Paul Hitz

Managing protected health information in distributed research network environments: automated review to facilitate collaboration

BackgroundMulti-site health sciences research is becoming more common, as it enables investigation of rare outcomes and diseases and new healthcare innovations. Multi-site research usually involves the transfer of large amounts of research data between collaborators, which increases the potential for accidental disclosures of protected health information (PHI). Standard protocols for preventing release of PHI are extremely vulnerable to human error, particularly when the shared data sets are large.MethodsTo address this problem, we developed an automated program (SAS macro) to identify possible PHI in research data before it is transferred between research sites. The macro reviews all data in a designated directory to identify suspicious variable names and data patterns. The macro looks for variables that may contain personal identifiers such as medical record numbers and social security numbers. In addition, the macro identifies dates and numbers that may identify people who belong to small groups, who may be identifiable even in the absences of traditional identifiers.ResultsEvaluation of the macro on 100 sample research data sets indicated a recall of 0.98 and precision of 0.81.ConclusionsWhen implemented consistently, the macro has the potential to streamline the PHI review process and significantly reduce accidental PHI disclosures.

Enhancing Risk Assessment in Patients Receiving Chronic Opioid Analgesic Therapy Using Natural Language Processing

Objectives Clinical guidelines for the use of opioids in chronic noncancer pain recommend assessing risk for aberrant drug-related behaviors prior to initiating opioid therapy. Despite recent dramatic increases in prescription opioid misuse and abuse, use of screening tools by clinicians continues to be underutilized. This research evaluated natural language processing (NLP) together with other data extraction techniques for risk assessment of patients considered for opioid therapy as a means of predicting opioid abuse. Design Using a retrospective cohort of 3,668 chronic noncancer pain patients with at least one opioid agreement between January 1, 2007, and December 31, 2012, we examined the availability of electronic health record structured and unstructured data to populate the Opioid Risk Tool (ORT) and other selected outcomes. Clinician-documented opioid agreement violations in the clinical notes were determined using NLP techniques followed by manual review of the notes. Results Confirmed through manual review, the NLP algorithm had 96.1% sensitivity, 92.8% specificity, and 92.6% positive predictive value in identifying opioid agreement violation. At the time of most recent opioid agreement, automated ORT identified 42.8% of patients as at low risk, 28.2% as at moderate risk, and 29.0% as at high risk for opioid abuse. During a year following the agreement, 22.5% of patients had opioid agreement violations. Patients classified as high risk were three times more likely to violate opioid agreements compared with those with low/moderate risk. Conclusion Our findings suggest that NLP techniques have potential utility to support clinicians in screening chronic noncancer pain patients considered for long-term opioid therapy.

Managing personal health information in distributed research network environments

BackgroundStudying rare outcomes, new interventions and diverse populations often requires collaborations across multiple health research partners. However, transferring healthcare research data from one institution to another can increase the risk of data privacy and security breaches.MethodsA working group of multi-site research programmers evaluated the need for tools to support data security and data privacy. The group determined that data privacy support tools should: 1) allow for a range of allowable Protected Health Information (PHI); 2) clearly identify what type of data should be protected under the Health Insurance Portability and Accountability Act (HIPAA); and 3) help analysts identify which protected health information data elements are allowable in a given project and how they should be protected during data transfer. Based on these requirements we developed two performance support tools to support data programmers and site analysts in exchanging research data.ResultsThe first tool, a workplan template, guides the lead programmer through effectively communicating the details of multi-site programming, including how to run the program, what output the program will create, and whether the output is expected to contain protected health information. The second performance support tool is a checklist that site analysts can use to ensure that multi-site program output conforms to expectations and does not contain protected health information beyond what is allowed under the multi-site research agreements.ConclusionsTogether the two tools create a formal multi-site programming workflow designed to reduce the chance of accidental PHI disclosure.

PS2-37: Design Considerations and Population Identification for Implementing a Dental Virtual Data Warehouse: Feasibility, Table Structure and Data Elements.

Background/Aims The population seeking dental care at the Marshfield Clinic has grown substantially in recent years, partly as a result of increased capacity with the construction of new Marshfield Clinic dental clinics. The influx of additional dental data presents a great opportunity to enhance our dental research within the Marshfield Clinic Research Foundation and facilitate external collaborations. In order to collaborate in multi-site dental research studies, creation of a standardized data structure (analogous to the HMORN Virtual Data Warehouse) will be an enormous asset. Methods An initial table structure and mapping was proposed based on researcher needs and available data elements. Through collaborations with couple of other HMORN sites, common elements between sites were identified and adjustments were made to the proposed table layout. Patients seen at Marshfield Clinic dental facilities were included in the patient population. Results The table structure is currently composed of data comprising dental enrollment, dental medication orders, treatment plans, intervention, prescription and medical histories, tooth surface, and risk assessment tables. The Marshfield Clinic served more than 3,500 dental patients in 2007 – this number has increased to approximately 45,000 seen during 2011; we anticipate that our dental patient population will likely continue to grow, as further dental facilities are added (1 facility in 2003; 7 as of 2011; 2 more are planned for 2012). Of the dental patients seen in 2011, 70.3% (31,880/45,322) of them also had a medical visit in Marshfield Clinic’s integrated electronic health record system. Conclusions This dental table structure that will be shared across the sites involved will allow collaboration with these sites in future dental studies. The data elements used herein are common to dental care delivery throughout the country. The substantial overlap between dental and medical care in the Marshfield Clinic system offers great opportunities for Oral-Systemic research.

PS1-31: Assessing the Potential for Research on Genetics of Drug Induced Liver Injury in the HMORN

Background/Aims Drug Induced Liver Injury (DILI) is a major cause of liver failure in the US and the leading reason for failure of investigational drugs in clinical trials, lack of drug approval, and post-market withdrawal of approved drugs. Recent genome-wide association studies have identified variations within the major histocompatibility complex in Caucasians to be linked with flucloxacillin and lumiracoxib-related liver injury. The need for replication of these findings and extension of these investigations to other drug exposures and other ethnic groups will require substantial case numbers with supporting medical record documentation. With support from The Serious Adverse Event Consortium, an international consortium led by the pharmaceutical industry in conjunction with the FDA, we conducted a feasibility study to evaluate the potential for using electronic clinical and administrative data from two HMORN sites to identify provisional DILI cases. Methods Building upon previous research, we developed data specifications for electronic searches of ICD-9 codes with time proximate laboratory results indicative of liver-related disease. Electronic criteria were used to ‘rule out’ other liver diseases and other co-morbid conditions indicative of systematic liver-related effects. For feasibility testing, two methods of population identification were incorporated: the VDW and EMR reporting. All records from 1/1/00–8/1/09 that were identified as ‘potential DILI cases’ were reviewed manually, and selected data were abstracted, including suspected implicated drug(s). Results Records for 1,123,173 individuals were screened for potential case status; 29,893 records with one or more diagnoses of interest were identified. After application of the “rule out’ exclusion criteria, 584 potential DILI cases were reviewed and 99 ultimately met provisional case status. Drugs commonly associated with provisional DILI cases were sulfa-containing antibiotics, anticonvulsants, isoniazid, and statins. Roughly one-half of provisional DILI cases were associated with a single implicated drug. Conclusions Electronic infrastructures currently available within many highly integrated health care delivery systems can be efficiently leveraged to support identification of provisional DILI events. Systems that have comprehensive episodic, historic and post-event data, including lab, diagnostic, treatment, imaging, and medication records that can be accessed electronically should be considered primary systems for expanding DILI case accumulation efforts.