Ramón Salazar
Head of Medical Oncology at the Catalan Institute of Oncology (ICO), head of the Colorectal Cancer Research Group, Oncobell programme (IDIBELL) and associate professor of Medicine at the University of Barcelona
Although the quality of data collection is high, this is a rather heterogeneous observational study in terms of the origin and nature of its databases.
I was unaware of the proposed link between Helicobacter pylori infection and colorectal cancer, but it may make sense if the bacterium is identified in the colon, given its pro-inflammatory properties; however, in this study it has only been quantified or detected in the stomach.
The study finds an association between Helicobacter pylori (detected only in the stomach) and colorectal cancer, but this does not mean that it has been proven that the bacterium is the cause of these tumours. Furthermore, the fact that the bacterium was detected in the stomach does not necessarily imply its presence in the colon mucosa – where it could indeed cause a pro-cancerous inflammatory reaction – and the strength of the correlation varies substantially depending on which studies are taken into account, suggesting that the result is not particularly robust.
This is one of the major problems with causal inference: coincidence is not causation. As Nassim Nicholas Taleb argues in his book Fooled by Randomness, we can construct highly convincing narratives based on associations that actually reflect chance, biases or factors we have not taken into account. In science, finding an association is only the beginning; there must then be a rational mechanism of action to explain causality, and this must ultimately be validated: the difficult part is proving that we are not being misled by chance.
In summary, although the study is well-conducted and the authors attempt to verify their results in various ways, it still relies on observational data, studies that differ greatly from one another, possible errors in measuring exposure to H. pylori, and several assumptions inherent in the model itself. For this reason, its results should be viewed as exploratory and hypothesis-generating, not as a definitive demonstration of causality.