1 Future NCHS-Linked Data Using Linked Data and Natural Language Processing to Support Patient-Centered Outcomes Research on Opioids Merianne Rose T. Spencer, M.P.H. National Center for Health Statistics Centers for Disease Control and Prevention
Presentation Outline 2 • Provide background about the HHS patient centered outcomes research trust fund (PCORTF) • Describe two 2018 PCORTF opioid projects led by NCHS • Highlight new data to be made available from these projects
Background of the PCORTF18 3 • The Patient Protection and Affordable Care Act of 2010 established the patient centered outcomes research trust fund (PCORTF) to help build the national capacity and infrastructure needed for research on outcomes and effectiveness of health care treatment and preventions for better healthcare decision-making • For the fiscal year 2018, HHS had approximately $25 million available for PCOR activities (i.e., PCORTF18)
NCHS OS-PCORTF Opioid Projects Selected FY18 4 • Strengthening the Data Infrastructure for Outcomes Research on Mortality Associated with Opioid Poisonings (CDC/NCHS) • Enhancing Identification of Opioid-Involved Health Outcomes with Linked Hospital Care and Mortality Data (CDC/NCHS)
5 Strengthening the Data Infrastructure for Outcomes Research on Mortality Associated with Opioid Poisonings (CDC/NCHS)
Introduction of Mortality Infrastructure 6 • The National Vital Statistics System (NVSS), which encompasses the processing and coding of vital records from 57 jurisdictions* • NCHS serves as a centralized hub to assign cause-of-death (ICD-10 codes) to death records received by states and disseminates the data for health statistics, surveillance and for research • Some data made available from NVSS includes: • NVSS public-use files (can be downloaded and accessed through online tools, such as CDC WONDER) • NVSS multiple cause of death mortality restricted use-files • National Death Index • Provisional estimates, including quarterly death rates and monthly drug overdose death rates *50 states, New York City, District of Columbia and 5 US territories (American Samoa, Guam, Norther Marianas, Puerto Rico, and Virgin Islands)
7 H IGH -L EVEL V IEW OF THE M ORTALITY S YSTEM D ATA I NFRASTRUCTURE Data collectio ion & exchange Data process ssing & data Dissemin inatio ion of mortalit lity y of death info forma rmatio ion coding data • Death occurs and death • NCHS receives birth and death • Releases data products (e.g., investigation may take place records from state vital records statistical files, provisional (e.g., interviews, toxicology, offices; estimates, National Death Index) ancillary tests), if injury-related • Processes and conducts data • Disseminates mortality statistics • Death is certified by death quality checks; via online platforms (e.g., CDC certifier (i.e., physicians, medical WONDER), conferences, • Codes and assigns cause(s) of examiners, coroners) presentations, statistical reports death; and other venues • State vital records offices submit • Develops and prepares statistical information to NCHS • Interacts with media, answers files and other data for analysis and responds to inquiries
8 O PPORTUNITIES FOR M ODERNIZING THE M ORTALITY S YSTEM I NFRASTRUCTURE Data collecti tion on & & Data proce ocessi ssing ng Disse semina mination on of exchang ange e of & data coding ng morta tality ty data death th informa rmatio tion • Improve data collection • Improve timeliness of how • Enhance public health mechanisms and exchange of NCHS processes and codes surveillance and reporting of death information at the death records received by states drug overdoses involving scene (i.e., between medical opioids • Develop capacity to better examiners/coroners, state vital identify and report on the • Increase access to data records offices, autopsy specific types of drugs involved necessary for public health centers, laboratory sites) in the deaths beyond ICD-10 preparedness and response • Improve infrastructure of how codes assigned for cause of • Support researchers and other states submit records to NCHS death end-users of data products
