Prediction in healthcare must come with laser-sharp accuracy. CAR-T therapy is an immunotherapy that is capable of treating certain types of blood cancer. This therapy can be very effective, capable of achieving long lasting remission or even cure, it can come with serious neurological side effects. Accurately predicting which patients would benefit from CAR-T therapy is an extremely important mission, where even a small increase in true positives could mean lives saved. Since only a small percentage of people can benefit from the therapy, machine learning models used for prediction of the utilizers have only limited data to learn from.
The machine learning models trained on synthetic data outperformed the target by +2-3%. The overall accuracy remained high even after rebalancing: close to 98%. As the result of the improved ML performance, new patients were identified who could benefit from the therapy. Besides the improved health outcomes for these people, an estimated $8M+ savings in cost was achieved. Privacy checks were also passes and as a result, the training of the machine learning model was carried out without a data privacy risk.