From MESH, NLM, Ontologies, and IR to Health Informatics

Javed recounts that his journey into health informatics has multifaceted reasons. Initially, as an information scientist interested in information organization and classification, he was fascinated by biomedical informatics in the mid-1990s, influenced by the National Library of Medicine’s work on medical subject headings and the PubMed system.

“Intellectually, this was the root of my interest, and I started attending the American Medical Informatics Association (AMIA) meetings early in my career. Personal experiences also played a role. Caring for my parents and dealing with my father-in-law’s hospitalizations exposed me to inefficiencies in the healthcare system. This motivated me to improve health IT systems. Thus, both professional interests and personal experiences led me to focus on health informatics.”

Controlled vocabularies for healthcare have been in existence for centuries. The International Classification of Diseases, the first standard vocabulary, traces its lineage back at least to 1893. Medical Subject Headings (MeSH), created in the 1960s and updated regularly by the United States National Library of Medicine (NLM), is a comprehensive controlled vocabulary for indexing journal articles and books in the life sciences. It serves as a thesaurus that facilitates searching and is used by the MEDLINE/PubMed article database and by NLM’s catalog of book holdings. MeSH is also used by ClinicalTrials.gov to classify which diseases are studied by trials registered in clinical trials.

Cimino & Zhu (2006) trace the evolution of medical vocabularies as new methods of representing terminologies along the precepts of a branch of philosophy known as “ontology” and review some of the work in the ontological approach to controlled biomedical terminologies. Beginning with the most venerable ontology projects, GALEN (General Architecture for Languages, Encyclopedias, and Nomenclatures in Medicine) by a consortium of European universities, agencies, and vendors, the paper enumerates major ontological projects and tools such as the Unified Medical Language System (UMLS) of the US National Library of Medicine, the MED (Medical Entities Dictionary) at Columbia University, SNOMED CT, the world’s most comprehensive clinical terminology, the LOINC (Logical Observation Identifiers Names and Codes), Foundational Model of Anatomy (FMA) as a frame-based domain ontology, The Gene Ontology (GO), ISO Reference Terminology Model for Nursing Diagnosis (now known as Health Informatics — Categorial Structures for Representation of Nursing Practice in Terminological Systems), National Drug File Reference Terminology (NDF-RT), a formalization of the original VHA National Drug File, RxNorm developed at the National Library of Medicine, NCI Thesaurus by the US National Cancer Institute (NCI), and Protégé, a free, open-source ontology editor and framework for building intelligent systems.

Some of the biomedical ontologies that support drug discovery include OBI (Ontology for Biomedical Investigations), Cell Line Ontology, Vaccine Ontology, Ontology of Adverse Events which led to the Ontology of Vaccine Adverse Events extending both OAE and VO, and an Ontology of Drug Adverse Events (ODAE). In the late 1990s, the International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use (ICH) developed MedDRA, a rich and highly specific standardized medical terminology to facilitate sharing of regulatory information internationally for medical products used by humans. 

Over the last couple of decades, automated methods for extracting and applying medical vocabularies in the biomedical domain have evolved and advanced. With the advent of the web and the concomitant challenges of selecting relevant documents, there has been considerable research focus on information filtering systems.. An information filtering system aims to remove redundant or unwanted information using (semi) automated or computerized methods prior to presentation to a human user. Foltz, and Dumais (1992) analyzed various information filtering methods to personalize information delivery. After examining the relationships between Information Retrieval and Information Filtering, Belkin and Croft conclude that they are the two sides of the same coin.

Javed, in a paper coauthored with Lam(2000), demonstrated that automatic classification using supervised learning, incorporating various procedures and schemes to conduct document classification, produced satisfactory results. They suggested this approach to improve biomedical information retrieval and information filtering systems.

Asim et al. (2019) review modern ontology-based information retrieval methods for textual, multimedia, and cross-lingual data types, comparing and categorizing the most recent approaches used in information retrieval methods, outlining their major drawbacks and advantages.

EHR Adoption and the HITECH Act: Transformative Impacts on Healthcare

The healthcare industry has undergone rapid transformation, with manual operational systems being replaced by digital healthcare technologies such as personal health records, electronic prescriptions, smart health devices, wearable technologies, artificial intelligence-enabled patient relationship management, and telemedicine.

