[Turing-Southampton] S3RI seminar: Simon White, Thursday 2-3pm

[Ogden H.E.]

Dear all,

On Thursday (2 May) at 2pm in 54 / 7035 (7B), we have an S3RI seminar from Simon White (University of Cambridge) on "Bayesian inference of disease incidence in two-phase two-wave studies including loss to follow-up and death". Details are given below.

The talk will be followed by tea and cake in the staff reading room on level 4 of building 54.

All are welcome!

The talk planned for next week has had to be postponed, so there will be no seminar on 9 May. The next talk after this one will be on 16 May.

Best wishes,

Helen

Bayesian inference of disease incidence in two-phase two-wave studies including loss to follow-up and death: illustrated using dementia incidence from the UK MRC CFAS study

Simon R. White
Department of Psychiatry, University of Cambridge
MRC Biostatistics Unit, University of Cambridge

Estimating disease incidence has long been an important research question, with a variety of methods proposed for different settings and study designs; for example using back-calculation methods based on cross-sectional data. However estimating incidence is inherently a longitudinal problem and requires repeated observations over time. Longitudinal studies suffer from the problem of incomplete follow-up, where subjects do not return for subsequent waves. Of interest in our work is the need to consider the reason for incomplete follow-up and whether the missingness is ignoreable. Specifically, we consider the case of mortality in studies of dementia incidence. Mortality is an issue in any study of an ageing population, failure to account for this can induce a so-called immortal cohort bias.

Population-based studies often employ a two-phase design, with an initial screening interview used to select a subset for a more detailed assessment; this reduces the burden of the study and, ideally, focuses on the individuals of interest. However, this induces a further missing data mechanism, distinct from loss to follow-up and item non-response, which must be accounted for in the inference.

In this talk we present a Bayesian approach to estimating incidence from a population-based two-phase two-wave design, accounting for missing data due to several missingness mechanisms. Using a Bayesian framework we can naturally incorporate missingness mechanisms within the model.

To illustrate our method we consider the UK Medical Research Council's (MRC) Cognitive Function and Ageing study (CFAS), a population-based two-phase two-wave study of dementia in the UK. The CFAS design includes an initial screening test, with dementia status being part of the second detailed assessment. Hence, the issue of missing data through the two-phase design compounds the missing data due to loss to follow-up and death. Importantly, consent was obtained to independently monitor whether individuals within the study died and hence we can differentiate loss to follow-up and death.

For the current schedule of S3RI seminars, see https://tinyurl.com/s3riseminar<https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Ftinyurl.com%2Fs3riseminar&data=01%7C01%7C%7Ce42b1ef891ed4090313b08d6cc711899%7C4a5378f929f44d3ebe89669d03ada9d8%7C0&sdata=L%2F1IvjVvhYt2YgZ7assxyWKrMTesCbDjCLnNwt%2Fjr9U%3D&reserved=0>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.ecs.soton.ac.uk/pipermail/turing-southampton/attachments/20190429/059ae834/attachment.html