Share Tweet

What if cancer screening was not determined primarily by your age, but by changes in your own health over time? Technology entrepreneur and engineer Pieter van Rooyen is exploring how artificial intelligence and existing health data could help shift cancer detection earlier – while making it more affordable for countries such as South Africa. His recent Pro-VC Open Lecture at Stellenbosch University, Ahead of the Tumor: Rewriting the Cancer Timeline, unpacked the thinking behind this approach.

For Pieter van Rooyen, the challenge of early cancer detection is not only about developing better technology. It is about asking a different question: who should we test, and when?

Van Rooyen, co-founder of xTenure and Extraordinary Professor in Stellenbosch University’s Department of Electrical and Electronic Engineering, has spent the past 26 years in San Diego, California, including around 15 years working in genomics. There, he worked alongside people developing cancer screening, diagnostic and treatment technologies, as well as the patients these technologies are designed to help.

But spending more time in South Africa in recent years brought another reality into focus.

“It became clear to me that a great deal of that progress will not reach this country – not because the science does not work here, but because of what it costs,” he says. “That was the genesis of the work: to develop, together with my colleagues, techniques and technologies that bring early cancer screening within reach of a country like ours.”

Looking before symptoms appear

One of the fundamental challenges in cancer care is that too many cancers are still found only once symptoms appear.

Organised screening currently exists for only a handful of cancers, including breast, cervical, colorectal and prostate cancer, and – for heavy smokers only – lung cancer. Together, these programmes cover cancers accounting for roughly 38% of cancer deaths. In other words, about six in ten cancer deaths occur in an organ that no organised screening programme looks at.

Pancreatic cancer provides a stark example. Van Rooyen points out that when it is caught while still confined to the pancreas, 44 in 100 people are alive five years later. Once it has spread elsewhere in the body, that falls to just three in 100. About half of South African cases are detected only after the cancer has spread.

“The same disease, caught early or caught late, is in survival terms two different diseases,” he notes. “You do not close that gap by diagnosing better once somebody arrives with symptoms. You close it by looking earlier, in people who feel perfectly well.”

A different approach

Multi-cancer early detection tests are already beginning to change this picture. Instead of separate screening procedures for individual cancers, these tests use a blood sample to search for signals associated with multiple cancers.

But cost remains a major obstacle. Many emerging tests rely heavily on genomic sequencing, with prices running from around US$950 to nearly US$3,000. For population-wide screening in a country such as South Africa, Van Rooyen argues, that is simply not economically feasible.

Cost, however, is only part of the problem. In any given year, cancer is relatively rare – roughly one person in 630. Test very large numbers of people for something rare and even an excellent test can produce far more false alarms than true findings, simply because there are so many more healthy people than people with cancer being tested.

How good the test is therefore matters, but so does how likely it is that the person being tested actually has cancer. That second factor depends on who is selected for testing in the first place.

This is why Van Rooyen and his team are starting somewhere different: with information the healthcare system may already have.

Using AI, the model examines an individual’s existing health record, including diagnoses and routine laboratory results accumulated over time. Rather than testing everyone of a certain age on a fixed annual schedule, the aim is to identify people for whom testing is most worthwhile at a particular point in time.

“You are read against yourself rather than against a population reference range,” Van Rooyen explains.

A laboratory result that falls within the accepted range for the general population might nevertheless represent a significant change for an individual whose own results have remained stable for years. Conversely, someone may consistently sit outside a population range while remaining healthy. Following an individual’s trajectory could therefore reveal information that conventional thresholds miss.

Importantly, Van Rooyen stresses that the technology is not intended to diagnose cancer.

“This decides who should be tested, and when. It does not diagnose anybody – it is a screening test, a scheduler rather than a predictor.”

Diagnosis remains with clinicians and confirmatory testing.

Why South Africa?

Perhaps counter-intuitively, Van Rooyen believes South Africa’s constraints could help make it an important environment for developing this approach.

“A country that cannot afford to test everybody is forced to solve the harder and more useful problem, which is how to spend a fixed budget on the right people at the right time,” he says. “This is a global problem; wealthier systems are simply able to postpone noticing it.”

South Africa also has valuable longitudinal patient data, alongside AI and data-science expertise, a strong clinical research community and an established clinical-trial industry.

The work will initially focus on South Africa’s private healthcare sector, where patient records are currently more structured and accessible. The longer-term ambition, however, is to develop an approach that can work within the public health system, where the overwhelming majority of South Africans receive their care.

Getting there will require evidence. Van Rooyen emphasises that prospective clinical trials will be essential, alongside careful attention to data quality, diagnostic capacity, regulation, patient privacy and integration into clinical practice.

From screening event to screening schedule

Looking ten years ahead, Van Rooyen’s ambition is straightforward: “That screening stops being an event and becomes a schedule.”

Instead of reaching a particular age and automatically being called for a test, screening could become a continuously updated, individual process based on information the healthcare system is already collecting.

“Detection would move from something done to a whole population on a fixed cycle, to something tracked for one person across their life.”

He is careful, however, not to get ahead of the evidence. The idea that this approach could move cancer detection years earlier remains a thesis to be tested through clinical trials.

Testing that thesis will require expertise from across disciplines. For Van Rooyen, Stellenbosch University offers an unusually fertile environment in which to bring those disciplines together. Engineering and machine learning, clinical research, biostatistics, trial design and entrepreneurship all need to intersect if an idea is ultimately to become something a healthcare system can use.

“A university is the only place where the four things this problem needs sit close enough together to actually meet,” he says.

“At most institutions those are four buildings, four funding cycles and four vocabularies, and a project like this one dies in the gaps between them. Stellenbosch has an unusual amount of it in a single place, which is precisely why the conversation is happening here.”

By Katrine Anker-Nilssen

BELOW: Instead of one expensive test for everyone once a year, evidence is gathered in stages - the cheapest and broadest read first. Each step raises the chance that the person being tested actually has cancer, which is what makes the expensive test affordable.

News date: 2026-08-28

Links: