A transformer model for cause-specific hazard prediction.

BACKGROUD: Modelling discrete-time cause-specific hazards in the presence of competing events and non-proportional hazards is a challenging task in many domains. Survival analysis in longitudinal cohorts often requires such models; notably when the data is gathered at discrete points in time and the predicted events display complex dynamics. Current models often rely on strong assumptions of proportional hazards, that is rarely verified in practice; or do not handle sequential data in a meaningful way. This study proposes a Transformer architecture for the prediction of cause-specific hazards in discrete-time competing risks. Contrary to Multilayer perceptrons that were already used for this task (DeepHit), the Transformer architecture is especially suited for handling complex relationships in sequential data, having displayed state-of-the-art performance in numerous tasks with few underlying assumptions on the task at hand. RESULTS: Using synthetic datasets of 2000-50,000 patients, we showed that our Transformer model surpassed the CoxPH, PyDTS, and DeepHit models for the prediction of cause-specific hazard, especially when the proportional assumption did not hold. The error along simulated time outlined the ability of our model to anticipate the evolution of cause-specific hazards at later time steps where few events are observed. It was also superior to current models for prediction of dementia and other psychiatric conditions in the English longitudinal study of ageing cohort using the integrated brier score and the time-dependent concordance index. We also displayed the explainability of our model's prediction using the integrated gradients method. CONCLUSIONS: Our model provided state-of-the-art prediction of cause-specific hazards, without adopting prior parametric assumptions on the hazard rates. It outperformed other models in non-proportional hazards settings for both the synthetic dataset and the longitudinal cohort study. We also observed that basic models such as CoxPH were more suited to extremely simple settings than deep learning models. Our model is therefore especially suited for survival analysis on longitudinal cohorts with complex dynamics of the covariate-to-outcome relationship, which are common in clinical practice. The integrated gradients provided the importance scores of input variables, which indicated variables guiding the model in its prediction. This model is ready to be utilized for time-to-event prediction in longitudinal cohorts.

Introducing the BlendedICU dataset, the first harmonized, international intensive care dataset.

OBJECTIVE: This study introduces the BlendedICU dataset, a massive dataset of international intensive care data. This dataset aims to facilitate generalizability studies of machine learning models, as well as statistical studies of clinical practices in the intensive care units. METHODS: Four publicly available and patient-level intensive care databases were used as source databases. A unique and customizable preprocessing pipeline extracted clinically relevant patient-related variables from each source database. The variables were then harmonized and standardized to the Observational Medical Outcomes Partnership (OMOP) Common Data Format. Finally, a brief comparison was carried out to explore differences in the source databases. RESULTS: The BlendedICU dataset features 41 timeseries variables as well as the exposure times to 113 active ingredients extracted from the AmsterdamUMCdb, eICU, HiRID, and MIMIC-IV databases. This resulted in a database of more than 309000 intensive care admissions, spanning over 13 years and three countries. We found that data collection, drug exposure, and patient outcomes varied strongly between source databases. CONCLUSION: The variability in data collection, drug exposure, and patient outcomes between the source databases indicated some dissimilarity in patient phenotypes and clinical practices between different intensive care units. This demonstrated the need for generalizability studies of machine learning models. This study provides the clinical data research community with essential data to build efficient and generalizable machine learning models, as well as to explore clinical practices in intensive care units around the world.

Clinical utility of a deep-learning mortality prediction model for cardiac surgery decision making.

OBJECTIVES: The aim of this study using decision curve analysis (DCA) was to evaluate the clinical utility of a deep-learning mortality prediction model for cardiac surgery decision making compared with the European System for Cardiac Operative Risk Evaluation (EuroSCORE) II and to 2 machine-learning models. METHODS: Using data from a French prospective database, this retrospective study evaluated all patients who underwent cardiac surgery in 43 hospital centers between January 2012 and December 2020. A receiver operating characteristic analysis was performed to compare the accuracy of the EuroSCORE II, machine-learning models, and an adapted Tabular Bidirectional Encoder Representations from Transformers deep-learning model in predicting postoperative in-hospital mortality. The clinical utility of these models for cardiac surgery decision making was compared using DCA. RESULTS: Over the study period, 165,640 patients underwent cardiac surgery, with a mean EuroSCORE II of 3.99 ± 6.67%. In the receiver operating characteristic analysis, the area under the curve was significantly greater for the deep-learning model (0.834; 95% confidence interval, 0.831-0.838) than the EuroSCORE II (P < .001), the random forest model (P = .03), and the Extreme Gradient Boosting model (P = .03). In the DCA, the clinical utility of the 3 artificial intelligence models was superior to that of the EuroSCORE II, especially when the threshold probability of death was high (>45%). The deep-learning model showed the greatest advantage over the EuroSCORE II. CONCLUSIONS: The deep-learning model had better predictive accuracy and greater clinical utility than the EuroSCORE II and the 2 machine-learning models. These findings suggest that deep learning with Tabular Bidirectional Encoder Representations from Transformers prediction model could be used in the future as the gold standard for cardiac surgery decision making.

