Accuracy and Clinical Utility of Clinical Predictive Models for Identifying Dizziness with Central Causes; A Retrospective Diagnostic Accuracy Study

Hepatoblastoma (HB) is a prevalent form of liver cancer in pediatric patients, characterized by an embryonal malignant tumor. In the current study, a clinical prediction model was developed; that can effectively assess the likelihood of a patient's survival with HB. Data from the Surveillance, Epidemiology, and End Results (SEER) database for cases of HB between 2010 and 2019 were used in this retrospective research. Information on clinicopathologic characteristics, therapeutic interventions, and survival outcomes were included in the data. The HB patients were randomly assigned to the training or validation cohort in a 7:3 ratio. Using univariate and multivariate Cox proportional hazards regression models, the prognostic indicators for overall survival (OS) and cancer-specific survival (CSS) were identified. The area under the receiver operating characteristic curve (AUC-ROC), calibration plots, and concordance index (C-index) were used to evaluate the accuracy and calibration of these models. The clinical utility of the models was examined using decision curve analysis (DCA). The multivariate Cox regression analysis revealed multiple autonomous prognostic determinants for the OS and CSS, including age, surgical interventions, and chemotherapy administration. Significantly, tumor size was found to be a strong predictor of OS. AUC values of 0.915, 0.846, and 0.847 for 1-, 3-, and 5-year OS, respectively, indicated that the nomogram-based models were highly accurate at predicting outcomes. Similarly, the AUC values for CSS were 0.871, 0.814, and 0.825. The C-index measurements, which quantify the discriminatory performance of the models, produced CSS values of 0.836 and OS values of 0.864. Furthermore, the calibration plots accurately represented the actual survival rates. Concurrently, the DCA had validated the clinical relevance of the nomogram-based models. The present study successfully developed and validated user-friendly nomogram-based models, allowing for accurate assessment of OS and CSS in pediatric HB patients. These tools enable personalized survival predictions, enhance risk stratification, and strengthen clinical decision-making for managing HB.

