Health Services Research & Development

Veterans Crisis Line Badge
Go to the ORD website
Go to the QUERI website

IIR 11-292 – HSR&D Study

New | Current | Completed | DRA | DRE | Portfolios/Projects | Centers | Career Development Projects

IIR 11-292
National Surveillance of Acute Kidney Injury Following Cardiac Catheterization
Michael E Matheny MD MS MPH
Tennessee Valley Healthcare System Nashville Campus, Nashville, TN
Nashville, TN
Funding Period: September 2012 - August 2016

Cardiac catheterization represents a significant medical diagnostic or treatment exposure risk for the development of acute kidney injury (AKI). That risk varies widely depending on the patient's pre-procedural medical conditions as well as exposures during or immediately before or after the procedure. Post procedural AKI occurs in 1% to 31% of the patients, depending on the cohort studied, and is associated with a 30% one-year mortality rate. For outcomes further downstream, AKI increases the risk of progressing to chronic kidney disease, which can lead to dialysis, increased cardiovascular adverse outcomes, reductions in quality of life, and significant personal and health care costs. Cardiac catheterization is a high risk, closely observed, and intervenable clinical care window, and preventing an occurrence of AKI would have significant impact on veteran's health and VA costs of care, which has been estimated to be approximately $7,500 per patient. However, automated outcomes surveillance is not widely performed, and the VA does not currently have the informatics tools to conduct this surveillance for the 40,000 veterans a year undergoing the procedure.

The overall objective of this project is to develop and evaluate the informatics infrastructure and tools to perform national VA near real-time automated adverse event surveillance after cardiac catheterization, and to demonstrate the utility of the tools within the use case of post-procedural AKI. More specifically, we will 1) develop and validate near real-time natural language processing (NLP) tools using interactive learning techniques in order to extract information that is relevant to AKI but is collected in structured data, 2) develop and validate a robust family of logistic regression prediction models for AKI following cardiac catheterization for use in the identification of high risk patients and populations, 3) conduct automated national retrospective and prospective analyses of institutional care variation among veterans receiving cardiac catheterization using novel surveillance methods, and 4) to conduct a survey of current patterns of care with regards to risk stratification and risk mitigation for post-procedural AKI.

This proposal will analyze retrospective and prospective cohort data from the VA Cardiovascular Assessment, Reporting, and Tracking for Catheterization Laboratories (CART-CL) voluntary clinical registry and electronic health record system (CPRS) from 2009 to 2015. All adult patients who received a cardiac catheterization in the VA during this time period will be included. All variables will be extracted from the structured data elements of CART-CL, CDW, and ViSTA, with near real time NLP used to extract risk factors from unstructured data within CDW and ViSTA. Risk factors will be identified by comprehensive literature review, expert consensus, and discovery during evaluation of retrospective signals, and selected through the use of the Lasso regression variable selection technique. Logistic regression models were developed for each of the Acute Kidney Injury Network AKI stages, internally validated with bootstrapping, and externally validated with the Northern New England Cardiovascular Disease Study Group percutaneous coronary intervention registry. Surveys were administered to the director of each cardiac catheterization center within the VA to determine current strategies for identification and risk mitigation for patients at risk for AKI. Institutional surveillance analyses were conducted using maximized sequential probability ratio testing.

Specific Aim 1 (NLP): A total of 1344 documents were double annotated across a set of concepts related to acute kidney injury: ACe, arb, diuretic, and nsaid medication exposure, enteral and parenteral intake, renal anatomic status, contrast exposure, and dialysis and transplant status. The final inter-annotator agreement across all concepts and attributes were 0.926, micro-averaged across the corpus. The data were split into training and testing data. The final fully trained NLP system had a micro-averaged precision, recall, and f-measure of 0.917, 0.907, and 0.911 across all concepts in the corpus. Using 10 parallel processes employing the RapTAT tool deployed on the LEO/cTakes framework on 1.26 million documents in the overall 90 day pre-catheterization corpus took a total of 806 minutes, which shows scalability.
Specific Aim 2 (Risk Models): The structured retrospective risk prediction model has been accepted for publication in JAHA 2015;e002136.
Specific Aim 3 (Automated Surveillance of Institutional Variation for Cath AKI): We evaluated hospital variation and simulate prospective surveillance for risk adjusted institutional outliers for acute kidney injury following cardiac catheterization within a retrospective national Veteran cohort. We used the national retrospective data from CART and CDW were integrated for patients receiving cardiac catheterization from 1/1/2009 to 09/31/2013. The outcome was Acute Kidney Injury (AKI) Network Stage 1 or greater AKI determined using the most recent creatinine in the prior year and 7 days following the procedure. Risk adjusted sequential probability ratio testing (RA-SPRT) was performed in each year for each center using an alerting threshold of an odds ratio (OR) of 2.0 and 0.5, alpha 0.05, and beta 0.1. Risk adjustment was conducted by developing sequential logistic regression risk models from 12 months prior to the scored month using 42 pre- and peri- procedural variables from among patient demographics, laboratory tests, IV fluids, medications, administrative codes, and the proportion of missing post-procedural creatinine per site/month. Observed/Expected (O/E) ratios with 95% confidence intervals for the study period were calculated. This analysis was performed using an open source Java statistical engine that supports prospective surveillance and verified using SAS 9.4. A total of 71 institutions and 111,995 catheterizations were analyzed for risk-adjusted AKI. The overall AKI event rate was 14.2%. 22 institutions had a statistically significantly lower-than-expected O/E, 38 were within expectation, and 11 had a higher-than-expected O/E for all years. The RA-SPRT analysis revealed that a total of 21, 7, 4, 2, and 0 institutions were at <0.5 OR and 8, 8, 2, 2, and 1 institutions were at > 2.0 OR for risk-adjusted AKI events in 1, 2, 3, 4, or 5 calendar years, respectively. Institutional variation for AKI following cardiac catheterization was substantial aft er extensive risk-adjustment using electronic health record and clinical registry data. These analyses are hypothesis generating and warrant additional root cause analyses into causes of variation at the outlying institutions.
Specific Aim 4 (Cath Director Survey): A total of 21/77 (27.3%) sites responded to the survey, with noted practice variation among risk assessment, stratification, and risk mitigation strategies. Most clinical factors were significantly different between responders and non-responders, and respectively. Detailed findings will be referenced following manuscript acceptance for publication.

