Full course description
Video Series on Risk Prediction Models in Public Health
Training Launch Date: January 5, 2023
Berkeley Public Health and the California Department of Public Health have teamed up to provide advanced training in data science and biostatistics. This video series provides an introduction to risk prediction modeling. Learners should be familiar with multivariable logistic regression and basic epidemiologic measures to assess predictive ability of a diagnostic test (e.g., sensitivity, specificity, positive predictive value, negative predictive value).
Understand the role of risk prediction models in public health settings.
Learn to build and evaluate these models using logistic regression and techniques from machine learning.
Gain skills in model calibration, discrimination, and external validation.
Approximately 30 mins for the non-analyst version of the training and 1.5 hours for the analyst version of the training, which includes 45 minutes of hands-on practice material.
Tiered Certificates: Analyst and Non-Analyst Options
The training will offer tiered certificates to accommodate different levels of expertise and engagement with the material. The two tiers are designed to distinguish between those who engage with the course at a conceptual level (non-analyst) and those who apply the concepts in a hands-on, technical manner (analyst).
Learners pursuing the analyst certificate should be familiar with multivariable logistic regression and basic epidemiologic measures to assess the predictive ability of a diagnostic test (e.g., sensitivity, specificity, positive predictive value, negative predictive value).
Part 1: Risk prediction models: what and why
Upon completion of Part I, learners will be able to:
- Explain why risk prediction models are used in public health and clinical care settings.
- Distinguish between settings where risk prediction models are the appropriate analytical tool compared to causal/etiologic models.
- Provide examples of risk prediction models in different contexts.
Analyst Specific Objectives
Part 2: How to build a risk prediction model
- Upon completion of Part 2, learners will be able to:
- Build a risk prediction model using logistic regression in R.
- Use techniques from machine learning to inform a risk prediction model.
- Determine which types of variables, and how many, to include in a risk prediction model.
Part 3: How to determine if your risk prediction model is useful
Upon completion of Part 3, learners will be able to:
- Define the concepts of calibration and discrimination in the context of risk prediction models.
- Create a calibration plot to assess how well the model’s predictions perform.
- Calculate measures including area under the curve (AUC), sensitivity, specificity, positive predictive value, negative predictive value, positive likelihood ratio, and negative likelihood ratio, to assess how well the model discriminates between individuals who do and do not have an adverse outcome.
- Evaluate how well the model separates the population into distinct risk groups.
- Understand how external validation is used to evaluate a model’s performance.