Individual Patient Outcome Prediction

Table of Contents

• Individual Patient Outcome Prediction

  • Tabular Patient: Index

  • Sequential Patient: Index

  • Tabular: Example

  • Sequence: Example

Making individual outcome predictions is the basic AI for healthcare task. We need to specify the target response to predict, e.g., mortality , readmission , length of stay , transform the input trial patient records to either static descriptive features where is the number of features or sequential event features where is the number of visits, then predict the target using the processed data.

Such that, depending on the input patient data format: tabular or sequence, we have the following two subtasks: indiv_outcome.tabular and indiv_outcome.sequence.

Tabular Patient: Index

Here is the list of colab examples on each model for this task.

• Model 1: LogisticRegression

• Model 2: XGBoost

• Model 3: MLP

• Model 4: FTTransformer

• Model 5: TransTab

Sequential Patient: Index

Here is the list of colab examples on each model for this task.

• Model 1: RNN

• Model 2: RETAIN

• Model 3: RAIM

• Model 4: Dipole

• Model 5: StageNet

Tabular: Example

Here, we highlight the usage of indiv_outcome.tabular.transtab model for this task. Besides learning and predicting on a single tabular dataset, transtab shows promising performances on learning across different datasets. It sheds light on training a fundational model for clinical trials.

Sequence: Example

Here, we highlight the usage of indiv_outcome.sequence.rnn model for this task. Other more advanced models have the similar pipeline.