Early prediction of in-hospital mortality utilizing multivariate predictive modelling of electronic medical records and socio-determinants of health of the first day of hospitalization.

BACKGROUND: In France an average of 4% of hospitalized patients die during their hospital stay. To aid medical decision making and the attribution of resources, within a few days of admission the identification of patients at high risk of dying in hospital is essential. METHODS: We used de-identified routine patient data available in the first 2 days of hospitalization in a French University Hospital (between 2016 and 2018) to build models predicting in-hospital mortality (at ≥ 2 and ≤ 30 days after admission). We tested nine different machine learning algorithms with repeated 10-fold cross-validation. Models were trained with 283 variables including age, sex, socio-determinants of health, laboratory test results, procedures (Classification of Medical Acts), medications (Anatomical Therapeutic Chemical code), hospital department/unit and home address (urban, rural etc.). The models were evaluated using various performance metrics. The dataset contained 123,729 admissions, of which the outcome for 3542 was all-cause in-hospital mortality and 120,187 admissions (no death reported within 30 days) were controls. RESULTS: The support vector machine, logistic regression and Xgboost algorithms demonstrated high discrimination with a balanced accuracy of 0.81 (95%CI 0.80-0.82), 0.82 (95%CI 0.80-0.83) and 0.83 (95%CI 0.80-0.83) and AUC of 0.90 (95%CI 0.88-0.91), 0.90 (95%CI 0.89-0.91) and 0.90 (95%CI 0.89-0.91) respectively. The most predictive variables for in-hospital mortality in all three models were older age (greater risk), and admission with a confirmed appointment (reduced risk). CONCLUSION: We propose three highly discriminating machine-learning models that could improve clinical and organizational decision making for adult patients at hospital admission.

Cohort profile for development of machine learning models to predict healthcare-related adverse events (Demeter): clinical objectives, data requirements for modelling and overview of data set for 2016-2018.

PURPOSE: In-hospital health-related adverse events (HAEs) are a major concern for hospitals worldwide. In high-income countries, approximately 1 in 10 patients experience HAEs associated with their hospital stay. Estimating the risk of an HAE at the individual patient level as accurately as possible is one of the first steps towards improving patient outcomes. Risk assessment can enable healthcare providers to target resources to patients in greatest need through adaptations in processes and procedures. Electronic health data facilitates the application of machine-learning methods for risk analysis. We aim, first to reveal correlations between HAE occurrence and patients' characteristics and/or the procedures they undergo during their hospitalisation, and second, to build models that allow the early identification of patients at an elevated risk of HAE. PARTICIPANTS: 143 865 adult patients hospitalised at Grenoble Alpes University Hospital (France) between 1 January 2016 and 31 December 2018. FINDINGS TO DATE: In this set-up phase of the project, we describe the preconditions for big data analysis using machine-learning methods. We present an overview of the retrospective de-identified multisource data for a 2-year period extracted from the hospital's Clinical Data Warehouse, along with social determinants of health data from the National Institute of Statistics and Economic Studies, to be used in machine learning (artificial intelligence) training and validation. No supplementary information or evaluation on the part of medical staff will be required by the information system for risk assessment. FUTURE PLANS: We are using this data set to develop predictive models for several general HAEs including secondary intensive care admission, prolonged hospital stay, 7-day and 30-day re-hospitalisation, nosocomial bacterial infection, hospital-acquired venous thromboembolism, and in-hospital mortality.

A comparative study of two automated solutions for cross-sectional skeletal muscle measurement from abdominal computed tomography images.

BACKGROUND: Measurement of cross-sectional muscle area (CSMA) at the mid third lumbar vertebra (L3) level from computed tomography (CT) images is becoming one of the reference methods for sarcopenia diagnosis. However, manual skeletal muscle segmentation is tedious and is thus restricted to research. Automated solutions are required for use in clinical practice. PURPOSE: The aim of this study was to compare the reliability of two automated solutions for the measurement of CSMA. METHODS: We conducted a retrospective analysis of CT images in our hospital database. We included consecutive individuals hospitalized at the Grenoble University Hospital in France between January and May 2018 with abdominal CT images and sagittal reconstruction. We used two types of software to automatically segment skeletal muscle: ABACS, a module of the SliceOmatic software solution "ABACS-SliceOmatic," and a deep learning-based solution called "AutoMATiCA." Manual segmentation was performed by a medical expert to generate reference data using "SliceOmatic." The Dice similarity coefficient (DSC) was used to measure overlap between the results of the manual and the automated segmentations. The DSC value for each method was compared with the Mann-Whitney U test. RESULTS: A total of 676 hospitalized individuals was retrospectively included (365 males [53.8%] and 312 females [46.2%]). The median DSC for SliceOmatic vs AutoMATiCA (0.969 [5th percentile: 0.909]) was greater than the median DSC for SliceOmatic vs. ABACS-SliceOmatic (0.949 [5th percentile: 0.836]) (p < 0.001). CONCLUSIONS: AutoMATiCA, which used artificial intelligence, was more reliable than ABACS-SliceOmatic for skeletal muscle segmentation at the L3 level in a cohort of hospitalized individuals. The next step is to develop and validate a neural network that can identify L3 slices, which is currently a fastidious process.

