Insulin therapy and the risk of colorectal cancer in patients with type 2 diabetes: a meta-analysis of observational studies

Abstract

Aims

Several epidemiological studies have reported inconsistent associations between insulin therapy and the risk of colorectal cancer (CRC) in patients with type 2 diabetes mellitus. We performed this meta-analysis of observational studies to evaluate the effect of insulin therapy on the risk of CRC.

Methods

We carried out a systematic search of PubMed, Embase and the Cochrane Library Central database between January 1966 and August 2013. Fixed-effects and random-effects models were used to estimate the pooled relative risk (RR) and corresponding 95% confidence interval (CI).

Results

A total of 12 epidemiological studies were included in the present meta-analysis, involving a total of 7947 CRC cases and 491 384 participants. There was significant heterogeneity among the studies, but no publication bias. Insulin therapy significantly increased the risk of CRC [RR = 1.69, 95% CI (1.25, 2.27)]. When the various studies were stratified by study design, we found that insulin use was associated with a statistically significant 115% higher risk of CRC among case–control studies [RR = 2.15, 95% CI (1.41, 3.26)], but not among cohort studies [RR = 1.25, 95% CI (0.95, 1.65)]. Furthermore, a significant association was noted among studies conducted in USA [RR = 1.73, 95% CI (1.15, 2.60)] and Asia [RR = 2.55, 95% CI (2.14, 3.04)], but not in Europe [RR = 1.20, 95% CI (0.92, 1.57)].

Conclusions

The present meta-analysis suggests that insulin therapy may increase the risk of CRC. More prospective cohort studies with longer follow-up durations are warranted to confirm this association. Furthermore, future studies should report results stratified by gender and race and should adjust the results by more confounders.

WHAT IS ALREADY KNOWN ABOUT THIS SUBJECT

• The association between insulin therapy and cancer growth is linked at the biological level through hyperinsulinemia. Although several observational studies have investigated the association between insulin therapy and risk of colorectal cancer; however, the results were inconsistent.

WHAT THIS STUDY ADDS

• The present meta-analysis suggested that insulin therapy may increase the risk of colorectal cancer.

Introduction

Colorectal cancer (CRC) is the third most commonly diagnosed cancer worldwide in males and the second in females, with over 1.2 million new cases being diagnosed in 2008, accounting for 9.7% of all incident cancers [1]. Colorectal cancer is a multifactorial disease that involves both environmental and genetic factors. Epidemiological and experimental studies have provided strong evidence that smoking, physical inactivity, overweight and obesity can predispose to CRC [2,3].

The global prevalence of diabetes mellitus (DM) is rapidly increasing as a result of population ageing, urbanization and associated lifestyle changes [4,5]. Type 1 DM accounts for 5–10% of the total cases of DM, while type 2 DM accounts for 90–95%. Individuals with type 2 DM are at increased risk of developing various types of cancers [6], including CRC [7]. In type 2 DM, the pancreas secretes more insulin to compensate for insulin resistance [8]. This endogenous insulin helps to maintain normal blood glucose levels; however, insulin is also a known growth factor. Hyperinsulinaemia is hypothesized to promote cell growth and proliferation [9], and receptors for insulin are highly expressed on various types of cancer cells [10,11]. Whether insulin treatment increases the risk of cancers is an important issue because almost all patients with DM will eventually require insulin treatment [12]. The association between insulin therapy and cancer growth is linked at the biological level through hyperinsulinaemia. Several observational studies have investigated the association between insulin therapy and the risk of CRC, but their results were inconsistent. Some studies found that insulin therapy was associated with a significantly increased risk of CRC [13–21], whereas other studies found there was no significant association between insulin therapy and CRC risk [22–24]. As a result, whether insulin therapy is a risk factor for CRC remains unknown. Hence, we performed a detailed meta-analysis of observational studies to evaluate the effect of insulin therapy in patients with type 2 DM on the risk of developing CRC.

Methods

Literature search

This meta-analysis was conducted following guidance provided by the Cochrane Handbook [25] and reported according to the Preferred Reporting Items for Systematic reviews and Meta-Analyses guidelines (PRISMA) [26]. A literature search of PubMed, Embase and the Cochrane Library Central database, which contain manuscripts published between January 1966 and August 2013, was conducted to begin identifying potential studies to include in the meta-analysis. There was no restriction on origin and languages. Search terms included the following: ‘insulin’ or ‘glucose lowering therapy’ or ‘anti diabetic agent’, and ‘colorectal’ or ‘colorectum’ or ‘colon’ or ‘rectal’ or ‘rectum’, and ‘cancer’ or ‘neoplasm’ or ‘carcinoma’ or ‘tumor’. Reference lists of review articles and primary studies were also searched to identify additional relevant studies.

