South Africa: Updates to the Health Professions Council ethical guidelines

On 20 November 2025, the Health Professions Council of South Africa (HPCSA) announced, revisions to two booklets: Booklet 7 – Withholding and Withdrawing Treatment, and Booklet 20 – AI Guidelines on Ethical Use of Artificial Intelligence (AI).

All medical practitioners registered with the HPCSA are expected to familiarise themselves with these updated guidelines. Highlights of the changes are outlined below.

Booklet 7 – Guidelines for the Withholding and Withdrawing of Treatment

This booklet provides guidance for health practitioners who face the decision of whether to withhold or withdraw life-prolonging treatment. The cornerstone of these guidelines has always been the protection of patient dignity.

End-of-life decisions must, naturally, be made in the patient’s best interests and in consultation with the patient themselves or their nominated representative (a new concept) or close relatives of the patient.

The booklet also makes provision for withdrawing or withholding of treatment in the face of resource scarcity and where clinical need justifies giving priority to certain patients over others, but these aspects are unchanged.

The new revisions of these guidelines are set out below.

• The ‘definitions’ section sees the insertion of ‘Patient representative’, being defined as a person nominated by the patient as an appropriate substitute in making healthcare decisions. Such a person should be chosen based on their own willingness and ability to make decisions that reflect the patient’s most recent wishes if explicitly stated; otherwise, their likely wishes.
• The patient must be consulted before a decision to withdraw or withhold treatment is made, provided they are competent to participate in decision-making. The guidelines recognise, however, that a patient may not wish to know or discuss the details of such a decision. In that case, the health practitioner may now, in terms of the new guidelines, provide the patient’s nominated representative with sufficient information to enable an informed decision to be made.
• If a patient lacks the capacity to decide, decisions relating to their treatment can be made on their behalf. Although it was previously alluded to, the revised guidelines now explicitly state that a patient may, in writing, grant another person authority to consent on their behalf if they are unable to do so themselves. Alternatively, a court order or applicable law may authorise consent when a patient is unable to provide their own consent. If neither of the above circumstances is satisfied, consent should be given by the patient’s spouse or partner or, in the absence of such a spouse or partner, a parent, grandparent, adult child or sibling, in that specific order.
• These guidelines have long recognised the health practitioner’s right to withhold or withdraw treatment even where the patient or their family request continued treatment, if the health team deems treatment to be futile. The revised guidelines, however, now expressly include a proviso that such a decision can only be taken if it is in the best interests of the patient.

Booklet 20 – Guidelines on the Use of Artificial Intelligence

These guidelines are drafted with the view that innovation is welcomed where it improves healthcare delivery, whilst ensuring and promoting safe practice in the health profession. An open-minded, yet cautious approach is therefore adopted. AI tools can support, enhance and expedite clinical decision-making, but their use may also raise certain concerns, such as:

• Patient confidentiality and data protection cannot always be guaranteed. Anonymisation of data does not provide enough protection when machine-learning algorithms can identify an individual from complex data sets, which could put patient data privacy at risk.
• A risk of further marginalisation of minorities and disadvantaged populations, as there is less data available about their population dynamics and characteristics in AI platforms. Clinical decisions must therefore always be made with reference to the specific circumstances within which a practitioner works, rather than relying blindly on AI without critical discernment.
• A current lack of regulatory oversight to ensure safety and clinical efficacy, as well as a lack of widely accepted standards, clear laws and regulations or any shared understanding of the various AI tools.

Considering the above challenges and risks, the HPCSA’s ethical guidelines outline the following three ‘pillars of AI’:

1. Ethics – health practitioners should always be mindful that AI tools should support ethical matters such as respect for patient autonomy, privacy and confidentiality.
2. Legal – platforms should meet the standards of South African laws and legislation, including the Protection of Personal Information Act and the Promotion of Access to Information Act.
3. Technical – healthcare AI tools should have defined criteria, features and capabilities. Systems should be capable of meeting intended objectives and fulfilling practitioners’ and patients’ needs. Further, AI applications should be safe, reliable and sufficiently robust to meet the data quality management, interoperability and cybersecurity standards.

The primary rule highlighted within the guidelines is that any use of AI in healthcare must be patient-centred, for the benefit of the patient’s health and well-being and to improve health outcomes. There is limited room for AI that serves only to make practitioners’ lives easier.

The revised guidelines largely echo the principles set out in the previous version. However, they provide some additional direction and clarity on the use of AI within the healthcare profession. These changes are outlined below.

• The new guidelines make it clear that the use of AI may never usurp a health practitioner’s own expertise and discernment. The practitioner must always make the final decision on patient care, maintaining accountability and oversight. AI in healthcare must only occur with appropriate ethical oversight.
• The standard of healthcare must also remain consistent regardless of whether AI is used or not. The new guidelines state that ‘patients who decline the use of AI may not be disadvantaged or refused access to health services’.
  There are now also limitations on which forms of AI may be used, in that only AI that has been validated for reliability and cultural appropriateness should be used. In this context, it is noteworthy that the South African Health Products Regulation Authority (SAHPRA) outlined, on 25 September 2025, its regulatory requirements for AI and machine learning enabled medical devices.
• AI/ machine learning, according to SAHPRA, includes, without limitation, applications involving:

• • medical imaging analysis (eg, software using AI to detect tumours or fractures in radiological images);
  • predictive algorithms (eg, a machine learning model that forecasts the risk of patient deterioration);
  • clinical decision support systems (eg, AI-driven diagnostic aids or treatment recommendation systems for healthcare professionals); and
  • wearable health monitoring technologies (eg, wearables that analyse bio signals and alert to abnormalities).

Key takeaways

Regarding end-of-life decision-making, practitioners should be aware of the potential that the patient may have, in writing, a nominated representative and the guidelines that surround this.

Regarding the use of AI, this is a fast-moving, developing area, and practitioners should ensure that they do not rely on informal or unrecognised tools in the same way, we suggest that practitioners would not resort to the use of a medicine or medical device that is unrecognized and unstudied.