Proven Processes

A number of important principles serve as the blueprint for the establishment of the Vanderbilt Institute for Clinical and Translational Research (VICTR) in Nashville, Tennessee. First, the needs of investigators are at the forefront of all activities and planning. In the implementation of every new program, consideration is given toward meeting the objective of making research initiation and conduct easier, faster, or of higher quality for investigative teams. In particular, we recognize that directly supporting a researcher's goals at the right time can be critical in his or her career. Second, through substantive institutional investments in informatics, infrastructure, and expertise, we are able to streamline administrative activities, make resources accessible, and capture key metrics in the course of operations. These informatics capabilities are applied to nearly all functions of VICTR. Software and systems developed internally and found by faculty to be valuable are made available to other institutions. One important paradigm … is the way in which the program leadership views patients: it is worthwhile to remember that we all engage the health care system as patients, if not yesterday or today, then at some point in the future. Third, we have transparency and are willing to share results of programs, especially those considered “experiments.” Novel, enterprise-wide initiatives are often themselves testable hypotheses; this reality requires continuous evaluation to make progress and requires a willingness to make mid-course corrections when things are not going as expected. Lastly, we inspire and harvest a sense of urgency throughout the enterprise, regularly focusing on the ultimate aim of clinical and translational research, which is to improve societal health through the transformation of how ideas and fundamental discoveries make their way from inception to clinical practice. One important paradigm in this regard is the way in which the program leadership views patients: it is worthwhile to remember that we all engage the health care system as patients, if not yesterday or today, then at some point in the future. Meharry Medical College, the largest private, comprehensive, historically black institution for educating health professionals and scientists in the United States, is our focused partner for the Clinical and Translational Science Awards (CTSA) program. This partnership has proven effective in stimulating new research collaborations at both Vanderbilt and Meharry, which is also in Nashville. VICTR has benefitted greatly from the expertise among Meharry faculty, including those related to community engagement. It is the collective vision that this partnership and the joint programs and services will facilitate and propel research aimed at eliminating health disparities among minority populations. The traditional approach to genetic research is to identify a cohort of patients who participate in studies to address a disease-specific research question. There are some disadvantages as well as advantages to this approach. Regardless, the time required to complete enrollment undoubtedly slows the pace of translation of new discoveries. Electronic medical records (EMRs) contain large populations with diverse diseases. Thus, use of these systems for research can rapidly and inexpensively create large, inclusive patient sets. BioVU, the Vanderbilt DNA databank, represents an example of a biorepository linked to a de-identified EMR. BioVU has been propelled intangibly by CTSA principles and has benefitted directly from CTSA resources, including pilot funding as well as CTSA-supported professional program management staff. BioVU is an enabling resource for exploration of the relationships among genetic variation, disease susceptibility, and variable drug responses, and represents a key first step in moving genomics from research to clinical practice. A major goal of the resource is to generate data sets that incorporate de-identified information derived from medical records (see our discussion of the Synthetic Derivative database below) and genotype information to identify factors that affect disease susceptibility, disease progression, and/or drug response. The scale of BioVU is large (51,000 samples currently and about 200,000 by 2011). Rapid accrual is a result of Vanderbilt's ability to de-identify the samples, which in part leads to a “non-human subject” designation with the Institutional Review Board (IRB), thus allowing the use of blood samples collected for clinical care and otherwise scheduled to be discarded. The program has received approval from the IRB and was reviewed by the federal Office for Human Research Protections, which agreed with the “non-human subjects” regulatory designation for both the resource and subsequent research. Program planning, which included patient research, community engagement, and extensive ethics committee involvement, started in 2004, and sample accrual began at the end of February 2007.1 Although the physical infrastructure and processing systems needed for this project are extensive, an equal amount of effort has gone into the steps taken to inform the Vanderbilt patient community of the nature and purpose of this project so as to educate, garner community support for, and measure the favorability towards this mission.2,3 This is a large-scale, multidisciplinary operation aimed to rapidly and cost efficiently support translational genomic research. Our internally developed sample acceptance system consists of a series of software applications interacting with one another and with external software and hardware. Its role is to determine whether a given blood sample will be accepted into the databank. The system has been tested and validated. Once a sample passes all criteria, it is accepted by the program. Acceptance triggers the encryption program to assign a unique research ID number. The Secure Hash Algorithm is a published and verified hash function employed for the databank that has the property of generating a unique output for every unique input; the input cannot be inferred by cryptanalysis of the