How do clinicians bill for remote patient monitoring and digital health services?

Established CPT code families exist for remote physiologic monitoring (the 99453/99454 setup and device-supply codes plus 99457/99458 for management time), remote therapeutic monitoring (the 98975-98981 family), chronic care management, and telehealth evaluation and management visits. Coverage and payment rates vary by payer, plan, and year, so verify current codes and requirements with your billing team and payers before implementing.

What is a prescription digital therapeutic and how do I prescribe one?

A prescription digital therapeutic (PDT) is FDA-authorized software that delivers an evidence-based intervention for a specific condition and requires a clinician's order. Prescribing typically works through the product's hub or pharmacy pathway: you issue the order, the patient receives an access code, and many products report engagement data back to you. Coverage varies by payer and is evolving, so confirm benefits before prescribing.

If a patient shares wearable or app data with me, am I obligated to monitor it?

The prevailing guidance is to define scope in advance: tell patients explicitly whether their data will be reviewed continuously, at visits, or only when they flag a concern. Unbounded expectations create both clinical risk and workload burden. Formal remote monitoring programs should specify who reviews data, how often, and what triggers outreach.

How do I keep up with digital health and clinical AI?

A sustainable approach combines one or two peer-reviewed journals in digital medicine, a small set of curated newsletters, and one conference per year matched to your specialty. Digital.Health maintains curated directories of digital health journals, newsletters, conferences, podcasts, and reports to make this efficient.

A Digital Health and AI Guide for Clinicians

Q: What is the difference between FDA clearance and FDA approval for digital health tools?

Clearance (510(k)) means a device was shown substantially equivalent to an existing predicate device. De Novo authorization creates a new classification for novel low-to-moderate-risk devices. Approval (PMA) is the most rigorous pathway, reserved for high-risk devices and requiring clinical evidence of safety and effectiveness. Most digital health and AI tools reach market via 510(k) or De Novo. General wellness products require no premarket review at all.

Q: Am I liable if I recommend a health app to a patient?

Recommending digital tools is increasingly considered part of ordinary clinical practice, and documented, evidence-based recommendations are the norm rather than an outlier risk. Reasonable practice includes preferring tools with published evidence and appropriate regulatory status, documenting the rationale, setting patient expectations about what the tool does and does not do, and clarifying whether you will review incoming data. For specific concerns, consult your malpractice carrier or counsel - this guide is educational, not legal advice.

Q: How should I evaluate an AI tool before using it clinically?

Ask six questions: What population was it trained and validated on, and does that match your patients? Is there prospective or external validation, not just retrospective performance? How does it handle edge cases and communicate uncertainty? What monitoring exists for performance drift after deployment? Where does a human review or override its output? And does it fit your workflow, or add clicks and burden?

Q: What should I do when a patient brings me an app or AI tool they found?

Treat it as engagement, not an interruption. A brief structured response works: ask what they want it to help with, check the tool's evidence and regulatory status if it makes medical claims, flag privacy considerations for consumer apps, and either endorse it, suggest a validated alternative, or explain your reservations. Pointing patients to a plain-language evaluation framework saves visit time.

A working guide to digital health and AI in clinical practice for physicians, pharmacists, nurses, advanced practice providers, and care teams, from Digital.Health by Daniel Kraft, MD. Topics: appraising evidence for digital health tools including randomized controlled trials and real-world evidence; FDA regulatory pathways including general wellness exemption, enforcement discretion, 510(k) clearance, De Novo authorization, premarket approval (PMA), Software as a Medical Device (SaMD), and Predetermined Change Control Plans (PCCP) for AI/ML devices; reimbursement including remote patient monitoring (RPM) CPT code families 99453, 99454, 99457, 99458, remote therapeutic monitoring (RTM) codes 98975-98981, chronic care management, and telehealth billing; evaluating clinical AI including ambient documentation, clinical decision support, imaging AI, algorithmic bias, and model drift; prescribing digital therapeutics and prescription digital therapeutics (PDTs); building a digital formulary; and a five-step adoption framework for integrating digital health tools into clinical workflow.

Orientation

Why this belongs in your practice, now.

The working definitionDigital health, in clinical terms: software, sensors, and AI that extend what you can measure, monitor, and deliver beyond the visit — from remote physiologic data and validated digital therapeutics to AI that documents, triages, and supports decisions.

Three shifts have moved digital health from conference-keynote material to Tuesday-afternoon clinical reality. Patients arrived first: a substantial share of your panel is already using wearables, health apps, and consumer AI assistants — and increasingly bringing that data and those conclusions to visits. Reimbursement followed: established CPT code families now pay for remote physiologic and therapeutic monitoring, and telehealth is a routine covered modality. The tools matured: a growing set of products carries FDA authorization and peer-reviewed evidence, while AI has moved into documentation, imaging, and decision support inside mainstream EHR workflows.

