Categories of Machine Learning Applications in Healthcare

Categories of Machine Learning Applications in Healthcare #

In the Stanford course and broader healthcare ML ecosystem, “practice of care” is just one of several conceptual categories used to frame ML applications. Each category reflects a different purpose and clinical context.

🧠 1. Diagnosis #

Question answered: What is wrong with the patient?

Machine learning identifies diseases or conditions based on input data.

Examples:

  • Detecting pneumonia from a chest X-ray (image classification)
  • Identifying arrhythmias from ECG waveforms
  • Classifying skin lesions as malignant vs benign

📈 2. Prediction / Prognosis #

Question answered: What will happen to the patient?

ML estimates future risk or outcomes to guide decision-making.

Examples:

  • Predicting 30-day readmission for heart failure
  • Estimating sepsis onset 6 hours ahead
  • Predicting 6-month stroke risk

⚙️ 3. Practice of Care #

Question answered: How should care be delivered?

Focuses on workflow, timing, triage, and care operations — not just predictions.

Examples:

  • Forecasting ICU nurse staffing needs
  • Scheduling follow-ups based on predicted readmission risk
  • Prioritizing emergency department queues

🧬 4. Discovery #

Question answered: What unknown patterns or mechanisms can we find?

ML is used as a research tool to uncover new relationships.

Examples:

  • Discovering novel genetic associations with disease
  • Identifying new disease subtypes through clustering
  • Finding gene-drug interaction patterns

🧩 Comparison Table #

Category Question Answered Data Types Used Output Example
Diagnosis What condition does this patient have? Imaging, labs, clinical notes Disease label
Prediction What is likely to happen next? Time-series EHR, vitals, labs Risk score
Practice of Care How should we act or plan care? EHR, operational data, predictions Workflow alert
Discovery What unknown pattern exists? Omics, clustering, NLP Scientific insight