Statistical Models for "Prioritization"?

Question

Are there any general classes of statistical models that are able to perform "ranking and prioritization" tasks?

For instance, suppose a hospital has:

• data on patients (e.g. age, height, weight, etc.)
• the reason why they came to the hospital (e.g. chest pain, car accident, etc.)
• the triage decision associated with each patient who came to the hospital (e.g. ICU, sent home, etc.)
• what happened to that patient (e.g. died, almost died, full recovery, released from hospital, no real problem, etc.)

Are there any statistical models that can be used to "triage" patients?

The closest thing I have seen to this is "Learn to Rank" models - but I am not sure if this can be used for the problem I have described. Can a Machine Learning approach be used for this kind of problem?

Thanks!

Answer 1

Your problem setup calls for a causal prediction setup. In particular for unit-level ( individual) treatment effect (ute) estimation.

In your case

• data about incident and patients are features (X)
• triage decision is treatment (T)
• and 'what happened' are outcomes (Y)

Triage incoming patients according to predicted treatment effect and budget constraints ( rank ute).

A recent probabilistic neural network algorithm for this learning problem has been developed in "Predictive state propensity subclassification (PSPS)" by Kelly et al (2022) with implementation available at https://github.com/gmgeorg/pypsps.