The purpose of this K25 proposal is to provide Dr. Benjamin Goldstein Ph.D., M.P.H., with the necessary protected time and additional training to develop as an independent, clinical biostatistician. This proposal has two key components: (1) an innovative research plan and (2) a comprehensive training plan. It is well recognized that patients undergoing hemodialysis (HD) are at increased risk of cardiac related events which often prove fatal. While substantive research has identified risk factors for these events, little work has been performed on forecasting their occurrence. The proposed research proposes to use existing electronic health record (EHR) data available through a collaborating dialysis center, DaVita Inc, to derive such a prediction model. EHRs contain detailed information on both a patient's health history (e.g. comorbidities, medications) as well as their evolving health statu (i.e. changes in health). A particularly unique aspect of the DaVita EHR system is the availability of real-time measures of health (e.g. blood pressure, pulse) available over the course of an HD session. Through our ongoing collaboration we will have data on 10,000s of individuals each with 100s of HD sessions, presenting the opportunity to analyze millions of dialysis sessions. Within this wealth of data two particular questions will be addressed: (1) How does a patient's hemodynamics vary over the course of and across HD sessions? (2) Can we derive a predictor for the near term onset of a cardiac event? To answer question 1, sophisticated statistical methodology, referred to as functional data analysis (FDA), will be utilized. Patterns of hemodynamic measures will be compared during and across HD sessions with key features extracted. For question 2, machine learning methodology will be used to derive a prediction model for the onset of cardiac events.
The final aim will be to assess the feasibility of applying such models within a clinical environment. As a Ph.D. biostatistician, Dr. Goldstein has many of the methodological and computational skills necessary to perform the proposed analyses. The proposed methods, while established, also have ample room for statistical investigation and will provide the basis for methodological research. He will be mentored by Dr. Bradley Efron, professor in the Stanford Department of Statistics, and a world recognized expert in statistical methodology. Serving as a consultant will be Drs. Trevor Hastie and John Ioannidis, fellow members of the department of statistics and experts in FDA and prediction evaluation respectively. The focus of Dr. Goldstein's training will be on developing his clinical expertise. This will be performed through a combination of didactic courses, one-on- one tutorials and clinical exposure. Dr. Wolfgang Winkelmayer, a clinical nephrologist and close collaborator of Dr. Goldstein, will supervise Dr. Goldstein's clinical knowledge development. He will be joined by Dr. Mark Hlatky, a research cardiologist, who will also provide mentorship with regards to the cardiac substance of the project. Additional consultants across the department of medicine will be used as needed. The proposed project will have a tremendous impact on Dr. Goldstein's career prospects. At the end of the 5 year period he will have begun the process of developing a research program in the analysis of EHR data. There will be ample avenues to pursue future studies, through the analysis of other predictor variables (e.g. biomarkers, psycho-social factors), outcomes (e.g. hospitalization, cost) and most importantly, implementation of the prediction models in the clinic. The clinical training period will provide him with the necessary background to succeed as a clinically-oriented biostatistician and develop as a leader in the field.

Public Health Relevance

Electronic health records have become increasingly common in clinical settings but to this point have been underutilized for research purposes. This project will analyze real-time EHR data that are generated during a dialysis session. The focus will be on understanding patients'hemodynamic changes over the course of a three hour dialysis session and using that information to derive a prediction model for severe cardiac events.

Agency
National Institute of Health (NIH)
Institute
National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK)
Type
Mentored Quantitative Research Career Development Award (K25)
Project #
1K25DK097279-01
Application #
8425985
Study Section
Diabetes, Endocrinology and Metabolic Diseases B Subcommittee (DDK)
Program Officer
Rankin, Tracy L
Project Start
2013-09-01
Project End
2018-07-31
Budget Start
2013-09-01
Budget End
2014-07-31
Support Year
1
Fiscal Year
2013
Total Cost
$161,180
Indirect Cost
$10,458
Name
Stanford University
Department
Internal Medicine/Medicine
Type
Schools of Medicine
DUNS #
009214214
City
Stanford
State
CA
Country
United States
Zip Code
94305
Phelan, Matthew; Bhavsar, Nrupen A; Goldstein, Benjamin A (2017) Illustrating Informed Presence Bias in Electronic Health Records Data: How Patient Interactions with a Health System Can Impact Inference. EGEMS (Wash DC) 5:22
Winkelmayer, Wolfgang C; Goldstein, Benjamin A; Mitani, Aya A et al. (2017) Safety of Intravenous Iron in Hemodialysis: Longer-term Comparisons of Iron Sucrose Versus Sodium Ferric Gluconate Complex. Am J Kidney Dis 69:771-779
Goldstein, Benjamin A; Pomann, Gina Maria; Winkelmayer, Wolfgang C et al. (2017) A comparison of risk prediction methods using repeated observations: an application to electronic health records for hemodialysis. Stat Med 36:2750-2763
Spratt, Susan E; Pereira, Katherine; Granger, Bradi B et al. (2017) Assessing electronic health record phenotypes against gold-standard diagnostic criteria for diabetes mellitus. J Am Med Inform Assoc 24:e121-e128
Goldstein, Benjamin A; Navar, Ann Marie; Carter, Rickey E (2017) Moving beyond regression techniques in cardiovascular risk prediction: applying machine learning to address analytic challenges. Eur Heart J 38:1805-1814
Goldstein, Benjamin A; Pencina, Michael J; Montez-Rath, Maria E et al. (2017) Predicting mortality over different time horizons: which data elements are needed? J Am Med Inform Assoc 24:176-181
Goldstein, Benjamin A; Navar, Ann Marie; Pencina, Michael J et al. (2017) Opportunities and challenges in developing risk prediction models with electronic health records data: a systematic review. J Am Med Inform Assoc 24:198-208
Assimes, Themistocles L; Goldstein, Benjamin A (2016) Genetic cardiovascular risk prediction: are we already there? Eur Heart J 37:3279-3281
Goldstein, Benjamin A; Thomas, Laine; Zaroff, Jonathan G et al. (2016) Assessment of Heart Transplant Waitlist Time and Pre- and Post-transplant Failure: A Mixed Methods Approach. Epidemiology 27:469-76
Goldstein, Benjamin A; Polley, Eric C; Briggs, Farren B S et al. (2016) Testing the Relative Performance of Data Adaptive Prediction Algorithms: A Generalized Test of Conditional Risk Differences. Int J Biostat 12:117-29

Showing the most recent 10 out of 50 publications