Risk Institute Showcase Conference 2022

The accurate and timely detection of adverse drug-drug interactions (DDIs) during the post-marketing phase is an important yet complex task with potentially major clinical implications. The development of data mining methodologies that scan healthcare databases for drug safety signals requires appropriate reference sets for performance evaluation. By automatically extracting and aggregating information from multiple clinical resources, we have provided a scalable approach for generating a reference set for DDIs that could support research in post-marketing safety surveillance.

The COVID-19 pandemic has impacted surveillance activities for multiple pathogens. Since March 2020, there was a decline in the number of reports of norovirus and Campylobacter recorded by England’s national laboratory surveillance system. Our results show that reporting of norovirus was more adversely impacted than Campylobacter.

Logistic regression is an important statistical tool for assessing the probability of an outcome based upon some predictive variables and is a popular supervised learning algorithm used for classification. Standard methods can only deal with precisely known data, however many data sets have uncertainties which traditional methods either reduce to a single point or completely disregarded. In this talk we show that it is possible to include these uncertainties by considering an imprecise logistic regression model using the set of possible models that can be obtained from values from within the intervals. This has the advantage of clearly expressing the epistemic uncertainty removed by traditional methods.

Simulation codes allow to reduce the high conservativism in nuclear reactor design improving the reliability and sustainability associated to nuclear power. Full core coupled reactor physics at the fuel pin level are not provided by most simulation codes. This has led in the UK to the development of a multiscale and Multiphysics software development focused on LWRS. Main simulation codes that are being coupled within this multiscale and Multiphysics software development include: The subchannel code CTF and the transport code LOTUS to provide full coupled reactor physics at the fuel pin level. The nodal code DYN3D to provide simplified coupled reactor physics at both the fuel assembly and fuel pin levels. One of the aims within this multiscale and Multiphysics software development is the coupling between the subchannel code CTF, the nodal code DYN3D and the transport code LOTUS to provide either improved or full coupled reactor physics at the fuel pin level. Several objectives within this aim include: Thermal hydraulics validations and verifications which have been performed to evaluate the accuracy and methodology available to obtain thermal hydraulics at the fuel pin level in DYN3D and CTF. A 1 way coupling between DYN3D and CTF which has been performed through the transfer of the power distributions from DYN3D to CTF to obtain partially verified improved coupled reactor physics at the fuel pin level. A 2 ways coupling between DYN3D and CTF which has been performed through the transfer of converged power and feedback distributions between DYN3D and CTF to obtain fully verified improved coupled reactor physics at the fuel pin level. The coupling between LOTUS and CTF is still pending which will allow to obtain full coupled reactor at the fuel pin level.

In this talk, Adolphus will present an overview of his PhD topic as well as the tools he has developed over the course of his research journey. His presentation will first introduce the concept of Bayesian Model Updating, followed by the short-comings in existing strategies from which the research questions are discussed. After which, he will provide a summary and description to the sampling algorithms he developed before concluding this talk with a short remark on his research plans going forward in this PhD and drawing the presentation to a close.

A two-level framework is demonstrated for stochastic model updating. At the first level, variance based global sensitivity analysis is carried out with the purpose of identifying those parameters with significant uncertainty and those that might be considered deterministic and can be eliminated as inputs to the metamodel. Then, at the second level, an inverse problem is solved to determine the statistics of the parameters of a modification that causes numerical metamodel results to converge on experimental data. The framework methodology is applied to a simulated three degrees of freedom representation of an experimental rig.

I argue that, when performing sensitivity analysis, sometimes it is desirable to assume as less as possible about the nature of the input variables in a mathematical model. In this talk I will introduce a method that requires only assuming the input uncertainty to be an interval.

In this talk, we present an add-on to the R-package GAMLSS, to model nonstationary fields whose margins are extended-GPD (EGPD) distributed, in the sense of Naveau et al. The GAMLSS package provides univariate distributional regression models, where parameters of the assumed distribution for the response can be modeled as additive functions of the explanatory variables. It is flexible enough to allow the encoding of novel families of distribution, as long as their density, cumulative, quantile and random generative function and, optionally, first and second (cross-) derivatives of the likelihood can be computed. We seize this potential to implement the parametric EGPD family in a generic way, allowing to test any new parametric form satisfying the requirement of the EGPD family.

In this talk, a framework based on Credal Networks is presented as a general and efficient manner for uncertainty quantification in risk and resilience assessments. These techniques are applied in areas such as safety critical engineering and infrastructure systems as an alternative to traditional methods. This approach adopts probabilistic intervals to represent the epistemic uncertainty, due to the lack of data, of the components of complex systems. In addition to this, the proposed technique allows to perform prediction, diagnostic and what-if analyses which are paramount for analysis of different scenarios. The technique is extended to dynamic scenarios for resilience assessment of the system performance before, during and after the occurrence of a disruption.