Electronic health records (EHR), previously called electronic medical records (EMR), are fundamental to e-health and have been discussed for decades. EHRs are crucial to advancing healthcare, improving efficiency and care coordination, and making it easier for health information to be shared between different entities involved. EHRs are real-time, patient-centered records that make information available instantly and securely to authorized users. EHRs are a vital part of health IT and are built to share information with other healthcare providers and organizations—such as laboratories, specialists, medical imaging facilities, pharmacies, emergency facilities, and school and workplace clinics—so they contain information from all clinicians involved in a patient’s care. While EHR adoption was quite low before HITECH, the Health Information Technology for Economic and Clinical Health (HITECH) Act of 2009 significantly accelerated the adoption of EHRs. The HITECH Act encouraged healthcare providers to adopt electronic health records and improve privacy and security protections for healthcare data. This was achieved through financial incentives for adopting EHRs and increased penalties for violations of the HIPAA Privacy and Security Rules.

According to Javed, the adoption of any technology, including EHRs, is influenced by various factors. He believes that the HITECH Act led to a rapid increase in EHR adoption, reaching nearly 100% in large medical centers. Historically, digitizing medical records dates to the 1960s when large academic centers first adopted EHRs to reduce errors and improve efficiency. The adoption of computing in healthcare also includes early applications in AI, such as the MYCIN system, a knowledge-based expert system for infectious diseases.

Javed further elaborated that the use of a “carrot and stick” approach proved highly effective. This approach, combining incentives and penalties, significantly boosted EHR adoption from just about 10% before 2009 to near-universal use in large centers. Different countries might require different models for EHR adoption, and a top-down approach might not always be best. Understanding the legislative framework’s role in promoting adoption is crucial, and interoperability remains a significant issue for EHR systems.

Javed highlights the importance of legislation like the HITECH Act, funding, and standards in driving the adoption of electronic health records (EHRs). He emphasizes the need for different approaches tailored to each country’s healthcare system.

Interoperability: Beyond Standards – The Economic and Political Dimensions

Over the last couple of decades, healthcare service providers worldwide have transitioned to electronic health information systems. A health information system integrates data collection, processing, reporting, and the use of information necessary for improving health service effectiveness and efficiency through better management at all levels. Patient care benefits from the availability of actionable, data-driven decision support standardized across care-delivering organizations. The sharing of information among different levels of healthcare is directly linked to the quality, efficiency, and safety of patient care.

Interoperability refers to the ability of systems to connect and exchange information without limitations, either in implementation or access. Torab-Miandoab et al. (2023) systematically studied the interoperability of health information systems, investigating the requirements for heterogeneous health information systems. They concluded that true interoperability in healthcare has not yet been achieved. The lack of interoperability between healthcare systems reinforces information silos akin to those in today’s paper-based medical files, leading to private ownership of health data. Achieving interoperability requires a large, multidisciplinary expert team and encompasses technical, syntactical, semantic, and organizational dimensions.

Responding to my question about specific strategies and technologies that have shown promise in addressing interoperability challenges in EHRs, Javed underscores the importance of standards. He states, “Without standards, you cannot have exchange and interoperability.” He observed, “The importance of standards in health informatics and biomedical informatics cannot be overstated. Researchers and practitioners in this field place a high value on standards development, adoption, and use. Standards such as Medical Subject Headings (MeSH) and SNOMED allow for consistent terminology and understanding, which are essential for interoperability. However, simply adopting standards isn’t enough. The success of information exchange also depends on the organizational, economic, and political contexts within which these systems operate. Even with well-developed standards, projects can fail due to these additional factors. Thus, while standards are the core of interoperability, attention must also be paid to the broader environment.”

Data Abundance, Genomic Data, and Precision Medicine

Biomedical research and medical practice are at an inflection point where the ability to describe and collect data is rapidly expanding. However, the efficiency of compiling, organizing, and analyzing this data to gain a true understanding of biological processes and insights into health and disease has not kept pace.

The widespread adoption of EHRs enables the collection of extensive data on demographics, symptoms, diagnoses, test results, treatments, and more. This includes high-throughput data from genomic, proteomic, and metabolomic analyses. Public datasets like the Cancer Genome Atlas and the 100,000 Genomes Project offer valuable baselines for comparison, although such comparisons are rarely made. The volume, depth, and speed of data collection are unprecedented in human history.

