Existing Electronic Health Record (EHR) systems have limited functionality to find patients that can benefit by particular interventions. Hence, clinicians are unable to proactively reach-out to these patients, leading to an ever-widening evidence-care gap. The goal of this proposal is to improve healthcare delivery. The objective is to develop a data-framework for accurately finding patients that can benefits from large/population scale interventions. We will investigate two dimensions for improving the patient search?increasing the resolution of the search (aim 1) and increasing the quality of the data that the search is performed on (aim 2).
In aim 1 we will test the hypothesis that a rule-based model of the relevant clinical guideline, will find patients with higher accuracy than the conventional queries. The rational is that the rule-based approach will allow complex groupings of the eligibility criteria and will thereby provide a higher resolution of search than conventional query.
In aim 2 we will investigate whether machine learning (ML) can improve the accuracy of patient data and contribute to the accuracy of search results obtained through the conventional query and the guideline rule-base. In addition to the above aims, in aim 3 we will automate deployment of the rule-base and ML models and minimize the manual effort for developing ML models. This proposal builds on the lipid management program at Brigham and Women's hospital (BWH). Our work will be focused on finding patients for lipid- management; however, our methodology and tooling will be generalizable to other medical areas and institutions. The study team includes national experts in cardiology, machine learning, health information technology (HIT) and open-source software development. We will create an open-source software platform (i2b2-ML) by extending the popular `Informatics for Integration Biology and the Bedside' (i2b2) platform that we have developed and supported over the past 10 years and that is used by over 200 medical centers. I2b2-ML will extend i2b2's proven ability to characterize patient cohorts for research into the clinical realm. Our study will yield methodology for accurately finding patients that can benefit by clinical intervention and will thereby enable cost-effectiveness of population programs. It will directly help scale the lipid management program at BWH to benefit a wider patient population. Our detailed characterization of the gaps in lipid management at a large healthcare system will potentially inform improvements in the lipid management guideline. The methodology and tooling developed in this project will be disseminated in open-source for potential incorporation in the clinical data infrastructure at other institutions, which will facilitate implementation of population-scale clinical programs across the nation. In addition, the resultant infrastructure will serve as a platform for development of artificial intelligence-based applications to improve clinical care. These outcomes are expected to have a positive impact on health care delivery so that more patients will get the optimal care.
We propose developing Informatics for Integrating Biology and the Bedside (i2b2), a well-established, open source, platform that is currently used at over 200 hospitals and medical centers to use machine learning for identifying patient groups receiving suboptimal care. The proposed work will enable healthcare institutions across the United States to proactively reach-out to these patients, for delivering the appropriate clinical intervention. This pioneering research directly impacts public health by improving the quality of care.
