It is well known that Electronic Health Records (EHR) data contain inconsistent and inaccurate data, the effect of which on predictive model performance and risk/benefit factor identification are often neglected. This study investigates how varying levels of random and non-random binary differences, often referred to as "noise", affect modeling tools, such as logistic regression, support vector machines, and gradient boosting models. Using curated data from the All of Us database, we simulated different noise levels to mimic real-world variability. Across all models and noise types, increased noise consistently reduced classification accuracy. More importantly, noise diminished the variance of variable impact scores while leaving their means unchanged, suggesting a muted ability to identify key predictors. These findings imply that even modest noise levels can obscure meaningful signals. Measures like accuracy and hazard ratios may thus be misleading in noisy data contexts. The consistency of effects across models and noise mechanisms suggests this issue stems from inherent data variability rather than model brittleness, with broad implications for EHR data analyses.

Objective International Classification of Diseases (ICD) codes recorded in electronic health records (EHRs) are frequently used to create patient cohorts or define phenotypes. Inconsistent assignment of codes may reduce the utility of such cohorts. We assessed the reliability across time and location of the assignment of ICD codes in a US health system at the time of the transition from ICD-9-CM (ICD, 9th Revision, Clinical Modification) to ICD-10-CM (ICD, 10th Revision, Clinical Modification). Materials and methods Using clusters of equivalent codes derived from the US Centers for Disease Control and Prevention General Equivalence Mapping (GEM) tables, ICD assignments occurring during the ICD-9-CM to ICD-10-CM transition were investigated in EHR data from the US Veterans Administration Central Data Warehouse using deep learning and statistical models. These models were then used to detect abrupt changes across the transition; additionally, changes at each VA station were examined. Results Many of the 687 most-used code clusters had ICD-10-CM assignments differing greatly from that predicted from the codes used in ICD-9-CM. Manual reviews of a random sample found that 66% of the clusters showed problematic changes, with 37% having no apparent explanations. Notably, the observed pattern of changes varied widely across care locations. Discussion and conclusion The observed coding variability across time and across location suggests that ICD codes in EHRs are insufficient to establish a semantically reliable cohort or phenotype. While some variations might be expected with a changing in coding structure, the inconsistency across locations suggests other difficulties. Researchers should consider carefully how cohorts and phenotypes of interest are selected and defined.

Objective We developed a method to assess the consistency of the assignment of ICD codes, using coding performed at a United States health system at the time of the transition from ICD-9CM to ICD-10CM. Methods Using clusters of equivalent codes derived from the US Centers for Disease Control General Equivalence Mapping (GEM) tables, ICD assignments occurring during the ICD-9CM to ICD-10CM transition were evaluated in EHR data from the US Veterans Administration Central Data Warehouse, using a deep learning model based on 860 covariates. The model was then used to detect abrupt changes across the transition; additionally changes at each VA station were examined. Results Many of the 687 most-used code clusters had ICD-10CM assignments differing greatly from that predicted by the GEM from the codes used in ICD-9CM. Notably, the observed transition patterns varied widely across care locations. Conclusion Machine learning can model variability across time and across location, enabling an assessment of coding consistency. Expert review is not scalable, deep learning model applied to a large dataset of EHR records provides an approximation of ground truth.### Competing Interest StatementThe authors have declared no competing interest.### Funding StatementThis work was supported by VA HSRD grant 1I21HX003278-01A1, and by AHRQ grant R01 HS28450-01A1.### Author DeclarationsI confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained.YesThe details of the IRB/oversight body that provided approval or exemption for the research described are given below:The IRBs of the Veterans Administration Health Services Research Division and the George Washington University School of Medicine and Health Sciences have determined this research is exempt from review, as it involves deidentified data.I confirm that all necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived, and that any patient/participant/sample identifiers included were not known to anyone (e.g., hospital staff, patients or participants themselves) outside the research group so cannot be used to identify individuals.YesI understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance).YesI have followed all appropriate research reporting guidelines, such as any relevant EQUATOR Network research reporting checklist(s) and other pertinent material, if applicable.YesThe data is available through the US Veterans Administration.

Donald A.B. Lindberg M.D. arrived at the U.S. National Library of Medicine in 1984 and quickly launched the Unified Medical Language System (UMLS) research and development project to help computers understand biomedical meaning and to enable retrieval and integration of information from disparate electronic sources, e.g., patient records, biomedical literature, knowledge bases. This chapter focuses on how Lindberg's thinking, preferred ways of working, and decision-making guided UMLS goals and development and on what made the UMLS markedly "new and different" and ahead of its time.