Limited data are available on the diagnostic accuracy of blood RNA biomarker signatures for extrapulmonary TB (EPTB). We addressed this question among people investigated for TB lymphadenitis and TB pericarditis, in Cape Town, South Africa. We enrolled 440 consecutive adults referred to a hospital for invasive sampling for presumptive TB lymphadenitis (n=300) or presumptive TB pericarditis (n=140). Samples from the site of disease underwent culture and/or molecular testing for Mycobacterium tuberculosis complex (Mtb). Discrimination of patients with and without TB defined by microbiology or cytology reference standards was evaluated using seven previously reported blood RNA signatures by area under the receiver-operating characteristic curve (AUROC) and sensitivity/specificity at predefined thresholds, benchmarked against blood C-reactive protein (CRP) and the World Health Organization (WHO) target product profile (TPP) for a TB triage test. Decision curve analysis (DCA) was used to evaluate the clinical utility of the best performing blood RNA signature and CRP. Data from 374 patients for whom results were available from at least one microbiological test from the site of disease, and blood CRP and RNA measurements, were included. Using microbiological results as the reference standard in the primary analysis (N=204 with TB), performance was similar across lymphadenitis and pericarditis patients. In the pooled analysis of both cohorts, all RNA signatures had comparable discrimination with AUROC point estimates ranging 0.77-0.82, superior to that of CRP (0.61, 95% confidence interval 0.56-0.67). The best performing signature (Roe3) achieved an AUROC of 0.82 (0.77-0.86). At a predefined threshold of 2 standard deviations (Z2) above the mean of a healthy reference control group, this signature achieved 78% (72-83%) sensitivity and 69% (62-75%) specificity. In this setting, DCA revealed that Roe3 offered greater net benefit than other approaches for services aiming to reduce the number needed to investigate with confirmatory testing to <4 to identify each case of TB. RNA biomarkers show better accuracy and clinical utility than CRP to trigger confirmatory TB testing in patients with TB lymphadenitis and TB pericarditis, but still fall short of the WHO TPP for TB triage tests. South African MRC, EDCTP2, NIH/NIAID, Wellcome Trust, NIHR, Royal College of Physicians London. Evidence before this study: Blood RNA biomarker signatures and CRP measurements have emerged as potential triage tests for TB, but evidence is mostly limited to their performance in pulmonary TB. Microbiological diagnosis of extrapulmonary TB (EPTB) is made challenging by the need for invasive sampling to obtain tissue from the site of disease. This is compounded by lower sensitivity of confirmatory molecular tests for EPTB compared to their performance in pulmonary disease. We performed a systematic review of diagnostic accuracy studies of blood RNA biomarkers or CRP measurements for EPTB, which could mitigate the need for site-of-disease sampling for the diagnosis of TB. We searched PubMed up to 1 st August 2023, using the following criteria: "extrapulmonary [title/abstract] AND tuberculosis [title/abstract] AND biomarker [title/abstract]". Although extrapulmonary TB was included in several studies, none focused specifically on EPTB or included an adequate number of EPTB cases to provide precise estimates of test accuracy. Added value of this study: To the best of our knowledge, we report the first diagnostic accuracy study of blood RNA biomarkers and CRP for TB among people with EPTB syndromes. We examined the performance of seven previously identified blood RNA biomarkers as triage tests for TB lymphadenitis and TB pericarditis compared to a microbiology reference standard among people referred to hospital for invasive sampling in a high TB and HIV prevalence setting. Multiple blood RNA biomarkers showed comparable diagnostic accuracy to that previously reported for pulmonary TB in both EPTB disease cohorts, irrespective of HIV status. All seven blood RNA biomarkers showed superior diagnostic accuracy to CRP for both lymphadenitis and pericarditis, but failed to meet the combined >90% sensitivity and >70% specificity recommended for a blood-based diagnostic triage test by WHO. Nonetheless, in decision curve analysis, an approach of using the best performing blood RNA biomarker to trigger confirmatory microbiological testing showed superior clinical utility in clinical services seeking to reduce the number needed to test (using invasive confirmatory testing) to less than 4 for each EPTB case detected. If acceptable to undertake invasive testing in more than 4 people for each true case detected, then a test-all approach will provide greater net benefit in this TB/HIV hyperendemic setting.Implications of all the available evidence: Blood RNA biomarkers show some potential as diagnostic triage tests for TB lymphadenitis and TB pericarditis, but do not provide the level of accuracy for blood-based triage tests recommended by WHO for community-based tests. CRP has inferior diagnostic accuracy to blood RNA biomarkers and cannot be recommended for diagnostic triage among people with EPTB syndromes referred for invasive sampling.

Outcome prediction by serum calprotectin in patients with COVID-19 in the emergency department

