Background

Diabetes mellitus (DM) is the fourth or fifth leading cause of death in developed countries and one of the biggest threats to human health worldwide [1]. More than 90 % of all DM is type 2 diabetes mellitus (T2DM) [2, 3]. Colorectal cancer (CRC) is the third most common cancer in the world. Colorectal adenoma (CRA) (also known as adenomatous polyp and always found by colonoscopy screen [4]) is a prevalent precancerous lesion that can lead to CRC through the adenoma–carcinoma sequence [5].

Research on risk factors for CRA has focused on several epidemiological factors, including smoking [6], alcohol consumption [5], body mass index [7], physical activity [8], and calcium intake [9]. Recent research on patients with diabetes suggested that insulin therapy and diabetes itself may increase the risk of CRC [10–12]. However, the association between T2DM and the risk of CRA risk has not yet been fully established. Some researchers asserted that there were no overall associations between T2DM and CRA risk [13–16], while others reported a higher risk [17–20]. To further examine these findings and provide evidence of association between T2DM and risk of CRA risk, we performed a meta-analysis about T2DM on the risk of CRA.

Methods

Literature search

Two investigators (FY and YG) independently conducted a systematic literature searches on January 10, 2016 in PubMed, Embase, the Cochrane Library and Web of Science without limiting the publication date range. The following search terms were used: (diabetes mellitus OR diabetes OR diabetic OR glucose) AND (colorectal OR colon OR rectal) AND (adenomas OR adenoma OR adenomatous OR polyp). No language restrictions and any other limitations were imposed. Conference abstracts on the websites of American Society of Clinical Oncology’s (ASCO) and the European Society for Medical Oncology’s (ESMO) annual meetings were also searched, along with the reference lists of the identified publications. Additional file 1 includes the complete searching process.

The titles and abstracts of all of the studies from the searches were screened independently by three reviewers (FY, YG and JF). To be included in this meta-analysis, studies had to be at least one of the following criteria: (1) retrospective or perspective observational study of the association between diabetes mellitus and CRA, or (2) a study reporting the relative risks (RRs) or odds ratios (ORs) for T2DM on CRA with 95 % confidence intervals (95 % CIs) adjusted for gender, age, or other factors. Studies reporting on the CRA recurrence were excluded.

Data extraction

Data extraction was performed by three reviewers (FY, YG and WH), and verified independently for accuracy by a forth reviewer (JH). The following information was collected for each study: title and author, publication year, population, location, sample size, proportion of males and covariates controlled for by matching or multivariate analysis. For studies that reported several multivariate adjusted ORs, the effect estimate that adjusted for the maximum potential confounders was selected. Two investigators (FY and ZJ) conducted a quality assessment using the 9-star Newcastle-Ottawa Scale (NOS) [21], which was verified by a third investigator (YG). We considered studies with a NOS score of seven or more to be high-quality studies. The study selection process was based on the Meta-analysis of Observational Studies in Epidemiology (MOOSE) guidelines [22] and is described in Additional file 2.

Statistical analysis

We examined the relationship between T2DM and CRA risk on the basis of the adjusted RRs and ORs and corresponding 95 % CI published in each study. A fixed effects model was used to estimate the pooled RR and OR with 95 % CIs if there was no evidence of heterogeneity; otherwise, a random effect model was used [23, 24]. Because the incidence of CRA is low, the ORs in retrospective studies approximate the RRs [25, 26]. Heterogeneity between the studies was evaluated by the chi-square test and I-squared (I2) statistic [23]. Statistical heterogeneity was considered significant when p < 0.10 [27].

Several methods were used to test and adjust for potential publication bias. Visual inspection of funnel plots was performed, and the Egger’s regression test [28] and Begg’s test [29] were used. Where publication bias existed, we used the trim and fill method to correct it [30]. Subgroups analyses by gender, adenoma subsite, and study type were performed to explore the potential heterogeneity among the included studies. Sensitivity analyses were performed in two ways: (1) by excluding each study iteratively from the meta-analysis and (2) by keeping high-quality studies only.

