VET412 Veterinary epidemiology II – Data analysis and interpretation

This course introduces key statistical methods used in veterinary epidemiology. Students learn to create descriptive summaries, assess diagnostic tests, evaluate data quality, and analyze prevalence surveys. The course covers basic linear and logistic regression, including checking model assumptions, interpreting results, and handling simple clustering structures.

Students are also introduced to essential concepts in causal effect estimation and prediction modelling, with emphasis on practical applications. Students learn to recognize different types of epidemiological data and select appropriate analytical methods.

The overall learning objective for this course is to enable students to design, conduct, and interpret relevant and sound statistical analyses for the most common veterinary epidemiological studies, and to recognize when more advanced methods are needed. This entails that:

The student is able to produce relevant and appropriate descriptive statistics and plots for epidemiological studies.

For studies evaluating measurements or diagnostic tests, the student should be able to

• apply statistical methods for evaluating agreement between binary, categorical, and continuous measurements.
• discuss accuracy and precision based on evaluations of agreement.
• use simple methods to estimate sensitivity and specificity.
• demonstrate knowledge of Bayesian methods for diagnostic test evaluation and discuss when their application is appropriate.

For studies evaluating the quality of databases, the student must be able to estimate measures of representativeness, completeness, timeliness, and validity.

For prevalence surveys, the student should be able to

• use simple techniques to adjust for selection bias.
• apply estimators that adjust for the accuracy of the diagnostic test used.

For regression linear and logistic models, the student will be able to

• understand how the models are fit (least squares and maximum likelihood).
• evaluate whether model assumptions are met and apply appropriate solutions when they are not.
• interpret model coefficients.
• estimate and interpret conditional and marginal predictions.
• model simple clustering structures or repeated measurements using random effects.

In studies aimed at causal effect estimation, the student should be able to

• assess whether the positivity assumption holds and discuss pragmatic solutions when it does not.
• construct, fit, and interpret regression-based estimators of estimands for time-invariant/single-time point exposures, and continuous and binary outcomes.
• understand the principles of sensitivity analysis and quantitative bias analysis, and discuss when such analyses are relevant.

For studies aiming to predict health, welfare, or productivity outcomes, the student should be able to

• obtain and interpret predictions with prediction intervals for continuous and binary variables using regression models.
• apply frameworks for training, modifying, and testing prediction models.
• compare different models based on fit statistics (e.g., R2R^2R2, information criteria, p-values).
• present and interpret measures of prediction accuracy and precision.
• demonstrate knowledge of more advanced algorithms (e.g., machine-learning algorithms).
• discuss when advanced algorithms are appropriate to apply.
• incorporate causal considerations into the construction of prediction models and discuss when this is relevant.

Students should be able to:

• recognize outcomes that are categorical (ordinal or nominal), counts, or time-to-event measurements, as well as time-varying exposures or complex hierarchical/repeated-measures structures
• identify methods appropriate for these types of data.