In the 1950s, Kodak sold almost all of the colour photography film in the United States. As printing services became more widespread, they needed to ensure that their machines were correctly calibrated and produced the correct skin tones, shadows and light in photographs. Kodak developed a solution. They started providing printing labs with kits containing colour prints and negatives of Shirley Page, one of their studio models. These 'Shirley cards' quickly caught on and were soon being used all over the world, becoming a byword for colour photography calibration. They worked well for their target population: largely white, middle-class Americans. However, people with black or brown (or even tanned) complexions found their images were blotchy and washed out, or that they blended into the backgrounds. Photographs of Black and White people in the same shot often turned out partially over- or under-exposed. The problem was so severe that filmmaker Jean-Luc Godard refused to use Kodak film to shoot in Mozambique, declaring it “racist”.

It is tempting to consider this story nothing more than a historical oddity, or to assume that such mistakes would not be made today. Yet, over three-quarters of patients in global clinical studies between 2005 and 2019 were European, with just 11% from Asian and 7% from African populations. Genomic studies fare even worse: almost 90% of all genome-wide association study participants are of European ancestry. The majority of patients come from just three countries: the UK, the US and Iceland. Asian populations contribute around 5%; less than one percent are of African ancestry.

All for one...

The harms to individual patients are obvious. Simplistic assumptions around ethnicity meant kidney function was overestimated in Black patients, potentially resulting in prescription of toxic drug doses. These were only changed in 2021. Similarly, in acute lymphoblastic leukaemia, the frequency of genes responsible for metabolising thiopurines varies dramatically across populations, meaning that a standard dose in Europeans (where these variants are rare, at <0.5%) could be toxic and potentially fatal in East Asians (where their frequency is 10-15%). These effects will only be amplified in the personalised prediction era. Technologies such as polygenic risk scores tacitly imply universality; in other words, that they are equally effective for everyone. Yet, we already know that is not the case. By creating tools generated in largely White populations, we create a situation where ethnic minorities who already are more likely to suffer worse outcomes are unable to benefit from improved early detection. Put another way, we further entrench health inequalities, where those who most require help are least able to benefit from technological advancements.

 …and one for all

The harms to our overall understanding of cancer are less obvious. Current designs implicitly assume that all patients can be treated in the same way as one smaller group. But focusing on a single population risks masking the true complexity of cancer and missing critical opportunities for better treatment. In the early 2000s, the drug gefitinib had received early approval in the US for treatment of non-small cell lung cancer, on the basis of some promising early trials. However, a landmark multi-country trial found almost no benefit, throwing it into jeopardy. In amongst the cohort, though, were a small number of patients who had massive improvements in survival. Unlike the majority White, smoker population, these patients were overwhelmingly young, female, Asian and non-smokers. A subsequent clinical trial in Asian populations found that gefitinib did have a strong effect after all. This was because gefitinib targeted mutations in the EGFR gene, a major driver of cancer growth. EGFR mutations, as it turned out, were around two to five times more common in Asian populations than in their European counterparts. If the trial had used a sample of convenience, and limited itself to European patients, the signals might have been too weak to detect. We would have discarded this drug as ineffective, when in reality, it was so effective that this class of drugs is now the standard of care for lung cancers harbouring these mutations. Luckily, diligent research in diverse populations allowed scientists to glimpse the truth, and yield benefits for patients around the world.

Sharpening the focus

Gefitinib should have served as a huge red flag. But twenty years later, we continue to do the same thing. The scientific community designs studies that represent only a small segment of the global population. In doing so, we fail both individuals and societies. On one hand, we risk worsening health outcomes and entrenching health inequalities in under-represented groups. On the other, we risk missing effective treatments. It is hard to know how many drugs like gefitinib have been incorrectly labelled as failures because they were tested in the wrong populations. Instead, we should be taking pains to design clinical trials that accurately reflect the populations they serve; and building diverse patient biobanks to allow us to truly unravel the complexity of cancer.

Kodak did eventually release a more diverse version of the Shirley card in the 1970s. These were not, as you might hope, due to an outcry from individual customers. Rather, the Shirley card was changed because of complaints from chocolate and furniture companies annoyed by the poor rendering of their products. Change was driven not by moral altruism but by economic necessity. Digital photography ultimately rendered the Shirley card obsolete, although the underlying problems persist. In an age where politicians openly question the benefits, or even the need, for diversity, the scientific community needs to remember the lessons from Kodak fifty years ago. It is up to us to ensure that everyone looks sharp in the picture. If not for individual benefit, then for the collective clarity it provides. Cancer doesn't discriminate; neither should we.

This piece won the 2024 Mel Greaves Science Writing Prize.

Dr Avirup Chowdhury is a public health doctor and postdoctoral clinical research fellow in the Cancer Dynamics group at the ICR. Prior to his current role, he studied medicine at the University of Glasgow and completed foundation training in Scotland, before commencing public health specialty training in the East of England. He has worked in specialist roles across key national organisations including the National Disease Registration Service and NICE.

He holds an MSc in Human Molecular Genetics from Imperial College London and an MPhil in Public Health from the University of Cambridge. He received his PhD at the ICR investigating treatment response in soft tissue sarcomas as part of the Molecular & Systems Oncology group.

His research evaluates the added value of integrating molecular data into established risk prediction tools. He focuses on developing enhanced prediction models that are both clinically translatable and generalisable across global populations.