OBJECTIVES The study sought to learn if it were possible to develop an ontology that would allow the Food and Drug Administration approved indications to be expressed in a manner computable and comparable to what is expressed in an electronic health record. MATERIALS AND METHODS A random sample of 1177 of the 3000+ extant, distinct medical products (identified by unique new drug application numbers) was selected for investigation. Close manual examination of the indication portion of the labels for these drugs led to the development of a formal model of indications. RESULTS The model represents each narrative indication as a disjunct of conjuncts of assertions about an individual. A desirable attribute is that each assertion about an individual should be testable without reference to other contextual information about the situation. The logical primitives are chosen from 2 categories (context and conditions) and are linked to an enumeration of uses, such as prevention. We found that more than 99% of approved label indications for treatment or prevention could be so represented. DISCUSSION While some indications are straightforward to represent, difficulties stem from the need to represent temporal or sequential references. In addition, there is a mismatch of terminologies between what is present in an electronic health record and in the label narrative. CONCLUSIONS A workable model for formalizing drug indications is possible. Remaining challenges include designing workflow to model narrative label indications for all approved drug products and incorporation of standard vocabularies.

INTRODUCTION:Computable biomedical knowledge artifacts (CBKs) are digital objects conveying biomedical knowledge in machine-interpretable structures. As more CBKs are produced and their complexity increases, the value obtained from sharing CBKs grows. Mobilizing CBKs and sharing them widely can only be achieved if the CBKs are findable, accessible, interoperable, reusable, and trustable (FAIR+T). To help mobilize CBKs, we describe our efforts to outline metadata categories to make CBKs FAIR+T.METHODS:We examined the literature regarding metadata with the potential to make digital artifacts FAIR+T. We also examined metadata available online today for actual CBKs of 12 different types. With iterative refinement, we came to a consensus on key categories of metadata that, when taken together, can make CBKs FAIR+T. We use subject-predicate-object triples to more clearly differentiate metadata categories.RESULTS:We defined 13 categories of CBK metadata most relevant to making CBKs FAIR+T. Eleven of these categories (type, domain, purpose, identification, location, CBK-to-CBK relationships, technical, authorization and rights management, provenance, evidential basis, and evidence from use metadata) are evident today where CBKs are stored online. Two additional categories (preservation and integrity metadata) were not evident in our examples. We provide a research agenda to guide further study and development of these and other metadata categories.CONCLUSION:A wide variety of metadata elements in various categories is needed to make CBKs FAIR+T. More work is needed to develop a common framework for CBK metadata that can make CBKs FAIR+T for all stakeholders.

This study set out to analyze questions about type 2 diabetes mellitus (T2DM) from patients and the public. The aim was to better understand people's information needs by starting with what they do not know, discovered through their own questions, rather than starting with what we know about T2DM and subsequently finding ways to communicate that information to people affected by or at risk of the disease. One hundred and sixty-four questions were collected from 120 patients attending outpatient diabetes clinics and 300 questions from 100 members of the public through the Amazon Mechanical Turk crowdsourcing platform. Twenty-three general and diabetes-specific topics and five phases of disease progression were identified; these were used to manually categorize the questions. Analyses were performed to determine which topics, if any, were significant predictors of a question's being asked by a patient or the public, and similarly for questions from a woman or a man. Further analysis identified the individual topics that were assigned significantly more often to the crowdsourced or clinic questions. These were Causes (CI: [-0.07, -0.03], p < .001), Risk Factors ([-0.08, -0.03], p < .001), Prevention ([-0.06, -0.02], p < .001), Diagnosis ([-0.05, -0.02], p < .001), and Distribution of a Disease in a Population ([-0.05,-0.01], p = .0016) for the crowdsourced questions and Treatment ([0.03, 0.01], p = .0019), Disease Complications ([0.02, 0.07], p < .001), and Psychosocial ([0.05, 0.1], p < .001) for the clinic questions. No highly significant gender-specific topics emerged in our study, but questions about Weight were more likely to come from women and Psychosocial questions from men. There were significantly more crowdsourced questions about the time Prior to any Diagnosis ([(-0.11, -0.04], p = .0013) and significantly more clinic questions about Health Maintenance and Prevention after diagnosis ([0.07. 0.17], p < .001). A descriptive analysis pointed to the value provided by the specificity of questions, their potential to disclose emotions behind questions, and the as-yet unrecognized information needs they can reveal. Large-scale collection of questions from patients across the spectrum of T2DM progression and from the public-a significant percentage of whom are likely to be as yet undiagnosed-is expected to yield further valuable insights.

