# Workshop on bias in observational studies

## 11th -15th FEBRUARY  2019

This course covers 2 topics essential for producing valid results from observational data:

• bias (and quantitative bias analysis)
• use of Bayesian methods to incorporate bias correction into the study analyses

In the first section, we will review the three fundamental types of bias (confounding, selection bias, and information bias) including causes of the bias, approaches to preventing the bias and an evaluation of the potential impact these biases could have on study results. Given that not all biases can be prevented, it is important to know how to deal with biases which may affect a study. Two general approaches will be presented. Probabilistic quantitative bias analysis is a post-hoc approach which allows an investigator to apply knowledge about factors which may have biased a study in order to adjust observed estimates of effects (eg odds ratios) to remove the bias effects. While it does allow for adjustment for multiple biases and for uncertainty in bias parameters estimates, probabilistic quantitative bias analysis is usually applied to models that can be summarized by a 2x2 table.

Bayesian methods allow for incorporation of bias parameters directly during the analysis phase and, consequently, can be applied to more complex models. For instance multivariable logistic regression model (ie models with more than one predictor, including continuous predictors), mixed models (ie models with random effects), etc can be run with these. Using the Bayesian methods allows an investigator to compute unadjusted and bias-adjusted point estimate and 95% CI that will, hopefully, be closer to the true counterfactual effect. At the very least, it would allow for estimating the biases direction and magnitude.

Dr Ian Dohoo and Dr Simon Dufour

#### WHERE:

AHC 1.04 lab , Massey University, Palmerston North.

#### WHEN:

11th - 15th February 2019

#### Participant Presentations

One component of this course is to have participants present the results of either a bias analysis of a study of particular interest to them, or of a Bayesian analysis based on a data set of their own. Working on your own project is, by far, the best way to come to grips with new material. Although we will not have time for all participants to present, we would encourage everyone to do some work on a study / data set of interest to them.

Preparation

• Bring to the course either:
• a manuscript describing a risk factor study in an area of interest to you (this might be one that you have written or one from other authors). Note: it is best if you can identify a study with both a dichotomous exposure and outcome, but if this isn't possible we can work around that.
• or
• a data set in which you would like to incorporate some bias adjustment into the analysis Ideally, this should be a data set in which you have already fit a linear or logistic model)

Activity During the Course
During the course you will be given some time to work on either a bias analysis of your chosen study or a Bayesian analysis. You will be given some guidelines to facilitate this activity. Given the extent of the material we are hoping to cover in this course, most (but not all) of this available time will fall between 2 am and 4 am ... but who needs sleep anyway?

Presentations
Participants will be given the chance to discuss their findings with the class on Friday afternoon. These presentations are NOT supposed to be conference quality presentations but rather an opportunity to discuss what you have found and what obstacles you have encountered along the way.

#### Assumed knowledge:

This an advanced course that would best suit someone who has post-graduate qualifications, or experience, in the fields of public health or epidemiology. We will assume participants:

1. Understand the concepts of bias and error in the context of epidemiological studies,
2. Can run and interpret outputs from multivariable logistic regression models, and
3. Have a working knowledge of spreadsheet and either R or STATA.

#### Cost :

• Regular \$1560  GST inclusive
• Student (Non Massey) \$1350  GST inclusive
• Massey staff \$1355  (rate when paying by internal transfer)
• Massey student \$1000  (rate when paying by internal transfer)

Inclusions 5 days tuition, course material, morning and afternoon tea, lunch

#### Workshop Convenor

Enquiries for further information can be made to the Workshop Convenor, Chris Compton C.W.Compton@massey.ac.nz.

#### Timetable

(Tentative schedule)

