By Eric Andrew Kristof, RN, CAHIMS | Healthcare IT Professional | HIMSS Arkansas Chapter Member
About This Series: The HIMSS Word of the Day is a running blog series drawing from the HIMSS Dictionary of Health Information and Technology Terms, Acronyms, and Organizations, 6th Edition (2025) — the authoritative reference for healthcare IT professionals and the primary study resource for CAHIMS and CPHIMS certification. Each post paraphrases the term in plain language, grounds it in real-world healthcare IT context, and adds a bedside nursing perspective you won’t find in a textbook.
📖 The Term: Machine Learning
Source: HIMSS Dictionary of Health Information and Technology Terms, Acronyms, and Organizations, 6th Edition (2025)
🔍 Plain-Language Definition
Machine learning (ML) is a branch of artificial intelligence in which software improves at a task by learning from data rather than by following rules a person wrote in advance. Show a model enough examples, and it learns the patterns that connect inputs to outcomes.
The HIMSS Dictionary frames machine learning as AI that lets computers learn in two directions. Inductive learning starts with many examples and works toward a general rule, such as finding the patterns in thousands of patient records that precede clinical deterioration. Deductive learning starts from premises assumed to be true and reasons to conclusions that must follow from them. Most of the machine learning in healthcare today is inductive: it generalizes from large data sets.
In practice, machine learning is usually grouped by how the model learns:
| Type | How It Learns | Healthcare Examples |
|---|---|---|
| Supervised Learning | Trains on labeled examples where the right answer is known | Readmission risk, sepsis and deterioration prediction, image classification |
| Unsupervised Learning | Finds structure in unlabeled data on its own | Grouping patients into risk clusters, detecting unusual billing or access patterns |
| Reinforcement Learning | Learns by trial and feedback toward a goal | Research uses in treatment sequencing and operations; rare in routine care |
| Deep Learning | Uses many-layered neural networks to learn complex features | Radiology and pathology image analysis, speech recognition, ambient documentation |
⚠️ Exam Alert: Machine learning is a subset of artificial intelligence, not a synonym, and deep learning is a subset of machine learning. Distinguish ML from rule-based clinical decision support, which follows explicit logic written by people. Know the vocabulary of model development: training data builds the model, validation tunes it, testing checks it on data it has not seen, and model drift is the decline in performance when real-world data changes over time.
🌐 Why It Matters: Real-World Healthcare IT Context
Machine learning already runs inside most hospital EHRs and imaging systems, usually without patients or staff knowing which outputs come from a model.
The FDA maintains a public list of AI-enabled medical devices it has authorized, and the list is dominated by imaging applications. The FDA, Health Canada, and the U.K.’s medicines regulator have also published Good Machine Learning Practice principles for developing these products. For predictive tools built into certified EHRs, ONC’s HTI-1 Final Rule added transparency requirements so that users can learn how a model was developed and what it is intended to do.
The strongest argument for transparency comes from real-world evidence. In 2021, a JAMA Internal Medicine study externally validated a widely implemented proprietary sepsis prediction model at a large academic health system. The model’s discrimination was modest, with an area under the curve of 0.63. It failed to flag 67% of patients who developed sepsis, while generating alerts on 18% of all hospitalized patients. A model can perform well where it was built and poorly somewhere else.
Failure modes of healthcare machine learning:
- Training data that underrepresents some patient groups, producing biased predictions
- Performance that degrades as populations, workflows, or documentation habits change
- Models deployed without local validation against the organization’s own patients
- Alert thresholds set without regard to the alert burden on clinicians
- Outputs presented without explanation, which invites blind trust or blanket dismissal
- No owner responsible for monitoring the model after Go-Live
Strengths of mature machine learning programs:
- An AI governance committee that reviews each model’s intended use, evidence, and risks
- Local validation, often in silent mode, before a model influences care
- Ongoing monitoring of accuracy, alert volume, and performance across demographic groups
- Clear documentation of what a model predicts, for whom, and how to respond
- Workforce education on what a risk score means and what it does not
- Alignment with frameworks such as the NIST AI Risk Management Framework
🏥 The Nurse’s Perspective: Clinical, Bedside, & Workflow
Machine learning often reaches the bedside as a single number, and nurses are frequently the first clinicians asked to act on it.