Many risk stratification models have been developed to improve the accuracy of sonographic evaluation of adnexal lesions. Although these models have been validated across various adult populations, their accuracy in the pediatric population remains underexplored. This study aimed to evaluate and compare the diagnostic accuracy of three sonography-based risk models in assessing pediatric adnexal lesions. This was a single-center retrospective diagnostic accuracy study conducted at a tertiary hospital in China, which included consecutive patients aged 17 years or younger with adnexal lesions managed either surgically or conservatively between January 2015 and January 2024. Each adnexal lesion was assessed using the Ovarian-Adnexal Reporting and Data System (O-RADS) ultrasound risk stratification and management tool, the International Ovarian Tumor Analysis (IOTA) Simple Rules (SR) followed by expert assessment for inconclusive cases in a two-step strategy and the Pediatric Risk of Malignancy Index (PRMI) model. The reference standard was either surgical intervention or conservative management with ultrasound surveillance. The accuracy of the different sonography-based risk models was evaluated using the area under the receiver-operating-characteristics curve (AUC), sensitivity, specificity and positive and negative predictive values. The clinical utility of each model was evaluated using decision-curve analysis. This study included 364 patients with a total of 375 adnexal lesions. The median age of the patients was 15.0 (interquartile range, 14.0-17.0) years, with a malignancy rate of 2.1% (8/375). The AUC for the O-RADS tool, IOTA two-step strategy and the PRMI model were 0.989 (95% CI, 0.980-0.999), 0.992 (95% CI, 0.985-0.998) and 0.806 (95% CI, 0.626-0.985), respectively. The IOTA two-step strategy demonstrated the highest accuracy and provided the greatest net benefits at malignancy risk thresholds between 1% and 30% compared with the other two models. In this retrospective pediatric cohort, the O-RADS ultrasound model and the IOTA-SR followed by expert assessment for inconclusive cases in a two-step strategy exhibited high diagnostic accuracy in characterizing adnexal lesions, comparable to that observed in adults. These results support the applicability of both of these risk-stratification methods to the pediatric population. The PRMI scoring system showed lower discriminative capacity compared with the other models; therefore, it is not recommended to serve as a standalone diagnostic tool. © 2025 International Society of Ultrasound in Obstetrics and Gynecology.

This study aims to evaluate the predictive utility of multi-slice spiral computed tomography (MSCT) in assessing the response to neoadjuvant chemotherapy (NAC) in patients with muscle-invasive bladder cancer (MIBC). The clinical data of patients with MIBC who received treatment at our institution between January 2022 and January 2024 were retrospectively analysed. The patients were assigned to an observation group or a control group according to whether pathological downstaging was achieved following NAC with a gemcitabine and cisplatin regimen. A multivariate logistic regression analysis was conducted to develop a clinical prediction model, and a receiver operating characteristic (ROC) curve was generated to assess its diagnostic performance. For the evaluation of the model's stability, all patients were randomly assigned to either a training set or a validation set in a 7:3 ratio. A total of 86 patients were included in the study, among whom 46 achieved pathological downstaging with a downstaging rate of 53.49%. Multivariate logistic regression analysis identified blood flow (BF; odds ratio (OR), 95% confidence interval (CI): 0.63 (0.48-0.81)) and blood volume (BV; OR, 95% CI: 0.04 (0.00-0.46)) measured by MSCT as independent predictors of NAC efficacy (p < 0.05). In the training set, the area under the ROC curve (AUC) was 0.94 (95% CI: 0.89-0.99), with a specificity of 0.86 and a sensitivity of 0.87. In the validation set, the AUC was 0.98 (95% CI: 0.93-1.00), with a specificity of 0.84 and a sensitivity of 0.88. The Hosmer-Lemeshow goodness-of-fit test indicated good calibration, with p-values of 0.504 and 0.915 in the training and validation sets, respectively. The calibration curve demonstrated strong concordance with the ideal 45° reference line, indicating close alignment between predicted and observed probabilities. Decision curve analysis further confirmed the clinical utility of the nomogram, showing a favourable net benefit across a wide range of threshold probabilities. Pre-chemotherapy BF and BV parameters measured by MSCT were identified as significant predictors of NAC efficacy in patients with MIBC. The predictive model constructed using these parameters demonstrated strong performance in forecasting pathological downstaging following NAC.

