For stomach cancer, previous case–control studies reported either null or positive associations with spicy food intake,6 of which only three studies (conducted in Mexico and Korea) had quantitative assessment of consumption, comprehensive adjustment for confounders and >200 cases of stomach cancer.17,18,35 With almost 15 times the number of stomach cancer cases, we found a weak inverse association that was attenuated towards the null after excluding the first 3 years of follow-up, suggesting that the inverse association may be due in part or wholly to reverse causation. As for colorectal cancer, only three case–control studies were identified, of which two (each with <200 cases) reported positive associations.36,37 The largest study, with 400 cases and conducted in Sichuan, however, found no significant association.38 Similarly to stomach cancer, the weak inverse association with colorectal cancer in our study appeared to be partly explained by reverse causation, to be confirmed in other large prospective studies. Since no previous study has examined spicy food intake with sub-sites of stomach and colorectal cancers, further evidence is needed to determine whether associations truly differ between cardia and non-cardia stomach cancer, and between colon and rectal cancer.
Capsaicin, the main bioactive constituent of spicy food, has exhibited various carcinogenic effects in animal studies, e.g. through inducing mucosal damage.39 In contrast, capsaicin has also demonstrated anti-carcinogenic effects, through altering GI cancer risk factors such as inhibiting the growth of Helicobacter pylori (H. pylori)40 and reducing body fat.41,42 Specifically for adiposity, cross-sectional epidemiological studies in China have also reported inverse associations of chilli intake with prevalence of obesity and serum cholesterol levels12,13 but these were not replicated in CKB and we also did not find clear evidence of mediation by adiposity. Overall, it is possible that any carcinogenicity or anti-carcinogenicity of capsaicin is dependent on dose, and there may be a threshold beyond which the harms start to outweigh the benefits (or vice versa), but further epidemiological studies with quantitative assessment are needed to clarify this.
The strengths of our study included the prospective design, large numbers of cases, assessment of multiple aspects of spicy food intake and adjustment for a wide range of confounders. However, limitations exist. First, spicy food consumption was self-reported and we did not have objective indicators such as capsaicin concentration extracted from participants’ food to validate self-reported preferences for spice strength. To mitigate this issue and to better distinguish between mild-, moderate- and strong-spice consumers, interviewers were instructed to monitor the coherence of participants’ answers across the different spicy food questions (since daily spice consumers or those who consume chillies directly tend to prefer stronger spice) and to clarify with participants when there were important contradictions. Second, the quantity of spicy food intake was not available for more accurate quantification of the observed relationships or investigation of potential threshold effects. Third, we were unable to explore the effects of spice type and strength in the two regions where consumption levels were very high (Hunan and Sichuan) as participants in these regions consumed almost exclusively fresh chilli, at moderate/high intensity. Although the high consumption levels in these two regions might have distorted the overall associations, our sensitivity analyses restricted to the other eight areas showed broadly consistent results with the main findings. Fourth, although we have attempted to control for a wide range of known and suspected confounders, residual confounding from age, regions (e.g. urban–rural differences), suboptimally measured factors (e.g. other dietary factors) or unmeasured factors (e.g. H. pylori infection for stomach cancer) may still be present. For example, residual confounding from age and urban–rural residency, if any, could have biased the associations with oesophageal and stomach cancers away from and towards the null, respectively. Nonetheless, there were no significant subgroup differences by urban–rural areas (pheterogeneity > 0.05) and associations remained directionally-consistent after excluding individuals with a prior history of peptic ulcers (a proxy for symptomatic H. pylori infection).43Finally, even as the single largest study on this topic, we had limited power to draw conclusions on the interaction between spicy food intake and smoking or alcohol-drinking.