Time varying confounding in medical research

Friday 11th March 2011

Venue
The meeting will be held at:

University Place,
The University of Manchester,
Oxford Road,
Manchester,
M13 9PL.

Directions to University Place at The University of Manchester
The University of Manchester is located on the south side of Manchester city centre and University Place is the focal building on the central Oxford Road campus (formerly The Victoria University of Manchester campus).

There are two large multi-storey car parks on site as well as a ground level car park and the campus is well served by public transport, being on the busiest bus route in Europe, and only 10 minutes walk from Manchester Piccadilly and Manchester Oxford Road stations.

Further directions to The University of Manchester by road, rail and air are available at:
http://www.manchester.ac.uk/aboutus/travel/

and University Place is Building 37 on the campus map at:
http://www.manchester.ac.uk/aboutus/travel/maps/az/

Registration
Pre-registration is ESSENTIAL. To register, please email Richard.emsley@manchester.ac.uk with the subject heading: “Time varying confounding”, and include your name, contact telephone number and whether you are vegetarian (for the lunch menu).

The cut-off for registrations is noon, Monday 7th March.

Cost
BIR members: £20; non-members £40; Students: £10.

Payment: post a cheque (payable to Biometric Society) to:

Dr Richard Emsley
Biostatistics, Health Sciences Research Group
School of Community Based Medicine
The University of Manchester
4.304 Jean McFarlane Building
Oxford Road
Manchester, M13 9PL

or pay on-the-door by cheque or cash

The cost includes buffet lunch and refreshments plus afternoon tea/coffee.

Programme		Document downloads for IBS members. Join us now.
12.00 - 13.00	Lunch
13.00 - 13.40	Methods for dealing with time-varying confounding Rhian Daniel (London School of Hygiene and Tropical Medicine)       Longitudinal studies, where data are repeatedly collected on subjects over a period of time, are common in medical research. When estimating the effect of a time-varying treatment or exposure on an outcome of interest measured at a later time, standard methods fail to give consistent estimators in the presence of time-varying confounders if those confounders are themselves affected by the treatment. Robins and colleagues have proposed several alternative methods which, provided certain assumptions hold, avoid the problems associated with standard approaches. They include the g-computation formula, inverse probability weighted estimation of marginal structural models and g-estimation of structural nested models. In this talk, we introduce the problem and give a brief description of each of the three methods.	Slides - Rhian Daniel
13.40 - 14.20	G-estimation in practice Kate Tilling (School of Social and Community Medicine, University of Bristol)     There is much current interest in the long-term effects of risk factors for cardiovascular disease and mortality, including the effects of factors such as smoking status which may change over time. The usual analysis of such associations may be biased because of confounding occurring over time. For example, obese people may be advised to quit smoking. Quitters therefore appear at higher risk than non-quitters. The confounder may also be on the causal pathway - smoking tends to decrease weight, which may decrease blood pressure. In this situation, exposure effects may instead be estimated using g-estimation. In this talk we show how G-estimation can be applied to repeated measures data, addressing issues of planned and unplanned censoring.	Slides - Kate Tilling
14.20 - 14.40	Tea/Coffee
14.40 - 15.20	Inferring the Effect of Dynamic Treatment Strategies from Observational Data Vanessa Didelez (School of Mathematics, University of Bristol) Sequential treatment strategies can essentially be of two types: either fixed in advance, e.g. "start treatment at t=3 and stop treatment at t=5", or dynamic, i.e. a function of the patient's disease progress, e.g. "start treatment as soon as the CD4 count drops below a given value". In this talk I will review the assumptions required to allow estimation of these two types of strategies from observational data. It turns out that assumptions for fixed strategies are slightly less restrictive than for dynamic ones. Directed acyclic graphs (DAGs) can be used to check these. However, usually we are interested in finding an "optimal" strategy for a given patient, and it is more plausible that this should be dynamic. Marginal structural models target marginal causal effects, and therefore lend themselves better to evaluate fixed sequential strategies; but they can be twisted to deal with dynamic strategies. This will be compared with a sequential Cox model, which targets a conditional causal parameter, and can easily be evaluated for dynamic strategies by including suitable interactions. The comparison will be illustrated with an application to data from the Swiss HIV Cohort Study. Dawid, A.P., Didelez, V. (2010).  Identifying the consequences of dynamic treatment strategies: A decision theoretic overview, Statistics Surveys, 4, 184-231. Gran, J. M.; Roysland, K.; Wolbers, M.; Didelez, V.; Sterne, J.A.C.; Ledergerber, B.; Furrer, H.; von Wyl, V.; Aalen, O.O. (2010). A sequential Cox approach for estimating the causal effect of treatment in the presence of time-dependent confounding applied to data from the Swiss HIV Cohort Study, Statistics in Medicine.	Slides - Vanessa Didelez
15.20 - 16.00	To Be Confirmed
16.00 - 16.30	Discussion Facilitated by Richard Emsley (University of Manchester)