What is pmsampsize?

 

1. What is pmsampsize?

pmsampsize (Prediction Model Sample Size) is a statistical procedure used to calculate the minimum sample size needed to develop a reliable multivariable prediction model.

It was developed from research by Riley and colleagues on sample-size requirements for clinical prediction models.

Unlike traditional sample-size methods that focus on testing a single hypothesis, pmsampsize considers whether the dataset is sufficiently large to develop a model with an acceptable risk of overfitting.

It supports three types of prediction outcomes:

- Binary: Two possible outcomes, such as PHQ-9 ≥ 10 versus PHQ-9 < 10.

- Continuous: A numerical outcome, such as total PHQ-9 score.

- Time-to-event: An outcome involving the time until an event occurs.

For your NHANES capstone, you used the binary-outcome planning procedure because your machine-learning models predict whether a participant has elevated depressive symptoms.

2. Why is pmsampsize important for your capstone?

Your Research Question 3 investigates whether demographic, socioeconomic, lifestyle, and health-related factors can predict elevated depressive symptoms.

You are comparing Logistic Regression, Random Forest, and XGBoost.

Before developing these models, you need to answer:

Do we have enough participants and enough positive cases to develop a reasonably stable prediction model?

Simply having thousands of records is not sufficient. For example, if only 20 participants have the outcome, fitting a model with many predictors could produce unreliable results.

pmsampsize helps address this by considering outcome prevalence, model complexity, expected predictive performance, and overfitting.