In healthcare, data refers to raw, unprocessed facts: a blood pressure reading of 142/91, a lab value of 6.8, a timestamp from a heart monitor, a diagnostic code in a billing record. Information is what emerges when those facts are organized, contextualized, and interpreted so that someone can act on them. That blood pressure reading becomes information when it is linked to a specific patient, compared against their baseline, and flagged as elevated for their age and medical history. The distinction sounds abstract, but it has concrete consequences for patient safety, clinical research, billing accuracy, and how well different parts of a health system communicate with one another.
What Counts as Data and What Counts as Information at the Bedside
A hospital generates enormous quantities of data every hour. Vital signs flow from bedside monitors. Lab instruments spit out numerical results. Clinicians type notes into electronic health records. Pharmacies log medication orders. Imaging machines produce files. All of this is data in its rawest form: discrete facts that have not yet been assembled into a picture that helps anyone make a decision.
Information appears when a clinician, a software system, or a report pulls those facts together and gives them meaning. A single potassium level of 5.9 mmol/L is a data point. That same value becomes information when the system flags it as dangerously high, cross-references the patient’s kidney function, checks whether they are on a medication that raises potassium, and alerts the care team. The nursing informatics field has long used a framework called the data-information-knowledge-wisdom hierarchy to describe this progression, tracing how raw observations build into understanding and eventually guide clinical judgment.1PubMed. The Evolution of Data-Information-Knowledge-Wisdom in Nursing Informatics
The practical takeaway is that healthcare workers swim in data all day. The challenge is rarely collecting more of it. The challenge is converting what already exists into information that reaches the right person at the right time in a form they can use.
Why So Much Clinical Data Stays Unprocessed
A large share of the data sitting in electronic health records is unstructured, meaning it lives as free-text paragraphs rather than neatly coded fields. A physician’s progress note, an operative report, a radiology impression, a patient’s own description of symptoms: all of these are typed or dictated as narrative text. The majority of clinical data takes this form.2Advances in Healthcare Information Systems and Administration. NLP for Clinical Data Analysis Structured data, by contrast, sits in defined fields: a drop-down menu for smoking status, a numeric slot for heart rate, a standardized code for a diagnosis. Computers handle structured data easily because the format tells them what each value means.
Unstructured text is far harder to work with at scale. A note reading “Patient reports worsening chest tightness over the past two weeks, especially with exertion” contains clinically valuable detail, but a computer cannot automatically extract and compare it the way it can with a coded field. Natural language processing, or NLP, is the set of techniques designed to bridge that gap by pulling structured facts out of free text so they can be searched, aggregated, and fed into alerts or research databases.3PubMed Central. Natural Language Processing of Clinical Notes on Chronic Diseases: Systematic Review Until that extraction happens, a narrative note is data with enormous potential but limited computational reach. It is information to the clinician who reads it, but it remains raw data to any system trying to use it across thousands of patients.
Standardization as the Bridge
Even structured data is only useful across institutions if everyone speaks the same language. If one hospital records “Type 2 diabetes mellitus” as a text string and another records it as ICD-10 code E11, a computer trying to merge those datasets has to figure out they mean the same thing. This is where clinical terminologies and interoperability standards come in. SNOMED CT is the world’s largest clinical terminology, offering standardized codes for diseases, procedures, findings, and more. LOINC provides standardized identifiers for laboratory and clinical observations.4PubMed Central. Recent Developments in Clinical Terminologies – SNOMED CT, LOINC, and RxNorm Together, these systems ensure that a hemoglobin A1c result at one lab means the same thing as a hemoglobin A1c result at another.
On the transport side, HL7 FHIR (Fast Healthcare Interoperability Resources) provides a modern framework for packaging and exchanging clinical data between systems. When FHIR is paired with SNOMED CT, both the structure and the meaning of the data are preserved in transit, which is the difference between sending a box of unsorted lab slips and sending a neatly organized patient summary.5PubMed Central. HL7 FHIR with SNOMED-CT to Achieve Semantic and Structural Interoperability in Personal Health Data: A Proof-of-Concept Study Research on cancer synoptic reports has demonstrated that clinical datasets can be fully represented using SNOMED CT to make data elements computable and semantically faithful to their intended meaning, while FHIR enables standardized exchange of that clinical information across institutions.6PubMed Central. Building a Standardized Cancer Synoptic Report With Semantic and Syntactic Interoperability: Development Study Using SNOMED CT and Fast Healthcare Interoperability Resources (FHIR)
Without these standards, healthcare data remains locked inside whichever system generated it. Standardization does not change the underlying facts, but it transforms them from isolated data points into information that can flow between a primary care clinic, a specialist office, and a hospital emergency department without losing its meaning along the way.
Alarm Fatigue and the Cost of Data Without Context
One of the most vivid examples of the data-versus-information problem in healthcare is alarm fatigue. Modern intensive care units are full of monitors that track heart rate, oxygen levels, blood pressure, respiratory rate, and more. Each monitor generates data continuously, and each is programmed to sound an alarm when a reading crosses a threshold. The trouble is that most of those alarms are not clinically actionable. A patient shifts in bed and dislodges a sensor, triggering an oxygen alarm. An artifact in the heart rhythm tracing sets off a cardiac alert. The data behind each alarm is technically real, but it does not represent a meaningful change in the patient’s condition.