Precision medicine (also referred to as personalized medicine) is an innovative approach that uses information about an individual’s genomic, environmental, and lifestyle information to guide decisions related to their medical management. The goal of precision medicine is to provide a more precise approach for the prevention, diagnosis, and treatment of disease.

This brings us to the question of how much genomic data should be included in EHRs and how to effectively integrate genomic data with clinical data. In response to my query, Javed expresses mixed feelings about further integrating genomic data into EHRs despite being a technologist at heart. He highlights challenges not only with the integration process itself but also with utilizing the integrated data effectively. In the current fast-paced healthcare environment, especially in the US and Canada, the time allocated for diagnosis and treatment planning is very short. This raises concerns about whether additional data will be effectively used in practice. While genomic information is crucial, integrating it seamlessly into EHRs poses significant challenges. Genomic information is undeniably important for health. For instance, studies on the drug Metformin revealed adverse reactions in certain populations through digital records, leading to the discovery of genetic reasons for these reactions. Immediate benefits of integrating genomic data might include catching bad drug interactions during prescription entry. However, there are significant practical challenges, such as ensuring accurate collection and integration of genetic data into EHRs.

Despite moderate successes in areas like cancer research, especially breast cancer and genomics, the widespread and effective use of genomic data in routine clinical practice remains a long way off, concluded Javed.

Smart Healthcare: The Privacy and Precision Paradox in Medicine

The rapid development of Internet of Medical Things (IoMT) technology has advanced smart healthcare by enhancing real-time services and patient monitoring. However, privacy and data protection concerns, particularly involving multiple stakeholders throughout the data lifecycle, hinder its widespread adoption. Following a systematic review of the literature, El Majdoubi et al. (2022) propose a blockchain-based solution for privacy preservation in IoMT, addressing these limitations. Privacy in the age of medical big data poses many challenges including the legal and ethical challenges big data brings to patient privacy.

Precision medicine (PM) aims to offer more targeted treatments and prevention strategies, reducing human suffering, health disparities, and healthcare costs. Genetic risk profiling, imaging technologies, and mass spectrometry promise to enhance individualized disease prevention by detecting pre-disease stages before symptoms appear. Many countries are investing in technologies and data infrastructures to advance PM, aiming to better tailor disease treatment and prevention to individual patients.

The Precision Medicine Initiative (PMI), launched in January 2015 in the US, aims to enhance individualized patient care through biomedical discoveries and new clinical tools. The All of Us Research Program, part of PMI, emphasizes privacy and trust, focusing on governance, transparency, participant empowerment, respect for preferences, data sharing, access, use, and quality. The International Consortium for Personalised Medicine (ICPerMed) brings together almost 50 European and international partners, representing ministries, funding agencies, and the European Commission (EC). Together, they work on coordinating and fostering research to develop and evaluate personalized medicine approaches.

Javed discusses the paradox of balancing privacy with the benefits of precision medicine. He opines that this is an enormously complex issue extending beyond medicine to any type of information use, such as education. He said, “As someone who has extensively worked on personalization, I’ve pondered this paradox for a long time. It’s a significant challenge because you can’t have both maximum personalization and maximum privacy at the same time.”

Addressing this paradox requires social, organizational, and cultural changes. Pragmatically, there must be an adaptation to the idea that some privacy trade-offs are necessary to gain certain benefits. However, this cultural shift requires time, understanding, and education.

Javed noted that people frequently share personal information without realizing it yet become more protective of their privacy when it comes to health and education.

“In terms of precision and personalized medicine, while there’s tremendous promise, we must be cautious about creating unintentional barriers. These advanced medical approaches can be costly and potentially make healthcare less accessible to many people. This could inadvertently widen the digital divide. It’s essential to consider the digital divide and ensure these innovations don’t exacerbate existing inequalities. While precision medicine holds great potential, we must think carefully about how to make it broadly available and consider the cultural balance required. It’s important to be cautiously optimistic about the benefits of precision medicine rather than overly enthusiastic.”