output. Using the Vanderbilt medical record number as an input, the system generates a unique, 512-bit (128-character) code that serves as a Research Unique Identifier (RUI) and is used to relate samples with de-identified computer data. Validation protocols ensure that our version is performing to published encryption standards (i.e., it is not possible to infer or compute from the RUI the medical record number that generated it). Upon acceptance, DNA is extracted. DNA extraction relies on Autopure extraction protocols and systems currently utilized in the Vanderbilt DNA Core Resource. The sample acceptance system also recognizes if there are fewer than 10 μg DNA left in storage and selectively accepts further blood samples from these individuals, thereby allowing replenishment. Samples accrued to date have a wide range of chronic and acute conditions, based on associated ICD-9 codes that are extracted from the corresponding de-identified medical records. The top 10 chronic conditions of records (based only on ICD-9 codes) are shown in Figure 1. The extraction of these data takes only minutes. The most common diagnoses are hypertension (22.1%), hyperlipidemia (14.1%), type II diabetes (13.6%), anemia (12.4%), and hypercholesterolemia (10.2%). Most Common ICD-9 Diagnosis Codes in BioVU Notably, approximately 55% of samples currently in the repository (all are obtained from outpatients) also have clinical records associated with an inpatient stay. The data are shown here to characterize the varied composition of the overall resource, as well as the efficiency of structured data extraction (they do not relate specifically to the scientific program being proposed). Of those currently collected samples, 43% are male and 57% are female. The samples reflect the surrounding community; 83% Caucasians and 11% African American. Asian and Hispanic populations comprise 1% each. Records have a mean of 6.5 ± 6.2 years of history. Eighty-eight percent of records have at least one medication indicated, with an average of 8.0 ± 6.8 medications per record. In addition, 98% have one or more procedure codes. The Synthetic Derivative (SD) database is a research tool developed to enable studies with de-identified clinical data. The SD collection includes information extracted from the EMR systems and indexed by the same one-way RUI used to track samples. Content is changed by deletion or permutation of all identifiers contained within each record. The SD contains 1.7 million total records, with highly detailed longitudinal clinical data for approximately 1 million subjects, an average clinical record size of 106,727 bytes (about 30 pages of text), and an average of 13 distinct diagnostic codes (ICD-9) per record. The database incorporates data from multiple sources and includes diagnostic and procedure codes (ICD-9 and CPT); basic demographics (age, gender, race); text from clinical care including discharge summaries, nursing notes, progress notes, history and physical, problem lists, and multidisciplinary assessments; laboratory values; ECG diagnoses; clinical text and electronically derived trace values; and inpatient medication orders. All clinical data are updated regularly to include patients new to Vanderbilt University Medical Center (and therefore the SD), and to append new data to clinical records of existing patients as they continue to access care at Vanderbilt. Thus, the resource is entirely suitable for mining information relative to disease progression over time. Simple queries to the SD and BioVU resources (for example, how many records meet certain criteria and how many have DNA samples) are available by a user interface hosted on the Vanderbilt University Medical Center intranet. Search criteria may include structured data fields such as ICD-9 and CPT codes, demographics such as male or female or race, and even specific age ranges. Search criteria include the ability to search by specific medications or lab values, each of which can be used to include or exclude records returned. Unstructured text can be searched using specific keywords. With IRB approval, the user interface will also return data from individual records and provide text excerpts from the clinical record relative to user-specified keywords. In this way, records can be included or excluded from further analysis. An assessment of the regulatory review and approval process at Vanderbilt identified up to 20 applications, authorizations, or agreements potentially required prior to initiating research. Streamlining this process for investigators was an important area for improvement. An interactive informatics tool collecting the specific characteristics that generate the need for an associated approval was developed. The Customized Action Plan provides users with a series of questions about a research study; upon answering, they are presented with a printable list of required approvals and an electronic link to each application. The system uses PHP (version 5.2.6) and JavaScript scripting languages to present interview questions, customized help screens, and project-specific recommendations required for regulatory approval. User activity is automatically logged, and selected elements are transferred to dashboards for real-time evaluation.4 Screen shots can be seen at http://www.mc.vanderbilt.edu/victr/pub/newspub/vcap.html. In just this first year of existence, we are very pleased with the improvements we have made locally, some of which are beginning to be exported across the CTSA consortium (Table 1). In the true spirit of the CTSA initiative, the research infrastructure at Vanderbilt has substantially improved such that the investigators get their projects initiated more quickly and can focus on improving quality. There remains much to be done, especially in overall research management, implementation of the consortiums strategic goals, and conducting research on the research process itself. Vanderbilt is proud to be part of this grand experiment contributing to the national effort to transform the clinical and translational landscape. This work was supported in part by Vanderbilt CTSA grant 1 ULl RR024975from NCRR/NIH.