Tunable Privacy Risk Evaluation of Generative Adversarial Networks.

Generative machine learning models such as Generative Adversarial Networks (GANs) have been shown to be especially successful in generating realistic synthetic data in image and tabular domains. However, it has been shown that such generative models, as well as the generated synthetic data, can reveal information contained in their privacy-sensitive training data, and therefore must be carefully evaluated before being used. The gold standard method through which such privacy leakage can be estimated is simulating membership inference attacks (MIAs), in which an attacker attempts to learn whether a given sample was part of the training data of a generative model. The state-of-the art MIAs against generative models, however, rely on strong assumptions (knowledge of the exact training dataset size), or require a lot of computational power (to retrain many "surrogate" generative models), which make them hard to use in practice. In this work, we propose a technique for evaluating privacy risks in GANs which exploits the outputs of the discriminator part of the standard GAN architecture. We evaluate our attacks in terms of performance in two synthetic image generation applications in radiology and ophthalmology, showing that our technique provides a more complete picture of the threats by performing worst-case privacy risk estimation and by identifying attacks with higher precision than the prior work.

Evaluating Synthetic Data Augmentation to Correct for Data Imbalance in Realistic Clinical Prediction Settings.

Predictive modeling holds a large potential in clinical decision-making, yet its effectiveness can be hindered by inherent data imbalances in clinical datasets. This study investigates the utility of synthetic data for improving the performance of predictive modeling on realistic small imbalanced clinical datasets. We compared various synthetic data generation methods including Generative Adversarial Networks, Normalizing Flows, and Variational Autoencoders to the standard baselines for correcting for class underrepresentation on four clinical datasets. Although results show improvement in F1 scores in some cases, even over multiple repetitions, we do not obtain statistically significant evidence that synthetic data generation outperforms standard baselines for correcting for class imbalance. This study challenges common beliefs about the efficacy of synthetic data for data augmentation and highlights the importance of evaluating new complex methods against simple baselines.

Development of a Predictive Model for Hospital-Acquired Pressure Injuries.

Hospital-acquired pressure injuries are a challenge for healthcare systems, and the nurse's role is essential in their prevention. The first step is risk assessment. The development of advanced data-driven methods based on machine learning techniques can improve risk assessment through the use of routinely collected data. We studied 24 227 records from 15 937 distinct patients admitted to medical and surgical units between April 1, 2019, and March 31, 2020. Two predictive models were developed: random forest and long short-term memory neural network. Model performance was then evaluated and compared with the Braden score. The areas under the receiver operating characteristic curve, the specificity, and the accuracy of the long short-term memory neural network model (0.87, 0.82, and 0.82, respectively) were higher than those of the random forest model (0.80, 0.72, and 0.72, respectively) and the Braden score (0.72, 0.61, and 0.61, respectively). The sensitivity of the Braden score (0.88) was higher than that of long short-term memory neural network model (0.74) and the random forest model (0.73). The long short-term memory neural network model has the potential to support nurses in clinical decision-making. Implementation of this model in the electronic health record could improve assessment and allow nurses to focus on higher-priority interventions.

National Early Warning Score (NEWS) Outperforms Quick Sepsis-Related Organ Failure (qSOFA) Score for Early Detection of Sepsis in the Emergency Department.

