Having scanned the original paper (https://journals.plos.org/plosone/article/file?id=10.1371/jo...) this seems like p-value hacking. If you check table 3 they report the p-values for the chi-squared test that they did across disease categories and dietary habits, it seems they are using a threshold of p<0.05 as being significant. The problem is that they have repeated this test across ALL the different conditions without adjusting this threshold of significance, and then handpicking the ones that satisfy p<0.05. At the very least they should correct for this by using something like a Bonferroni correction (https://en.wikipedia.org/wiki/Bonferroni_correction).