Colorectal cancer is the third most common cancer and the fourth leading cause of cancer death worldwide, and the fourth and second, respectively, in Europe (1, 2). Among established risk factors for this neoplasm are high consumption of red and processed meat, heavy alcohol consumption, body fatness, and family history of colorectal cancer, whereas physical activity, regular aspirin use, statin use, and, probably, a diet rich in fiber, especially from fruit and vegetables, appear to have a protective role against this neoplasm (3–6).

Type 2 diabetes mellitus has been associated with an excess risk of colorectal cancer (7–10), as well as with of cancers related to metabolic factors (11, 12). Although the magnitude of the excess risk appears to be modest (about 30%), it may have important public health implications given the high prevalence of diabetes. The association has been found to be consistent between study designs (i.e., cohort versus case–control studies), sex, and cancer subtypes (i.e., colon versus rectum) (7, 9, 13, 14). However, only a few studies considered the timing of the disease in relation to colorectal cancer diagnosis, reporting no clear trend in risk (9).

Diabetes medications have also been associated with colorectal cancer risk in some studies which suggested that insulin use may increase the risk (15), while metformin (15–17) and thiazolidinedione (15, 18) use may reduce it. The evidence on diabetes medications is, however, still inconsistent, and only scanty studies considered time–risk relationships (15).

We further examined the association between type 2 diabetes, antidiabetic medications, and colorectal cancer risk using data from two companion case–control studies from Italy and Spain, where information on duration of exposures was considered.

Table 1 gives the distribution of colorectal cancer cases and controls, according to selected characteristics. Cases and controls had a similar distribution by sex, tobacco smoking, alcohol drinking, body mass index, leisure-time physical activity, and regular aspirin use. Cases were somewhat older and heavier alcohol drinkers, and reported a lower level of education, and a lower use of statins than controls.

The distribution of colorectal cancer cases and controls by history of diabetes and the corresponding ORs by country and overall are shown in Table 2. Overall, 159 (13.9%) of cases and 188 (11.8%) of controls reported a diagnosis of type 2 diabetes, corresponding to a multivariable OR of 1.21 (95% CI 0.95–1.55). The OR was 1.12 in the Italian study and 1.20 in the Spanish study. The OR of colorectal cancer was higher for a duration of diabetes of ≥15 years (OR 1.49, 95% CI 0.97–2.29), but the trend in risk for increasing duration was not significant (p-value = 0.10). Corresponding values were 1.68 in Italy and 1.47 in Spain. In sensitivities analyses excluding patients with a diagnosis of diabetes within 2 or 3 years prior to cancer diagnosis, the OR were 1.16 (95% CI 0.89–1.50) and 1.15 (95% CI 0.88–1.50), respectively, in the two studies combined (data not shown). Further allowance for selected dietary factors (e.g., meat, fruit or vegetables) or total energy intake did not meaningfully change our risk estimates (data not shown).

No significant heterogeneity was observed in risk estimates across strata of sex, age, education, tobacco smoking, alcohol consumption, body mass index, physical activity, and statin and aspirin use (Table S1 in Supplementary Material).

Table 3 shows the distribution of anatomical subsites of colorectal cancer cases and corresponding controls, and the OR estimates according to history of diabetes by country and overall. In both countries combined, the OR was 1.15 (95% CI 0.89–1.55) for colon cancer overall, 1.53 (95% CI 1.06–2.19) for proximal colon, 0.94 (95% CI 0.66–1.36) for distal colon, and 1.32 (95% CI 0.94–1.87) for rectal cancer. Corresponding estimates were 1.04, 1.32, 0.94, and 1.46 in Italy and 1.15, 1.64, 0.88, 1.27 in Spain. The OR for overlapping and NOS colon cancer was not shown due to the limited number of patients.

