Biostatistics Group Seminar, Thursday 14th March

Advertisements for Seminars taking place on campus.
Forum rules
This forum is now unmoderated.

Biostatistics Group Seminar, Thursday 14th March

Postby rsc97 » Wed Mar 06, 2019 4:37 pm

Title: Meta-analysis of diagnostic test accuracy across the full range of possible cut-offs.

Speaker: Dr Hayley Jones, Bristol Medical School, University of Bristol.

Date: Thursday 14th March 13:00-14:00

Venue: Dinwoodie Lecture Theatre, David Weatherall Building
https://www.keele.ac.uk/connect/howtofi ... colour.pdf

Abstract: The optimum threshold or cut-off at which to operate a diagnostic test is usually a key question for clinical practice. Standard methods for meta-analysis of test accuracy don’t facilitate answering this question, since they (i) don’t provide summary estimates of accuracy across the full range of thresholds, and (ii) can only synthesise a single pair of sensitivity and specificity from each study, despite studies often reporting data at more than one threshold. Several models have recently been proposed for a unified meta-analysis of all available data, via assumptions about the underlying distributional form of test results. I will describe a multinomial random effects model, fitted in WinBUGS, which synthesises test accuracy data across studies reporting at different and varying numbers of cut-offs. The model allows for a flexible range of underlying distributions of test results, through estimation of a Box-Cox transformation parameter. I will present examples including meta-analysis of the accuracy of B-type natriuretic peptide in the diagnosis of acute heart failure. In a recent collaboration with researchers at McGill University and with Gerta Rücker (University of Freiburg), we compared results from three alternative models fitted to data from 45 studies reporting on accuracy of PHQ-9 in diagnosing major depression. Each model was fitted to two alternative versions of the data: (i) that available from the study publications, (ii) the full individual participant data (IPD). I will present results from this study, in particular showing that our model fitted to the ‘published’ dataset well approximated the IPD.
rsc97
 
Posts: 6
Joined: Fri Jul 31, 2015 3:17 pm

Return to Seminars

Who is online

Users browsing this forum: No registered users and 0 guests

cron