NCHS’ S A PPROACH TO PCOR P ROJECT , S TRENGTHENING THE D ATA I NFRASTRUCTURE 9 FOR O UTCOMES R ESEARCH ON M ORTALITY A SSOCIATED WITH O PIOID P OISONINGS Data col ollectio lection n & & Data process ocessin ing g & Dissemin inatio ion of excha hang nge of death h data codin ding mortali lity data inform ormatio ion • TASK 3 – Improve the • TASK 1 – Modernize the coding • TASK 1&2 (subtask) – Make exchange of drug information system with IT infrastructure supplemental drug data between state vital registration improvements (e.g., NLP, available to researchers systems and medical machine learning) • TASK 4 – Enhance the Vital examiner/coroner case • TASK 2 – Develop supplemental Statistics Rapid Release (VSRR) management systems drug information that would be program for improved public captured from death certificate health surveillance literal text • TASK 6 – Study effects of opioids (including birth outcomes) TASK 5 – Engage researchers throughout process to ensure that mortality data infrastructure improvements are • aligned with researchers ’ needs
End-Products 10 • Enhancements to improve data quality and to provide more specific information about drugs involved in the death: • NCHS’s public -use and restricted mortality statistical files released • National Death Index • Provisional estimates provided by the Vital Statistics Rapid Release program
11 Enhancing Identification of Opioid-Involved Health Outcomes with Linked Hospital Care and Mortality Data (CDC/NCHS)
Enhancing Identification of Opioid-involved Health Outcomes 12 with Linked Hospital Care and Mortality Data (NCHS) Project Goals: • Develop methods that identify specific opioids (e.g., fentanyl and heroin) involved in outcomes such as drug-related hospital visits and drug poisoning deaths using NCHS data: • National Hospital Care Survey (NHCS), the National Death Index (NDI), and the National Vital Statistics System restricted mortality data (NVSS-M) • Make these newly linked data files available to researchers • Share and disseminate about these data products and seek input from the research community
Data Sources 13 • National Hospital Care Survey (NHCS) • Nationally representative sample of 581 hospitals inpatient and emergency department visits that includes patient demographics, diagnoses, procedures, lab tests, medications, and emergency department clinical notes • National Death Index (NDI) • Confirms death occurred with select patients and includes cause-of-death information (International Classification of Diseases, tenth edition, ICD-10) • National Vital Statistics System Mortality Drug Overdose File (NVSS-M-DO) • A supplemental data file developed in collaboration with the FDA and NCHS, which contains identified specific substances mentioned on death certificate literal text and causes of death
Existing Gaps for Understanding Opioid-Involved 14 Hospital Visits • Hospital visits and cause-of-death information on opioid-involved outcomes are from two independent data sources • Opioid-involved hospital visits rely on claims diagnosis and procedure codes, which underrepresent the number of hospital visits resulting from opioid use • Cause of death is available to researchers using ICD-10 codes, which does not track specific drugs such as fentanyl or novel substances • Drug vocabulary for identifying new drugs and nomenclature is outdated
15 Solutions for Addressing Gaps • Link files in-house and leverage natural language processing to • Identify specific opioid agents involved in hospital visits and deaths • Build upon and establish an updated drug vocabulary for identifying new drugs and nomenclature that classifies novel substances • Analyze data files, create data files and publish findings about these new enhancements • Provide documentation and make datafiles available to researchers on the Research Data Center
16 Overview of Data Linkage & NLP Tasks National Hospital Care Survey (NHCS), National Death Index (NDI) and National Vital Statistics System Restricted Mortality File on Drug Overdose Deaths (NVSS-M-DO) will be linked in an iterative process: NHCS/NDI file (already linked as part of a previously funded PCORTF project) 1. will be linked to NVSS-M-DO (2014 & 2016) 2016 NHCS data contains EHR clinical notes. Use NLP on clinical notes to 2. identify opioid-involved ED visits not identified in medical codes. The NHCS/NDI/NVSS-M-DO file will be analyzed and used to establish 3. methods to better identify hospital visits involved with opioid poisonings. Findings will be published.
End Products 17 Enhanced data will be made available after the completion of the PCOR projects including: • Data files containing hospitalizations, cause of death and information on specific drugs involved (2014, 2016) • Enhanced drug vocabulary and methodologies for identifying opioid-involved outcomes (i.e., hospitalizations and death)
18 Thank You! Merianne Rose T. Spencer, M.P.H. Health Statistician, National Center for Health Statistics Centers for Disease Control and Prevention Email: kvd1@cdc.gov Phone: 301-458-4377
Recommend
More recommend