All statistical tests were two-sided and regarded as statistically significant at p < 0.05 Stata (Version 11.0; Stata Corp, College Station, TX) was used for all analyses.

Results

Study characteristics

Until January 10, 2016, 2522 records were retrieved by using our search strategy. After reviewing the titles and abstracts, 113 articles were further evaluated by reviewing the full texts. Of those remaining articles, we excluded studies that : (1) reported the data of adenoma recurrence were excluded [31, 32], (2) did not reported the RRs of getting CRA separately but mixed CRC and CRA patients [31], and (3) discussed the relationship between metformin [33] or insulin use [34] and CRA. We identified 17 studies that met all of our criteria [13–20, 35–44], including four conference abstracts [36, 37, 43, 44]. Figure 1 provides a flow chart of study selection. The final studies included 28,999 participants and 6798 CRA cases and 11 were rated as high-quality. Four of the conference abstracts rated less than seven stars due to insufficient information about their research. Table 1 includes the general characteristics of the included studies.

Fig. 1
figure 1

Flow chart of article selection process

Full size image
Table 1 Characteristic of studies included in the meta-analysis
Full size table

Diabetes and risk of colorectal adenoma

The summary RR of diabetes on CRA was statistically significant (RR: 1.52; 95 % CI: 1.29–1.80). Evidence of the heterogeneity was identified (I2 = 65.6 %, P < 0.001). Figure 2 shows the results.

Fig. 2
figure 2

Forest plot of relative risk estimates of diabetes on risk of colorectal adenoma

Full size image

Subgroup analysis

As shown in Table 2, we conducted subgroup analyses based on multiple factors, including sub-site of adenoma, geographic region, gender, and study type. The results showed that advanced adenoma was significantly associated with T2DM (RR: 1.41; 95 % CI: 1.06–1.87). However, a similar effect was not detected for proximal, distal, or colon adenoma. No evidence indicated significant associations between T2DM and CRA by gender, i.e., males (RR: 1.36; 95 % CI: 0.99–1.80) or females (RR: 1.29; 95 % CI: 0.76–2.17). he relationships between T2DM and CRA risk was significant in Europe (RR: 1.27, 95 % CI: 1.02–1.57), the USA (RR: 1.69; 95 % CI: 1.14–2.51) and Asia (RR: 1.57; 95 % CI: 1.21–2.05). A significant increase in risk was found in retrospective studies (RR: 1.57; 95 % CI: 1.30–1.89) and not in prospective studies (RR: 1.27; 95 % CI: 0.77–2.10).

Table 2 Subgroup analyses for the effect of diabetes on risk of colorectal adenoma
Full size table

Sensitivity analysis

Sensitivity analysis indicated that no single study dramatically influenced the pooled RR. The results are shown in Fig. 3. Regardless of which study was omitted, the summary RRs were always greater than one. Similarly, Table 2 shows that excluding low-quality studies yielded results comparable with including all studies (RR: 1.64; 95 % CI: 1.26–2.14).

Fig. 3
figure 3

Result of sensitivity analyses by omitting one study in each turn

Full size image

Publication bias

The Begg’s rank correlation test (p = 0.001) and Egger’s regression test (p = 0.003) results showed potential publication bias that is described in Fig. 4. Once corrected by the trim and fill method [30], the result indicated that the pooled effect size did not changed.

Fig. 4
figure 4

Filled funnel plot of log relative risk vs. standard error of log relative risks

Full size image

Discussion

This study indicated that patients with diabetes, especially type 2, have about 50 % increased relative risk of developing CRA than non-diabetic individuals, regardless of their geographic location. Although sample size was small in the newly diagnosed T2DM subgroup, the heterogeneity was also small and a significant risk relationship between T2DM and CRA was still detected. A similar result was only found in the advanced adenoma subgroup, not in the proximal, distal, colon or multiple adenoma subgroups. When low-quality studies were excluded, the positive association still existed. These results suggested that T2DM patients should pay more attention to their risk of CRA.

