Machine learning applications using OMOP-CDM

Alexandros Rekkas

Introduction

https://rekkasa-presentations.github.io/ml-in-omop-summerschool-2024/

Introduction

Introduction

PatientLevelPrediction

Introduction

Problem

For patients hospitalized with pneumonia, what is their probability of death within the next 60 days?

Cohort definition

Cohort is a set of patients that fulfill a prespecified set of criteria for a duration of time.

Prediction setting

• Target cohort (T)
• Outcome cohort (O)
• Time at risk (TAR)

Settings

• Database details: The connection details to the database.
• Population: Further restrictions on our dataset.
• Split: The settings for the split (train/test).

Settings

• Restrict: Mostly temporal restrictions.
• Sample: Settings for sampling from the train set.
• Feature engineering: Settings for modifying covariates.

Settings

• Covariates: Covariates to be considered.
• Model: Settings for the model(s) to be trained.

Performance evaluation

Discrimination

How well can the model separate lower risk from higher risk patients?

Performance evaluation

Calibration

The agreement between observed and estimated probabilities of the occurence of the event or outcome.