Loading field map
Health informatics emerged from a practical pressure that remains its central challenge: how to systematically use data and knowledge to improve health care, when the relevant information is scattered across paper records, professional expertise, and a growing array of digital systems. Since the 1960s, the field has developed a dozen major frameworks, each responding to a specific gap in the previous generation of tools and ideas. The story of health informatics is not a simple succession of replacements but a layered accumulation in which earlier frameworks continue to operate as infrastructure for later ones.
The first three frameworks appeared together in the late 1960s and co-evolved over the following decades. Clinical Decision Support (CDS) began as an effort to encode medical expertise into rule-based systems that could assist clinicians at the point of care. Early CDS systems, such as MYCIN, used if-then rules derived from expert consensus to suggest diagnoses or treatments. At the same time, Clinical Information Systems and Electronic Health Records (EHRs) aimed to digitize the patient record itself, replacing paper charts with structured data that could be stored, retrieved, and shared. The earliest EHRs were little more than digital abstracts of paper records, and their adoption was slow because the technology was expensive and the benefits uncertain.
These two frameworks depended on a third: Clinical Knowledge Representation and Terminology Engineering. Without a standardized way to name diseases, medications, and procedures, both CDS rules and EHR data would remain local and incomparable. Terminology systems such as SNOMED and ICD provided the controlled vocabularies that made it possible to encode clinical meaning in a machine-readable form. This framework supplied the terminological foundation that allowed CDS rules to fire on data stored in EHRs, creating the first integrated clinical information systems. The relationship among these three was one of mutual dependence: EHRs needed terminologies to structure data, and CDS needed both terminologies and structured data to apply its rules.
By the late 1980s, it had become clear that isolated systems could not communicate with each other. Health Data Standards and Interoperability emerged as a framework dedicated to creating the technical and semantic bridges between systems. Standards such as HL7 and DICOM defined message formats for exchanging clinical data, while later efforts like FHIR provided modern web-based APIs. This framework did not replace the earlier ones; it provided the infrastructure that allowed EHRs, CDS, and terminology systems to work together across institutions. Without interoperability, the vision of a seamless digital health ecosystem remained out of reach.
As clinical systems matured, the field began to expand its focus beyond the hospital and the individual patient. Public Health Informatics (formalized around 1995) applied informatics methods to population-level questions: disease surveillance, outbreak detection, and the analysis of vital statistics. Where clinical informatics looked at one patient at a time, public health informatics aggregated data across populations to identify trends and guide policy. This framework coexisted with clinical systems by drawing on the same data sources—EHRs, registries, laboratory reports—but analyzing them at a different scale.
Shortly afterward, Consumer Health Informatics (emerging around 2000) shifted the focus to the patient as an active participant. Instead of treating the patient as a passive recipient of care, this framework studied how information technologies—patient portals, mobile apps, online communities—could support self-management, shared decision making, and health behavior change. Consumer health informatics broadened the user base of health information systems from clinicians alone to include patients and caregivers, creating new design requirements and new ethical questions about access, literacy, and privacy.
The widespread adoption of EHRs produced a new resource: large collections of structured clinical data. Health Data Analytics and Predictive Modeling (emerging around 2000) took advantage of this data to build statistical and machine-learning models that could predict outcomes, stratify risk, and identify patterns that rule-based CDS had missed. Where early CDS relied on expert-authored rules, health data analytics learned from the data itself, using regression, clustering, and early neural networks. This shift from top-down knowledge engineering to bottom-up data mining represented a fundamental change in method. The two approaches did not replace each other; they complemented each other, with rule-based systems handling well-understood clinical logic and data-driven models capturing complex, non-obvious associations.
Clinical Research Informatics (formalized around 2007) applied informatics methods to the conduct of clinical trials and observational studies. It addressed the inefficiencies of traditional research—slow recruitment, manual data collection, fragmented datasets—by leveraging EHR data for cohort identification, protocol management, and adverse event monitoring. This framework absorbed many of the tools of health data analytics but directed them toward the specific goal of generating new evidence rather than supporting real-time decisions.
