What is ambient AI?

“Ambient AI scribes passively listen to clinical encounters and automatically generate documentation, helping reduce workload, documentation time, and burnout while improving patient-physician interaction and workflow efficiency,” explains a narrative review on transforming clinical documentation with ambient artificial intelligence (AI) scribes.

As a result, ambient AI scribes combine automated speech recognition (ASR), natural language processing (NLP), and generative AI to capture clinician-patient conversations and automatically generate clinical documentation. “Ambient AI scribes show promise in reducing workload, improving efficiency, and decreasing burnout, but current systems still generate high omission rates and intermittent factual inaccuracies that may affect clinical decision-making.”

 

How ambient AI works

The scribing technology identifies relevant information and organizes it into documentation formats, like SOAP notes, which include subjective information, objective findings, assessment, and plan.

Ambient AI can interpret conversational language and generate a structured draft, where the provider doesn’t have to dictate the notes. This can then be transferred into an EHR for review and approval, as the system itself does not establish whether a patient's examination findings or treatment plan have been interpreted correctly. Healthcare providers are still responsible for reviewing the generated documentation.

 

Five ways ambient AI is being used in healthcare

1. Automating clinical documentation

Ambient AI tools can draft histories, examination findings, assessments, and treatment plans from conversations. Healthcare providers can then correct omissions and add information that was not captured during the encounter.

The above-mentioned review specifically discusses that providers who used an ambient AI called Nuance Dragon Ambient eXperience (DAX) Copilot reduced their time spent documenting each visit from 10.3 to 8.2 minutes, a 20.4% decrease. The review also mentions how such improvements could give providers more time for patient care and clinical decision-making.

 

2. Reducing provider burnout

“One potential barrier to optimal healthcare may be provider burnout or occupation-related stress in the workplace,” explains Predictors of burnout among US healthcare providers.

Administrative workload can contribute to this burnout, especially if the provider’s documentation must be done during each consultation. Ambient AI helps reduce this burden, so providers can concentrate on patient care. The review on transforming clinical documentation discusses how this process can support cognitive offloading, a concept describing the transfer of some mental tasks to an external system. In this context, clinicians may feel less pressure to remember every detail of a consultation while simultaneously listening and making clinical decisions.

Nevertheless, the findings on burnout were not uniform, suggesting a productivity paradox where some providers worry that time saved through automation will simply be used to schedule more appointments. Consequently, healthcare organizations should first assess whether ambient AI improves their specific working conditions.

 

3. Improving patient-provider interactions

The review says that ambient AI “removes the computer as the third party,” referring to the possibility of restoring a more direct interaction between the patient and provider.

For example, the Evaluation of an Ambient Artificial Intelligence Documentation Platform for Clinicians “found that ambient AI was associated with improved overall experience and time in notes for clinicians,” as full attentiveness increased from 57.9% to 93.0% after implementation “with clinicians feeling less distracted and more engaged.”

 

4. Supporting specialty-specific clinical workflows

Since documentation requirements differ between primary care, surgery, cardiology, oncology, and other clinical settings, the use of ambient AI may not be equally effective across all specialties.

More specifically, the review reports that primary care clinicians generally experienced favorable workflow outcomes while medical subspecialists sometimes experienced smaller benefits. Additionally, some of these subspecialists needed to make more extensive edits because the generated notes did not adequately reflect their documentation requirements.

For example, a system intended for a cardiology consultation must accurately capture examination findings and the clinician's assessment, as misrepresenting a finding could affect how subsequent providers interpret the patient's condition.

Ultimately, healthcare organizations must determine which ambient AI would be most effective in each specialty.

 

5. Supporting AI-assisted clinical workflows

The review describes a future model with an “AI-augmented clinician or co-pilot, where the AI… documents [and] actively supports care by alerting clinicians to missed screening questions, suggesting differential diagnoses based on dialogue, and assisting with ordering and referral workflows.”

These capabilities could help clinicians organize information and identify gaps during a consultation. For example, if a patient with cardiovascular disease mentions dizziness during an AI-assisted consultation, the system could prompt the clinician to assess for orthostatic hypotension or review medications that may contribute. Such prompts would remain advisory, with the clinician responsible for interpreting them in context and deciding whether further assessment is warranted. This is a future direction of advanced clinical support, which would still require provider oversight and appropriate validation, like a specific clinical decision-support tool embedded in the electronic health record (EHR).

 

The accuracy and safety challenges of ambient AI

According to an Instrument Validation Study on the Accuracy and Safety of AI-Enabled Scribe Technology, “AI-enabled ambient digital scribes may have many potential benefits, [yet] there are errors that must be evaluated to mitigate safety risks.”