At the Bedside
Deterioration indexes, sepsis scores, fall-risk predictions, and readmission flags increasingly appear in nursing worklists and banners. A score is not a diagnosis. It is a statistical estimate built from data that may be incomplete, delayed, or different from the patient in the bed.
Nurses add what the model cannot see: how the patient looks, what changed since the last set of vitals, and what the family just reported. When a score and the bedside assessment disagree, the assessment deserves a voice, and escalation pathways should make room for nursing judgment in both directions.
In Clinical Workflow
Nursing documentation is also training data. Vital signs entered late, assessments copied forward, and free-text notes in place of structured fields all shape the data future models learn from. Accurate, timely documentation improves both today’s care and tomorrow’s predictions.
The workflow question is what a nurse is expected to do with a prediction. A high score that triggers nothing is noise; a high score that triggers a reassessment, a rapid response consult, or a provider notification is a tool. Clear response protocols are what turn machine learning into safer care rather than another source of alert fatigue.
In EHR Implementation (Go-Live Perspective)
When predictive models are activated during or after an Epic Go-Live, the key build decisions are thresholds, target populations, display locations, and who receives the alert. Running a model in silent mode first, recording its scores without showing them, lets the organization compare predictions to outcomes before anyone acts on them.
End-user support analysts hear the first reactions: a score that seems wrong, an alert that fires on stable patients, or a banner no one understands. Logging those reports separately from routine break-fix tickets and routing them to the AI governance team helps distinguish a model problem from a training gap.
🎓 CAHIMS / CPHIMS & HIMSS Perspective
CAHIMS Exam Domain Mapping
Machine learning appears increasingly across the CAHIMS exam domains:
| CAHIMS Domain | How Machine Learning Appears |
|---|---|
| Clinical Informatics | Predictive analytics in clinical decision support; deterioration and sepsis models; human oversight; alert burden |
| Healthcare & Technology Environments | FDA oversight of AI-enabled devices; HTI-1 transparency requirements; ethical and equity considerations |
| Healthcare Information & Systems Management | Data quality for training; validation and testing; monitoring for drift; integration with the EHR |
| Management & Leadership | AI governance structures; vendor evaluation; workforce education; risk management |
Key Terms to Know Alongside Machine Learning
- Artificial Intelligence (AI) — the broader field of computer systems that perform tasks associated with human intelligence
- Deep Learning — machine learning that uses many-layered neural networks to learn complex patterns
- Predictive Modeling — using historical data to estimate the likelihood of a future event
- Model Drift — the loss of model performance as real-world data shifts away from the training data
- Algorithmic Bias — systematic differences in model performance across patient groups
- External Validation — testing a model on data from a different setting than the one it was built in
HIMSS Organizational Position
HIMSS supports responsible use of machine learning through public policy engagement with federal agencies, research from HIMSS Global Insights, and AI-focused programming at its Global Health Conference. Members can find AI governance guidance and case studies through the HIMSS Resource Center.
🔗 Explore Further: External References
- Wong A, et al. External Validation of a Widely Implemented Proprietary Sepsis Prediction Model in Hospitalized Patients. JAMA Internal Medicine, 2021 https://pubmed.ncbi.nlm.nih.gov/34152373/
- FDA — List of Artificial Intelligence-Enabled Medical Devices https://www.fda.gov/medical-devices/artificial-intelligence-enabled-medical-devices/list-artificial-intelligence-enabled-medical-devices
- FDA — Good Machine Learning Practice for Medical Device Development: Guiding Principles https://www.fda.gov/medical-devices/artificial-intelligence-enabled-medical-devices/good-machine-learning-practice-medical-device-development-guiding-principles
- ONC — HTI-1 Final Rule https://healthit.gov/regulations/hti-rules/hti-1-final-rule/
- NIST — AI Risk Management Framework https://www.nist.gov/itl/ai-risk-management-framework
- HIMSS — Resource Center https://www.himss.org/resource-center
Term source: HIMSS Dictionary of Health Information and Technology Terms, Acronyms, and Organizations, 6th Edition (2025). All definitions in this series are paraphrased for editorial purposes. Readers are encouraged to consult the primary source for exact language.
Eric Andrew Kristof, RN, CAHIMS is a Healthcare IT professional and HIMSS Arkansas Chapter member based in Hot Springs Village, AR, with hands-on Epic Go-Live experience and a diverse healthcare and IT background. He writes at kristof.org at the intersection of clinical care and healthcare technology.
Leave a Reply