Monday 17 February 2020
Access to administrative and secondary data for public health research is increasing. Application of analytical methods that cope with and harness the complexity of these data needs to keep pace. Machine learning is a technique that fits models algorithmically by adapting to patterns in the data itself. It is increasingly used by health researchers for surveillance, risk prediction and to enhance currently used methods (including those used to make causal inference).
Interesting recent examples of its application include detection of disease from scans, biopsies and other patient data, and analyses of the web search activity to provide real-time information on non-communicable disease population risk, potentially useful to assess the effect of policy implementation in real-time.
Machine learning applications are of graded complexity (i.e. from classification models to neural networks and deep learning). They are most useful to health researchers when data are:
- Complex (e.g. images, non-linear variables, text)
- Dynamic (e.g. data changes or can be updated, pooled from multiple sources)
This course will aim to increase participant’s understanding of machine learning, its relevance to research in the health sector and practical challenges to its application, so as to enable participants to work in conjunction with people with technical skills in machine learning. We will outline what can and cannot be solved with machine learning models and provide a basic overview of machine learning techniques and their use. It is expected neither that you have a good understanding nor that you have used machine learning before; this course aims to give you a conceptual overview.
Style of course
Small group – teaching and discussions in a group of up to 30 people.
Who should attend?
- Quantitative health researchers (e.g. Epidemiologists, biostatistician, economists, etc)
- Graduate research students in epidemiology, biostatistics and economics
- Public health and health sector analysts
By the end of this course participants should have the knowledge/skills to
- Understand the concept and potential application of machine learning in health care research
- Understand how machine learning can be applied as a tool in research (how to ask machine learning questions)
- Understand the limitations of machine learning
- Be familiar with some mathematical concepts underlying the models
- Become familiar with the latest development in the application of machine learning in the health context
Timetable
| Time | Session | Presenter(s) |
|---|---|---|
| 9:00am | Introduction and welcome An overview of key machine learning concepts | Rebecca Bentley / Nhung Nghiem |
| 10:30am | Morning Tea | |
| 11:00am | Contemporary applications of machine learning in health | Rebecca Bentley / Nhung Nghiem |
| 12:30pm | Lunch | |
| 1:30pm | Case studies: | |
| 1. Using machine learning to predict cardiovascular disease | Nhung Nghiem | |
| 2. Using machine learning to improve causal inference | Tony Blakely | |
| 3:00pm | Afternoon Tea | |
| 3:30pm | Understanding the limits of machine learning: lessons from equity focussed research | Rebecca Bentley / Nhung Nghiem |
| 4:00pm | Group discussion and conclusion | |
| 5:00pm | Finish |
Teaching staff
- Associate Professor Rebecca Bentley, University of Melbourne
Rebecca is a Principal Research Fellow in Social Epidemiology in the Centre for Health Equity, Melbourne School of Population and Global Health. Over the past ten years, Rebecca has developed a research program exploring the role of housing and residential location in shaping health and wellbeing in Australia, using advanced quantitative methods. Rebecca currently leads a project that seeks to understand and scope applications of machine learning in epidemiological research. - Dr Nhung Nghiem, University of Otago, Wellington
Nhung is a Senior Research Fellow with expertise in mathematical modelling and health economics with the Burden of Disease Epidemiology, Equity and Cost-Effectiveness Programme (BODE3) research group. Nhung is currently using machine learning (such as classification trees, L1 regularisation, and support vector machine) to predict cardiovascular disease incidence in New Zealand using health and social datasets in the IDI (Statistics New Zealand Integrated Data Infrastructure).
Course cost and registration
$300 early bird, $400 after 19 December 2019.
A 50% discount is available to full-time students, those unwaged and University of Otago staff.
