AMARILLO, TX – The term AI (artificial intelligence) is meant to conjure up science fiction concepts of androids and robots, but let’s start with some caution. For the last few years, AI is most fundamentally a marketing term used by IT companies to sell products (“technical solutions”) to other companies.
Still, behind that marketing there are some pretty exciting developments. Although “AI” is used to capture a great variety of enhancements to old-fashioned computer programs, what AI boils down to is the ability of a digital algorithm (at core, a computer) to mimic certain features of ordinary human intelligence.
New forms of technology require less up-front specialization to get started and those new forms also typically generate results that are closer to user expectations. The result is that our computers can perform a larger variety of tasks, and much of the work of “training” the computers to perform those tasks can be done in advance. Computing is typically done off-site at large data-centers, not handled by on-site computing hardware.
Types of AI
The predominant four types of AI are:
- Rule-Based Systems – fixed decision trees and deterministic outputs
- Machine Learning – learning patterns from data and predictive decision making
- Computer Vision – image interpretation and document recognition
- Generative AI – creates text, summarizes information, generates code, and produces images
Generative AI
Generative AI predicts likely outputs, does not think like people, does not understand context the way people do, does not have anything quite like a conceptual map of the world, can produce convincing mistakes, and requires human review. It is important that AI be treated like any other tool, not an autonomous decision maker. Major players include OpenAI, Microsoft, Anthropic, Google, and Meta.
State and Federal Legislative Attempts at Definition
Under the National Artificial Intelligence Initiative, Artificial Intelligence is defined as follows: The term “artificial intelligence” means a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations or decisions influencing real or virtual environments. Artificial intelligence systems use machines and human-based inputs to—
(A) perceive real and virtual environments;
(B) abstract such perceptions into models through analysis in an automated manner; and
(C) use model inference to formulate options for information or action.
(5 U.S. Code § 9401.)
For comparison, here’s California’s general legal definition: “Artificial Intelligence” means an engineered or machine-based system that varies in its level of autonomy and that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.
(CA Govt Code § 11546.45.5(1).)
Benefits of AI for DME Suppliers
Potential benefits of AI for DME suppliers include labor savings, faster processing, improved documentation, better patient engagement, more accurate inventory forecasting, reduced denial rates, and scalable operations.
AI can assist with reading prescriptions, extracting demographic information, identifying missing records, categorizing referrals, and prioritizing work queues.
Other potential AI uses include medical necessity review, collection of missing documents, appeal drafting, prior-auth package assembly, and workflow monitoring. At the same time, human review remains essential.
AI may compare coverage criteria, analyze plan requirements, flag missing documentation, and review payer rules. Benefits include faster qualification decisions and reduced rework.
AI can assist with multiple documentation functions, including draft call summaries, draft intake notes, meeting summaries, draft correspondence, and creation of policy summaries. However, be aware that AI frequently introduces factual inaccuracies.
Regarding PAP and sleep therapy, AI can assist with compliance monitoring, resupply prediction, adherence outreach, patient communication, and predictive scheduling. These AI functions can result in better adherence rates and reduced staff workload.
Regarding the provision of oxygen, AI can assist with equipment utilization, delivery forecasting, patient follow-up, route optimization, and maintenance scheduling.
Regarding the provision of infusion, AI can assist with inventory forecasting, operational task automation, supply-chain analysis, workflow optimization, and risk identification.
For delivery and logistics, AI may optimize vehicle routing, driver schedules, dispatching timing, fuel utilization, and equipment retrieval. Benefits include lower transportation costs and faster service.
Potential applications of AI for revenue cycle management include claim review, denial prediction, collection prioritization, payment forecasting and coding assistance.
Regarding patient engagement, AI may support appointment reminders, equipment reminders, therapy education, refill notifications and chatbot communication. Be aware that communication laws still apply.
Regarding fraud detection, AI is increasingly used to identify billing anomalies, utilization outliers, suspicious referral behavior, prescribing trends and potential fraud schemes.
Hallucinations occur when AI invents facts, creates fake citations, misstates legal requirements and generates incorrect patient information. DME suppliers must verify outputs.
Automation bias occurs when humans assume software is correct, skip independent and accept recommendations without scrutiny. This creates legal exposure.
Bias and discrimination include unequal outcomes, hidden training-set bias, disparate impact and algorithmic discrimination. Regulators increasingly focus on these issues. Examples of bias and discrimination include:
- Racial bias and clinical risk – AI sometimes uses healthcare spending to predict healthcare needs; groups that spend less will be disadvantaged by the use of this proxy;
- Underrepresentation in training data – Training data typically has more data from Caucasians than minority groups. Training data typically includes more data from men than women. Both make for a larger error rate in correct interpretation of the less-represented groups
- Socioeconomic bias – Insurers profiling for readmission risk or future can be driven by cost reduction rather than improving clinical outcomes.
From a cybersecurity risk standpoint, questions providers should ask include:
- Where is data stored?
- Who can access the data?
- Is data used for training?
- Is encryption used?
- Are audit logs maintained?
AI-generated errors can result in liability for the submission of false claims. Examples include unsupported claims, improper coding, false documentation, and invalid medical necessity statements. DME suppliers are responsible for AI-generated errors.
Government agencies increasingly use AI for data mining, outlier analysis, fraud detection, audit targeting, and investigations.
Hallucinations occur when AI invents facts, creates fake citations, misstates legal requirements, and generates incorrect patient information. DME suppliers will need to carefully review and verify any adverse claims allegations from a government or commercial reviewer.
Potential Software as a Medical Device (“SaMD”) issues arise when software diagnoses disease, recommends treatment, influences clinical judgment. There is an important distinction between supporting decisions and making decisions.
An SaMD risk pertains to reimbursement compliance. AI-driven tools are increasingly used to evaluate patient eligibility, support medical necessity documentation, monitor compliance, and generate information relied upon in claims submissions. If the software produces inaccurate conclusions that affect billing or coverage determinations, suppliers may face audit exposure, overpayment demands, or False Claims Act scrutiny.
Another risk pertains to product liability. Traditionally, supplier liability focused on equipment setup, maintenance, and patient instruction. With AI, plaintiffs may instead challenge the reliability of software-generated recommendations, predictive analytics, or automated clinical assessments. Questions concerning responsibility become more complex when providers and suppliers rely on algorithmic outputs to make healthcare decisions.
Jeffrey S. Baird, Esq. is chairman of the Health Care Group at Brown & Fortunato, PC, a law firm with a national health care practice based in Texas. He represents pharmacies, infusion companies, HME companies, manufacturers and other health care providers throughout the United States. Baird is Board Certified in Health Law by the Texas Board of Legal Specialization and can be reached at (806) 345-6320 or [email protected].
Blinn E. Combs, Esq. is a healthcare attorney at Brown & Fortunato, a national, full-service law firm delivering exceptional quality and personalized service. Combs advises pharmacies, DME suppliers, and other healthcare providers on regulatory compliance, risk mitigation, and state licensure matters. His practice focuses on complex state and federal healthcare regulations, with experience in pharmacy law, pharmaceutical compounding, and operational compliance. Blinn provides practical guidance to healthcare businesses seeking to strengthen compliance, manage risk, and support growth in a highly regulated industry. Combs can be reached at (806) 345-6355 or [email protected].
