Dynamic predictions of postoperative complications from explainable, uncertainty-aware, and multi-task deep neural networks

Abstract

Accurate prediction of postoperative complications can inform shared decisions regarding prognosis, preoperative risk-reduction, and postoperative resource use. We hypothesized that multi-task deep learning models would outperform conventional machine learning models in predicting postoperative complications, and that integrating high-resolution intraoperative physiological time series would result in more granular and personalized health representations that would improve prognostication compared to preoperative predictions. In a longitudinal cohort study of 56,242 patients undergoing 67,481 inpatient surgical procedures at a university medical center, we compared deep learning models with random forests and XGBoost for predicting nine common postoperative complications using preoperative, intraoperative, and perioperative patient data. Our study indicated several significant results across experimental settings that suggest the utility of deep learning for capturing more precise representations of patient health for augmented surgical decision support. Multi-task learning improved efficiency by reducing computational resources without compromising predictive performance. Integrated gradients interpretability mechanisms identified potentially modifiable risk factors for each complication. Monte Carlo dropout methods provided a quantitative measure of prediction uncertainty that has the potential to enhance clinical trust. Multi-task learning, interpretability mechanisms, and uncertainty metrics demonstrated potential to facilitate effective clinical implementation.

Figure 1

Classification accuracy compared with baseline models. Shown are area under the receiver operating characteristic curve (AUROC) results for random forest and XGBoost models, individual deep learning models independently trained on each outcome, and a combined multi-task jointly trained on all outcomes, using only preoperative features (a), only intraoperative features (b), and both preoperative and intraoperative features (c). A comparison of multi-task deep learning results at three stages of prediction is shown in (d).

Figure 2

Temporal integrated gradients feature attributions for example patient experiencing prolonged mechanical ventilation. The multi-task deep learning model correctly predicted elevated risk of prolonged mechanical ventilation after integrating multivariate intraoperative time series. Physiological time series labeled by variable (left) and value range (right). Implementation of integrated gradients highlighted physiological patterns important for updated risk prediction, including a rapid increase in heart rate and ETCO2, fluctuations in PIP, and changes in SPO2. ETCO2, end-tidal carbon dioxide; PIP, peak inspiratory pressure; SPO2, blood oxygen saturation.

Figure 3

Data processing pipeline and deep learning model architecture. Patient-level input variables were split into static preoperative data and temporal intraoperative data. Preoperative variables were split into continuous, binary, and high-cardinality features and followed variable-specific preprocessing procedures. Deep learning model architecture utilized a data fusion design combining latent representations of high-frequency intraoperative data (from a bidirectional recurrent neural network) and static preoperative patient data (from fully connected layers) for eventual multi-task prediction of nine postoperative complications.