Advanced Epidemiology Short Course - Sydney, Australia 25-28 September 2017
Lecturers: Professor Tony Blakely (Otago University and Melbourne University) and Professor John Lynch (University of Adelaide)
The Advanced Epidemiology Short Course is being held in Sydney, Australia this year at Sydney University (preceding the AEA 2017 conference).
What does the course include?
- An introduction to causal inference using contemporary approaches such as a potential outcomes approach and directed acyclic graphs (DAGs).
- A comprehensive overview of systematic error (confounding, selection and information biases).
- An introduction to quantitative bias analysis methods to correct for systematic error in epidemiological studies. (Sometimes called sensitivity analyses.) Methods taught range from simple to probabilistic methods.
- Quantitative bias analysis exercises using Excel spreadsheets. Understanding and applying bias analyses not only enables you to undertake your own analyses in the future, but also means you have a deeper understanding of systematic error.
- Selected specific topics such as regression model building strategies, direct and indirect effects (i.e. mediation analysis), propensity scores, instrument variables.
- Applications of some of these methods in disease and cost effectiveness simulations (e.g. Markov and multistate lifetable models), and G methods such as marginal structural models (MSMs) and causal mediation analysis.
The course will be similar to previous years with the option to attend the four-day course, or just the final day on Thursday 28 September which focuses on addressing policy questions using advanced epidemiological methods, e.g. disease and cost-effectiveness modelling, and causal mediation analysis.
Course details and registration can be found at:
Half day workshops - $80pp:
The following half-day workshops will be held on Thursday 28 September. To register for a workshop, click here.
Please note there are minimum numbers for each workshop to proceed. You will be notified approximately three weeks prior to the workshop if it is not proceeding.
Morning Session - 0930-1230
TITLE: Relative risk estimation - methods and recent developments
CONVENOR/S: Leigh Blizzard and Petr Otahal
The risk ratio (relative risk) is the ratio measure of choice for summarising the impact of exposure in epidemiologic studies (Greenland, 1987). Fitting a log binomial regression model with a logarithmic link to binary outcome data makes it possible to estimate risk and risk ratios, and prevalence and prevalence ratios in cross-sectional studies. Standard methods for fitting the model can result in numerical difficulties including failure of the fitting algorithm to converge, however. This has prompted some practitioners to resort to one of several improvised methods. None of these ad hoc approaches is satisfactory.
Fortunately, relatively straightforward modifications to the fitting algorithms provided in standard statistical software, including optional variations to their default settings, make it possible to overcome numerical instability. An alternative is to perform estimation with a fitting algorithm due to Marschner and Gillett (2012).
This workshop provides a demonstration of these methods. The presenters will provide participants with purpose-written code for R and Stata statistical software, and assist those with laptops to apply it to example data. It will be shown that two of the most common work-around methods – estimating relative risk by Poisson regression, and approximating relative risk by odds ratios – can produce seriously misleading estimates.
WHAT TO BRING: Laptop with R or STATA software installed
TITLE: Extracting Insights from Unstructured Text
CONVENOR: David White
There are significant amounts of unstructured data that surround us and the structured data that we are comfortable analysing. This workshop introduces concepts, approaches and tools for extracting insight from unstructured data. The workshop is designed for those interested in learning about the value of insights in unstructured data and those new to unstructured data analysis.
WHAT TO BRING: Course participants bring their own laptop and should download Wordstat for Stata or QDAMiner and Wordstat.
Afternoon session - 1330-1630
TITLE: Assessing risk of lead time bias in studies of overdiagnosis
CONVENOR/S: Gemma Jacklyn, Katy Bell, Alex Barratt
Overdiagnosis is recognised as a common problem in cancer screening, but estimates of its frequency depend on reliable estimates of lead time. Studies of screening mammography that do not allow for lead time may be biased and overestimate overdiagnosis. The methodology for dealing with lead time is diverse, complex and challenging, especially for non-randomised studies. Currently, there is no agreed systematic method to assess the risk of lead time bias in studies of overdiagnosis. Our workshop seeks to address this knowledge gap by discussing key considerations when evaluating lead time and exploring potential criteria to assess the risk of lead time bias in studies that estimate overdiagnosis due to screening mammography for breast cancer. The workshop will include two presentations, an open, group-based discussion, and identification of next steps.
WHAT TO BRING: Laptop, notepad and pen.
TITLE: Multiple imputation of missing data in longitudinal cohort studies
CONVENORS: Dr Margarita Moreno-Betancur and Assoc Prof Katherine Lee
Modern epidemiological studies collect a wide range of time-varying exposures, outcomes and other factors repeatedly over the course of follow-up. Examples include physical/biological characteristics (body mass index (BMI), blood pressure, etc.) and behavioural/socio-demographic characteristics (smoking, marital status, etc.). These longitudinal measurements are key to evaluate etiological hypotheses, including understanding of the pathways by which different exposures affect health outcomes. However, the prolonged observation of individuals, over repeated waves of follow-up, exacerbates the occurrence of missing data as participant engagement decreases with time. Statistical approaches for handling missing data in longitudinal studies vary in complexity, with principled methods such as multiple imputation becoming increasingly popular as they yield valid inferences under a broad range of scenarios regarding the causes for missing data. Multiple imputation methods for handling missing data in numerous variables are widely available in mainstream statistical software. Nonetheless, there are limitations, both computational and conceptual, regarding their use in the longitudinal setting. In this workshop, the speakers will first provide an overview of multiple imputation, followed by lectures with computer demonstrations in Stata and R (the code will be provided to students) focusing on methods available for multiple imputation of longitudinal data and guidance on good practice. Detailed illustrations will be based on the Longitudinal Study of Australian Children (LSAC), a large-scale Australian observational cohort.
TARGET AUDIENCE and PRE-REQUISITES: This course is suitable for quantitative epidemiologists and applied statisticians working in health research. It is assumed that participants will have a sound working familiarity with Stata or R, and with statistics to the level of multivariable logistic regression models, prior to the course. Some familiarity with multiple imputation is desirable.
WHAT TO BRING: Students are welcome to bring their laptop with Stata or R installed but this is not necessary.
Full Day Workshop - $120pp
TITLE: Systematic Reviews, Meta-analysis and Meta-regression
CONVENORS: Dr Janni Leung & Dr Gary Chan
Systematic reviews, meta-analyses and meta-regression analyses are the key techniques for summarising existing evidences on a research topic and answering research questions based on existing literature.
The goal of this workshop is to enable attendees to use techniques. This workshop will consist of three components:
Part 1) Systematic reviews: Planning for challenges in research question formation, keeping track of screening, data-extraction template designs for meta-analyses.
Part 2) Meta-analysis: When to use what? Fixed effects, random effects, and quality effects models, forest plots, pooling binary and continuous data, sensitivity analyses.
Part 3) Meta-regression analyses: Preparing the data and running the meta-regression analyses.
The workshop is designed to be hands-on and interactive. Attendees will complete the workshop with the tools and practical skills required to conduct their own systematic review, meta-analysis, and meta-regression analyses. It is assumed that the workshop attendees will have basic skills in excel.
This workshop also aims to serve as a standard AEA Clincal Epidemiology and Research Synthesis Methods Special Interest Group (CERSSIG) event that usually occurs at the Annual Scientific Meeting and a special invitation has been sent to the CERSSIG members.
WHAT TO BRING: Attendees will need to have Microsoft Excel, meta-XL, and R pre-installed on their windows computers.