BackgroundAcute kidney disease (AKD) affects more than half of critically ill elderly patients with acute kidney injury (AKI), which leads to worse short-term outcomes.ObjectiveWe aimed to establish 2 machine learning models to predict the risk and prognosis of AKD in the elderly and to deploy the models as online apps.MethodsData on elderly patients with AKI (n=3542) and AKD (n=2661) from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database were used to develop 2 models for predicting the AKD risk and in-hospital mortality, respectively. Data collected from Xiangya Hospital of Central South University were for external validation. A bootstrap method was used for internal validation to obtain relatively stable results. We extracted the indicators within 24 hours of the first diagnosis of AKI and the fluctuation range of some indicators, namely delta (day 3 after AKI minus day 1), as features. Six machine learning algorithms were used for modeling; the area under the receiver operating characteristic curve (AUROC), decision curve analysis, and calibration curve for evaluating; Shapley additive explanation (SHAP) analysis for visually interpreting; and the Heroku platform for deploying the best-performing models as web-based apps.ResultsFor the model of predicting the risk of AKD in elderly patients with AKI during hospitalization, the Light Gradient Boosting Machine (LightGBM) showed the best overall performance in the training (AUROC=0.844, 95% CI 0.831-0.857), internal validation (AUROC=0.853, 95% CI 0.841-0.865), and external (AUROC=0.755, 95% CI 0.699–0.811) cohorts. In addition, LightGBM performed well for the AKD prognostic prediction in the training (AUROC=0.861, 95% CI 0.843-0.878), internal validation (AUROC=0.868, 95% CI 0.851-0.885), and external (AUROC=0.746, 95% CI 0.673-0.820) cohorts. The models deployed as online prediction apps allowed users to predict and provide feedback to submit new data for model iteration. In the importance ranking and correlation visualization of the model’s top 10 influencing factors conducted based on the SHAP value, partial dependence plots revealed the optimal cutoff of some interventionable indicators. The top 5 factors predicting the risk of AKD were creatinine on day 3, sepsis, delta blood urea nitrogen (BUN), diastolic blood pressure (DBP), and heart rate, while the top 5 factors determining in-hospital mortality were age, BUN on day 1, vasopressor use, BUN on day 3, and partial pressure of carbon dioxide (PaCO2).ConclusionsWe developed and validated 2 online apps for predicting the risk of AKD and its prognostic mortality in elderly patients, respectively. The top 10 factors that influenced the AKD risk and mortality during hospitalization were identified and explained visually, which might provide useful applications for intelligent management and suggestions for future prospective research.

Background: Beyond prostate-specific antigen (PSA), other biomarkers for prostate cancer (PCa) detection are available and need to be evaluated for clinical routine. Objective: The aim of the study was to evaluate the Prostate Health Index (PHI) density (PHID) in comparison with PHI in a large Caucasian group >1,000 men. Methods: PHID values were used from available patient data with PSA, free PSA, and [−2]pro­PSA and prostate volume from 3 former surveys from 2002 to 2014. Those 1,446 patients from a single-center cohort included 701 men with PCa and 745 with no PCa. All patients received initial or repeat biopsies. The diagnostic accuracy was evaluated by receiver operating characteristic (ROC) curves comparing area under the ROC curves (AUCs), precision-recall approach, and decision curve analysis (DCA). Results: PHID medians differed almost 2-fold between PCa (1.12) and no PCa (0.62) in comparison to PHI (48.6 vs. 33; p always <0.0001). However, PHID and PHI were equal regarding the AUC (0.737 vs. 0.749; p = 0.226), and the curves of the precision-recall analysis also overlapped in the sensitivity range between 70 and 100%. DCA had a maximum net benefit of only ∼5% for PHID versus PHI between 45 and 55% threshold probability. Contrary, in the 689 men with a prostate volume ≤40 cm³, PHI (AUC 0.732) showed a significant larger AUC than PHID (AUC 0.69, p = 0.014). Conclusions: Based on DCA, PHID had only a small advantage in comparison with PHI alone, while ROC analysis and precision-recall analysis showed similar results. In smaller prostates, PHI even outperformed PHID. The increment for PHID in this large Caucasian cohort is too small to justify a routine clinical use.