Around the same time, Learning Health Systems (LHS) emerged as a meta-framework that aimed to close the loop between research and practice. An LHS is a system in which data from every patient encounter is captured, analyzed, and fed back into care delivery and research, creating a continuous cycle of improvement. This framework did not introduce a new technology so much as a new organizational vision: it called for integrating clinical informatics, data analytics, and research informatics into a single, self-improving system. The LHS concept influenced the goals of later frameworks by insisting that informatics should serve both immediate care and long-term knowledge generation.
As health informatics systems became more complex, researchers recognized that technical solutions alone were insufficient. Sociotechnical Health Informatics (emerging around 2008) argued that health information systems must be understood as sociotechnical systems—combinations of people, workflows, organizational culture, and technology. This framework critiqued the assumption that simply installing an EHR would improve care; instead, it showed that successful implementation required attention to how clinicians and patients actually use technology, how power and trust shape adoption, and how unintended consequences arise. Sociotechnical informatics coexists with the more technical frameworks as a corrective lens, reminding the field that data and algorithms operate within human contexts.
Translational Bioinformatics (also emerging around 2008) extended informatics methods to the molecular level. It focused on integrating genomic, proteomic, and other -omic data with clinical data to enable precision medicine. Where earlier frameworks worked with diagnoses, medications, and lab results, translational bioinformatics added DNA sequences, gene expression profiles, and protein interactions. This framework narrowed the gap between basic biomedical research and clinical application, creating tools that could identify genetic variants linked to disease or predict drug responses based on a patient's genome.
Artificial Intelligence and Machine Learning in Health Informatics (emerging around 2012) represents the most recent major framework. It derives directly from Health Data Analytics and Predictive Modeling, but with a distinct emphasis on deep learning, natural language processing, and other advanced AI techniques. The LHS framework influenced AI/ML by providing a systems-level rationale: AI models are most valuable when they are embedded in a learning cycle that continuously updates them with new data and feeds their outputs back into care. AI/ML has transformed areas such as medical imaging, where deep learning models can detect abnormalities with accuracy rivaling specialists, and clinical text mining, where natural language processing extracts structured information from unstructured notes. This framework does not replace earlier approaches; it coexists with rule-based CDS, statistical analytics, and sociotechnical analysis, each suited to different tasks.
Today, the leading frameworks—EHRs, CDS, health data analytics, AI/ML, sociotechnical informatics, and learning health systems—are all active, and their relationships are best described as a division of labor. They agree on several core principles: data should be structured and interoperable; systems should be evaluated for their impact on outcomes, not just their technical performance; and the ultimate goal is to improve health at both individual and population levels.
But they also disagree on fundamental questions. One persistent tension is between top-down knowledge (expert rules, guidelines, curated terminologies) and bottom-up learning (data mining, AI models). Proponents of rule-based CDS argue that expert knowledge is transparent, explainable, and grounded in evidence, while advocates of AI/ML counter that data-driven models can capture patterns too complex for human experts to articulate. A second disagreement concerns the role of the human user: sociotechnical informatics insists that technology must adapt to human workflows and social contexts, while some AI/ML approaches aim to automate decisions with minimal human intervention. A third debate revolves around the learning health system ideal: critics argue that the LHS vision underestimates the difficulty of integrating research and practice, the privacy risks of continuous data reuse, and the organizational changes required.
These disagreements are not signs of fragmentation; they are the productive tensions that drive the field forward. Health informatics today is a pluralistic discipline in which different frameworks address different parts of the problem, and the most successful projects draw on multiple frameworks in combination. The early foundations of terminologies and EHRs remain essential infrastructure; the data-driven turn has opened new analytical possibilities; and the sociotechnical and learning systems perspectives ensure that the field remembers the human and organizational dimensions of its work.