Vital Signs

Latent Dirichlet allocation, or LDA, is a statistical topic model that explains a collection of documents as mixtures of hidden themes. This post explains the HIMSS Dictionary definition, then reviews six sources: the original 2003 model, a 2006 application to a biomedical bibliography, topic modeling of 2.9 million clinical notes for people living with HIV, topics drawn from codes and prescriptions in 1,102 patients with rheumatoid arthritis, psychotherapy notes from veterans stratified by suicide risk, and a topic model of 11,674 online complaints that mention nurses.

Natural language is the everyday speech and writing people use, set against formal computer languages. This post explains the HIMSS Dictionary definition and how it differs from natural language processing, then reviews five sources: an AI chatbot that answered 1,000 natural-language questions about ophthalmic records with 96 percent accuracy, a natural-language query platform for 1.2 million ECG records, multilingual documentation workflows in a Swiss hospital, a voice-guided colonoscopy documentation pilot, and a study of documented interpreter use among emergency department patients with limited English proficiency.

An expert system is an artificial intelligence-based system that converts the knowledge of an expert in a specific subject into software code, usually built from a knowledge base, an inference engine, and a user interface. This post explains the HIMSS Dictionary definition, then reviews six sources: a hybrid rule-based and language model decision support system tested in two German nursing homes, the development of a knowledge base with 138 care problems for older adult care, an inspectable rule-based tool for diabetic menu planning, the evolution of the Arden Syntax standard, a historical review of knowledge-based expert systems in health informatics, and a retrieval-augmented architecture for post-surgical monitoring.

Computer-assisted coding is software that uses natural language processing to analyze whole medical charts and pre-assign procedure and diagnosis codes for a coder to review. This post explains the HIMSS Dictionary definition and the HIPAA code sets involved, then reviews five sources: a crossover randomized trial in Norway and Sweden, a study of evaluation metrics for ICD coding systems, a systematic review of 54 automated ICD coding studies, an audit that found errors in widely used benchmark datasets, and a pilot of a GPT-4 based coding assistant.

A decision tree is a tree-shaped diagram or model that works from a starting decision through branches of possible actions to leaves that show outcomes, and in data analysis it splits a large dataset into groups with similar values to understand likely results. This post explains the HIMSS Dictionary definition and reviews five sources: a bleeding-risk tree for women with intracerebral hemorrhage, a prehospital tree built on the mSOFA score, a prognosis model for palliative radiotherapy, a conversational dementia screening tool, and a comparison of machine learning models for Cushing’s syndrome.

A chatbot is an AI program that simulates conversation with people, using pre-set phrases and text or audio inputs, and is often used to provide round-the-clock service. This post explains the HIMSS Dictionary definition and distinguishes rule-based from generative chatbots, then reviews five sources: a qualitative review of how chatbots change patient-clinician communication, a systematic review in mental health screening, a review of pediatric chatbots, a review of randomized trials on preoperative anxiety, and a randomized trial in urogynecology.

An artificial neural network, or ANN, is a computational model based on the structure and function of biological neural networks that learns relationships between inputs and outputs. The HIMSS Dictionary spells it Artificial Neuron Network. This post explains the definition and reviews four sources: a meta-analysis comparing ANN and logistic regression for trauma outcomes, a review in digestive disease diagnosis, a nurse-collected image study of peristomal skin, and a stroke length-of-stay model.

Computer-aided detection, or CAD, combines artificial intelligence and image processing to help radiologists find findings they might otherwise overlook. This post explains the HIMSS Dictionary definition, including the related term computer-aided diagnosis, and reviews four sources: a systematic review of CAD in mammography screening, a 2026 meta-analysis of AI across 26 screening studies, a 109-facility study of an AI workflow, and a pediatric CT study of lung nodule CAD.

Cognitive computing is a type of computing aimed at complex, ambiguous, and uncertain problems, where data are dynamic and may conflict, and where the goal is a decision that is best for the situation rather than a single right answer. This post explains the HIMSS Dictionary description and reviews six sources: two systematic reviews, a colon cancer study of 656 patients with 48.9 percent absolute concordance, a meta-analysis of 9 studies and 2,463 patients, a urologic oncology study with 92.7 percent concordance, and a pilot of an automated problem list.

An algorithm is a step-by-step procedure for solving a problem or calculating a result, and in data mining it fixes the settings that shape a model. This post explains how simple clinical algorithms differ from machine learning models, what a widely cited study found about racial bias in a commercial risk algorithm, what an external validation found for a widely implemented sepsis prediction model, and what governance questions follow.

Voice recognition is the computer analysis of human speech to interpret words and phrases or to identify a particular voice. This post explains how traditional dictation differs from ambient AI scribes, what a controlled evaluation of eight systems found about clinically significant transcription errors, what two emergency department studies found about documentation time, and why interpreter-mediated visits are a risk.

Computer vision is artificial intelligence that enables computers to identify objects and content in images and video. This post explains how it is used in medical imaging, screening, and hospital monitoring, what a pneumothorax meta-analysis and a screening review found, and why privacy, bias, generalizability, and workflow integration need governance.

ChatGPT is a large language model chatbot that generates fluent text from patterns learned in massive training data. This post explains what it is, what studies found when it drafted answers to patient questions, where accuracy, privacy, and bias risks arise, and how nurses and informaticists can help govern its use.

Predictive modeling uses historical and current data to build statistical or machine learning models that estimate the likelihood of future events, such as deterioration, sepsis, or readmission. This post explains how models are built and validated, why external validation and ongoing monitoring matter, concerns about bias and alert fatigue, federal transparency requirements, and how nurses use predictions.

A foundation model is a large AI model pretrained on vast amounts of data that can be adapted to many different tasks. This post explains how pretraining and fine-tuning work, how foundation models power generative AI tools in healthcare, the risks of hallucination, bias, and privacy exposure, and the governance and clinical oversight they require.

Deep learning is a form of machine learning that uses many-layered neural networks to learn complex patterns directly from data such as images, signals, and text. This post explains how deep learning works, landmark healthcare studies in retinal and skin imaging, where deep learning appears in practice, and why validation, bias, and oversight remain essential.

Natural language processing (NLP) is the branch of artificial intelligence that helps computers understand and work with human language. This post explains how NLP extracts meaning from clinical notes, where it is used in healthcare, how large language models and ambient documentation are changing the field, and why accuracy, bias, and human review still matter.

Machine learning is the part of artificial intelligence that learns patterns from data instead of following hand-written rules. This post explains the main types of machine learning, why models must be validated locally, what can go wrong, and why nursing judgment remains the safety layer.

Artificial Intelligence in healthcare is not a single technology; it is a family of techniques — rule-based systems, machine learning, natural language processing, and generative AI — each with distinct capabilities, risks, and governance requirements. This post breaks down the four categories, the ONC HTI-1 transparency rule, the FDA framework, and how AI shows up at the bedside and on the CAHIMS exam.

By Eric Andrew Kristof, RN | March 2026 I recently completed the CompTIA AI Essentials v2 course. Among the course material was a section covering…