Wearable Devices, Remote Monitoring of Health, and Ambient Intelligence

Remote patient monitoring utilizes wearable devices to collect and transmit real-time health data to healthcare providers. This enables professionals to monitor conditions remotely, enhancing access to care, early detection of abnormalities, and timely interventions. Patients benefit from the convenience of staying connected to their healthcare teams from home. Research has demonstrated that wearable systems have great potential in disease diagnosis, therapy, and drug delivery. In another episode of InfoFire, Maybury cited Foresite as an example. Foresite provides patient care and eldercare using various inputs, including depth-sensor technology, under-mattress pads, and motion detectors to capture a range of information, such as respiratory rate, bed restlessness, gait, motion, and activity.

There are myriad applications of Ambient Intelligence in Healthcare as it aims to make explicit input and output data collection devices redundant in smart cities. Instead, sensors, processors, and actuators embedded in everyday objects will capture and process data. Advances in machine learning and contactless sensors have led to ambient intelligence—physical spaces sensitive and responsive to human presence. Passive, contactless sensors embedded in the environment can create an ambient intelligence that is aware of people’s movements and can adapt to their ongoing health needs. Despite the promise of ambient intelligence to improve the quality of care, the continuous collection of large amounts of sensor data in healthcare settings presents numerous challenges. These challenges range from sensor technologies, data management, technical model development, and causal inference to ethical, privacy, bias, and fairness issues. The Ambient Intelligence for Healthcare (AmI4HC) workshop held in Vancouver, Canada in 2023, organized by Stanford University, aimed to shed light on this novel technology.

In response to my request to share potential challenges or limitations associated with the widespread adoption of wearable devices and remote monitoring technologies, Javed observed that advancements in wearable devices and ubiquitous sensing technologies are remarkable. He noted that beyond wearables, we now have ambient intelligence—environments that can sense and respond to us. For example, certain cars can sense and adjust the seat and temperature as you enter, and some offices can adjust settings based on the number of people present.

He said: “These sensor-based technologies hold great promise. The symbiosis comes from the digital platform for care delivery, which includes the widespread adoption of EHRs. However, we’re still in the early stages of realizing the full potential of these technologies. Data from research repeatedly shows that the promise of these technologies often doesn’t match the actual health outcomes.

There’s a lot of optimism around using these technologies for mental health, weight management, and pregnancy support, among others. Many promising apps and tools are being developed. Yet, when rigorously studied for health outcomes, the results aren’t as promising. This doesn’t mean they won’t eventually show benefits, but we are still in the early days.”

Wearable devices and remote monitoring have significantly impacted health informatics by collecting vast amounts of health data, enabling continuous monitoring, and potentially improving health outcomes. However, their effectiveness depends on rigorous evaluation and ensuring they provide measurable benefits.”

Unstructured Healthcare data, Analytics, AI and Telemedicine

Healthcare is a data-driven business, with the average hospital producing roughly 50 petabytes of data every year—more than twice the amount housed in the Library of Congress, amounting to 137 terabytes per day. The healthcare industry generates vast amounts of data daily from various sources like electronic health records (EHRs), medical imaging, and wearable devices. To maximize its utility, this data needs thorough analysis. Data analytics is crucial for uncovering hidden patterns and trends in clinical data, which can enhance patient care, treatment effectiveness, cost savings, and innovation. However, collecting and utilizing healthcare data presents numerous challenges.

The rapid growth of unstructured healthcare data holds immense potential for extracting valuable insights, enhancing healthcare services, ensuring quality patient care, and securing data management. However, technological advancements are needed to fully realize these benefits. The heterogeneity, diversity of sources, and varied representations of unstructured data present more challenges compared to structured data. A literature review by Adnan et al. (2019) identifies the challenges and issues in data-driven healthcare arising from the unstructured nature of the data. The main challenge is to classify data and bring order to it with the help of NLP and AI technologies.

When asked to provide some case studies or examples illustrating how data analytics has significantly improved healthcare outcomes and operational efficiency, Javed opines that there’s a lot of promise, particularly in streamlining workflows and record-keeping. He elaborated: Practical needs, such as converting unstructured notes into analyzable formats, summarizing them, and linking this data with other records, are being addressed. Similarly, analyzing images, animations, and videos to impose structure and make them computable is becoming increasingly feasible.

Positive results have been seen in image analysis, particularly in radiology, where it’s being used to detect cancerous tissues. These advancements indicate strong potential for applications in managing and utilizing unstructured data, not just text but also images and videos.

