A recent article published in BMJ Oncology, claimed that;
“Global incidence of early-onset (< 50 years of age) cancer increased by 79.1% and the number of early-onset cancer deaths increased by 27.7% between 1990 and 2019”
this would be worrying if it were true, thankfully, it’s not. Inevitably, this received a fair bit of attention from the press and other academics who obligingly relayed the message.
The age-standardised rates - the ones that matter - are quite different (see figure).
The claim could have been
“Global age standardised incidence of early-onset cancer has increased a little (~6% over 30 years) whereas age-standardised early-onset cancer deaths have decreased by 25%.”
But that is less of a headline - even though the data for mortality are really positive. Why are the two sets of figures so different?
The authors of the paper compared the number of early-onset cancers in 1990 (1.82 million) with the number of early-onset cancers in 2019 (3.26 million). This is an increase of 79.1% but their calculation omits something important - context - and a denominator. For the purposes of comparison we must relate the number of new cases or deaths to the number in the population in which they occur - this is really basic stuff! So to make a fair comparison we divide cases by the number in the population.
We call this kind of operation in statistics standardising - putting things on a common scale for the purpose of comparison. But just dividing by the number in the population isn’t rarely good enough. We also need to take into account age.
Here is another example. When we teach statistics to the undergraduate medical students in Oxford, we show them mortality rates in Peru (6 per 1000) and the UK (9 per 1000). We get them to think about this - does this mean Peruvians live longer, eat better, have a better health care system - all these are possible. We then get the students to think about factors that affect mortality - and whether a direct comparison is fair. We then give them some relevant information - the median age in Peru is about 29 years and the median age in the UK is about 41 years. The UK has a much older population than Peru - this is likely to explain the difference. We then show the students mortality rates after standardising by age - the order is reversed with Peru’s age-standardised mortality rate higher than the UK. All talk of a failing NHS and lifestyle factors explaining the difference, vanish into thin air. Context is all important as are denominators.
The strange thing is, that the authors of the paper knew this and did the correct, age-standardised analysis, but then shelved it in the supplementary material section1. For anyone unfamiliar with modern academic publishing, the supplementary material section is like the cutting room floor of the editing suite. It contains all of the analyses that didn’t quite make it. Sometimes however, sifting through the debris of the supplementary material section yields something very interesting.
With thanks to Victoria Stern at MedScape for alerting me to this issue
Table S7 - Global age standardised incidence and mortality rates between 1990 and 2019. Zhao J, Xu L, Sun J, et al Global trends in incidence, death, burden and risk factors of early-onset cancer from 1990 to 2019 BMJ Oncology 2023;2:e000049. doi: 10.1136/bmjonc-2023-000049