhance Survival database from 2013-2021. All non-traumatic OHCA occurring in adults (≥65 years) who survived to hospital admission were included. The primary outcome measure was neurologically intact survival defined as a cerebral performance category (CPC) score of 1 or 2 at hospital discharge. Our original logistic regression model was compared to Boosted Trees and Decision Tree ML algorithms and assessed using accuracy, Area Under the Receiver Operating Characteristic Curve (AUC-ROC) and Area Under the Precision-Recall Curve (AUC-PR). The model variables included age, gender, race/ethnicity, location, witnessed status, who initiated cardiopulmonary resuscitation, whether an automated external defibrillator was applied prior to emergency medical services arrival, and first monitored rhythm. Results: A total of 83,561 OHCA were included. Median age was 75 years (IQR 69-82), 58.9% were male, 53% were White, 67% experienced an OHCA at home, 52% were witnessed by a bystander, 44% received bystander CPR, and 34% were found in shockable rhythm. Neurologically intact survival (CPC 1 or 2) occurred in 23% of patients. The logistic regression model had an accuracy of 79.6%, AUC-PR 0.542, and AUC-ROC of 0.774. The Boosted Trees model had an accuracy of 79.2%, AUC-PR 0.555, and AUC-ROC of 0.774. The Decision Tree Model had an accuracy of 79.3%, AUC-PR 0.490, and AUC-ROC of 0.717. Conclusion: Boosted Trees and Decision Tree ML models performed comparably to the logistic regression model in predicting favorable versus unfavorable neurological outcomes following successful resuscitation from OHCA in advanced-age adults. Given their similar performance, logistic regression may be preferred for its simplicity and interpretability over alternative ML models. Future efforts to improve and ultimately externally validate the model are warranted.

Abstract Su705: Predictive Model and Factors for Out-of-Hospital Cardiac Arrest in Geriatric Patients with insomnia: A Longitudinal Study

Background: Previous studies have shown that geriatric insomnia can impact the quality of life and lead to various health risks and comorbidities, which may further result in out-of-hospital cardiac arrest (OHCA). Therefore, assessing the risk of OHCA in elderly patients with insomnia is crucial. Hypothesis: We hypothesized that a predictive model incorporating demographic, clinical, and treatment adherence factors can effectively estimate the risk of OHCA among geriatric patients with insomnia. Aim: This study aimed to develop and validate a predictive model for OHCA in geriatric patients with insomnia using a large-scale population-based analysis. Method: A retrospective cohort study was conducted using data from Taiwan's Health and Welfare Data Science Center spanning from 2011 to 2020. The study included elderly patients diagnosed with insomnia who received medication. Predictors of OHCA, such as age, sex, comorbidities, medication adherence, and psychotherapy, were analyzed using multivariable logistic regression. Model performance was assessed using the area under the receiver operating characteristic curve (AUROC) with both internal and temporal validation. Results: Among 438,147 eligible elderly patients with insomnia, 7,666 experienced OHCA. Using data from 2011 to 2018 for model development, the AUROC for predicting OHCA within 7 days, 30 days, and 90 days ranged from 0.749 to 0.762 in the training set and 0.768 to 0.784 in the validation set. Key predictors included older age, male sex, prior medical resource use, hemodialysis, pre-existing comorbidities, medication possession ratio, medication changes, and recent psychotherapy (Figure 1). The model was also tested in 2019 and 2020 dataset (began of the COVID-19 in Taiwan), respectively, demonstrating consistent predictive performance (Figure 2). Specifically, the 7-day model in 2019 showed a sensitivity of 0.776, specificity of 0.632, and accuracy of 0.632, with a positive likelihood ratio (PLR) of 2.11, and negative likelihood ratio (NLR) of 0.36. In 2020, it achieved a sensitivity of 0.739, specificity of 0.661, accuracy of 0.661, PLR of 2.18, NLR of 0.40. Conclusion: The study developed a reliable predictive model for OHCA in geriatric patients with insomnia, highlighting significant predictors such as age, comorbidities, medication adherence, and recent psychotherapy. This model can aid clinicians in identifying high-risk patients and improving preventative care strategies.

Sequential Diagnostic Approach Using FIB-4 and ELF for Predicting Advanced Fibrosis in Metabolic Dysfunction-Associated Steatotic Liver Disease

Background and Aims: Multiple non-invasive tests (NITs) for identifying advanced fibrosis in patients with non-alcoholic fatty liver disease (NAFLD) are available, but, due to the limitations of single NITs, the American Association for the Study of Liver Disease (AASLD) guidelines suggest a two-step strategy, combining the Fibrosis-4 Index (FIB-4) score with the Enhanced Liver Fibrosis (ELF) test to improve diagnostic accuracy and minimize unnecessary liver biopsies. However, few real-world studies have used such a sequential approach. We here evaluated the diagnostic accuracy of the ELF test in patients with recently established metabolic dysfunction-associated steatotic liver disease (MASLD) and assessed the clinical utility of applying a two-step strategy, including the ELF test following the FIB-4 score assessment, in patients with MASLD. Methods: We enrolled 153 patients diagnosed with MASLD who underwent liver biopsy at the Dong-A University Hospital between June 2018 and August 2023. The degree of fibrosis was determined based on liver biopsy results. Various NITs were used, including the Aminotransferase-to-Platelet Ratio Index (APRI), FIB-4 score, NAFLD Fibrosis score (NFS) and ELF test. The diagnostic efficacy of these NITs was evaluated based on the area under the receiver operating characteristic curve (AUROC). Additionally, the performance of each test was further examined both when applied individually and in a two-step approach, where FIB-4 was used followed by ELF testing. Key metrics such as sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy were used for this analysis. Results: Overall, 153 patients with MASLD (mean age: 46.62 years; 52.3% men; 28.1% with type 2 diabetes) were included. The performance of the NITs in identifying advanced fibrosis was as follows: the AUROC of the APRI, FIB-4, NFS, and ELF tests were 0.803 (95% confidence interval (CI), 0.713–0.863), 0.769 (95% CI, 0.694–0.833), 0.699 (95% CI, 0.528–0.796), and 0.829 (95% CI, 0.760–0.885), respectively. The combination of the FIB-4 score ≥ 1.30 and the ELF score ≥ 9.8 showed 67.86% sensitivity, 90.40% specificity, a PPV of 75.18%, an NPV of 86.78%, an accuracy of 83.64%, and an AUROC of 0.791 for predicting the diagnosis of advanced fibrosis. This approach excluded 28 patients (71.8%) from unnecessary liver biopsies. Conclusions: Our study demonstrated that ELF testing maintained diagnostic accuracy in assessing liver fibrosis in patients with MASLD in real-world practice. This test was used as a second step in the evaluation, reducing clinically unnecessary invasive liver biopsies and referrals to tertiary institutions. This approach allows assessment of MASLD severity in primary care settings without requiring additional equipment.