The practical problem is no longer whether to engage, but how to separate the small set of tools worth adopting from a very large market of tools that will cost you time, add inbox burden, or expose patients to unvalidated claims. This guide is organized around that filtering problem.

A note on the word clinician: throughout this guide it means the whole care team — physicians, advanced practice providers, nurses, and pharmacists alike. The evidence, regulatory, and AI frameworks here apply across professions. Pharmacists will also find a dedicated companion, The Pharmacist’s Guide to Digital Health & AI-Enabled Care, which goes deep on pharmacy-specific services and payment models.

First principles

Appraising evidence when the intervention is software.

Digital health evidence doesn’t map neatly onto the pharma model, and applying drug-trial expectations uncritically will lead you both to reject useful tools and to accept weak ones. Four adjustments matter.

Product velocity outpaces trial timelines. The version studied in a two-year RCT is rarely the version in the app store today. Weight evidence recency, and look for developers who publish updated real-world performance — not just a single foundational trial from several versions ago.

Engagement is an effect modifier. A digital intervention only works in patients who use it. Interrogate retention curves, not just intention-to-treat outcomes: a tool with strong per-protocol effects and 90% three-month attrition will not reproduce its trial results in your panel.

Check the comparison arm. “Better than waitlist” is a low bar. Superiority or non-inferiority against active treatment, usual care, or an attention-matched digital control is far more informative.

Distinguish evidence tiers. Peer-reviewed RCTs and prospective studies sit at the top; registries and real-world evidence in the middle; white papers, posters, and testimonials at the bottom. A company that can only produce the bottom tier for a medical claim is telling you something.

Reading the label

Regulatory pathways, decoded.

A tool’s regulatory status tells you how much premarket scrutiny it received — and what the manufacturer was allowed to claim. Here’s the map, from least to most oversight.

Pathway	What it means	Typical products	Scrutiny
General wellness	No premarket review; product makes only low-risk wellness claims, not disease claims.	Fitness trackers, meditation apps, habit tools	None
Enforcement discretion	Technically a device function, but FDA has stated it won’t enforce requirements due to low risk.	Many symptom trackers, some clinical decision support where clinicians can review the basis	Minimal
510(k) clearance	Demonstrated substantial equivalence to a predicate device already on the market.	Most SaMD, ECG features, many imaging-AI tools	Moderate
De Novo	Novel low-to-moderate-risk device with no predicate; creates a new classification.	First-of-kind digital therapeutics and AI diagnostics	Moderate–high
PMA approval	Full premarket approval with clinical evidence of safety and effectiveness; highest-risk class.	A small set of high-risk devices and software	Highest

AI/ML-specific: because adaptive algorithms change after clearance, FDA’s Predetermined Change Control Plan (PCCP) framework lets manufacturers pre-specify permitted model updates. When evaluating an AI device, ask whether it has a PCCP and how post-market performance is monitored — the clearance date tells you about the model as it was, not necessarily as it is.

Getting paid

Reimbursement: the code families that matter.

Digital health billing is more established than many clinicians assume. These are the major categories — use them as a map, and verify current-year codes, requirements, and payer policies with your billing team.

Category	What it covers	Example code family
Telehealth E/M	Evaluation & management visits delivered by video or audio, widely covered across commercial plans, Medicare, and Medicaid.	Standard E/M codes with telehealth modifiers
Remote patient monitoring (RPM)	Collection and management of physiologic data — blood pressure, glucose, weight, oximetry — from home devices.	CPT 99453/99454 (setup & device supply); 99457/99458 (management time)
Remote therapeutic monitoring (RTM)	Non-physiologic therapeutic data: medication adherence, respiratory status, musculoskeletal therapy response.	CPT 98975–98981 family
Chronic care management (CCM)	Non-face-to-face care coordination for patients with multiple chronic conditions — often the wrapper around digital programs.	CCM code family
Prescription digital therapeutics	Coverage is payer-by-payer and evolving; some state Medicaid programs and commercial plans cover specific PDTs.	Product- and payer-specific; verify benefits

Practical caution: codes, thresholds (such as required monitoring days and minutes), and payment rates change with each fee schedule cycle. Treat this table as orientation, not a billing reference — and confirm requirements with payers before building a program around them.

The AI layer

Clinical AI: where it works, and how to vet it.

AI is entering practice through four main doors, each with a different risk profile.