The positive relationship between T2DM and CRA may be linked to insulin resistance or an increased insulin-like growth factor 1 (IGF-1) might take effect in the adenoma–carcinoma process. High insulin levels could promote tumor growth [31, 45, 46]. Also, diabetes may lead to slower bowel transit, which would increase the probability of exposure to potential carcinogens for colonic mucosa [47–49]. It is worth noting that there might be some confounding effects because of the similar risk factors for both T2DM and CRA, such as physical inactivity, obesity, and an unhealthy diet habit [12, 50]. For example, a case–control study reported that higher red meat intake could significantly increase the risk of colon adenoma [51]. At the same time, obese people also tend to consume more red meats and have a higher risk of diabetes. Therefore, dietary habits might be a confounding factors. Finally, some researchers also report that obesity might be a confounder in the association between T2DM and colorectal disease [52].

Some studies reported a difference in risk between males and females [12, 39, 53–55]; however, the results of our subgroup analysis showed no difference. One possible explanation involves the redistribution of body fat that can occur when women experience menopause. The increase in visceral body mass fat could lead to hyperinsulinemia so that women, especially post-menopausal women, are more susceptible to colorectal diseases. However, the existence of menopause in some women cannot explain the different CRC risks for males and females [56–59]. Discrepancies among these studies and ours and the insignificant results by adenoma sub-site might be attributed to the limited sample sizes and insufficient statistical power. For the prospective studies, varied different follow-up procedures and mix of ethnicities different study populations might be the sources of heterogeneity.

Our analysis revealed that with T2DM have about a 5 % higher risk of CRA than newly diagnosed diabetes patients, revealing the duration of T2DM as a risk factor for CRA. A possible explanation is that known T2DM patients’ bowels are exposed to hyperinsulinemia or a hyperglycemic environment for a longer time, and such hormonal or metabolic abnormalities (according to former study [60]) could affect tumour growth. However, some studies reported that metformin use was a protective factor of CRA [33] and cancer [61]. If this is true, diagnosed diabetes patients should have a lower risk of adenomas than new patients, which is counter to our results. On the other hand, the severity of T2DM, which was not confirmed in the included studies, may affect colorectal disease risk and contribute to the mixed results. In sum, there might be a dose–response relationship between insulin and CRA, and further studies should include this as an important potential confounding factor.

Several limitations of in this meta-analysis that should be taken into consideration. First, results for several subgroups, such as gender and adenoma sub-site subgroup, were based on a limited number of studies. Therefore, we cannot rule out the possibility that insufficient statistical power is present. Second, in the present analysis, some small studies with inverse associations between T2DM and risk of CRA risk seemed to be suppressed. The presence of possible publication bias could have led to an overestimation of the effect of T2DM on CRA risk. However, the adjusted result was comparable after trim and fill method corrections. Third, we could not account for all of the confounding factors in the meta-analysis, though most confounders were adjusted in the original RRs. Many factors might induce the adenomas, such as age, ethnicity, inactivity, regular aspirin use, obesity, and family history of CRA, and menopausal status. We could not control for these covariates because of lack of relevant data. Relevant studies with additional data on these other factors may be found by searching by searching beyond the sources used for this study. Furthermore, we could not determine whether using insulin as a therapy for T2DM is an important factor because CRA risk might be altered by hyperinsulinemia, thought to be an important promoter of carcinogenesis [62, 63]. At the same time, metformin may have a direct anti-proliferative effect [64]. Finally, most of the existing studies did not discuss the influences of T2DM severity level on CRA risk. Thus, more cohort studies about these topics should be conducted.

Conclusions

In conclusion, the results of our meta-analysis indicated that patients with T2DM have higher risks for the development of CRA, which is an important inducement for colorectal cancer. Our study has important implications for clinical and public health. Because T2DM and CRA are prevalent in the developed and developing countries [65], medical workers should increase the rate of CRA screening for T2DM patients so that they can benefit from behavioural interventions that can help prevent CRA [38]. Large prospective studies that investigate the interactions among environmental and behavioral factors, medications, and functional polymorphisms are also needed to further clarify the etiology of CRA.