LIDA Seminar Series 26th September 2019

This seminar will be held in the LIDA Boardroom 11.87, Worsley Building, at 12.30 on Thursday 26^th September.

Seminars are free and open for all to attend. No prior booking is required.

Each presentation will be followed by a short Q&A session.

#LIDAseminar

 

Presentation 1: Machine Learning Classification on Blood Pressure Variability: from SPRINT Trial to Community eHealth Program in Hong Kong

By: Professor Kelvin Tsoi

Abstract: Hypertension has long been considered a modifiable major risk factor of cardiovascular diseases. However, visit-to-visit blood pressure variability (BPV) has recently been gaining attention and studies have suggested that it is independent of blood pressure (BP) and associated with cardiovascular risks. However, many aspects of BPV such as measurement frequency and interaction with drugs remain unclear. In this work, we begin by presenting our post hoc analysis on BPV based on the SPRINT trial. This is followed by a discussion on a community-based BP-monitoring programme, named eHealth, as a way forward to the SPRINT study.

Speaker Biography: Professor Kelvin Tsoi is an associate professor of the School of Public Health and Primary Care, Institute of Ageing, and a research associate professor at the Big Data Decision Analytics Research Centre of the Chinese University Hong Kong (CUHK). He developed a strong interest in evidence-based medicine, digital epidemiology, and big-data research for healthcare application. He leads an interdisciplinary team on Big Data research for Healthcare including, for example, understanding the digital behavior on dementia patients and personal profiling of blood pressure variability.

He has published over 80 full scientific articles in leading academic journals. Some of his research has also published in short papers in systems engineering conferences. A recent paper to compare dementia screening was published in JAMA Internal Medical. He is also the assistant editor (Digital Health) in BMJ Innovations and Mini-track Chair for Big Data on Healthcare Application in HICSS 2019.

 

Presentation 2: Synthetic population estimation and scenario projection - The English Future Elderly Model

By: Luke Archer

What I will be talking about: Around the world, virtually all developed and developing countries are experiencing population ageing. For example, the number of people aged 60 years and over has tripled since 1950, and it is projected that the combined global senior and geriatric population will reach 2.1 billion by 2050. The implications of this in terms of government initiatives and meeting growing healthcare demands raises many questions about how best to plan for the future.

To answer some of these questions, this project uses a dynamic microsimulation model, which is a data-driven forecast of individuals who are aged and who develop certain chronic conditions during their life.  The model is used to project larger population-level trends. By predicting future outcomes reflected in longitudinal panel surveys, microsimulation can help to answer “what if” questions on the health of our future elderly population, and help to influence policy in an evidence-based manner.

Our model is based on an already established model called the Future Elderly Model (FEM). The FEM was created at the Schaeffer Center for Health Policy & Economics, based at the University of Southern California, who we have been collaborating with to adapt their original model to an English context. The primary data source of the US-FEM is the Health and Retirement Study (HRS), a large longitudinal study of people aged 50+ in America, collecting data on individuals’ health, economic status and demography. We have based the English-FEM on the English Longitudinal Study of Ageing (ELSA), which has a similar format to the HRS and allows us to conduct cross-country comparisons.

Our model simulates elderly populations into the future, and can predict not only levels of chronic disease but also a broad range of related factors, like the impact on disability free living and median retirement age. The real power of the model however is the ability to assess hypothetical interventions. For example, we can ask what would be the economic and public health impact in reducing the incidence of diabetes by 20% in people aged 50+? These “what if” scenarios are only limited by the input data.

 

Presentation 3: Quantifying Uncertainty in Agent Based Models for Smart Cities Forecasts

By: Robert Clay

What I will be talking about: Smart Cities technology aims to harness big data to predict traffic flows and plan urban infrastructure. Much of the data used in smart cities is poor due to uncertain noisy measurements from both unreliable, partial sources and low resolution aggregates. High performance computing has given rise to agent based models (ABMs) allowing us to simulate crowds based on real data at a microscopic scale. In this talk I will demonstrate how combining an ABM with our uncertain measurements using data assimilation techniques allows us to quantify and incorporate said uncertainty to improve behaviour prediction.

 

Presentation 4: Can we teach robots to drive like humans?

By: Dr Kevin Minors

What I will be talking about: In this talk, we will apply inverse reinforcement learning to sampled trajectories of individual vehicles on a motorway in order to produce a reward function that will allow reinforcement agents to reproduce the observed behaviour. After first introducing the concepts of reinforcement learning and inverse reinforcement learning, we will examine the motorway data, describe the inverse reinforcement learning method in more detail, and finish with early results from the research.