Bayesian Epidemiologic Screening Techniques

Evaluation and Application of Diagnostic Tests
(VME 217)

Instructor: Ian Gardner

Course Objectives: The course is directed towards graduate academic (PhD / MS) students in the health sciences. The course is designed to facilitate understanding of diagnostic test selection, design, evaluation, and interpretation, and is especially suitable for students who plan to use diagnostic tests in their research. At the completion of the course, students will be able to:

1. Design a study to estimate sensitivity and specificity of a new diagnostic test and statistically analyze the data from the study.
2. Critically evaluate the strengths and limitations of published papers involving diagnostic test evaluation and application.
3. Describe the factors that contribute to biased estimation of sensitivity and specificity.
4. Estimate receiver-operating characteristic curves and likelihood ratios using quantitative test data.
5. Determine optimal cutoff points for individual and aggregate test interpretation.
6. Critically evaluate testing strategies in a variety of clinical and epidemiologic settings.
7. Differentiate characteristics of tests at the individual and aggregate levels.
8. Understand the effects of test dependence in series and parallel testing schemes.
9. Adjust estimates from prevalence and risk factor studies for imperfect tests.
10. Estimate test accuracy and prevalence using latent class methods that don’t require a gold-standard.

The course will focus on serologic tests for infectious diseases but the principles and methods that will be discussed apply equally well to all types of tests.

 

Lectures:

(1) Introduction to test evaluation, review of basic concepts, measures of accuracy, etc.

(2-3) Test evaluation with a gold standard

• Case definitions (gold standards)
• Sampling designs and study design issues
• Bias, precision and sample size
• Logistic modeling of sensitivity and specificity
  

(4-5) Data analysis for studies with a gold standard

• Binary tests, incl. confidence intervals, statistical comparison of 2 binary tests
• Ordinal and continuous tests incl. methods for selection of a cutoff point, ROC curve estimation, statistical comparison of 2 or more ROC curves

(6-7) Evaluation of multiple tests with a gold standard

• Conditional independence and conditional dependence
• Measures of test dependence
• Effect of dependence on series and parallel test interpretation
  

(8) Critical evaluation of published studies in the medical literature

(9-10) Herd (cluster)-level test interpretation

• Factors affecting results of herd tests
• Effects of imprecision and bias in individual tests
• Herd classification: individual and pooled samples
• Critical evaluation of published studies
  

(11-15) Latent class methods for test evaluation without a gold standard

• Theory and assumptions
• Frequentist methods: 2 test in 2 population model
• Bayesian methods: 2 test in 2 population model
• Models for k tests in k populations
• Sample size guidelines

(16-17) Application of test results

• Clinical diagnosis:
• Predictive values of tests;
• Use of likelihood ratios in individual test interpretation
• Prevalence estimation from individual and pooled samples
• Frequentist methods
• Bayesian methods

 

Computer Laboratory Session:

1. Descriptive statistics, ROC analysis, confidence interval estimation
2. Frequentist and Bayesian approaches to estimation of sensitivity and specificity
3. Frequentist and Bayesian approaches to estimation of prevalence