BMC anesthesiology · 2026
Early prediction of intraoperative hypotension: development and validation of the HypoBridCast hybrid deep learning model.
СтатьяВалидационное исследованиеНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Wang Youran, Zeng Xian, Chen Lingjue, Li Bin, Shen Kaiqiang, Xu Jun, Sun Jie
Аннотация
BACKGROUND: Intraoperative hypotension (IOH) is a frequent and clinically important complication associated with adverse postoperative outcomes. Early prediction may facilitate timely intervention, although existing models have limitations in integrating multimodal physiological data and patient-specific characteristics.
METHODS: We developed Hypotension Hybrid Forecast (HypoBridCast), a hybrid architecture integrating 1D Convolutional Layer (Conv1D) and Transformer (TF) modules. The model integrates intraoperative waveform data (arterial pressure, electrocardiogram, photoplethysmogram, and capnography) with preoperative clinical variables (including demographic characteristics, medical history, laboratory results, and surgical and anesthetic variables). Data from the VitalDB database (n = 3,369) were used for model development and internal evaluation, and an independent cohort from Zhongda Hospital (n = 437) was used for external evaluation. Model performance for predicting IOH (MAP ≤ 65 mmHg for ≥ 1 min) 5, 10, and 15 min in advance was assessed using AUROC, AUPRC, and calibration metrics.
RESULTS: HypoBridCast achieved strong discriminative performance across prediction horizons in both internal and external evaluations (e.g., internal 5-min AUROC 0.9442 [95% CI 0.9427-0.9456] and AUPRC 0.9387 [0.9376-0.9398]). Compared with Mono-ART models, performance was improved, whereas differences versus multi-channel waveform models were modest. The addition of preoperative variables provided limited and inconsistent gains across datasets. In contrast, performance gains over simple MAP-based baseline models were more pronounced. Calibration was acceptable overall, with some reduction observed in the external cohort, particularly at longer prediction horizons.
CONCLUSIONS: The proposed hybrid deep learning framework achieved strong performance for short-term prediction of intraoperative hypotension using routinely collected clinical data. Multimodal integration and preoperative variables provide incremental improvements. Further work is needed to improve generalizability and calibration before clinical deployment.
TRIAL REGISTRATION: ChiCTR2500099041.
METHODS: We developed Hypotension Hybrid Forecast (HypoBridCast), a hybrid architecture integrating 1D Convolutional Layer (Conv1D) and Transformer (TF) modules. The model integrates intraoperative waveform data (arterial pressure, electrocardiogram, photoplethysmogram, and capnography) with preoperative clinical variables (including demographic characteristics, medical history, laboratory results, and surgical and anesthetic variables). Data from the VitalDB database (n = 3,369) were used for model development and internal evaluation, and an independent cohort from Zhongda Hospital (n = 437) was used for external evaluation. Model performance for predicting IOH (MAP ≤ 65 mmHg for ≥ 1 min) 5, 10, and 15 min in advance was assessed using AUROC, AUPRC, and calibration metrics.
RESULTS: HypoBridCast achieved strong discriminative performance across prediction horizons in both internal and external evaluations (e.g., internal 5-min AUROC 0.9442 [95% CI 0.9427-0.9456] and AUPRC 0.9387 [0.9376-0.9398]). Compared with Mono-ART models, performance was improved, whereas differences versus multi-channel waveform models were modest. The addition of preoperative variables provided limited and inconsistent gains across datasets. In contrast, performance gains over simple MAP-based baseline models were more pronounced. Calibration was acceptable overall, with some reduction observed in the external cohort, particularly at longer prediction horizons.
CONCLUSIONS: The proposed hybrid deep learning framework achieved strong performance for short-term prediction of intraoperative hypotension using routinely collected clinical data. Multimodal integration and preoperative variables provide incremental improvements. Further work is needed to improve generalizability and calibration before clinical deployment.
TRIAL REGISTRATION: ChiCTR2500099041.