Data-Mining Clinical Decision Support from Electronic Health Records Public Health Motivation: National healthcare quality is compromised by undesirable variability, reflected in different locales having anywhere from 20-80% compliance with evidence-based guidelines. Much of this is due to uncertainty, with half of clinical practice guidelines lacking adequate evidence to confirm their efficacy. This is unsurprising when clinical trials cost >$15 million to answer individual clinical questions. The result is medical practice routinely driven by individual opinio and anecdotal experience. While Big Data has revolutionized how society processes internet scale information, the status quo in clinical decision making remains the manual interpretation of literature and isolated decision aids. The adoption of electronic health records (EHR) creates a new opportunity to answer a grand challenge in clinical decision support (CDS). In a learning health system, we could automatically adapt knowledge from the collective expertise embedded in the EHR practices of real clinicians and close the loop by disseminating that knowledge back as executable decision support. Candidate Goals and Objectives: The unifying goal of this BD2K K01 proposal is the mentored career development of Jonathan H. Chen, MD, PhD. This proposal will accelerate his transition into an independent physician scientist, towards his long-term goals to produce Big Data technologies that answer such grand challenges in clinical decision support. His near-term objective is developing methods to translate EHR data into useful knowledge in the form of patient- specific, point-of-care clinical order recommendations for acute medical hospitalizations. His doctoral background in computer science gives him the technical capability to achieve these objectives, while his medical training will ensure clinically meaningful results. His preliminary work to build an order recommender, analogous to commercial product recommenders, demonstrates the proposal's overall feasibility. Institutional Environment and Career Development: The research facilities and training opportunities at Stanford University provide the ideal environment to achieve these objectives, with established and growing Centers for Biomedical Informatics Research, the Biomedical Data Science Initiative, and the first Clinical Informatics Fellowship accredited in the nation. Prof. Russ Altman, Director of the Biomedical Informatics Training Program, will lead a collaborative team of mentors with expertise in clinical decision support (Mary Goldstein), implementation science (Steven Asch), data-mining electronic health records (Nigam Shah), statistical learning algorithms (Lester Mackey), and healthcare statistics (Michael Baiocchi). Combined with respective didactic training, this mentorship will enable Dr. Chen to achieve his objectives through a series of research aims.
Research Aims : The overriding hypothesis of the proposal is that clinical knowledge reflected in clinical order patterns from historical EHR data can improve medical decision making when adapted into functional clinical decision support.
The specific aims each address components of this concept, as they seek to: (1) Develop the algorithms to learn clinical order patterns from historical EHR data, building on a preliminary recommender system; (2) Assess how underlying clinician proficiency affects the quality of those learned clinical order patterns through observational data inference against external standards; and (3) Determine the impact of automatically learned clinical decision support (CDS) on (simulated) clinical workflows through a randomized controlled crossover trial of human-computer interfaces with real clinicians. Expected Results and General Significance: By the completion of the proposed work, Dr. Chen will answer the grand challenge in clinical decision support (CDS) by automating much of the CDS production process, and have direct translational impact with a prototype system. This will advance the field with new paradigms of generating and disseminating clinical knowledge, which can then improve the consistency and quality of healthcare delivery. Additional benefits will include methods to identify and monitor areas of high practice variability for targeted optimization and improve predictive models that inform precision medicine. With this applied research experience and career development, Dr. Chen can compete for R01 funding and become an independent physician scientist developing Big Data approaches to solve national healthcare problems in clinical decision making.

Public Health Relevance

National healthcare quality is compromised by undesirable practice variability and medical uncertainty, with most medical practice routinely driven by individual opinions and anecdotal experience. With methods analogous to commercial product recommender systems, the proposed project will automatically learn patterns in raw clinical transaction data to capture the undocumented knowledge of real-world clinicians, and close the loop in a learning health system by disseminating that knowledge back as clinical decision support to improve patient care.

Agency
National Institute of Health (NIH)
Institute
National Institute of Environmental Health Sciences (NIEHS)
Type
Research Scientist Development Award - Research & Training (K01)
Project #
1K01ES026837-01
Application #
9044538
Study Section
Special Emphasis Panel (ZRG1-GGG-R (50))
Program Officer
Shreffler, Carol K
Project Start
2015-09-30
Project End
2020-07-31
Budget Start
2015-09-30
Budget End
2016-07-31
Support Year
1
Fiscal Year
2015
Total Cost
$178,606
Indirect Cost
$13,230
Name
Stanford University
Department
Internal Medicine/Medicine
Type
Schools of Medicine
DUNS #
009214214
City
Stanford
State
CA
Country
United States
Zip Code
94304
Li, Ron C; Garg, Trit; Cun, Tony et al. (2018) Impact of problem-based charting on the utilization and accuracy of the electronic problem list. J Am Med Inform Assoc 25:548-554
Marafino, Ben J; Dudley, R Adams; Shah, Nigam H et al. (2018) Accurate and interpretable intensive care risk adjustment for fused clinical data with generalized additive models. AMIA Jt Summits Transl Sci Proc 2017:166-175
Wang, Jason K; Hom, Jason; Balasubramanian, Santhosh et al. (2018) An evaluation of clinical order patterns machine-learned from clinician cohorts stratified by patient mortality outcomes. J Biomed Inform 86:109-119
Chavez, Gustavo; Richman, Ilana B; Kaimal, Rajani et al. (2018) Reversals and limitations on high-intensity, life-sustaining treatments. PLoS One 13:e0190569
Ouyang, David; Tisdale, Rebecca; Ashley, Euan et al. (2018) Acetaminophen or Tylenol? A Retrospective Analysis of Medication Digital Communication Practices. J Gen Intern Med 33:1218-1220
Garmire, Lana X; Gliske, Stephen; Nguyen, Quynh C et al. (2017) THE TRAINING OF NEXT GENERATION DATA SCIENTISTS IN BIOMEDICINE. Pac Symp Biocomput 22:640-645
Chen, Jonathan H; Asch, Steven M (2017) Machine Learning and Prediction in Medicine - Beyond the Peak of Inflated Expectations. N Engl J Med 376:2507-2509
Chen, Jonathan H; Goldstein, Mary K; Asch, Steven M et al. (2017) Predicting inpatient clinical order patterns with probabilistic topic models vs conventional order sets. J Am Med Inform Assoc 24:472-480
Chen, Jonathan H; Alagappan, Muthuraman; Goldstein, Mary K et al. (2017) Decaying relevance of clinical data towards future decisions in data-driven inpatient clinical order sets. Int J Med Inform 102:71-79
Ouyang, David; Chen, Jonathan H; Krishnan, Gomathi et al. (2016) Patient Outcomes when Housestaff Exceed 80 Hours per Week. Am J Med 129:993-999.e1

Showing the most recent 10 out of 14 publications