Hybrid analysis framework for prediction of outcomes in clinical trials

1. A method in a computing system, comprising:

obtaining information describing one or more completed clinical trials;

extracting features from the obtained clinical trial information;

splitting the extracted features into a first dataset of exclusively time-series data and a second dataset of exclusively non-time-series data;

using the first dataset of exclusively time-series data to train a time-series data model for predicting clinical outcomes;

using the second dataset of exclusively non-time-series data to train an interpretable non-time-series data model for predicting clinical outcomes;

obtaining information describing a subject patient;

splitting the information describing a subject patient into time-series patient data and non-time-series patient data;

applying the time-series data model to the time-series patient data to predict a first clinical outcome for the subject patient;

applying the interpretable non-time-series data model to the non-time-series data to predict a second clinical outcome for the subject patient;

generating a dashboard used to present one or more of: at least a portion of the information describing the subject patient and at least a portion of the obtained clinical trial information;

generating a combined clinical outcome by combining the first clinical outcome with the second clinical outcome; and

presenting the combined clinical outcome within the generated dashboard.

2. The method of claim 1 wherein predicting a clinical outcome for the subject patient comprises merging clinical outcome predictions produced by applying the trained models.

3. The method of claim 1 wherein the time-series data model comprises a recurrent neural network.

4. The method of claim 1 wherein the time-series data model comprises a long short term memory.

5. The method of claim 1 wherein the time-series data model comprises a hierarchical neural attention encoder.

6. The method of claim 1 wherein the interpretable non-time-series data model comprises a forest of gradient-boosted trees.

7. The method of claim 1 , further comprising causing the predicted clinical outcome to be displayed.

8. One or more memories collectively storing a compound patient survival prediction statistical model data structure, the data structure comprising:

a dataset of survival observations obtained from one or more cancer treatment clinical trials, the dataset of survival observations being split into time-series data and non-time-series data;

a first submodel trained to predict patient survival using survival observations obtained from one or more cancer treatment clinical trials, the first submodel being a cyclic neural network, the first submodel exclusively utilizing the time-series data; and

a second submodel trained to predict patient survival using survival observations obtained from one or more cancer treatment clinical trials, the second submodel being a forest of gradient-boosted trees, the second submodel exclusively utilizing the non-time-series data, such that the models represented in the data structure can be applied to information about a patient to predict survival of the patient; and

a dashboard presenting first output generated from applying the time-series data to the first submodel, second output generated from applying the non-time-series data to the second submodel, and one or more of: clinical trial data and patient data.

9. The one or more memories of claim 8 wherein the first submodel is a recurrent neural network.

10. The one or more memories of claim 8 wherein the first submodel is a long short term memory.

11. The one or more memories of claim 8 wherein the first submodel is a hierarchical neural attention encoder.

12. One or more memories collectively having contents configured to cause a computing system to perform a method, the method comprising:

obtaining information describing one or more completed clinical trials;

extracting features from the obtained clinical trial information;

splitting the extracted features into a first dataset of exclusively time-series data and a second dataset of exclusively non-time-series data;

using the first dataset of exclusively time-series data to train a time-series data model for predicting clinical outcomes;

using the second dataset of exclusively non-time-series data to train an interpretable non-time-series data model for predicting clinical outcomes;

obtaining information describing a subject patient;

splitting the information describing a subject patient into time-series patient data and non-time-series patient data;

applying the time-series data model to the time-series patient data to predict a first clinical outcome for the subject patient;

applying the interpretable non-time-series data model to the non-time-series data to predict a second clinical outcome for the subject patient;

generating a dashboard used to present one or more of: at least a portion of the information describing the subject patient and at least a portion of the obtained clinical trial information;

generating a combined clinical outcome by combining the first clinical outcome with the second clinical outcome; and

presenting the combined clinical outcome within the generated dashboard.

13. The one or more memories of claim 12 wherein predicting a clinical outcome for the subject patient comprises merging clinical outcome predictions produced by applying the trained models.

14. The one or more memories of claim 12 wherein the time-series data model comprises a recurrent neural network.

15. The one or more memories of claim 12 wherein the time-series data model comprises a long short term memory.

16. The one or more memories of claim 12 wherein the time-series data model comprises a hierarchical neural attention encoder.

17. The one or more memories of claim 12 wherein the interpretable non-time-series data model comprises a forest of gradient-boosted trees.

18. The one or more memories of claim 12 , the method further comprising causing the predicted clinical outcome to be displayed.