When nurses and physicians face hundreds of alarms per shift, many of them false or trivial, they start tuning them out. This desensitization is alarm fatigue, and it poses a serious patient safety risk because the one genuinely critical alarm gets lost in the noise.7PubMed. Alarm Fatigue: Using Alarm Data from a Patient Data Monitoring System on an Intensive Care Unit to Improve the Alarm Management High alarm frequency, persistent false alarms, multitasking demands, and night shifts all compound the problem.8PubMed Central. Mechanism, contributing factors, and coping strategies of alarm fatigue in intensive care nursing: a qualitative study
Alarm fatigue is essentially a failure to convert data into information. The monitors are doing their job: they are reporting data points that cross pre-set thresholds. But without additional context (is this a new finding, is the sensor properly attached, does this match the patient’s trend over the last hour), the alarm carries no actionable meaning. Smarter alarm systems that filter and prioritize based on clinical context are an active area of work, and they represent exactly the kind of data-to-information transformation that healthcare needs more of.
When Patients Get Data Without Interpretation
Patient portals have made it common for people to see their own lab results, sometimes before their doctor has reviewed them. This is generally a positive development, but it also creates a situation where patients receive raw data without the clinical interpretation that turns it into information. A scoping review of patient and provider perspectives on portal access to test results found that the most frequently cited concern was patients learning alarming results before a clinician had a chance to explain them, especially when the reports lacked any accompanying interpretation.9PubMed Central. Patient and Health Care Provider Perspectives on Patient Access to Test Results via Web Portals: Scoping Review
Research specifically examining the experience of receiving abnormal results through portals tells a similar story: patients wanted access to all types of results, including abnormal ones, but they frequently reported difficulty interpreting the relevance of what they saw.10Patient Experience Journal. The patient portal and abnormal test results: An exploratory study of patient experiences A slightly elevated liver enzyme, for example, might be completely benign in context but looks frightening when presented as a red flag on a results page with no explanation. The data is accurate. What is missing is the contextual information that a clinician brings: “This is a minor bump, likely from the new medication you started, and we will recheck it in a month.”
This gap highlights something important about the data-information distinction in healthcare. The same number can be terrifying data or reassuring information depending entirely on context. Portal design that pairs results with plain-language interpretations or automated contextual notes is one way to close that gap, though most systems are still catching up.
Wearables and the Flood of Sensor Data
Wearable devices like smartwatches, continuous glucose monitors, and fitness trackers have added an entirely new dimension to healthcare data. These devices generate constant streams of physiological readings: heart rate sampled every few seconds, step counts, sleep stages, blood oxygen estimates, skin temperature. The volume is staggering, and most of it is noise from a clinical standpoint.
The real value emerges when artificial intelligence models process these raw sensor streams and extract clinically meaningful patterns. Research on AI-integrated wearable systems describes how these tools can transform raw sensor data into clinical insights that support early disease detection and personalized care.11International Journal of Research and Innovation in Applied Science. Wearable Sensor-Based Health Monitoring Using Artificial Intelligence: A Smart Healthcare Framework for Continuous Patient Monitoring One approach uses deep learning models to identify specific physiological events, such as changes in heart rate variability or respiratory patterns, from raw sensor samples, improving prediction accuracy over simpler methods.12PubMed Central. Hybrid deep learning for IoT-based health monitoring with physiological event extraction
For the average person wearing a fitness tracker, the implication is straightforward: the thousands of heart rate measurements your watch collects overnight are data. The notification that says “your resting heart rate has been trending upward over the past two weeks” is information. And if an algorithm connects that trend to other signals and suggests you may be developing an irregular rhythm, that starts to approach clinical knowledge. Each step requires aggregation, filtering, and interpretation layered on top of the raw numbers.
Data Quality Problems That Change Clinical Conclusions
The distinction between data and information matters just as urgently when it comes to quality. Healthcare data originates from different workflows, and those workflows introduce systematic differences that can distort conclusions if you are not careful. A striking example comes from pediatric cardiac surgery, where researchers compared outcomes derived from a clinical registry (where trained abstractors reviewed charts) to outcomes derived from administrative billing data for the same patients.
The agreement between the two data sources on the complexity category assigned to each operation was only about 68%. Those differences were not trivial: they translated into a roughly 26% underestimate of mortality for one procedure and a 31% overestimate for another, depending on which data source you used.13PubMed Central. Differential case ascertainment in clinical registry versus administrative data and impact on outcomes assessment for pediatric cardiac operations The underlying data, the codes recorded for each patient, looked perfectly fine in isolation. The problem was that billing codes capture a simplified version of what happened, optimized for reimbursement rather than clinical precision. Treating billing data as if it carried the same clinical meaning as a registry abstraction is a classic case of mistaking data for information.
