Clare O’Donnell: “AI is not about replacing clinicians but about optimising capacity”

I believe AI is set to expand teleoptometry, supporting remote screening, monitoring, and patient follow-up. As this develops, AI could help extend care to patients outside traditional clinic settings.

With this brings opportunities for patient care. By deploying AI tools capable of detecting more serious conditions with higher sensitivity and specificity, clinicians can spend more of their time on complex case management. This aligns with wider goals in population health management to enhance capacity, reduce disparity in provision, reduce waiting times, and improve risk-stratification capabilities.

However, as with all opportunities, the barriers and threats should be considered, and this includes the safe integration of AI, which requires an evidence-based approach and an awareness of issues relating to bias and data protection that should be considered.

There is also a need to address any regulatory and medicolegal concerns as AI innovation is often progressing faster than the regulatory frameworks. Naturally it is also important that educational institutions and professional bodies continue to provide AI training.

Optegra

Professor Clare O’Donnell pictured with two Optegra surgeons

Artificial intelligence in practice

In my view, AI in optometry should be viewed first and foremost as a clinical tool. Productivity gains are also important as they will ultimately help in supporting better care. Commercial differentiation may well occur, but this is unlikely to be viewed as the primary driver. We need to maintain professional integrity, build patient trust, and ensure that AI adoption strengthens the clinical role of the optometrist.

Looking at how tools have been assessed is important to appreciate whether bias and discrimination/health inequalities could be a concern. AI algorithms depend on having adequate datasets for training, in terms of sample size and how representative they are of the population they are designed for. Clinicians need to appreciate that bias can arise with AI-supported clinical decision making. It is important to recognise that we may be more likely to agree with AI when it matches our clinical judgment and to disagree with it when it does not.

When it comes to clinicians using AI tools in practice, adequate training is essential to enable the team to understand when and how to use AI tools.

Essential is ensuring all requirements are met from a data governance, security, privacy and safety perspective. We need to document how AI is being used and take into consideration the evidence base and patient preferences. Naturally auditing any safety issues or incidents and the actions taken should also feature, as for other areas of practice. This is key to ensuring ongoing safety and efficacy.

If used appropriately, I believe that AI is much more likely to free optometrists to practice at a higher clinical level (more specialised, more analytical) than to erode clinical skills.

However, we should be transparent with patients about how decisions are made and our use of AI. Patients should be informed how and when AI tools are being used and where possible, given the opportunity to opt out if they wish.

As AI becomes more embedded in practice, skills that will be valuable for practitioners will include an ability to critique the outputs of AI tools, combine AI generated data with more traditional history, symptoms and clinical data, explaining AI technologies and findings to patients whilst considering AI recommendations in the context of patient preferences.

Skills likely to matter less will include basic image interpretation, ability to take standardised clinical measures (which can be often performed well by smart devices with minimal supervision), and perhaps routine administrative tasks.

Optegra

Professor Clare O’Donnell presenting at UKISCRS

If used appropriately, I believe that AI is much more likely to free optometrists to practice at a higher clinical level (more specialised, more analytical) than to erode clinical skills

Fast-forward five years

In five years’ time, I predict that there will be more integration of multi-modal data from various sources to generate risk scores for common ocular conditions – which will enhance prognosis and develop more personalised intervention and treatment plans. While AI performance in image analysis has advanced rapidly, its ability to deliver meaningful clinical decision support in optometry depends on access to longitudinal, contextual patient data.

Today, for me, the biggest misconception about AI in eye care that needs correcting is the fear that adoption of AI means that we will all be out of work. With the ageing population, associated increase in chronic eye disease, and relative shortage of health professionals globally, existing models of care are increasingly unsustainable. We need to innovate to ensure that patients receive care in a timely fashion. AI is not about replacing clinicians but about optimising capacity by supporting earlier detection, improving triage and referral accuracy and allowing optometrists and ophthalmologists to focus their time on complex value-added clinical decision making.