Background: Prompt recognition of sepsis is critical to improving patients' outcomes. We compared the performance of NEWS and qSOFA scores as sepsis detection tools in patients admitted to the emergency department (ED) with suspicion of sepsis. Methodology: A single-center 12-month retrospective study comparing NEWS using the recommended cut-off of ≥5 and qSOFA as sepsis screening tools in a cohort of patients transported by emergency medical services (EMS) to the Lausanne University Hospital (LUH). We used the Sepsis-3 consensus definition. The primary study endpoint was the detection of sepsis. Secondary endpoints were ICU admission and 28-day all-cause mortality. Results: Among 886 patients admitted to ED by EMS for suspected infection, 556 (63%) had a complete set of vital parameters panel enabling the calculation of NEWS and qSOFA scores, of whom 300 (54%) had sepsis. For the detection of sepsis, the sensitivity of NEWS > 5 was 86% and that of qSOFA ≥ 2 was 34%. Likewise, the sensitivities of NEWS ≥ 5 for predicting ICU admission and 28-day mortality were higher than those of qSOFA ≥ 2 (82% versus 33% and 88% versus 37%). Conversely, the specificity of qSOFA ≥ 2 for sepsis detection was higher than that of NEWS ≥ 5 (90% versus 55%). The negative predictive value of NEWS > 5 was higher than that of qSOFA ≥ 2 (77% versus 54%), while the positive predictive value of qSOFA ≥ 2 was higher than that of NEWS ≥ 5 (80% versus 69%). Finally, the accuracy of NEWS ≥ 5 was higher than that of qSOFA ≥ 2 (72% versus 60%). Conclusions: The sensitivity of NEWS ≥ 5 was superior to that of qSOFA ≥ 2 to identify patients with sepsis in the ED and predict ICU admission and 28-day mortality. In contrast, qSOFA ≥ 2 had higher specificity and positive predictive values than NEWS ≥ 5 for these three endpoints.

Generation and Evaluation of Synthetic Data in a University Hospital Setting.

In this study, we propose a unified evaluation framework for systematically assessing the utility-privacy trade-off of synthetic data generation (SDG) models. These SDG models are adapted to deal with longitudinal or tabular data stemming from electronic health records (EHR) containing both discrete and numeric features. Our evaluation framework considers different data sharing scenarios and attacker models.

[Artificial Intelligence in internal medicine†: development of a model predicting length of stay for non-elective admissions]. / Intelligence artificielle en médecine interne†: développement d'un modèle prédictif des durées de séjour.

Efficient management of hospitalized patients requires carefully planning each stay by taking into account patients' pathologies and hospital constraints. Therefore, the ability to accurately estimate length of stays allows for better interprofessional tasks coordination, improved patient flow management, and anticipated discharge preparation. This article presents how we built and evaluated a predictive model of length of stay based on clinical data available upon admission to a division of internal medicine. We show that Machine Learning-based approaches can predict lengths of stay with a similar level of accuracy as field experts.

Une prise en charge efficiente des patients nécessite une planification minutieuse des soins en fonction de la pathologie et des contraintes hospitalières. Dans ce contexte, une estimation de la durée de séjour permet de mieux coordonner les tâches interprofessionnelles, de gérer le flux des patients et d'anticiper la préparation à la sortie. Cet article présente la construction et l'évaluation d'un modèle prédictif de la durée de séjour à l'aide de données cliniques présentes à l'admission dans un service de médecine interne universitaire. Nous démontrons que les approches basées sur le Machine Learning sont capables de prédire des durées de séjour avec une performance similaire à celle des professionnels.

Design and control of compliant tensegrity robots through simulation and hardware validation.

To better understand the role of tensegrity structures in biological systems and their application to robotics, the Dynamic Tensegrity Robotics Lab at NASA Ames Research Center, Moffett Field, CA, USA, has developed and validated two software environments for the analysis, simulation and design of tensegrity robots. These tools, along with new control methodologies and the modular hardware components developed to validate them, are presented as a system for the design of actuated tensegrity structures. As evidenced from their appearance in many biological systems, tensegrity ('tensile-integrity') structures have unique physical properties that make them ideal for interaction with uncertain environments. Yet, these characteristics make design and control of bioinspired tensegrity robots extremely challenging. This work presents the progress our tools have made in tackling the design and control challenges of spherical tensegrity structures. We focus on this shape since it lends itself to rolling locomotion. The results of our analyses include multiple novel control approaches for mobility and terrain interaction of spherical tensegrity structures that have been tested in simulation. A hardware prototype of a spherical six-bar tensegrity, the Reservoir Compliant Tensegrity Robot, is used to empirically validate the accuracy of simulation.