Transforming real-world data into reliable evidence requires understanding the quality, provenance, and organizational context of the source data, and addressing differences across institutions when combining records from multiple sites.14PubMed Central. Developing real-world evidence from real-world data: Transforming raw data into analytical datasets Data quality evaluation needs to be an ongoing process rather than a one-time check, especially when records are going to be used for research or predictive analytics beyond the original care encounter.15PubMed Central. Real-World Evidence of COVID-19 Patients’ Data Quality in the Electronic Health Records
Billing and Coding as a Parallel Data-to-Information Pipeline
Healthcare billing illustrates the data-information gap in a domain that most patients never think about. Every clinical encounter generates documentation, and that documentation gets translated into standardized codes (ICD-10 for diagnoses, CPT for procedures, E&M codes for the level of the visit) so that insurers can determine how much to pay. The codes themselves are data. The reimbursement decision that results from matching those codes to a fee schedule, applying payer rules, and checking for documentation requirements is information used to run a practice’s finances.16PubMed Central. A systematic review of outpatient billing practices
Where this gets tricky is that billing codes are a lossy compression of clinical reality. A complex encounter involving three diagnoses, a medication change, and a counseling conversation gets reduced to a handful of codes that may not capture the nuances. When those codes are later repurposed for clinical research or quality measurement, the gap between the original clinical data and the billing-derived “information” can be large, as the pediatric cardiac surgery example illustrated. The lesson is that data collected for one purpose does not automatically serve as good information for a different purpose. Context and intent matter.
Secondary Use of Health Data for Research
One of the most promising and most fraught applications of the data-information distinction is the secondary use of clinical records for research. When a hospital collects vitals, lab results, imaging, and notes as part of routine care, that data was generated for the purpose of treating an individual patient. Researchers increasingly want to reuse those records to study disease patterns, evaluate treatments, and train predictive models across large populations.
This transition from clinical data to research-grade information is not straightforward. Germany’s Medical Informatics Initiative, for example, has developed a national portal to enable access to distributed clinical data across the country’s federated healthcare system, but the project contends with challenges including data heterogeneity and strict data-protection requirements.17PubMed. Germany’s Approach for the Secondary Use of Federated Real-World Data in the German Portal for Medical Research Data The better that research protocols, contracts, and budgets are written with these data complexities in mind, the better the opportunity to generate evidence from electronic health records without compromising privacy.18Journal of the Society for Clinical Data Management. Key Considerations When Designing Real World Evidence Research Involving The Secondary Use of U.S.-Based Electronic Health Record Data
In practice, this means researchers cannot simply download a database and start running analyses. They need to understand how each field was captured, what it was intended to represent, how missing values were handled, and whether the definitions used at one hospital match those at another. The raw records are data. The carefully curated, harmonized, quality-checked analytical dataset is information, and building one is a significant undertaking in its own right.
Large Language Models and the Shift in How Data Becomes Information
The latest technological push in this space involves large language models. Clinicians spend a significant amount of their time sifting through text in electronic health records, and LLMs are increasingly being tested as tools for summarizing that material. Research applying domain-adapted LLMs to clinical summarization tasks, including radiology reports, progress notes, and doctor-patient dialogues, has shown that these models can perform well across multiple types of clinical text.19PubMed Central. Clinical Text Summarization: Adapting Large Language Models Can Outperform Human Experts
On the patient-generated side, early work exploring how LLMs can help clinicians make sense of data from wearables and health apps found that AI-generated summaries provided quick overviews that anchored clinical exploration, while conversational interaction with the data helped bridge gaps in data literacy.20arXiv.org. Exploring AI-Augmented Sensemaking of Patient-Generated Health Data: A Mixed-Method Study with Healthcare Professionals in Cardiac Risk Reduction In both cases, the underlying function is the same: taking a mass of raw or semi-processed data and producing a concise, contextualized summary that a human can act on. That is, by definition, turning data into information.
The risks here are real and worth noting. An LLM that summarizes a chart incorrectly, hallucinating a medication that was never prescribed or omitting a critical allergy, does not just fail to create information. It creates misinformation in a setting where the stakes are high. The technology is promising, but it is not yet a replacement for clinical judgment in most workflows.
Health Data as a Commercial Commodity
The data-information distinction also has an economic dimension that many people do not think about. Patient demographics, clinical findings, laboratory results, and genomic data have become a premium commodity with value for both private and public entities. The push to commercialize this material has spawned a multibillion-dollar industry centered on collecting, analyzing, and selling health data.21The Lancet Digital Health. Commercial health data and biomedical research: the need for transparency, reproducibility, and generalisability
What is being sold in most cases is not raw data. Companies that aggregate and sell health data typically clean, link, de-identify, and enrich the records before packaging them for pharmaceutical companies, insurers, or researchers. The raw EHR extract is data. The curated, linked, analytics-ready product is information, and the value added during that transformation is what justifies the price tag. This raises important questions about who benefits from that value, whether patients understand how their records are used, and whether the resulting research is transparent and reproducible enough to justify the enterprise. These are governance questions that sit squarely at the intersection of data, information, and trust.