Therapeutic intent, the reason behind the choice of a therapy and the context in which a given approach should be used, is an important aspect of medical practice. There are unmet needs with respect to current electronic mapping of drug indications. For example, the active ingredient sildenafil has 2 distinct indications, which differ solely on dosage strength. In progressing toward a practice of precision medicine, there is a need to capture and structure therapeutic intent for computational reuse, thus enabling more sophisticated decision-support tools and a possible mechanism for computer-aided drug repurposing. The indications for drugs, such as those expressed in the Structured Product Labels approved by the US Food and Drug Administration, appears to be a tractable area for developing an application ontology of therapeutic intent.

When patients cannot get answers from health professionals or retain the information given, increasingly they search online for answers, with limited success. Researchers from the United States, Ireland, and the United Kingdom explored this problem for patients with type 2 diabetes mellitus (T2DM). In 2014, patients attending an outpatient clinic (UK) were asked to submit questions about diabetes. Ten questions judged representative of different types of patient concerns were selected by the researchers and submitted to search engines within trusted and vetted websites in the United States, Ireland, and the United Kingdom. Two researchers independently assessed if answers could be found in the three top-ranked documents returned at each website. The 2014 search was repeated in June, 2016, examining the two top-ranked documents returned. One hundred and sixty-four questions were collected from 120 patients during 12 outpatient clinics. Most patients had T2DM (95%). Most questions were about diabetes (N = 155) with the remainder related to clinic operation (N = 9). Of the questions on diabetes, 152 were about T2DM. The 2014 assessment found no adequate answers to the questions in 90 documents (10 questions, 3 websites, 3 top documents). In the 2016 assessment, 1 document out of 60 (10 questions, 3 websites, 2 top documents) provided an adequate answer relating to 1 of the 10 questions. Available online sources of information do not provide answers to questions from patients with diabetes. Our results highlight the urgent need to develop novel ways of providing answers to patient questions about T2DM.

This chapter will not present a complete accounting of all terminology implementations. However, we will provide a survey of some of the more highly adopted terminologies. We will describe their scope and purpose, and we will discuss their implementation and the risks and benefits associated with employing each of these terminologies. In so doing we will provide examples of terminologies that can be used by healthcare informaticians to represent health knowledge and to use those representations for clinical decision support.

Although many see William Farr from the nineteenth century as the father of terminology and classification, we can find evidence in the work of Hippocrates that earlier efforts were underway and meaningful.

BACKGROUND CANCER SIGNIFICANCE AND QUESTION BioProspecting is a novel approach that enabled our team to mine genetic marker related data from the New England Journal of Medicine (NEJM) utilizing Systematized Nomenclature of Medicine-Clinical Terms (SNOMED CT) and the Human Gene Ontology (HUGO). Genes associated with disorders using the Multi-threaded Clinical Vocabulary Server (MCVS) Natural Language Processing (NLP) engine, whose output was represented as an ontology-network incorporating the semantic encodings of the literature. Metabolic functions were used to identify potentially novel relationships between (genes or proteins) and (diseases or drugs). In an effort to identify genes important to transformation of normal tissue into a malignancy, we went on to identify the genes linked to multiple cancers and then mapped those genes to metabolic and signaling pathways. FINDINGS Ten Genes were related to 30 or more cancers, 72 genes were related to 20 or more cancers and 191 genes were related to 10 or more cancers. The three pathways most often associated with the top 200 novel cancer markers were the Acute Phase Response Signaling, the Glucocorticoid Receptor Signaling and the Hepatic Fibrosis/Hepatic Stellate Cell Activation pathway. MEANING AND IMPLICATIONS OF THE ADVANCE This association highlights the role of inflammation in the induction and perhaps transformation of mortal cells into cancers. MAJOR FINDINGS BioProspecting can speed our identification and understanding of synergies between articles in the biomedical literature. In this case we found considerable synergy between the Oncology literature and the Sepsis literature. By mapping these associations to known metabolic, regulatory and signaling pathways we were able to identify further evidence for the inflammatory basis of cancer.

The majority of questions that arise in the practice of medicine relate to drug information. Additionally, adverse reactions account for as many as 98,000 deaths per year in the United States. Adverse drug reactions account for a significant portion of those errors. Many authors believe that clinical decision support associated with computerized physician order entry has the potential to decrease this adverse drug event rate. This decision support requires knowledge to drive the process. One important and rich source of drug knowledge is the DailyMed product labels. In this project we used computationally extracted SNOMED CT™ codified data associated with each section of each product label as input to a rules engine that created computable assertional knowledge in the form of semantic triples. These are expressed in the form of "Drug" HasIndication "SNOMED CT™". The information density of drug labels is deep, broad and quite substantial. By providing a computable form of this information content from drug labels we make these important axioms (facts) more accessible to computer programs designed to support improved care.

