Estimation of prevalence and incidence for use in descriptive epidemiology and risk assessment

 by

David Vose

Annual 3-day NOSOVE course

Wednesday, January 25^th – Friday, January 27^th, 2006.

 For a Printer-friendly version click here

Background

Prevalence and incidence of disease are key characteristics of the risk to, or from, a population in animal and human health risk assessment. They are used: to help determine the risk of animal (e.g. meat, semen) or human (e.g. blood) products; to demonstrate ‘freedom’ from a specific disease; and to track the health status of a population. The quantitative estimation of prevalence and incidence are derived from data in a variety of forms, and are complicated by factors like imperfect tests, non-random samples, regional variation, conflicting data, missing data, and weighting historic data.

 Risk analysis recognises that we will generally have incomplete information about the state of a population, and therefore need to describe the statistical uncertainty about our estimates. Standard epidemiological methods give a rich set of mathematical tools for providing point estimates of prevalence and incidence, and confidence intervals for some situations. However, the estimates of prevalence and incidence are usually used in more involved risk calculations requiring that we are able to specify the complete uncertainty distribution(s) of the estimated parameters in order to combine that uncertainty with others in a risk calculation to produce a realistic view of the risk.

 The course material

This course will provide you with a range of tools for estimating prevalence and incidence (and their uncertainty) appropriate for answering specific risk management questions such as: 

• What is the prevalence (incidence) in the region?
• Has the prevalence (incidence) increased (decreased), and by how much?
• Is the prevalence greater in one region than another?
• Can the region be characterised as enjoying ‘freedom’ from disease by OIE standards?

 We begin with an explanation of the necessary probability theory, focusing on the types of random behaviour (stochastic processes) relevant to estimating prevalence and incidence. A number of fairly simple class problems will help you see their relevance and introduce you to some useful software tools (@RISK/Excel and WinBUGS).

 We then take a look at some general statistical methods, starting with classical statistics (things like exact binomial and Poisson estimates, F-test and t-tests) which are most prevalent in the epidemiology texts), some common misunderstandings and how these tests can be reinterpreted to produce uncertainty distributions.

 Then we look at Bayesian methods, which offer us a more flexible and perhaps intuitive set of tools, and compare how these methods perform against classical statistics.

 Finally, we will apply this new knowledge with some diverse data sets to answer specific risk management questions.

 Who should attend?

Epidemiologists, statisticians, animal health or microbial or toxicological food safety risk analysts and risk managers who have some basic knowledge of spreadsheets and simulation modelling.

 Prerequisites

Most models are developed using Excel and @RISK because of the familiarity of Excel, though the material taught is equally applicable in other simulation environments. Given the short course duration, it is essential that all participants are reasonably proficient in Excel and have made themselves familiar with the basic principles of @RISK by going through the on-line tutorial available at http://www.palisade.com/training/risk45.html. WinBUGS will also be used to illustrate a couple of examples. It is freeware, not as user-friendly but rather unique in its capabilities. It is available for download (OpenBUGS homepage: http://mathstat.helsinki.fi/openbugs/ includes a link to the WinBUGS original pages) and has a help file that is worth browsing through. A short video tutorial on how to set a WinBUGS model running is also available (“WinBUGS – the movie”: http://www.statslab.cam.ac.uk/~krice/winbugsthemovie.html).

 Teaching philosophy

Vose Consulting courses aim to help participants understand (rather than 'learn') all the concepts covered, which can only be achieved through a relaxed, informal and interactive environment, through plenty of examples and hands-on exercises where course participants apply and adapt what they have learned.

------    -----

 

Course Schedule

 

Day 1 - (1/2 day)

Background and probability theory

• Definition of prevalence (a probability, a fraction?)

• Definition of incidence (a rate, probabilistic, a concentration?)

• Using estimates to answer risk management questions

• Some basic probability ideas

• Software 1: Monte Carlo simulation with @RISK/Excel

• The important stochastic processes in brief:

• The binomial process (for population prevalence)

• The hypergeometric process (for within-herd prevalence)

 

Day 2

Modelling and statistics 

• The important stochastic processes in brief (cont):

• The Poisson process (for incidence)

• Central Limit Theorem (to understand some statistical tests)

• Classical statistics methods

• Some examples

 

Day 3

More modelling with data

• Bayesian statistics methods

• Some examples

• Comparison of some Bayesian and classical methods

• Some examples

• Software (2): Markov Chain Monte Carlo with WinBUGS

• Some examples

• More example problems to try out these techniques with different data sets

 

 

 

 

About David Vose

 Risk analyst since 1988

Director of the Vose Consulting Group

Author: Risk analysis, ModelAssist, Vose Toolpaks

Worked in animal health and microbial risk for 10+ years in 26 countries

Main author OIE antimicrobial risk analysis guidelines

OIE animal health risk modelling guidelines based on Vose’s work

Editor/author of WHO/FAO microbial risk characterisation guidelines (currently under peer review)