Table 4 shows the distribution of colorectal cancer cases and controls, and the corresponding ORs, according to use of antidiabetic medications in diabetic patients from the Spanish study. Diabetic patients using any oral antidiabetic drug had a non-significant reduced risk of colorectal cancer as compared to those not using them (OR 0.59, 95% CI 0.18–1.92). The OR was significantly below unity in diabetic patients using metformin (OR 0.47, 95% CI 0.24–0.92) as compared with those not using it, with a trend of decreasing risk with increasing duration of use (OR 0.36, 95% CI 0.15–0.85, for ≥10 years of use; p-value for trend 0.005). Among other oral antidiabetic drugs, sulfonylurea derivatives were not significantly associated with colorectal cancer risk (OR 1.14, 95% CI 0.63–2.10). Diabetic patients using insulin had an increased risk of colorectal cancer as compared with those not using it (OR 2.20, 95% CI 1.12–4.33); the OR increased with increasing duration of insulin use and was 8.18 (95% CI 2.06–32.50) for ≥10 years of use (p-value for trend 0.002). Risk estimates did not meaningfully change when we adjusted the models for duration of diabetes. When we mutually adjusted our estimates for insulin and metformin, the OR was 0.52 (95% CI 0.26–1.03) for metformin and 2.00 (95% CI 1.00–3.99) for insulin (data not shown). Using as reference category non-diabetic patients besides patients using other antidiabetic drugs, the OR was 1.11 (95% CI 0.82–1.53) for any oral medications, 0.98 (95% CI 0.69–1.40) for metformin, 1.18 (95% CI 0.75–1.86) for sulfonylurea derivatives, and 2.04 (95% CI 1.18–3.55) for insulin use, with significant trend in risk by increasing duration of insulin use (p-value for trend 0.003).

As compared with diabetic patients using other antidiabetic medications, the OR of colorectal cancer was 0.69 (95% CI 0.37–1.28) for diabetic patients using metformin in monotherapy, 4.29 (95% CI 1.61–11.43) for those using sulfonylureas derivatives in monotherapy, and 1.70 (95% CI 0.52–5.51) for those using insulin in monotherapy; for combinations of therapies, the OR was 0.50 (95% CI 0.23–1.07) for diabetic patients using both metformin and sulfonylurea derivatives, 1.64 (95% CI 0.67–4.00) for those using both metformin and insulin, 0.97 (95% CI 0.19–4.91) for those using both sulfonylurea derivatives and insulin, and 3.41 (95% CI 0.56–20.71) for those using metformin, sulfonylurea derivatives, and insulin (Table S2 in Supplementary Material). Moreover, these estimates were consistent when we included in the reference category also non-diabetic patients. However, they should be interpreted cautiously since they were based on a small number of patients.

In our study, we find a 20% increased risk of colorectal cancer (though of borderline significance) in relation to type 2 diabetes, the excess risk being higher for a longer diabetes history and for proximal colon cancer. We provided further evidence of a possible inverse association with metformin and a positive association with insulin use, with significant duration–risk relationships.

Previous case–control and cohort studies showed an approximately 30% excess colorectal cancer risk in patients with diabetes as compared with those without diabetes (7–9). We found a somewhat lower association in our study which was, however, consistent in both the Italian and Spanish study. We found a higher risk of colorectal cancer for a history of diabetes of 10 or more years, although in the absence of a significant trend in risk. Similarly, a meta-analysis of six epidemiological studies reported no trend in risk for longer time since diabetes diagnosis (9). Furthermore, we found no differences by sex, again in agreement with the available evidence (7, 9, 13, 14). The 50% increased risk of proximal colon cancer is consistent with findings from previous studies reporting stronger associations with proximal than with distal colon cancer (7, 9, 21).

In the present study, we found a twofold excess risk of colorectal cancer in diabetic patients using insulin as compared with users of oral antidiabetic medications, with a significant increased risk with increasing duration of exposure. A similar increase was observed when the reference category also included non-diabetic patients and when we considered insulin use as monotherapy. This is consistent with the evidence from a meta-analysis − based on seven case–control studies and five cohort studies published before 2014 − which reported an overall RR of colorectal cancer of 1.69 (95% CI 1.25–2.27) for insulin users (22). However, the results of subsequent cohort studies were somewhat inconsistent (23, 24). Insulin has been shown to increase the risk of colorectal and other cancers by influencing the insulin-like growth factor system, which, in turn, may stimulate cellular proliferation and inhibit apoptosis (25). Insulin injection has also been shown to stimulate the growth of colorectal cancer precursors in animal models (26).

