The objective of this presentation is to explore the utility of a cloud-based system for automated collection of de-identified medical images, clinical reports, and patient outcomes for multicenter research projects.
Target Audience
Researchers interested in collaborating on multicenter studies related to cardiovascular MRI and/or CT.
Outcome/Objectives
Understand the underlying design and practical uses of a cloud-based system for automated collection of de-identified images, clinical reports, and patient outcomes.
Background
Reimbursement for cardiovascular MR and CT are strongly dependent on published evidence of patient outcome, such as predicting patient mortality. Assessing mortality has traditionally included the time-consuming task of contacting the patient directly, and growing concerns about patient privacy have made multicenter collaboration even more challenging. As a consequence, the evidence base for MR/CT patient outcome remains limited to small studies with short follow-up times and few hard events. Fortunately, however, U.S. Federal Law 45 CFR 46 46.101(b) provides that fully de-identified data is exempt from patient privacy regulations, and therefore can be aggregated across multiple centers.
Purpose
To create a fully-automated cloud-based platform to: 1) record structured data from electronically-signed clinical reports, such as calcium scores and infarct sizes; 2) continuously assess patient mortality based on the U.S. social security death index (SSDI); and 3) de-identify and transmit the full DICOM and clinical reporting system datasets to a cloud-based system to create a continuously-growing large-scale research database across multiple centers.
Methods
At each hospital, locally-installed software was used to author all final clinical reports, and data from the reports was stored in an internal relational database, such as ejection fraction, infarct size, and calcium scores. Patient mortality was evaluated by programmatically comparing the local patient identifiers to the United States Social Security Death Index (SSDI). Local systems then automatically de-identified all local data and securely transmitted the searchable, structured data to a cloud-based system. All DICOM datasets were also de-identified and transmitted to the cloud system.
Results
As of March 14, 2019, a total of 64,471 de-identified final reports and over 52 terabytes of de-identified DICOM images had been transmitted to the cloud system from 14 different hospitals. The total number of patients was 53,389 (some had multiple scans), of which 5,106 had subsequently died (9.7%). The browser-based user interface can be used to interactively search the structured reporting system data for research and educational purposes. Data from this system was used to publish a multicenter study of patient prognosis predicted by stress cardiac MRI consisting of 9,151 patients and 1,517 deaths [1].
Discussion
Because data are automatically harvested from routine clinical work, the number of patients grows quickly with little or no additional human effort needed to collect research data. Importantly, cloud-based data cannot be accessed without the approval of the originating site, meaning that multiple research groups can use the platform even if some groups prefer not to work with other groups. This platform facilitates the publication of large-scale publications that provide evidence needed to improve reimbursement for cardiovascular CT and MR.
Funding Sources
NIH-NHLBI R42 HL080843:
A Worldwide Research Network of Dynamic Cardiac Images
NIH-NHLBI R42 HL106864:
Software as a Service Cardiovascular PACS Based on Web 2.0 Technologies
NIH-NHLBI R42 HL117397:
Medical Images, HTML5, and Clinical Trial Remote Collaboration