Hypergraph Based Data Model for Complex Health Data Exploration and Its Implementation in PREDIMED Clinical Data Warehouse.

Within the PREDIMED Clinical Data Warehouse (CDW) of Grenoble Alpes University Hospital (CHUGA), we have developed a hypergraph based operational data model, aiming at empowering physicians to explore, visualize and qualitatively analyze interactively the complex and massive information of the patients treated in CHUGA. This model constitutes a central target structure, expressed in a dual form, both graphical and formal, which gathers the concepts and their semantic relations into a hypergraph whose implementation can easily be manipulated by medical experts. The implementation is based on a property graph database linked to an interactive graphical interface allowing to navigate through the data and to interact in real time with a search engine, visualization and analysis tools. This model and its agile implementation allow for easy structural changes inherent to the evolution of techniques and practices in the health field. This flexibility provides adaptability to the evolution of interoperability standards.

COVID-19 Geographical Maps and Clinical Data Warehouse PREDIMED.

PREDIMED, Clinical Data Warehouse of Grenoble Alps University Hospital, is currently participating in daily COVID-19 epidemic follow-up via spatial and chronological analysis of geographical maps. This monitoring is aimed for cluster detection and vulnerable population discovery. Our real-time geographical representations allow us to track the epidemic both inside and outside the hospital.

ODIASP: Clinically Contextualized Image Analysis Using the PREDIMED Clinical Data Warehouse, Towards a Better Diagnosis of Sarcopenia.

Big Data and Deep Learning approaches offer new opportunities for medical data analysis. With these technologies, PREDIMED, the clinical data warehouse of Grenoble Alps University Hospital, sets up first clinical studies on retrospective data. In particular, ODIASP study, aims to develop and evaluate deep learning-based tools for automatic sarcopenia diagnosis, while using data collected via PREDIMED, in particular, medical images. Here we describe a methodology of data preparation for a clinical study via PREDIMED.

Assessing Adherence to Healthy Dietary Habits Through the Urinary Food Metabolome: Results From a European Two-Center Study.

Background: Diet is one of the most important modifiable lifestyle factors in human health and in chronic disease prevention. Thus, accurate dietary assessment is essential for reliably evaluating adherence to healthy habits. Objectives: The aim of this study was to identify urinary metabolites that could serve as robust biomarkers of diet quality, as assessed through the Alternative Healthy Eating Index (AHEI-2010). Design: We set up two-center samples of 160 healthy volunteers, aged between 25 and 50, living as a couple or family, with repeated urine sampling and dietary assessment at baseline, and 6 and 12 months over a year. Urine samples were subjected to large-scale metabolomics analysis for comprehensive quantitative characterization of the food-related metabolome. Then, lasso regularized regression analysis and limma univariate analysis were applied to identify those metabolites associated with the AHEI-2010, and to investigate the reproducibility of these associations over time. Results: Several polyphenol microbial metabolites were found to be positively associated with the AHEI-2010 score; urinary enterolactone glucuronide showed a reproducible association at the three study time points [false discovery rate (FDR): 0.016, 0.014, 0.016]. Furthermore, other associations were found between the AHEI-2010 and various metabolites related to the intake of coffee, red meat and fish, whereas other polyphenol phase II metabolites were associated with higher AHEI-2010 scores at one of the three time points investigated (FDR < 0.05 or ß ≠ 0). Conclusion: We have demonstrated that urinary metabolites, and particularly microbiota-derived metabolites, could serve as reliable indicators of adherence to healthy dietary habits. Clinical Trail Registration: www.ClinicalTrials.gov, Identifier: NCT03169088.

Cohort Creation and Visualization Using Graph Model in the PREDIMED Health Data Warehouse.

Grenoble Alpes University Hospital (CHUGA) is currently deploying a health data warehouse called PREDIMED [1], a platform designed to integrate and analyze for research, education and institutional management the data of patients treated at CHUGA. PREDIMED contains healthcare data, administrative data and, potentially, data from external databases. PREDIMED is hosted by the CHUGA Information Systems Department and benefits from its strict security rules. CHUGA's institutional project PREDIMED aims to collaborate with similar projects in France and worldwide. In this paper, we present how the data model defined to implement PREDIMED at CHUGA is useful for medical experts to interactively build a cohort of patients and to visualize this cohort.

PREDIMED: Clinical Data Warehouse of Grenoble Alpes University Hospital.

Grenoble Alpes University Hospital (CHUGA) currently deploys a clinical data warehouse PREDIMED to integrate and analyze for research, education and institutional management the data of patients treated at CHUGA. In this poster, we present the methodology used to implement PREDIMED and illustrate its functionality through three first research use cases.