Clinicians involved in clinical care generate daily volumes of important data. This data is important for continuity of care, referrals to specialists and back to the patient's medical home. The same data can be used to generate alerts to improve the practice and to generate care activities to ensure that all appropriate care services are provided for the patient given their known medical histories using electronic quality (eQuality) monitoring. For many years we have used patient records as a data source for human abstraction of clinical research data. With the advent of electronic health record (EHR) data we can now make use of computable EHR data that can perform retrospective research studies more rapidly and lower the activation energy necessary to ask the next important question using electronic studies (eStudies). Barriers to these eStudies include: the lack of interoperable data between and among practices, the lack of computable definitions of measures, the lack of training of health professionals to use Ontology based Informatics tools that allow the execution of this type of logic, common methods need to be developed to distribute computable best practice rules to ensure rapid dissemination of evidence, better translating research into practice.

The authors sought to evaluate how well the Systematized Nomenclature of Medicine-Clinical Terms (SNOMED-CT) controlled vocabulary represents terms commonly used clinically when documenting posttraumatic stress disorder (PTSD). A list was constructed based on the PTSD criteria in the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV; American Psychiatric Association, 1994), symptom assessment instruments, and publications. Although two teams mapping the terms to SNOMED-CT differed in their approach, the consensus mapping accounted for 91% of the 153 PTSD terms. They found that the words used by clinicians in describing PTSD symptoms are represented in SNOMED-CT. These results can be used to codify mental health text reports for health information technology applications such as automated chart abstraction, algorithms for identifying documentation of symptoms representing PTSD in clinical notes, and clinical decision support.

BioProspecting is a novel approach that enabled our team to mine data related to genetic markers from the New England Journal of Medicine (NEJM) utilizing SNOMED CT and the Human Gene Onotology (HUGO). The Biomedical Informatics Research Collaborative was able to link genes and disorders using the Multi-threaded Clinical Vocabulary Server (MCVS) and natural language processing engine, whose output creates an ontology-network using the semantic encodings of the literature that is organized by these two terminologies. We identified relationships between (genes or proteins) and (diseases or drugs) as linked by metabolic functions and identified potentially novel functional relationships between, for example, genes and diseases (e.g. Article #1 ([Gene - IL27] = > {Enzyme - Dipeptidyl Carboxypeptidase 1}) and Article #2 ({Enzyme - Dipeptidyl Carboxypeptidase 1} < = [Disorder - Type II DM]) showing a metabolic link between IL27 and Type II DM). In this manuscript we describe our method for developing the database and its content as well as its potential to assist in the discovery of novel markers and drugs.

BACKGROUND:Optimal timing and treatment of patients with concomitant head, thoracic, or abdominal injury and femoral shaft fracture remain controversial. This study examines acute patient outcomes associated with early total care with intramedullary nailing (ETC group) versus damage control external fixation (DCO group) for multiple-injured patients with femoral shaft fractures. We propose DCO as a safe initial treatment for the multiple-injured patient with femur shaft fractures.METHODS:This study was a retrospective review of the trauma registry and multisystem organ failure registry data at a Level I trauma center. Two cohorts were identified to compare multiple-injured patients with femoral shaft fractures treated with early total care and damage control orthopaedic surgery. Primary outcome measures included mortality, pulmonary complications (adult respiratory distress syndrome [ARDS] score), transfusion requirements, and multiple organ failure (MOF score). Operative time, estimated blood loss, intensive care unit length of stay (LOS), and hospital length of stay (LOS) were also compared.RESULTS:During the study period, 462 patients with 481 femoral shaft fractures were identified. Of 462 patients with femoral shaft fractures, 97 met the inclusion criteria (42 ETC and 55 DCO). The DCO group had a significantly shorter operative time (22 minutes vs. 125 minutes) and less estimated blood loss from their operative procedure (37 mL vs. 330 mL). There was no significant difference between the groups for ARDS, lung scores, MOF, MOF score, intensive care unit LOS, or hospital LOS.CONCLUSION:Fracture fixation method did not have an impact on the incidence of systemic complications in multiple-injured patients with femoral shaft fractures. Although minimal differences were noted between DCO and ETC groups regarding systemic complications, DCO is a safer initial approach, significantly decreasing the initial operative exposure and blood loss.