Dr. Geoffrey Hinton, regarded as the “godfather of deep learning,” famously stated that society should stop training radiologists, as it was obvious they would be replaced by software in the next five years. Though this might not happen, this statement reflects the potential of Image Analytics. Today’s AI software, driven by deep learning algorithms, employs artificial neural networks like convolutional neural networks (CNN) to efficiently analyze imaging findings, reducing the need for manual labeling required by older algorithms. With high-quality annotated training data, these advanced algorithms can effectively interpret vast diagnostic imaging datasets, achieving accuracy comparable to or surpassing that of practicing radiologists.

According to Javed, AI’s influence spans the entire healthcare journey—from a prospective patient seeking care to engagement with a healthcare provider, developing and executing a treatment plan, and follow-up care. AI can assist at every level. However, he cautions: “we must be mindful of deployment and context. As a technologist, it’s easier to develop large-scale systems and publish research. In healthcare, though, the focus must be on practical application, translation, and real-world impact.”

“AI can streamline data collection, enhance diagnostic accuracy, and personalize treatment plans. However, deploying AI effectively requires addressing challenges like information overload and ensuring transparency and fairness. Despite some skepticism, particularly regarding provider burnout and the complexity of healthcare environments, AI’s potential benefits make it a promising tool for improving healthcare delivery.

Telemedicine benefits both health and convenience. The role of AI in healthcare delivery includes tele-assessment, tele-diagnosis, tele-interactions, and tele-monitoring. AI has expanded its use cases to include remote patient monitoring, patient diagnosis and medical image analysis, treatment plans, patient engagement, and chronic disease management. A 2019 report by MIT, based on a survey conducted by MIT Technology Review Insights in association with GE Healthcare, found that AI is humanizing healthcare as more than 82% of healthcare business leaders report that their AI deployments have already created workflow improvements in their operational and administrative activities—giving clinicians time back to work with their patients more closely and with more insight.

Javed echoes similar sentiments about telemedicine. He said:

There’s considerable promise to enhance telemedicine with AI. My work in virtual care, particularly in online triaging through voice and phone interaction, highlights this. Telehealth has a long history, but its success depends on socio-economic and political contexts. For instance, in the U.S., wider adoption of telehealth payments only occurred during the COVID-19 pandemic.

Telehealth complements one-to-one care, expanding reach and providing basic care interactions. In a country like India, there’s fantastic potential for telemedicine to enhance healthcare access. However, successful implementation requires considering incentives, economic, and political factors. Doctors in India hold significant political and economic influence, so engaging with these groups and ensuring proper incentives are crucial.

Heath Informatics and Multidisciplinarity: Breaking Down Silos

While it is recognized that addressing global challenges requires multidisciplinary, multi-partner, and multi-sector collaborations, overcoming the entrenched barriers within our institutions is a significant challenge. The longstanding structures we have operated in must be reevaluated to find new ways of working and sharing effectively. Establishing a collaborative research center to break down these institutional silos demands innovative approaches. Grönqvist et al. (2017) identified 15 major challenges, some general to establishing a new research environment and others specific to multidisciplinary eHealth programs. These challenges were organized into six themes: Organization, Communication, Implementation, Legislation, Software Development, and Multidisciplinarity. In an episode of InfoFire  , Gautam Menon discusses the crucial need for interdisciplinary research, citing examples such as climate change and infectious diseases like COVID-19.

Javed shares his views and experience in establishing the Carolina Health Informatics Program (CHIP) at the University of North Carolina:

Developing programs across different universities in a multidisciplinary field like health informatics is challenging. Integrating perspectives from computer science, healthcare, social sciences, epidemiology, and more requires the involvement and support of central institutions and political leaders.

Creating the Carolina Health Informatics Program (CHIP) at the University of North Carolina was possible due to support from the central provost office. Multidisciplinary initiatives need not only approval but also direct guidance and long-term support from central leadership. Political leaders also play a crucial role due to the regulatory and funding aspects of healthcare.

The key to successful multidisciplinary programs is to gain the respect and support of central leadership from the beginning and to maintain it long-term. Collaboration between different disciplines requires mutual respect for each other’s methods and theories.

In healthcare, achieving practical outcomes is essential. Universities should connect academics to real-world issues, making healthcare a valuable laboratory for interdisciplinary scholarship and community engagement.