An official website of the United States government
To test the hypothesis that accuracy, discrimination, and precision in predicting postoperative complications improve when using both preoperative and intraoperative data input features versus preoperative data alone. Models that predict postoperative complications often ignore important intraoperative physiological changes. Incorporation of intraoperative physiological data may improve model performance. This retrospective cohort analysis included 52,529 inpatient surgeries at a single institution during a 5 year period. Random forest machine learning models in the validated MySurgeryRisk platform made patient-level predictions for three postoperative complications and mortality during hospital admission using electronic health record data and patient neighborhood characteristics. For each outcome, one model trained with preoperative data alone and one model trained with both preoperative and intraoperative data. Models were compared by accuracy, discrimination (expressed as AUROC), precision (expressed as AUPRC), and reclassification indices (NRI). Machine learning models incorporating both preoperative and intraoperative data had greater accuracy, discrimination, and precision than models using preoperative data alone for predicting all three postoperative complications (intensive care unit length of stay >48 hours, mechanical ventilation >48 hours, and neurological complications including delirium) and in-hospital mortality (accuracy: 88% vs. 77%, AUROC: 0.93 vs. 0.87, AUPRC: 0.21 vs. 0.15). Overall reclassification improvement was 2.9-10.0% for complications and 11.2% for in-hospital mortality. Incorporating both preoperative and intraoperative data significantly increased accuracy, discrimination, and precision for machine learning models predicting postoperative complications.
more »
« less
Continuous Intraoperative Data Analysis Using Machine Learning Reveals Multiple Parameters to Predict Post-CABG Renal Failure
The purpose of this study is to utilize machine learning techniques to identify intraoperative parameters that contribute significantly to the development of postoperative renal failure following CABG and predict postoperative renal failure based on these parameters. Continuous intraoperative data were gathered retrospectively from the anaesthesia record and included hemodynamic information such as heart rate, arterial blood pressure, central venous pressure, pulmonary artery pressure, as well as additional information such as ventilator settings, temperature, and medication or fluid administration. Multiple machine learning algorithms were tested with this dataset using 10 fold cross validation with stratified folds and their classification performance was measured using area under the receiver operating characteristic curves (ROC AUC). Continuous intraoperative data gathered from patients undergoing CABG revealed potential targets for early, intraoperative intervention to prevent the development of postoperative renal failure.
more »
« less
- Award ID(s):
- 1950811
- PAR ID:
- 10339368
- Date Published:
- Journal Name:
- The Society of Thoracic Surgeons Annual Meeting
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
BackgroundAlthough conventional prediction models for surgical patients often ignore intraoperative time-series data, deep learning approaches are well-suited to incorporate time-varying and non-linear data with complex interactions. Blood lactate concentration is one important clinical marker that can reflect the adequacy of systemic perfusion during cardiac surgery. During cardiac surgery and cardiopulmonary bypass, minute-level data is available on key parameters that affect perfusion. The goal of this study was to use machine learning and deep learning approaches to predict maximum blood lactate concentrations after cardiac surgery. We hypothesized that models using minute-level intraoperative data as inputs would have the best predictive performance. MethodsAdults who underwent cardiac surgery with cardiopulmonary bypass were eligible. The primary outcome was maximum lactate concentration within 24 h postoperatively. We considered three classes of predictive models, using the performance metric of mean absolute error across testing folds: (1) static models using baseline preoperative variables, (2) augmentation of the static models with intraoperative statistics, and (3) a dynamic approach that integrates preoperative variables with intraoperative time series data. Results2,187 patients were included. For three models that only used baseline characteristics (linear regression, random forest, artificial neural network) to predict maximum postoperative lactate concentration, the prediction error ranged from a median of 2.52 mmol/L (IQR 2.46, 2.56) to 2.58 mmol/L (IQR 2.54, 2.60). The inclusion of intraoperative summary statistics (including intraoperative lactate concentration) improved model performance, with the prediction error ranging from a median of 2.09 mmol/L (IQR 2.04, 2.14) to 2.12 