🎤

Ambient documentation

AI scribes that listen to the encounter and draft the note. The most rapidly adopted category — and the lowest-risk, because every output passes through your review and signature. The evaluation question is accuracy and edit burden, not autonomy.

🖼️

Imaging & diagnostics

The most mature regulated category, with hundreds of cleared devices across radiology, cardiology, ophthalmology, and pathology. Typically deployed as a second reader or triage layer rather than an autonomous diagnostician.

⚠️

Decision support & risk prediction

Sepsis alerts, deterioration scores, care-gap flags. The category where validation gaps matter most: performance published on one health system’s data frequently degrades on another’s. Demand external or local validation before trusting thresholds.

💬

Patient-facing AI

Symptom checkers, AI health assistants, and chat-based coaching your patients use before and after they see you. You can’t control this layer — but you can shape it by recommending validated options and teaching patients how to use AI outputs as conversation starters.

1. Population match. What data was the model trained and validated on — and does that population resemble your panel in demographics, acuity, and setting?

2. Validation depth. Retrospective performance is a floor, not a ceiling. Is there prospective, external, or site-local validation?

3. Failure behavior. How does it handle edge cases, and does it communicate uncertainty — or fail silently with a confident answer?

4. Drift monitoring. Who watches performance after deployment, and what triggers retraining or rollback?

5. Human oversight. Where exactly does a clinician review, override, or sign off? If the answer is “nowhere,” the bar for evidence rises steeply.

6. Workflow cost. Does it remove clicks or add them? Alert burden and documentation overhead are patient-safety issues, not conveniences.

Beyond the pill

Prescribing & recommending digital tools.

Digital therapeutics extend your formulary. Evidence-based software interventions now exist for insomnia, substance use disorders, chronic pain, IBS, ADHD, and a growing list of conditions — some over-the-counter, some as prescription digital therapeutics (PDTs) requiring your order. Prescribing typically runs through the product’s hub or pharmacy pathway: you order, the patient receives access, and many products report engagement and outcomes back to you.

Recommendation is a clinical act — treat it like one. Prefer tools with published evidence and appropriate regulatory status, document your rationale, set expectations about what the tool does, and be explicit about whether and when you’ll review incoming data. Unbounded data-review expectations are the fastest way to turn a good tool into a liability and workload problem.

Build a digital formulary rather than deciding ad hoc. A short, maintained list of vetted tools for your most common scenarios — reviewed the way a P&T committee reviews medications — converts a recurring judgment call into a repeatable system. This is exactly what Digital.Health’s clinician tools support: search 5,000+ curated solutions, build a formulary of favorites for your practice, and share or “prescribe” via the virtual RxPad.

When patients bring you tools, meet them there. It’s engagement, not an interruption. For the patient side of that conversation, our plain-language consumer guide to digital health and 8-question evaluation checklist are built to hand off.

Putting it together

A five-step framework for adopting any digital tool.

Whether you’re a solo clinician adding one app or a group rolling out remote monitoring, the sequence is the same.

Define the clinical problem first

Start from a gap — uncontrolled hypertension between visits, CBT-I access, documentation burden — not from a product demo. Tools chosen problem-first get used; tools chosen demo-first get abandoned.

Appraise evidence & regulatory status

Apply the evidence lens from this guide: comparison arm, engagement data, evidence recency, and a regulatory status appropriate to the claims being made.

Map the workflow & data flow

Decide before launch: who enrolls patients, where data lands, who reviews it and how often, what triggers outreach, and how it’s documented and billed. If this map can’t be drawn simply, the tool isn’t ready for your practice.

Pilot with predefined outcomes

Run a small cohort for a defined period with success criteria written down in advance — clinical measures, patient engagement, and staff time. A pilot without predefined criteria is just a slow purchase.

Monitor, iterate — and be willing to drop it

Review outcomes on a schedule, watch for drift in both the tool and the workflow, and retire tools that stop earning their place. A digital formulary is a living document, like any formulary.

Reference

A clinical digital health glossary.

Eighteen terms you’ll encounter in product materials, FDA summaries, and the literature.