This validation study “identified a total of 127 errors, for an average of 2.9 errors per note,” where “70% of the notes generated contained at least one error.” Evidently, notes may appear fluent and professionally written while still containing an omission or factual mistake.

A recent comparative study used ChatGPT-4 to Create Structured Medical Notes From Audio Recordings of Physician-Patient Encounters, “classified these errors into three main categories. The most prevalent were omissions, accounting for 86.3% of all errors, where the AI frequently failed to include critical clinical information present in the original transcript, for example, omitting a patient’s loss of appetite in a hernia case or neglecting to mention echocardiogram results in a patient with congestive heart failure.”

“The second most common error type was additions or fabrications (10.5%), in which the AI generated content not present in the source, such as falsely stating that a patient’s weight loss was intentional or that they were noncompliant with medications. The third and least frequent category was incorrect facts (3.2%), where the AI recorded existing clinical data inaccurately, such as reporting a normal heart rate when it was tachycardic or misstating the timing of hospital admission.”

As a result, healthcare organizations must train providers how to recognize these common AI errors. They should also evaluate how systems perform with different accents, dialects, interpreters, and patient populations, since uneven performance could contribute to healthcare disparities.

As the review on transforming clinical documentation warns, “This raises substantial concerns that models trained predominantly on majority-group data may underperform for marginalized populations, potentially exacerbating existing healthcare disparities.” Ultimately, “Such efforts will help ensure that AI-driven documentation systems are both generalizable and equitable, supporting accurate, efficient, and patient-centered care…”

Related: How patient-centered communication improves patient outcomes

 

Privacy and HIPAA considerations for ambient AI

Healthcare organizations subject to HIPAA regulations must assess how their vendors collect, process, store, and transmit patients’ protected health information (PHI). When an AI vendor performs functions involving PHI on behalf of a covered entity, they must sign a business associate agreement (BAA). Organizations should also examine access controls, encryption, retention practices, audit logs, incident response procedures, and whether patient information may be used to train or improve AI models.

Healthcare organizations must understand the clinical implications of using certain AI platforms. As the comparative study on using ChatGPT-4 explains, “It is important to note that open AI platforms, such as ChatGPT-4, are not recommended for clinical use due to the many regulatory and privacy issues.” Healthcare organizations must therefore carefully assess whether an AI tool is appropriate for their clinical workflows and whether its implementation adheres to HIPAA laws.

These safeguards must be reflected in how the ambient AI handles patient information. For example, after an appointment documented with an ambient AI scribe, the provider can securely send a patient follow-up instructions and educational material related to the treatment plan. If these communications contain PHI, they must be transmitted through an appropriately configured and safeguarded communication system.

Paubox offers a HIPAA compliant email solution that helps healthcare organizations protect PHI in email communications, complementing the privacy and security measures required for ambient AI implementation. Paubox email integrates directly into existing workflows like Gmail, allowing providers to continue using the familiar Gmail interface. The solution also automatically encrypts all outgoing emails, so providers and their patients don’t have to navigate inconvenient patient portals.

Paubox also integrates with EHR systems, helping healthcare organizations connect ambient AI-assisted documentation with secure patient communication. For example, after an ambient AI scribe generates clinical notes that are reviewed and incorporated into the EHR, providers can use secure email to communicate follow-up instructions to their patients. Healthcare providers can use ambient AI to reduce documentation burdens and spend more time engaging with patients, while HIPAA compliant email keeps their subsequent communications confidential.

Read also: Ways to use HIPAA compliant email to increase patient engagement

 

FAQs

What are the privacy risks of ambient AI in healthcare?

Ambient AI may process patients’ protected health information (PHI), like their medical histories, symptoms, diagnoses, and treatment discussions. Healthcare organizations must evaluate how these AI vendors collect, store, transmit, and use this information to prevent unauthorized PHI access or a potential HIPAA violation.

 

Is ambient AI HIPAA compliant?

Ambient AI is not automatically HIPAA compliant. Healthcare organizations must verify whether the technology they use meets HIPAA requirements, including safeguards for PHI, appropriate access controls, encryption, and whether the vendor signs a business associate agreement (BAA).

 

Are there any limitations when using AI in HIPAA compliant emails?

Yes, healthcare providers must check that AI-powered features comply with HIPAA regulations and industry best practices for data security and privacy. Additionally, providers should evaluate the reliability of AI algorithms to avoid potential risks or compliance issues.

Read also: HIPAA compliant email API