mmol/L (IQR 2.06, 2.16). For two modelling approaches (recurrent neural network, transformer) that can utilize intraoperative time-series data, the lowest prediction error was obtained with a range of median 1.96 mmol/L (IQR 1.87, 2.05) to 1.97 mmol/L (IQR 1.92, 2.05). Intraoperative lactate concentration was the most important predictive feature based on Shapley additive values. Anemia and weight were also important predictors, but there was heterogeneity in the importance of other features. ConclusionPostoperative lactate concentrations can be predicted using baseline and intraoperative data with moderate accuracy. These results reflect the value of intraoperative data in the prediction of clinically relevant outcomes to guide perioperative management.more » « less
-
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.more » « less
-
Abstract Background Few interventions are known to reduce the incidence of respiratory failure that occurs following elective surgery (postoperative respiratory failure; PRF). We previously reported risk factors associated with PRF that occurs within the first 5 days after elective surgery (early PRF; E-PRF); however, PRF that occurs six or more days after elective surgery (late PRF; L-PRF) likely represents a different entity. We hypothesized that L-PRF would be associated with worse outcomes and different risk factors than E-PRF. Methods This was a retrospective matched case-control study of 59,073 consecutive adult patients admitted for elective non-cardiac and non-pulmonary surgical procedures at one of five University of California academic medical centers between October 2012 and September 2015. We identified patients with L-PRF, confirmed by surgeon and intensivist subject matter expert review, and matched them 1:1 to patients who did not develop PRF (No-PRF) based on hospital, age, and surgical procedure. We then analyzed risk factors and outcomes associated with L-PRF compared to E-PRF and No-PRF. Results Among 95 patients with L-PRF, 50.5% were female, 71.6% white, 27.4% Hispanic, and 53.7% Medicare recipients; the median age was 63 years (IQR 56, 70). Compared to 95 matched patients with No-PRF and 319 patients who developed E-PRF, L-PRF was associated with higher morbidity and mortality, longer hospital and intensive care unit length of stay, and increased costs. Compared to No-PRF, factors associated with L-PRF included: preexisiting neurologic disease (OR 4.36, 95% CI 1.81–10.46), anesthesia duration per hour (OR 1.22, 95% CI 1.04–1.44), and maximum intraoperative peak inspiratory pressure per cm H 2 0 (OR 1.14, 95% CI 1.06–1.22). Conclusions We identified that pre-existing neurologic disease, longer duration of anesthesia, and greater maximum intraoperative peak inspiratory pressures were associated with respiratory failure that developed six or more days after elective surgery in adult patients (L-PRF). Interventions targeting these factors may be worthy of future evaluation.more » « less
-
Abstract Levees are built to safeguard human lives, essential infrastructure, and farmland. However, failure of levees can have catastrophic impacts due to a fast rate of inundation in areas protected by levees. Earthen levees are prone to failure due to excessive moisture content that reduces the shear strength of the soil. The use of levee monitoring systems has demonstrated the ability to reduce the likelihood of failure by creating maps that depict the saturation levels of the surface of the levee, both in terms of space and time. By utilizing extensive sensor networks to continuously monitor these geo-infrastructure systems, the structural deterioration attributed to changing climate can be studied. Measuring environmental parameters surrounding such structures provides insight into the potential stressors that cause structural failure. Steps can then be taken to mitigate those effects on the levees and maintain structural integrity. However, the massive scale of levees makes it difficult to monitor with conventional wired sensors. This paper presents a preliminary investigation into the development and validation of UAV-deployable smart sensing spikes for soil conductivity levels in levees, which is a measurement modality for determining soil saturation levels. For this work, Gaussian process regression (also known as kriging) is used to model the soil saturation levels between sensing spikes obtaining a continuous moisture map of the levees. The expanded data is then categorized using a clustering-based machine learning approach with conductivity data from sensing spikes as model inputs. The machine learning model output is sorted into three categories: dry, partially saturated, and saturated soil. The findings of a laboratory study are presented, and the implications of the raw and expanded data are discussed. This work will aid in predicting potential levee failure risks and maintenance requirements based on the analysis of the soil conditions using a network of smart sensing spikes.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