PurposeTo develop and validate a combined model incorporating conventional clinical and imaging characteristics and radiomics signatures based on head and neck computed tomography angiography (CTA) to assess plaque vulnerability.MethodsWe retrospectively analyzed 167 patients with carotid atherosclerosis who underwent head and neck CTA and brain magnetic resonance imaging (MRI) within 1 month. Clinical risk factors and conventional plaque characteristics were evaluated, and radiomic features were extracted from the carotid plaques. The conventional, radiomics and combined models were developed using fivefold cross-validation. Model performance was evaluated using receiver operating characteristic (ROC), calibration, and decision curve analyses.ResultsPatients were divided into symptomatic (n = 70) and asymptomatic (n = 97) groups based on MRI results. Homocysteine (odds ratio, OR 1.057; 95% confidence interval, CI 1.001–1.116), plaque ulceration (OR 6.106; 95% CI 1.933–19.287), and carotid rim sign (OR 3.285; 95% CI 1.203–8.969) were independently associated with symptomatic status and were used to construct the conventional model and s radiomic features were retained to establish the radiomics model. Radiomics scores incorporated with conventional characteristics were used to establish the combined model. The area under the ROC curve (AUC) of the combined model was 0.832, which outperformed the conventional (AUC = 0.767) and radiomics (AUC = 0.797) models. Calibration and decision curves analysis showed that the combined model was clinically useful.ConclusionRadiomics signatures of carotid plaque on CTA can well predict plaque vulnerability, which may provide additional value to identify high-risk patients and improve outcomes.

Background: The primary objective of this study is to employ machine learning (ML) algorithms to develop predictive models for renal function recovery in critically ill patients undergoing continuous renal replacement therapy (CRRT) due to acute kidney injury (AKI). Methods: Data were retrospectively collected from patients with AKI who underwent CRRT during their first Intensive Care Unit (ICU) admission from the Medical Information Mart for Intensive Care-IV database and at Huzhou Central Hospital. The evaluation of model performance was conducted utilizing metrics including the area under the receiver operating characteristic curve (AUROC), calibration curves and decision curve analysis (DCA). Results: A total of 1078 AKI patients undergoing CRRT were included, with a renal function recovery rate of 18.18% (n = 196). Six features that significantly affect renal function recover rate were identified through least absolute shrinkage and selection operator (LASSO) regression with cross-validation. The Lasso-LR model was chosen as the definitive clinical risk prediction model for this study and served as the basis for constructing the nomogram. Its AUROCs in the training and external validation cohorts were 0.774 (95% CI: 0.735 ∼ 0.814) and 0.748 (95% CI: 0.685 ∼ 0.812), respectively. The calibration curve and ideal curve fit were generally found to be suboptimal. Simultaneously, the DCA curve suggested that the nomogram has clinical value. Conclusions: Among the ML models developed using data from the two datasets to predict renal function recovery in critically ill AKI patients undergoing CRRT, the Lasso-LR model demonstrated the best performance. It can serve as a valuable tool for more efficient clinical disease management and prognostic evaluation.

BackgroundWith the growing numbers of nonagenarians and centenarians admitted to intensive care, clinicians face complex decisions regarding care appropriateness and intensity. Prognostic tools such as APACHE III, SOFA, and the Australia and New Zealand Risk of Death (ANZROD) model are widely used, yet their validity in very old patients remains uncertain given the influence of frailty, multimorbidity, and age-related physiological changes. In this study, we evaluated and compared the performance of APACHE III, its Risk of Death score, SOFA, and ANZROD in a large binational cohort of ICU-admitted nonagenarians and centenarians, assessing discrimination, calibration, and predictive utility for short- and long-term mortality using AUROC, standardized mortality ratios, and decision curve analysis.MethodsWe conducted a multicenter retrospective cohort study of 40,910 patients aged 90 years and older who were admitted to ICUs across Australia and New Zealand from 2010 to 2024. Discrimination was assessed using time-dependent area under the receiver operating characteristic curve (AUROC). Calibration was evaluated with standardized mortality ratios (SMRs), and clinical utility was examined using decision curve analysis (DCA) across ICU, hospital, 1-month, 6-month, and 1-year mortality outcomes.ResultsANZROD consistently showed the highest discriminative ability, with AUROCs of 0.870 for ICU mortality and 0.768 for 1-month mortality. APACHEIII ROD and APACHE III followed, while SOFA had the lowest AUROCs across all timepoints. SMR analysis revealed overestimation of short term and underestimation of long term mortality by all models. ANZROD had the best calibration for hospital mortality. DCA showed ANZROD provided greater net benefit than the other scores at clinically relevant thresholds.ConclusionsIn this large binational cohort of over 40,000 nonagenarian and centenarian ICU patients, ANZROD outperformed APACHE III and SOFA in short-term mortality prediction, but all models showed poor calibration and declining discrimination over longer horizons. Current scores remain useful for benchmarking but are limited for individual decision-making, underscoring the need for age-specific prognostic tools that integrate frailty, function, and geriatric priorities.Trial registrationACTRN12625000775415 (23/07/2025).Supplementary InformationThe online version contains supplementary material available at 10.1186/s44158-026-00339-1.

