Abstract
This article discusses the establishment of interoperability among electronic medical records from 800 clinical sites and the use of machine learning for best practice discovery. A novel extraction-mapping algorithm is designed that accurately extracts, summarizes, and maps free text and content to concise structured medical concepts. Clinical decision processes and disease progression are also generated. The machine learning model (DAMIP) uncovers discriminatory feature sets that can predict the quality of treatment outcomes (blind prediction accuracies of 89% – 97%) for multiple diseases including heart, hypertension, and chronic kidney disease (CKD). For each disease, the best practice was used at fewer than 5% of the clinical sites, opening up excellent opportunities for knowledge sharing and rapid learning. This work led to the implementation of a new treatment policy for CKD pre-dialysis care management. The new policy offers better outcomes, saves lives, improves the quality of life, and reduces 35% of treatment costs. The system is scalable and generalizable.
TopIntroduction
Clinical decision making is complicated since it requires physicians to infer information from a given case and determine the best treatment based on their knowledge. Data from electronic medical records (EMRs) can reveal critical variables that impact treatment outcomes and inform the allocation of limited time and resources, allowing physicians to practice evidence-based treatment tailored to individual patient conditions. On a larger scale, realistically modifiable social determinants of health that will improve community health can potentially be discovered and addressed.
Although EMR adoption is spreading across the industry, many providers continue to document clinical findings, procedures, and outcomes with “free text” natural language in their EMRs. They have difficulty (manually) mapping concepts to standardized terminologies and struggle with application programs that use structured clinical data. This creates challenges for (multi-site) comparative effectiveness studies. Standardized clinical terminologies (e.g., SNOMED-CT, LOINC, RxNorm, UMLS) are essential to facilitate interoperability among EMR systems. They allow seamless sharing and exchange of healthcare information for quality care delivery and coordination among multiple sites. However, the volume and number of available clinical terminologies are large and expanding. Further, due to the increase in medical knowledge and the continued development of more advanced computerized medical systems, the use of clinical terminologies has extended beyond diagnostic classification.
This chapter summarizes our work in (1) designing an efficient, robust, and customizable information extraction and pre-processing pipeline for electronic medical records; (2) automatic mapping, standardization, and establishing interoperability; (3) uncovering best practices across multiple sites via machine learning; and (4) optimizing access timing and treatment decisions for chronic kidney disease patients Lee et al., 2016, 2019, 2021, 2022; Lee & Uppal 2020).
The work tackles over 800 clinical sites covering 9,000 providers and de-identified data for over 3.0 million patients with health records spanning the last 26 years. To the best of our knowledge, EMR data analysis across hundreds of sites and millions of patients has not been attempted previously. Such analysis requires effective database management, data extraction, preprocessing, and integration. In addition, temporal data mining of longitudinal health data cannot currently be achieved through statistically and computationally efficient methodologies and is still under-explored. This is a particularly important issue when analyzing outcome, health equity, and health conditions for chronic disease patients.
We first extract cohorts of patients from EMRs by disease / symptoms, and treatment features. Content discovery, concept mapping and interoperability are then established among EMRs from the 800+ clinical sites by developing a system that rapidly extracts and accurately maps free text to concise structured medical concepts. Multiple concepts and contents are extracted, and mapped, including patient diagnoses, laboratory results, medications, and procedures, which allows shared characterization and hierarchical comparison. A mixed integer programming-based machine learning model (DAMIP) is next applied to establish classification rules with relatively small subsets of discriminatory features that can be used to predict treatment outcomes for cardiovascular and chronic kidney diseases. Based on our results, optimal treatment design and associated new clinical practice guideline for chronic kidney disease pre-dialysis initiation is demonstrated. The results facilitate improved outcome, health quality, and cost-reduction for patients. Our findings can speed dissemination and implementation of best practice among all sites. Rapid learning across multiple sites show that improvement can be achieved within 12 months with a better health outcome, enhanced quality, and reduced cost. The next step will involve analyzing 60 million patients across the United States.
Key Terms in this Chapter
Health Insurance Portability and Accountability Act (HIPAA): A federal law that required the creation of national standards to protect sensitive patient health information from being disclosed without the patient's consent or knowledge.
Clinical Decision Support: An information technology system that provides clinicians, staff, patients, or other individuals with knowledge and person-specific information, intelligently filtered or presented at appropriate times, to enhance health and health care delivery.
Unstructured Text: Written content that lacks metadata and cannot readily be indexed or mapped onto standard database fields.
Evidence-Based Practice: A practice that has been rigorously evaluated via scientific evidence and shown to make a positive, statistically significant difference in important outcomes.
Electronic Medical Record: A systematized collection of patient and population electronically stored health and medical information in a digital format. These records can be shared across different health care settings.
Systematized Nomenclature of Medicine: A systematic, computer-processable collection of medical terms, in human and veterinary medicine, to provide codes, terms, synonyms and definitions which cover anatomy, diseases, findings, procedures, microorganisms, substances, etc.
Cardiovascular Diseases: Heart conditions that include diseased vessels, structural problems, and blood clots.
Chronic Kidney Disease: CKD, also known as chronic renal disease, is a condition characterized by a gradual loss of kidney function over time.
Machine Learning: A method of data analysis that automates analytical model building. It is a branch of artificial intelligence based on the idea that systems can learn from data, identify patterns, and make decisions with minimal human intervention.
Health Information Interoperability: The ability of two or more systems to exchange health information and use the information once it is received.
Dialysis: In medicine, dialysis is the process of removing excess water, solutes, and toxins from the blood in people whose kidneys can no longer perform these functions naturally. This is referred to as renal replacement therapy. The first successful dialysis was performed in 1943.
Standardization: The process of implementing and developing technical standards based on the consensus of different parties that include firms, users, interest groups, standards organizations, and governments.
Clinical Practice Guidelines: “Systematically developed statements to assist practitioner decisions about appropriate health care for specific clinical circumstances” (Field & Lohr, 1990 AU120: The in-text citation "Field & Lohr, 1990" is not in the reference list. Please correct the citation, add the reference to the list, or delete the citation. ) They can be used to reduce inappropriate variations in practice and to promote the delivery of high quality, evidence-based health care. They may also provide a mechanism by which healthcare professionals can be made accountable for clinical activities. The Institute of Medicine (IOM) (2012) defines clinical practice guidelines as “statements that include recommendations, intended to optimize patient care, which are informed by a systematic review of evidence and an assessment of the benefits and harms of alternative care options.”
Glomerular Filtration Rate: The kidney's filtration rate. It measures how well the kidney filters blood.