Ambient Voice Technology challenges in the NHS

Ambient Voice Technology (AVT), or AI scribes, holds significant potential to reduce the administrative burden on NHS staff, but its adoption faces several complex challenges across regulatory, clinical, and ethical domains.

 The main challenges in the NHS surrounding the use of AVT include:

1. Regulatory and Governance Challenges

A primary concern is the proper regulation and assurance of these tools, particularly as many suppliers initially failed to meet NHS standards:

Medical Device Classification: AVT solutions that generate summaries of consultations (rather than just simple transcription) are typically classified as medical devices and must be registered with the Medicines and Healthcare products Regulatory Agency (MHRA), usually as a minimum of a Class I medical device.

Compliance and Unregistered Tools: NHS England has issued warnings for staff to stop using unregistered AI scribe tools that do not meet essential assurance standards, posing risks to patient safety and data protection.

Safety Standards: NHS organisations must ensure compliance with rigorous safety standards like DCB0129 and DCB0160 (Digital Clinical Risk Management), and complete a comprehensive Data Protection Impact Assessment (DPIA) before deployment.

Supplier Compliance: Many vendors approaching the NHS have not complied with basic governance standards, such as providing MHRA clinical safety and technical/cyber platform assurance.

2. Clinical and Safety Risks

Even with compliant tools, the nature of the technology and its use in a clinical setting introduces new risks:

Accuracy and AI Errors: AI models, especially those using Generative AI, can introduce errors, inaccuracies, or even "hallucinations" (generating false or misleading clinical information).

Omission of Nuance: AVT can struggle with the nuances of human communication, such as sarcasm, significant silences, or subtext, particularly in fields like mental health, which can lead to clinically "vacuous" notes

 
Over-reliance and "Editing Fatigue": Clinicians face the risk of becoming overly reliant on the AI's output (automation bias). Furthermore, correcting lengthy, well-written but clinically inaccurate notes can sometimes be more draining than writing them from scratch, leading to "editing fatigue." 
 
Inconsistent Data Capture: Challenges remain with the technology's performance in varied real-world conditions, such as noisy clinical settings, different accents, and when multiple speakers are present.

3. Data Privacy and Consent Concerns

Handling highly sensitive patient conversations requires strict adherence to privacy legislation:

Informed Consent: Patients must be transparently informed that AVT is being used, why, and how their data will be processed, stored, and shared. Consent must be explicit and informed, not just a 'tick-box,' and patients must be offered an alternative or the right to stop the recording.

Biometric Data and Re-identification: Voice recordings are considered biometric data and are difficult to fully anonymise, as accent, pitch, and content can still lead to re-identification, raising significant data protection risks under UK GDPR.

Data Processing and Storage: There are major concerns over where confidential patient data is transferred, how long it is retained, and whether vendors might reuse identifiable data for training their AI models without an appropriate legal basis.

Cybersecurity: Organisations must ensure robust cybersecurity controls, including compliance with the Data Security and Protection Toolkit (DSPT), to protect sensitive recordings and transcripts.

4. Integration and Adoption Barriers
 
The practical implementation of AVT across the vast NHS presents organisational and technical hurdles:

Interoperability: Seamless integration with existing electronic patient records (EPRs), such as EMIS and SystmOne, remains a significant challenge. Without full integration, clinicians are still left with the administrative task of copying and pasting, reducing the technology's benefit.

ragmented Procurement: Procurement processes can be fragmented across different NHS trusts and Integrated Care Boards (ICBs), which can impede standardised, equitable, and widespread adoption.

Lack of Evidence and Capability: There is a call for further robust evidence of AVT's long-term impact on clinical workflows and patient outcomes. Additionally, NHS organisations need to build internal capability to assess and continually monitor these evolving tools.

Impact on Patient-Clinician Trust: The presence of a "listening" non-human "third party" in the room can risk eroding the delicate relationship of trust and psychological safety between a patient and their clinician, especially for vulnerable patients (e.g., those with paranoia or a history of trauma).

To discuss how Nelson Advisors can help your HealthTech, MedTech, Health AI or Digital Health company, please email [email protected]