The eighth version of the American Joint Committee on Cancer (8th AJCC) system for anaplastic thyroid carcinoma(ATC) added lymph node (LN) metastasis as the staging element. This study aimed to explore the association between LN status and ATC's prognosis, identify the optimal LN index and establish a novel prognostic model. Data of 199 ATC patients after surgery were collected from the Surveillance, Epidemiologyand End Results (SEER) database, then randomly divided into training and validation cohorts. We compared the prognostic value of AJCC N status, number of positive LN (PLNN), ratio of LN (LNR)and log odds of positive LN (LODDS). We conducted univariate and multivariate Cox analyses to determine the independent prognostic factors for ATC, and constructed a novel prognostic model. The concordance index(C-index), area under the receiver-operating characteristiccurve (AUC), calibration curvesand decision curve analysis (DCA) were used to assess the nomogram's predictive performance. LODDS showed the highest accuracy among four LN systems to predict overall survival (OS) for ATC. In the training cohort, the C-index of the LODDS-based nomogram was 0.738. The AUCs were 0.813, 0.850and 0.869 for predicting 1-, 2-and 3-year OS, respectively. The calibration plots and DCA indicated the great clinical applicability of the model. The above results were verified in the validation cohort. LODDS showed better predictive performance than other LN schemes in ATC. The LODDS-incorporated nomogram has the potential to more precisely predictthe prognosis for ATC patients than the AJCC system.

To establish a prognostic Nomogram model for predicting the risk of early death in polytrauma patients. Data extracted from a polytrauma study on Dryad, an open access database, was selected for secondary analysis. Patients from 18 to 65 years old with polytrauma in the original data were included. All patients with missing variables, such as blood lactic acid (Lac), Glasgow coma score (GCS) and injury severity score (ISS) at admission, were excluded. The differences of gender, age, Lac, ISS and GCS scores between the patients who died within 72 hours and those who survived were analyzed. The risk factors for 72-hour death were analyzed by Logistic regression, and the Nomogram prediction model was established using R software. The receiver operating characteristic (ROC) curve was used to evaluate the predictive ability of the model, and the Bootstrap method was used for internal verification by repeating sample for 1 000 times. Decision curve (DCA) was applied to analyze the clinical practical value of the model. A total of 2 315 polytrauma patients were included. Logistic regression analysis showed that Lac, GCS score and age > 55 years old were the risk factors for early death in polytrauma patients [Lac: odds ratio (OR) = 1.36, 95% confidence interval (95%CI) was 1.29-1.42, P < 0.001; GCS score: OR = 0.76, 95%CI was 0.73-0.79, P < 0.001; age > 55 years old: OR = 1.92, 95%CI was 1.37-2.66, P < 0.001]. The prediction model was established by using the above risk factors and displayed by Nomogram. ROC curve analysis showed that the area under the ROC curve (AUC) of Nomogram model to predict the risk of death within 72 hours was 0.858, and the predictive ability of Nomogram model was significantly higher than that of Lac (AUC = 0.743), GCS score (AUC = 0.774) and ISS score (AUC = 0.699), all P < 0.05. The model calibration chart showed that the predicting probability was consistent with the actual occurrence probability, and the DCA showed that Nomogram model presented excellent clinical value in predicting the 72-hour death risk for polytrauma patients. The prognostic Nomogram model presents significantly predictive value for the risk of death within 72 hours in polytrauma patients. Prognostic Nomogram model could offer individualized, visualized and graphical prediction pattern, and provide physicians with practical diagnostic tool for triage system and management of polytrauma according to precision medicine.