Conversely, metformin use was associated to a 50% reduced colorectal cancer risk as compared with use of other antidiabetic drugs, with a trend of decreasing risk with increasing duration of exposure. Consistently, a meta-analysis of five studies conducted up to 2011 reported an overall relative risk, RR, of 0.64 (95% CI 0.54–0.76) for metformin use (16). A weaker inverse association was reported in two subsequent meta-analyses that included at least nine epidemiological studies, with pooled RRs ranging between 0.55 and 0.90 depending on the reference category used (15, 17). Such inverse association was found in cohort studies (RR 0.82, 95% CI 0.71–0.94), but not in case–control studies (RR 1.04, 95% CI 0.94–1.15) (15). On the contrary, in a subsequent British cohort of patients with diabetes, metformin was not associated with colorectal cancer mortality (hazard ratio, HR, 1.06, 95% CI 0.80–1.40) (23). Metformin may reduce colorectal cancer risk by activating adenosine monophosphate-activated protein kinase and by consequent inhibition of the mammalian target of rapamycin (mTOR) (27). Moreover, it can suppress azoxymethane-induced formation of colorectal aberrant crypt foci (28, 29). A more widespread use of metformin as a first-line antidiabetic medication and its use in less severe cases of diabetes may also explain – at least in part – the lower association for history of diabetes observed in this study as compared with those conducted in 1980s and 1990s (10, 12). Moreover, in our study, the inverse relation with metformin was weaker when the reference category also included non-diabetic patients and in patients using metformin as monotherapy, suggesting that metformin may reduce the detrimental effect of insulin, rather than having a protective effect itself.

We found no consistent role of sulfonylurea derivatives on colorectal cancer risk, the OR being 1.14 when users of sulfonylureas were compared with diabetic patients using other antidiabetic medications and 1.18 when the reference category also included non-diabetic patients. Although we found a significant fourfold excess risk in diabetic patients using sulfonylurea derivatives as monotheraphy, this estimate was based on a small number of patients and should be cautiously interpreted. Similarly, a meta-analysis of three case–control and four cohort studies reported a colorectal cancer RR of 1.11 (95% CI 0.97–1.26) in users of sulfonylureas as compared with non-users (15). Similarly, in a subsequent cohort of patients with diabetes, sulfonylurea was not significantly associated with colorectal cancer mortality (HR 1.14, 95% CI 0.86–1.51) (23).

Limitations of this study are those typical of all case–control studies, including selection and information bias. However, in both countries, cases and controls came from comparable catchment areas, were interviewed by uniformly trained interviewers, and had a satisfactory participation rate for their study design. In particular, the high response rate in Italy reflects a traditionally high compliance of subjects approached for interview in the Italian case–control network; in addition, controls were hospital-based in Italy, while population-based in Spain, this explaining the higher response rates among controls in Italy as compared with Spain. Patients and interviewers were unaware of any hypothesis relating diabetes with colorectal cancer, thereby reducing the likelihood of potential selection and recall bias. Diagnosis of diabetes was self-reported, thus some potential misclassification is possible; however, the reliability of information on diabetes provided by hospital controls has been shown to be satisfactory (30) and the prevalence estimates of diabetes in our control groups were consistent with those from national population-based surveys (31, 32). Although in the Spanish study we could not distinguish between type 1 and type 2 diabetes, we excluded patients with a diagnosis of diabetes before age 30, who have been likely affected by type 1 diabetes. Another limitation is the lack of information on changes in medication type over time – which are frequent in diabetic patients – on glycemic level, and on the severity of the disease. Indeed, diabetic patients using insulin tend to have a more long-lasting and severe disease than non-insulin users. Among the strengths of our study is the allowance for major confounding factors for colorectal cancer, including body mass index and physical activity. Although in our dataset we did not have any information on household income and insurance, all estimates were adjusted for education, which has been shown to be a good proxy of socioeconomic status. Another strength of our study is that we were able to analyze the relation with duration of diabetes and use of various antidiabetic medications, exploring long-term exposures (i.e., >10 years), which have been examined in a limited number of studies.

Our study confirmed a positive association between diabetes and colorectal cancer, mainly proximal colon cancer. Among antidiabetic medications, it indicates a favorable effect for metformin use and a detrimental effect for insulin use, suggesting a need for enhanced screening measures for colorectal cancer in insulin-treated patients. It is, however, not easy to disentangle the effect of each antidiabetic medication since diabetic patients often use more than one medication, especially for more severe disease. Thus, it is difficult to establish whether these associations are causal or influenced by different severity of the disease and confounding.

VR took care of data management, statistical analysis, and manuscript draft; AT contributed in the result interpretation and revised the manuscript; EG-L and EG took care of data management for the Spanish center; GC-V, CV, MK, VM, JP, DS, FP, FB, CV contributed in the design of the HIWATE study and collection of the data; CB has the idea of the study and supervised the analysis and the manuscript drafting; she is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. All the authors revised the final manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.