IM-UFF: Extending the universal force field for interactive molecular modeling.

The universal force field (UFF) is a broadly applicable classical force field that contains parameters for almost every atom type of the periodic table. This force field is non-reactive, i.e. the topology of the system under study is considered as fixed and no creation or breaking of covalent bonds is possible. This paper introduces interactive modeling-UFF (IM-UFF), an extension of UFF that combines the possibility to significantly modify molecular structures (as with reactive force fields) with a broad diversity of supported systems thanks to the universality of UFF. Such an extension lets the user easily build and edit molecular systems interactively while being guided by physics based inter-atomic forces. This approach introduces weighted atom types and weighted bonds, used to update topologies and atom parameterizations at every time step of a simulation. IM-UFF has been evaluated on a large set of benchmarks and is proposed as a self-contained implementation integrated in a new module for the SAMSON software platform for computational nanoscience available at http://www.samson-connect.net.

Automatic molecular structure perception for the universal force field.

The Universal Force Field (UFF) is a classical force field applicable to almost all atom types of the periodic table. Such a flexibility makes this force field a potential good candidate for simulations involving a large spectrum of systems and, indeed, UFF has been applied to various families of molecules. Unfortunately, initializing UFF, that is, performing molecular structure perception to determine which parameters should be used to compute the UFF energy and forces, appears to be a difficult problem. Although many perception methods exist, they mostly focus on organic molecules, and are thus not well-adapted to the diversity of systems potentially considered with UFF. In this article, we propose an automatic perception method for initializing UFF that includes the identification of the system's connectivity, the assignment of bond orders as well as UFF atom types. This perception scheme is proposed as a self-contained UFF implementation integrated in a new module for the SAMSON software platform for computational nanoscience (http://www.samson-connect.net). We validate both the automatic perception method and the UFF implementation on a series of benchmarks.

Adaptively restrained particle simulations.

Interaction potentials used in particle simulations are typically written as a sum of terms which depend on just a few relative particle positions. Traditional simulation methods move all particles at each time step, and may thus spend a lot of time updating interparticle forces. In this Letter we introduce adaptively restrained particle simulations (ARPS) to speed up particle simulations by adaptively switching on and off positional degrees of freedom, while letting momenta evolve. We illustrate ARPS on several numerical experiments, including (a) a collision cascade example that demonstrates how ARPS make it possible to smoothly trade between precision and speed and (b) a polymer-in-solvent study that shows how one may efficiently determine static equilibrium properties with ARPS.

A comparison of neighbor search algorithms for large rigid molecules.

Fast determination of neighboring atoms is an essential step in molecular dynamics simulations or Monte Carlo computations, and there exists a variety of algorithms to efficiently compute neighbor lists. However, most of these algorithms are general, and not specifically designed for a given type of application. As a result, although their average performance is satisfactory, they might be inappropriate in some specific application domains. In this article, we study the case of detecting neighbors between large rigid molecules, which has applications in, e.g., rigid body molecular docking, Monte Carlo simulations of molecular self-assembly or diffusion, and rigid body molecular dynamics simulations. More precisely, we compare the traditional grid-based algorithm to a series of hierarchy-based algorithms that use bounding volumes to rapidly eliminate large groups of irrelevant pairs of atoms during the neighbor search. We compare the performance of these algorithms based on several parameters: the size of the molecules, the average distance between them, the cutoff distance, as well as the type of bounding volume used in the culling hierarchy (AABB, OBB, wrapped, or layered spheres). We demonstrate that for relatively large systems (> 100,000 atoms) the algorithm based on the hierarchy of wrapped spheres shows the best results and the traditional grid-based algorithm gives the worst timings. For small systems, however, the grid-based algorithm and the one based on the wrapped sphere hierarchy are beneficial.

Fast construction of assembly trees for molecular graphs.

A number of modeling and simulation algorithms using internal coordinates rely on hierarchical representations of molecular systems. Given the potentially complex topologies of molecular systems, though, automatically generating such hierarchical decompositions may be difficult. In this article, we present a fast general algorithm for the complete construction of a hierarchical representation of a molecular system. This two-step algorithm treats the input molecular system as a graph in which vertices represent atoms or pseudo-atoms, and edges represent covalent bonds. The first step contracts all cycles in the input graph. The second step builds an assembly tree from the reduced graph. We analyze the complexity of this algorithm and show that the first step is linear in the number of edges in the input graph, whereas the second one is linear in the number of edges in the graph without cycles, but dependent on the branching factor of the molecular graph. We demonstrate the performance of our algorithm on a set of specifically tailored difficult cases as well as on a large subset of molecular graphs extracted from the protein data bank. In particular, we experimentally show that both steps behave linearly in the number of edges in the input graph (the branching factor is fixed for the second step). Finally, we demonstrate an application of our hierarchy construction algorithm to adaptive torsion-angle molecular mechanics.