Machine Learning in Biostatistics · Lesson 1.1

What is machine learning in biostatistics?

Learn machine learning as a biostatistical prediction workflow: define the clinical question, identify the outcome, choose predictors, separate training and test data, run a first R model, interpret output and avoid causal overclaiming.

Download R script →Download dataset Open R lab

Conversational lecture

The first medical machine learning class begins.

Mr. R introduces machine learning through a clinical prediction problem rather than through algorithm names.

Scene: Mr. R walks into a computer lab where Emma, Oliver, James and Sophia are looking at a small patient dataset. The columns include glucose, BMI, age and diabetes status.

Emma

I keep hearing that machine learning can predict disease and identify high-risk patients. But in biostatistics, what does machine learning actually mean?

Mr. R

In biostatistics, machine learning means learning a prediction rule from health data. The model studies examples from patients whose outcomes are already known, then uses patterns in those examples to estimate an outcome for a new patient.

Oliver

So machine learning is mainly about prediction?

Mr. R

Very often, yes. We may ask whether routine clinical variables can predict diabetes status, hospital deterioration, treatment response, recurrence, survival or future complications.

James

Is that different from ordinary statistics?

Mr. R

There is overlap. Statistics and machine learning both learn from data. But the emphasis is often different. A traditional statistical model may ask how a variable is associated with an outcome. A machine learning workflow often asks whether a model can predict the outcome accurately for new observations.

Sophia

So if a model predicts diabetes well, does that prove what causes diabetes?

Mr. R

No. Prediction is not the same as causation. Glucose, BMI and age may help predict diabetes status, but this model alone does not prove that changing one predictor would cause the outcome to change.

Big idea

Machine learning in biostatistics is the use of data-driven models to predict health outcomes, classify patients, discover patterns or support medical decisions — but every model must be judged through clinical timing, validation, uncertainty, interpretation and usefulness.

Visual intuition

A prediction model maps patient information to estimated risk.

Patient data

GlucoseBMIAge

→

Model

Learns patternCombines predictors

→

Output

RiskClassUncertainty

The model does not magically understand medicine. It learns a mathematical pattern from examples. The biostatistical task is to check whether the pattern is valid, clinically timed, interpretable and useful.

Next lesson

Prediction, explanation and causal thinking.

Next, separate models that predict well from models that explain mechanisms or support causal claims.

Continue →