Software as a Medical Device (SaMD): Software intended for a medical purpose that functions as a medical device in its own right, without being part of hardware.
510(k) clearance: FDA premarket pathway demonstrating substantial equivalence to a legally marketed predicate device.
De Novo authorization: FDA pathway creating a new device classification for novel low-to-moderate-risk devices without a predicate.
Premarket approval (PMA): The FDA’s most stringent pathway, for high-risk devices, requiring clinical evidence of safety and effectiveness.
Enforcement discretion: FDA’s stated intent not to enforce requirements for certain low-risk software functions, including many wellness and some CDS tools.
Predetermined Change Control Plan (PCCP): FDA mechanism letting manufacturers pre-specify how an AI/ML device may be updated post-market without new submissions per change.
Remote patient monitoring (RPM): Clinician-ordered collection and review of physiologic data from patients outside the clinic, with dedicated CPT code families.
Remote therapeutic monitoring (RTM): Monitoring of non-physiologic therapeutic data — adherence, respiratory status, musculoskeletal response — with its own code family.
Digital therapeutic (DTx): Software delivering an evidence-based therapeutic intervention for a medical condition.
Prescription digital therapeutic (PDT): An FDA-authorized DTx requiring a clinician’s order, typically indicated for a specific condition.
Clinical decision support (CDS): Software providing knowledge or patient-specific assessments to inform decisions; regulatory treatment depends on whether clinicians can independently review its basis.
Ambient documentation: AI systems that listen to the clinical encounter and draft notes, orders, or summaries for clinician review and sign-off.
Real-world evidence (RWE): Clinical evidence derived from EHRs, claims, registries, and device data, increasingly used alongside trials in digital health.
Model drift: Degradation of an AI model’s performance over time as populations, practice patterns, or data inputs shift from training conditions.
Algorithmic bias: Systematic error in an AI tool’s outputs across patient subgroups, often traceable to unrepresentative training data.
Digital formulary: A curated, maintained list of vetted digital health tools a practice recommends or prescribes, analogous to a medication formulary.
FHIR: Fast Healthcare Interoperability Resources — the dominant standard for exchanging health data between EHRs, apps, and devices.
Patient-generated health data (PGHD): Health data created and recorded by patients outside clinical settings, from symptom logs to wearable streams.

Common questions

Clinicians frequently ask.

What’s the difference between FDA clearance and approval?

Clearance (510(k)) means substantial equivalence to an existing predicate device. De Novo authorization creates a new classification for novel low-to-moderate-risk devices. Approval (PMA) is the most rigorous pathway, reserved for high-risk devices with clinical evidence of safety and effectiveness. Most digital health and AI tools reach market via 510(k) or De Novo; general wellness products require no premarket review at all.

How do I bill for remote monitoring and digital health services?

Established CPT code families cover remote physiologic monitoring (99453/99454 for setup and device supply, 99457/99458 for management time), remote therapeutic monitoring (98975–98981), chronic care management, and telehealth E/M visits. Requirements and rates change with each fee schedule cycle — verify current codes and payer policies with your billing team before building a program.

Am I liable if I recommend an app to a patient?

Recommending digital tools is increasingly ordinary clinical practice. Reasonable practice includes preferring tools with published evidence and appropriate regulatory status, documenting your rationale, setting expectations about what the tool does, and clarifying whether you’ll review incoming data. For specific concerns, consult your malpractice carrier or counsel — this guide is educational, not legal advice.

How should I evaluate an AI tool before using it clinically?

Six questions: Does the training and validation population match your panel? Is there prospective or external validation, not just retrospective performance? How does it behave on edge cases and communicate uncertainty? Who monitors for drift after deployment? Where does a human review or override? And does it reduce workflow burden or add to it?

How do I prescribe a digital therapeutic?

Prescription digital therapeutics typically run through the product’s hub or pharmacy pathway: you issue the order, the patient receives an access code, and many products report engagement data back to you. Coverage varies by payer and is evolving — confirm benefits before prescribing, as you would for a specialty medication.

If a patient shares wearable data, must I monitor it continuously?

No — but define the scope explicitly and in advance. Tell patients whether their data will be reviewed continuously, at visits, or only when they flag a concern. Formal remote monitoring programs should specify who reviews data, how often, and what triggers outreach. Unbounded expectations create both clinical risk and inbox burden.

What do I do when a patient brings me an app or AI answer?

Treat it as engagement. Ask what they want it to help with, check evidence and regulatory status if it makes medical claims, flag privacy considerations for consumer apps, and either endorse it, suggest a validated alternative, or explain your reservations. Handing them our plain-language guide and evaluation checklist saves visit time.

How do I keep up with the field without drowning?

One or two peer-reviewed digital medicine journals, a small set of curated newsletters, and one conference a year matched to your specialty covers most of the signal.

Cite this page Kraft D. Digital Health & AI in Clinical Practice: A Working Guide for Clinicians. Digital.Health; 2026. Available at: https://digital.health/digital-health-ai-guide-for-clinicians

Build your digital formulary.

Search 5,000+ clinician-curated solutions across 40 specialty and care categories, create a formulary for your practice, and recommend tools via the virtual RxPad.

Explore clinician tools → Browse by specialty

Curated with clinical oversight · Founded by Daniel Kraft, MD