Introduction
Despite being underfunded for over a decade, and now in “critical condition”, the UK’s National Health System (NHS) has significantly invested in personalised medicine. These initiatives include the pioneering 100,000 Genomes Project, genetically based cancer screening, and personalised prevention. Proponents argue these programmes will improve outcomes and reduce pressure on the healthcare system. Critics say they are overhyped. Who’s right? And what are the lessons for other healthcare systems?
Holy Grail?
One of the NHS’s highest profile projects is the collaboration with Grail on the US company’s $950 multi-cancer screening blood test, Galleri. A £150 million trial is underway to assess the test’s value in detecting cancers such as pancreatic and ovarian malignancies that currently lack effective screening methods. If the trial is successful – defined by achieving a positive predictive value (the chance that a positive test is correct) above 30%, a 30% reduction in late-stage cancers, and a 75% higher detection rate than the control group – NHS is prepared to purchase a million tests.
Critics like Margaret McCartney, writing in the British Medical Journal, question Galleri’s effectiveness. They point to data showing low test sensitivity (27.5%) for early-stage cancers, the ones whose detection should save lives and money, and a 62% false-positive rate, which could lead to unnecessary treatments as well as patient anxiety. Members of the UK National Screening Committee have raised ethical concerns about the trial’s design and the close relationship between the government and Galleri’s developers. As one critic put it, “This is a clear-cut case of public risk and private profit.”
What is personalised medicine?
Personalised medicine uses an individual’s genetic profile and other personal information to guide decisions about prevention, diagnosis, and treatment of disease. It aims to provide the right treatment to the right person at the right time, moving away from a “one-size-fits-all” approach to healthcare.
Personalised prevention similarly is about tailoring preventive interventions based on genetic, environmental, and behavioural data that identify individuals at higher risk of disease. This is great. Prevention is better than cure, after all?
Appealing concepts but not necessarily easy to put into practice.
A recent Health Foundation report identified five assumptions that underpin the personalised approach – and that arguably might not hold up in reality. Let’s take a closer look.
1. Data and technology will enable personalised care
Data integration and health IT systems that enable personalised care should make health care more efficient and personalised, but the NHS’s past attempts to build a national data platform have failed spectacularly, costing multiple billions. And they still face resistance; concerns over data privacy and governance have stalled progress, with primary care data currently excluded from the system. Without full integration, and patient consent, another recent stumbling block, the potential for using data effectively, including in personalised prevention, is limited.
2. Polygenic risk scores (PRS) and AI tools will deliver superior clinical value
Thanks to new methods, and more genetic data, the predictive performance of polygenic risk scores has significantly improved. AI tools can help detect diseases like cardiovascular conditions from retinal imaging, or incidental findings on CT scans ordered for other reasons, but large-scale use has not yet arrived. The lack of high quality evidence from prospective randomised trials is a factor. Another reason is the concern about biases embedded in AI models which rely on data primarily from European populations.
Our Future Health is a UK project which uses AI to accelerate inferences from genomic analysis. An independent report suggests these types of “artificial intelligence-powered genomic health predictions” (AIGHP) could exacerbate societal issues including privacy and discrimination concerns but also offers 10 recommendations for mitigating risks and boosting benefits.
3. Personalised prevention will improve outcomes and change behaviour
PRS can separate populations into different risk levels. Typically, the top decile (10%) of people based on a risk score have 2-3 times higher than average odds of disease. The issue is how much better this prediction is than one based on already available risk factors; the additional predictive power, or extent of risk reclassification, may be too small to justify widespread use of the test.
Even when significant genetic risks are identified, translating this knowledge into meaningful behaviour change is a challenge. Behavioural interventions require opportunity, motivation, and capability for change which technology solutions do not necessarily provide.
Most health behaviour change depends on broader social and environmental barriers, which data alone cannot solve. For instance, without local access to healthy food or safe spaces for exercise, personalised prevention is unlikely to lead to improved outcomes.
4. People will access, interpret, and act on personal health data
Access to personalised health data is not a reality for all. In the UK, consumer-facing tools like the NHS App may offer access to health records but don’t yet provide personalised risk assessments or behaviour change advice. Even when people do receive information about their health risks, for example via the English National Childhood Measurement Programme which targets childhood obesity, behaviour change is not guaranteed; in this specific case, habits did not change.
5. Self-management tools will reduce healthcare pressure
Self-management tools, such as remote monitoring or wearables, could help reduce the strain on health systems. “Hospital at home” initiatives can avert or shorten admissions, and wearable devices make so much sense for patients with chronic disease in healthcare systems with long waiting lists. However, integrating wearable data into clinical systems has been challenging, and the accuracy of consumer devices varies widely. Furthermore, these tools often cater to younger, wealthier individuals, potentially widening health disparities rather than reducing them.
A South African perspective
When it comes to introducing personalising medicine, whether on the basis of genetic scores or other new biomarker, South Africa, with its fragmented healthcare system and more limited resources, faces significant challenges.
Low(er) tech solutions, such as use of existing clinical scores to estimate lifetime risk of a heart attack to guide long term prescription of a statin, could be more useful for public health than cutting-edge genetic tools, if properly implemented.
Nevertheless, the science is promising. Investment in digital infrastructure, including AI, is essential but should be coupled with improvements in basic healthcare services and workforce training.
Genetic screening of healthy individuals isn’t new to SA. For example, Discovery Health’s 2015 plan to offer DNA sequencing to its members never materialised – an interesting story for another day, and genetic testing remains expensive.
SA equivalents to the UK’s 100,000 genomes project have not had sufficient funding to achieve scale but H3 Africa has been successful and a large project to sequence 3 million African has been proposed. Polygenic scores do hold promise – if data gaps can be addressed, especially for under-represented populations. Most genetic research is conducted on European and North American populations, meaning that polygenic risk scores are less accurate for Africans.
Developing locally relevant AI and genetic tools will be crucial for ensuring that benefits of personalised medicine materialise are equitably distributed.
Weighing the cow doesn’t fatten It
Ultimately, identifying individual risk isn’t enough. Weighing a cow doesn’t fatten it.
Personalised prevention must come with actionable steps and support systems that can help people change their behaviour, and help clinicians make better decisions in partnership with their patients.
Implementing these approaches effectively requires more than data and technology. It demands an understanding of the clinical, social and environmental factors that drive health outcomes, and infrastructure that can support clinicians and patients.
Last, but not least, care should be personalised based not only on biomarkers but also by asking patients “what matters most?” and incorporating their preferences, needs and values into decision making.