Background and AimsThe 12-lead electrocardiogram (ECG) remains a cornerstone of cardiac diagnostics, yet existing artificial intelligence (AI) solutions for automated interpretation often lack generalizability, remain closed source, and are primarily trained using supervised learning (SL), which requires extensive labelled datasets and may limit adaptability across diverse clinical settings. Self-supervised learning (SSL) can potentially overcome these limitations by learning robust representations from unlabelled data. To address these challenges, this study developed and compared two open-source foundational ECG models: DeepECG-SL, a supervised multilabel ECG model, and DeepECG-SSL, a self-supervised model.MethodsBoth models were trained on over 1 million ECGs using a standardized preprocessing pipeline and automated free-text extraction from ECG reports to predict 77 cardiac conditions. DeepECG-SSL leveraged unlabelled data through self-supervised contrastive learning and masked lead modelling before fine-tuning for downstream tasks, while DeepECG-SL was trained directly on labelled diagnostic data in an end-to-end fashion. Performance was evaluated across seven private, multilingual healthcare systems and four public ECG repositories, with assessment of fairness by age and sex, and investigation of privacy vulnerabilities as well as memory and compute requirements.ResultsDeepECG-SSL achieved micro-averaged area under the receiver operating characteristic curves (AUROCs) across all 77 cardiac conditions for ECG interpretation of 0.990 [95% confidence interval (CI): 0.990, 0.990] on the internal dataset (MHI-ds), 0.981 (95% CI: 0.981, 0.981) on external public datasets (UKB, CLSA, MIMIC-IV and PTB), and 0.983 (95% CI: 0.983, 0.983) on external private datasets (UW, UCSF, JGH, NYP, MGH, CSH and CHUM), while DeepECG-SL demonstrated AUROCs of 0.992 (95% CI: 0.992, 0.992), 0.980 (95% CI: 0.980, 0.980), and 0.983 (95% CI: 0.983, 0.984), respectively. Fairness analyses revealed minimal disparities (true-positive rate and false-positive rate difference <0.1) across age and sex groups for both models. DeepECG-SSL demonstrated superior performance on limited-data digital biomarker tasks, with the largest improvements in long QT syndrome (LQTS) genotype classification (AUROC 0.931 vs 0.850, P = .026, n = 127 ECGs) and 5 year atrial fibrillation risk prediction (AUROC 0.742 vs 0.734, P < 0.001, n = 132 050 ECGs), while achieving superior performance in left ventricular ejection fraction ≤40% classification (AUROC 0.926 vs 0.917, P < 0.001, n = 25 252 ECGs) and comparable performance in LQTS detection (AUROC 0.767 vs 0.735, P = 0.117, n = 934 ECGs).ConclusionsThis study establishes SSL as a promising paradigm for ECG analysis, particularly in settings with limited annotated data, enhancing accessibility, generalizability, and fairness in AI-driven cardiac diagnostics. By releasing model weights, preprocessing tools, and validation code, this work aims to support robust, data-efficient AI diagnostics across diverse clinical environments and questions.

