Bayesian variable selection in generalized linear models under cost constraints

Add to your list(s) Download to your calendar using vCal

• David Draper, University of California, Santa Cruz, USA
• Friday 08 February 2008, 14:15-15:00
• Large Seminar Room, 1st Floor, Institute of Public Health, University Forvie Site, Robinson Way, Cambridge.

If you have a question about this talk, please contact Nikolaos Demiris.

In the field of quality of health care measurement, patient sickness at admission is traditionally assessed by using logistic regression of mortality within (say) 30 days of admission on a fairly large number of sickness indicators (on the order of 100) to construct a sickness scale, employing classical variable selection methods to find an “optimal” subset of 10-20 indicators. In a world where electronic medical records are only now being slowly phased in and manual data abstraction from patient charts will still be used for some time (particularly in countries that are not on the cutting edge in medical informatics), such “benefit-only” methods ignore the considerable differences among the sickness indicators in cost of data collection. This issue is crucial when admission sickness is used to drive programs (now implemented or under consideration in several countries, including the US and UK) that attempt to identify substandard hospitals by comparing observed and expected mortality rates (given admission sickness). When both data-collection cost and accuracy of prediction of 30-day mortality are considered, a large variable-selection problem arises in which costly variables that do not predict well enough should be omitted from the final scale.

In this talk I’ll argue that there are three main ways to solve this problem: (1) a decision-theoretic cost-benefit approach based on maximizing expected utility, (2) an alternative cost-benefit approach based on posterior model odds, and (3) a cost-restriction-benefit analysis that maximizes predictive accuracy subject to a bound on cost. I’ll present details of as many of these methods as time permits, in the context of a large quality of care study of American patients hospitalized under the Medicare program. This is joint work with Dimitris Fouskakis and Ioannis Ntzoufras.

This talk is part of the MRC Biostatistics Unit Seminars series.

This talk is included in these lists:

• All Talks (aka the CURE list)
• BHRU Annual Lecture 2015
• BHRU Annual Lecture 2016
• Biology
• Cambridge Centre for Data-Driven Discovery (C2D3)
• Cambridge Immunology
• Cambridge talks
• Cardiovascular Epidemiology Unit Special Seminars
• Chris Davis' list
• Department of Public Health and Primary Care
• Graduate-Seminars
• Interested Talks
• Large Seminar Room, 1st Floor, Institute of Public Health, University Forvie Site, Robinson Way, Cambridge
• Life Sciences
• Life Sciences
• ME Seminar
• MRC Biostatistics Unit Seminars
• Primary Care
• PublicHealth@Cambridge
• Trust & Technology Initiative - interesting events
• bld31
• my_list
• ndk22's list
• ob366-ai4er
• other talks
• primary care
• rp587

Note that ex-directory lists are not shown.