Lectures in bold

 Day Time Contents Monday 8:30-9:30 Introduction to Course, Assignments / Presentations 9:30 - 10:00 Causal thinking 10:00 - 10:30 Tea/Coffee 10:30 - 12:00 Confounding - brief review, DAGS, intervening variables, methods of control 12:00 – 1:00 Lunch 1:00 - 2:00 Bias Exc. 1 – DAG, intervening variables, methods of control 2:00 – 3:00 Selection bias - brief review, types of sel. bias, general structure (Hernan's), magnitude 3:00 - 3:30 Tea/Coffee 3:30 - 4:30 Bias Exc. 2 - selection bias - using VER non-response spreadsheet 4:30 - 5:00 Information bias - brief review, differential/non-diff., magnitude 5:00 - 6:00 Individual help Tues. 8:30 - 9:00 Information bias - (cont) 9:00 - 10:00 Bias Exc. 3 - information bias - using VER misclassification spreadsheet 10:00 - 10:30 Coffee 10:30 - 12:00 Quantitative Bias Analysis (QBA) - simple bias analysis (unmeasured confounder) 11:00 - 12:00 Bias Exc. 4  - simple bias analysis - using QBA spreadsheets and -episensr- 12:00 - 1:00 Lunch 1:00 -  2:00 QBA - simple bias analysis (selection and misclassification) 2:00 - 3:00 Bias Exc. 5 - simple bias analysis - using QBA sel./info. bias spreadsheets and -episensr- 3:00 - 3:30 Coffee 3:30 - 4:30 QBA - Where do we get estimates of bias parameters? 4:30 - 5:00 QBA - What if I am uncertain of the bias parameter estimates? 5:00 - 6:00 Individual help Wed. 8:30 - 9:30 Bias Exc. 6 - multidimensional and probabilistic bias analysis - using QBA spreadsheets and -episensr- (cont.) 9:30 - 10:00 QBA - What if there are multiple sources of bias? 10:00 - 10:30 Coffee 10:30 - 11:30 Incorporating bias parameters into analyses 11:30 - 12:00 Case Example – Fluorosis in Australian wildlife 12:00 - 1:00 Lunch 1:00 - 2:00 Bias Exc. 7 - multiple bias analysis - using QBA spreadsheets and -episensr- 2:00 - 3:00 Introduction to Bayesian methods 3:00 - 3:30 Coffee 3:30 - 4:30 Bias Exc. 8 - Running a logistic regression model using OpenBUGS 4:30 – 5:00 Bayesian analysis (BA)– simple bias analysis (misclassification) 5:00 - 6:00 Individual help Thurs. 8:30 - 9:30 Bias Exc. 9 – Extending the Bayesian logistic response model to adjust for outcome misclassification using OpenBUGS 9:30 - 10:00 BA - Differential misclassification 10:00 - 10:30 Coffee 10:30 - 11:00 Bias Exc. 10 – Extending to control for differential misclassification using OpenBUGS 11:00 – 12:00 BA – Studies where a gold standard test is used on a subset of the sample 12:00 - 1:00 Lunch 1:00 - 2:00 Bias Exc. 11 – Bayesian models for controlling misclassification in studies where a gold standard test is used on a subset of the sample using OpenBUGS 2:00 – 2:30 BA - Adjusting for selection bias 2:30 – 3:00 Bias Exc. 12 – Bayesian models for controlling selection using OpenBUGS 3:00 – 3:30 Coffee 3:30 – 4:00 BA – Adjusting for an unmeasured confounder 4:00-5:00 Bias Exc. 13 – Bayesian models for controlling for an unmeasured confounder using OpenBUGS 5:00 – 6:00 Individual help Fri. 8:30 - 9:00 BA – Multiple biases 9:00 – 10:00 Bias Exc. 14 – Bayesian models for controlling for multiple sources of bias using OpenBUGS 10:00 - 10:30 Coffee 10:30 – 12:00 work on own data 12:00 - 1:00 Lunch 1:00 - 2:00 work on own data 2:00 - 3:00 Presentations / Discussions of participants projects 3:00 - 3:30 Coffee 3:30 – 4:30 Presentations / Discussions of participants projects 4:30 – 5:00 Course wrap-up

## Ian R. Dohoo

Dr. Ian Dohoo is an internationally known veterinary teacher and researcher (loosely translated this means he has been at it a long time). He is a Professor Emeritus of epidemiology at the University of PEI and is the first author of the graduate level epidemiology texts “Veterinary Epidemiologic Research” and “Methods in epidemiologic Research”. Numerous students around the world have participated in epidemiology courses he has taught. Most survived the experience. He has a particular interest in the advancement of epidemiologic methods, including those used in analyses of hierarchical data, survival analyses, and meta-analyses.

He has served both as the Director of the Centre for Veterinary Epidemiologic Research (CVER at UPEI) and as an Associate Editor of Preventive Veterinary Medicine, (but don't blame him if your paper was rejected).

## Simon Dufour

Dr Simon Dufour completed a DVM degree at the Université de Montréal in 1998. Following 10 years of professional happiness as a dairy practitioner in Québec and British-Columbia he finally saw the light and completed a PhD in epidemiology at UdeM. He is currently associate professor of epidemiology at UdeM and scientific director of the Canadian Bovine Mastitis Research Network, an organization responsible for mobilization of national and international scientific and financial resources to decrease the incidence of mastitis and maintain milk quality.

Dr Dufour’s interest for the epidemiology of bovine infectious diseases is just a cover for attracting research funds from the dairy industry (but please do not tell that to Dairy Farmers of Canada). His main interest is teaching epidemiology and the development of epidemiological methods, including new methodologies that can be used for quantitative bias analyses and control, and the impact and modulation of sampling strategies for controlling biases.

Massey Contact Centre Mon - Fri 8:30am to 4:30pm 0800 MASSEY (+64 6 350 5701) TXT 5222 contact@massey.ac.nz Web chat Staff Alumni News Māori @ Massey