Introduction: Early and accurate diagnosis of transthyretin amyloid cardiomyopathy (ATTR-CM) is known to improve prognosis. However, it is sometimes challenging to distinguish ATTR-CM from other cardiac conditions that cause left ventricular hypertrophy (LVH), such as hypertrophic cardiomyopathy (HCM), hypertensive LVH (h-LVH), and aortic stenosis (AS). The Mayo ATTR-CM score is commonly used to distinguish ATTR-CM from these mimickers. Herein, we constructed a machine learning (ML) model and compared its performance to the Mayo score. Hypothesis: A ML model based on readily available clinical variables will outperform the Mayo score in distinguishing ATTR-CM from other cardiac conditions that cause LVH. Methods: We performed a multicenter case-control study of 1,605 patients, including 303 ATTR-CM cases and 1,302 controls – i.e., 935 with HCM, 331 with h-LVH, and 36 with AS. We allocated the former two-thirds of each disease group to the training set and the latter one-third to the prospective test set for validation. We developed a main ML model in the training set using XGBoost with 18 clinical variables including demographics, comorbidities, laboratory values, and echocardiographic data. We examined its performance in the test set using the area under the receiver-operating-characteristic curve (AUC) and decision curve analysis (DCA). In addition, we developed a simplified ML model using only the top 5 variables identified by feature importance and tested its performance. Results: The top 5 most important variables were age, LV diameter in systole, LV posterior wall thickness in diastole, LV ejection fraction, and left atrial diameter ( Figure 1 ). The main ML model (AUC: 0.96, 95% CI: 0.93–0.98) outperformed the Mayo score (AUC: 0.84, 95% CI: 0.79–0.89, DeLong’s P &lt;0.0001; Figure 2 ) in the test set. DCA demonstrated that the ML model provided greater net benefit across all threshold probabilities ( Figure 3 ). The simplified 5-variable model also outperformed the Mayo score (AUC: 0.94, 95% CI: 0.91–0.97, DeLong’s P &lt;0.0001). Conclusions: In this multicenter study, we developed ML models that effectively distinguish ATTR-CM from its mimickers that cause LVH using readily available clinical variables. Given the simplicity and accuracy of our models, the present study serves as an important initial step toward the development of a robust, data-driven tool that could be readily automated and integrated into electronic medical record system to screen for ATTR-CM.

The postoperative conversion of a protective stoma to a permanent stoma (PS) during sphincter-preserving surgery for rectal cancer can significantly impact the patient’s quality of life. No reliable tools exist to predict the likelihood of a protective stoma becoming permanent in clinical practice. This study aimed to develop an online, dynamic nomogram for predicting PS using multicenter, large-sample data. A total of 588 patients were enrolled following screening. Predictive variables were selected using the univariate analysis. A prediction model was then created using multivariate logistic regression and presented as an online nomogram. The model’s performance was evaluated using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). External validation was performed to assess the model’s accuracy within the validation cohort. Of the 588 patients, 81 (13.7%) had a PS. 507 paitients from our and another hospital were randomly assigned to a training cohort and internal testing cohort (7:3), and 81 patients from another hospital served as the external validation cohort. Critical predictive factors in the nomogram included ASA score ≥ 3, radiotherapy, diabetes and N་stage. The area under the ROC curve (AUC) was 0.784 [95% Confidence Interval (CI) 0.708–0.860] for the training cohort, 0.785 (95% CI 0.656–0.913) for the testing cohort, and 0.737 (95% CI 0.604–0.870) for the external validation cohort. The calibration curve demonstrated a strong correlation between predicted and observed outcomes, and decision curve analysis (DCA) confirmed the nomogram’s clinical utility in predicting PS following sphincter-preserving surgery for rectal cancer. This study developed an online nomogram to identify high-risk factors for PS following rectal sphincter-preserving surgery, using multicenter data (modeling and external validation) from three hospitals in China. This tool supports clinicians in risk assessment. For high-risk patients, health education should be provided, and it may be advisable to offer the option of a colostomy.