Image augmentation and automated measurement of endotracheal-tube-to-carina distance on chest radiographs in intensive care unit using a deep learning model with external validation.

BACKGROUND: Chest radiographs are routinely performed in intensive care unit (ICU) to confirm the correct position of an endotracheal tube (ETT) relative to the carina. However, their interpretation is often challenging and requires substantial time and expertise. The aim of this study was to propose an externally validated deep learning model with uncertainty quantification and image segmentation for the automated assessment of ETT placement on ICU chest radiographs. METHODS: The CarinaNet model was constructed by applying transfer learning to the RetinaNet model using an internal dataset of ICU chest radiographs. The accuracy of the model in predicting the position of the ETT tip and carina was externally validated using a dataset of 200 images extracted from the MIMIC-CXR database. Uncertainty quantification was performed using the level of confidence in the ETT-carina distance prediction. Segmentation of the ETT was carried out using edge detection and pixel clustering. RESULTS: The interrater agreement was 0.18 cm for the ETT tip position, 0.58 cm for the carina position, and 0.60 cm for the ETT-carina distance. The mean absolute error of the model on the external test set was 0.51 cm for the ETT tip position prediction, 0.61 cm for the carina position prediction, and 0.89 cm for the ETT-carina distance prediction. The assessment of ETT placement was improved by complementing the human interpretation of chest radiographs with the CarinaNet model. CONCLUSIONS: The CarinaNet model is an efficient and generalizable deep learning algorithm for the automated assessment of ETT placement on ICU chest radiographs. Uncertainty quantification can bring the attention of intensivists to chest radiographs that require an experienced human interpretation. Image segmentation provides intensivists with chest radiographs that are quickly interpretable and allows them to immediately assess the validity of model predictions. The CarinaNet model is ready to be evaluated in clinical studies.

A descriptive study of routine laboratory testing in intensive care unit in nearly 140,000 patient stays.

To describe the relationship between the use of laboratory tests and changes in laboratory parameters in ICU patients is necessary to help optimize routine laboratory testing. A retrospective, descriptive study was conducted on the large eICU-Collaborative Research Database. The relationship between the use of routine laboratory tests (chemistry and blood counts) and changes in ten common laboratory parameters was studied. Factors associated with laboratory tests were identified in a multivariate regression analysis using generalized estimating equation Poisson models. The study included 138,734 patient stays, with an ICU mortality of 8.97%. For all parameters, the proportion of patients with at least one test decreased from day 0 to day 1 and then gradually increased until the end of the ICU stay. Paradoxically, the results of almost all tests moved toward normal values, and the daily variation in the results of almost all tests decreased over time. The presence of an arterial catheter or teaching hospitals were independently associated with an increase in the number of laboratory tests performed. The paradox of routine laboratory testing should be further explored by assessing the factors that drive the decision to perform routine laboratory testing in ICU and the impact of such testing on patient.

Spatialized epidemiological forecasting applied to Covid-19 pandemic at departmental scale in France.

In this paper, we present a spatialized extension of a SIR model that accounts for undetected infections and recoveries as well as the load on hospital services. The spatialized compartmental model we introduce is governed by a set of partial differential equations (PDEs) defined on a spatial domain with complex boundary. We propose to solve the set of PDEs defining our model by using a meshless numerical method based on a finite difference scheme in which the spatial operators are approximated by using radial basis functions. Such an approach is reputed as flexible for solving problems on complex domains. Then we calibrate our model on the French department of Isère during the first period of lockdown, using daily reports of hospital occupancy in France. Our methodology allows to simulate the spread of Covid-19 pandemic at a departmental level, and for each compartment. However, the simulation cost prevents from online short-term forecast. Therefore, we propose to rely on reduced order modeling to compute short-term forecasts of infection number. The strategy consists in learning a time-dependent reduced order model with few compartments from a collection of evaluations of our spatialized detailed model, varying initial conditions and parameter values. A set of reduced bases is learnt in an offline phase while the projection on each reduced basis and the selection of the best projection is performed online, allowing short-term forecast of the global number of infected individuals in the department. The original approach proposed in this paper is generic and could be adapted to model and simulate other dynamics described by a model with spatially distributed parameters of the type diffusion-reaction on complex domains. Also, the time-dependent model reduction techniques we introduced could be leveraged to compute control strategies related to such dynamics.