With the advent of electronic health records, more data is continuously collected for individual patients and more data is available for review from past patients. Despite this, it has not yet been possible to successfully use this data to systematically build clinical decision support systems that can produce personalized clinical recommendations to assist clinicians in providing individualized healthcare. In this paper, we present a novel approach, Discovery Engine (DE) that discovers which patient characteristics are most relevant for predicting the correct diagnosis and/or recommending the best treatment regimen for each patient. We demonstrate the performance of DE in two clinical settings: diagnosis of breast cancer as well as personalized recommendation for a specific chemotherapy regimen for breast cancer patients. For each distinct clinical recommendation, different patient features are relevant; DE can discover these different relevant features and use them to recommend personalized clinical decisions. The DE approach achieves a 16.6% improvement over existing state-of-the-art recommendation algorithms in terms of kappa coefficients for recommending the personalized chemotherapy regimens. For diagnostic predictions, the DE approach achieves a 2.18% and 4.20% improvement over existing state-of-the-art prediction algorithms in terms of prediction error rate and false positive rate, respectively. We also demonstrate that the performance of our approach is robust against missing information, and that the relevant features discovered by DE are confirmed by clinical references.

Tremendous successes in machine learning have been achieved in a variety of applications such as image classification and language translation via supervised learning frameworks. Recently, with the rapid increase of electronic health records (EHR), machine learning researchers got immense opportunities to adopt the successful supervised learning frameworks to diverse clinical applications. To properly employ machine learning frameworks for medicine, we need to handle the special properties of the EHR and clinical applications: (1) extensive missing data, (2) model interpretation, (3) privacy of the data. This dissertation addresses those specialties to construct end-to-end machine learning frameworks for clinical decision support.

We focus on the following three problems: (1) how to deal with incomplete data (data imputation), (2) how to explain the decisions of the trained model (model interpretation), (3) how to generate synthetic data for better sharing private clinical data (synthetic data generation). To appropriately handle those problems, we propose novel machine learning algorithms for both static and longitudinal settings. For data imputation, we propose modified Generative Adversarial Networks and Recurrent Neural Networks to accurately impute the missing values and return the complete data for applying state-of-the-art supervised learning models. For model interpretation, we utilize the actor-critic framework to estimate feature importance of the trained model's decision in an instance level. We expand this algorithm to active sensing framework that recommends which observations should we measure and when. For synthetic data generation, we extend well-known Generative Adversarial Network frameworks from static setting to longitudinal setting, and propose a novel differentially private synthetic data generation framework.

To demonstrate the utilities of the proposed models, we evaluate those models on various real-world medical datasets including cohorts in the intensive care units, wards, and primary care hospitals. We show that the proposed algorithms consistently outperform state-of-the-art for handling missing data, understanding the trained model, and generating private synthetic data that are critical for building end-to-end machine learning frameworks for medicine.

Rationale

Objectives

Methods

Results

Conclusions