This proposal has generated generalizable knowledge that can improve veterans' care in a number of areas. We discovered an effective risk prediction model for AKI following cardiac catheterization, which is being deployed in CART operational work with our operational partner. We have detected strong variation in care in risk adjusted AKI and identified low and high performing centers, that point to future work and opportunities in implementation and QI. Finally, the informatics infrastructure and NLP development has the potential to be applied in a wide variety of exposures and outcomes beyond AKI for cardiac catheterization surveillance.


Journal Articles

  1. Sauer BC, Teng CC, Tang D, Leng J, Curtis JR, Mikuls TR, Harrison DJ, Cannon GW. Persistence With Conventional Triple Therapy Versus a Tumor Necrosis Factor Inhibitor and Methotrexate in US Veterans With Rheumatoid Arthritis. Arthritis care & research. 2017 Mar 1; 69(3):313-322.
  2. Hsu JC, Maddox TM, Marcus GM. Overtreatment of Low-Risk Patients With Atrial Fibrillation-The Quality Coin Has 2 Sides-Reply. JAMA cardiology. 2016 Oct 1; 1(7):849.
  3. Siew ED, Parr SK, Abdel-Kader K, Eden SK, Peterson JF, Bansal N, Hung AM, Fly J, Speroff T, Ikizler TA, Matheny ME. Predictors of Recurrent AKI. Journal of the American Society of Nephrology : JASN. 2016 Apr 1; 27(4):1190-200.
  4. Sauer B, Teng CC, Burningham Z, Cannon G. Errata to NLP study of infusion notes to identify outpatient infusions in the VA. Pharmacoepidemiology and drug safety. 2015 Nov 1; 24(11):1225-6.
  5. Cronin RM, VanHouten JP, Siew ED, Eden SK, Fihn SD, Nielson CD, Peterson JF, Baker CR, Ikizler TA, Speroff T, Matheny ME. National Veterans Health Administration inpatient risk stratification models for hospital-acquired acute kidney injury. Journal of the American Medical Informatics Association : JAMIA. 2015 Sep 1; 22(5):1054-71.
  6. Tsai TT, Patel UD, Chang TI, Kennedy KF, Masoudi FA, Matheny ME, Kosiborod M, Amin AP, Weintraub WS, Curtis JP, Messenger JC, Rumsfeld JS, Spertus JA. Validated contemporary risk model of acute kidney injury in patients undergoing percutaneous coronary interventions: insights from the National Cardiovascular Data Registry Cath-PCI Registry. Journal of the American Heart Association. 2014 Dec 1; 3(6):e001380.
  7. Matheny ME, Peterson JF, Eden SK, Hung AM, Speroff T, Abdel-Kader K, Parr SK, Ikizler TA, Siew ED. Laboratory test surveillance following acute kidney injury. PLoS ONE. 2014 Aug 12; 9(8):e103746.
  8. Gobbel GT, Reeves R, Jayaramaraja S, Giuse D, Speroff T, Brown SH, Elkin PL, Matheny ME. Development and evaluation of RapTAT: a machine learning system for concept mapping of phrases from medical narratives. Journal of Biomedical Informatics. 2014 Apr 1; 48:54-65.
  9. Tsai TT, Patel UD, Chang TI, Kennedy KF, Masoudi FA, Matheny ME, Kosiborod M, Amin AP, Messenger JC, Rumsfeld JS, Spertus JA. Contemporary incidence, predictors, and outcomes of acute kidney injury in patients undergoing percutaneous coronary interventions: insights from the NCDR Cath-PCI registry. JACC. Cardiovascular interventions. 2014 Jan 1; 7(1):1-9.
Journal Other

  1. Siew ED, Matheny ME. Choice of Reference Serum Creatinine in Defining Acute Kidney Injury. Nephron. 2015 Sep 2; 131(2):107-12.

DRA: Health Systems, Cardiovascular Disease, Kidney Disorders
DRE: Epidemiology, Prevention
Keywords: Natural Language Processing, Surveillance, Adverse Event Monitoring, Care Management Tools, Healthcare Algorithms
MeSH Terms: none