Study selection

The retrieved articles were reviewed independently by two investigators to determine whether they met the predefined criteria for inclusion in the meta-analysis. Disagreement between the two reviewers was settled by discussion with the third reviewer. Inclusion criteria were as follows: (i) used a case–control or cohort study design; (ii) evaluated the association between insulin therapy and CRC risk in patients with type 2 DM; and (iii) presented odds ratio (OR), relative risk (RR) or hazard ratio (HR) estimates with the 95% confidence interval (CI), or provided data for their calculation. There were no restrictions of origin, study size, language or publication type. Exclusion criteria were as follows: (i) lack of available data; and (ii) reviews, editorials, comments, reports from scientific sessions or discussions. When there were multiple publications from the same population, only data from the most recent report were included. Studies reporting different measures of RR, such as risk ratio, rate ratio, hazard ratio and odds ratio, were included in the meta-analysis. In practice, these measures of effect yield a similar estimate of RR, because the absolute risk of CRC is low.

Data extraction

Two investigators independently carried out data extraction of the following items: name of first author; date of publication; country of the population studied; study design; study period; sex of participants; number of cancer cases and subjects; data ascertainment methods; the study-specific adjusted ORs, RRs or HRs with their 95% CIs; and confounding factors for matching or adjustments. If the author reported the results of different durations of insulin exposure, we used only the data for the longest duration of insulin exposure.

Data synthesis and analysis

Heterogeneity was assessed using Cochran's Q test and I² statistics. For the Q statistic, a value of P < 0.10 was considered statistically significant for heterogeneity; for the I² statistic, heterogeneity was interpreted as absent (I² 0–25%), low (I² 25.1–50%), moderate (I² 50.1–75%) or high (I² 75.1–100%) [27]. For studies that reported results separately for males and females, but not combined, we pooled the results using a fixed-effects model to obtain an overall combined estimate before combining with the rest of the studies [28]. Subgroup analyses were carried out according to the following criteria: (i) study design (cohort vs. case–control studies); (ii) geographical location (Europe vs. USA vs. Asia); (iii) gender (male vs. female); (iv) number of adjustment factors (n ≥ 7 vs. n ≤ 6), adjustment for body mass index (BMI; yes vs. no) and adjustment for smoking status (yes vs. no). Pooled RR estimates and their corresponding 95% CIs were calculated using the inverse variance method. When substantial heterogeneity was detected (I² ≥ 50%), the summary estimate based on the random-effects model (DerSimonian–Laird method) [29] was reported, which assumed that the studies included in the meta-analysis had varying effect sizes. Otherwise, the summary estimate based on the fixed-effects model (the inverse variance method) [30] was reported, which assumed that the studies included in the meta-analysis had the same effect size. To test the robustness of association and characterize possible sources of statistical heterogeneity, sensitivity analysis was carried out by excluding studies one by one and analysing the homogeneity and effect size for all of the remaining studies. To investigate the possible sources of between-study heterogeneity better, a meta-regression analysis was performed [31]. A univariate model was established, and then variables with values of P ≥ 0.1 were entered into a multivariable model. Publication bias was assessed using Begg and Mazumdar adjusted rank correlation test and the Egger regression asymmetry test [32,33]. All analyses were performed using Stata version 11.0 (StataCorp, College Station, TX, USA).

Results

Literature search and study characteristics

The detailed steps of our literature search are shown in Figure 1. A total of 2657 citations were identified during the initial search. On the basis of the title and abstract, we identified 17 papers. After detailed evaluation, five papers were excluded because of a lack of data. Finally, the remaining 12 studies published between 2004 and 2013 were included in the present meta-analysis [13–24], involving a total of 491 384 participants and 7947 CRC cases. Of these 12 studies, seven were case–control studies [13,14,16,17,19–21] and five cohort studies [15,18,22–24]. Of the five cohort studies, only one study was a prospective study [22], whereas the others were retrospective. Five studies were conducted in USA [17–20,22], three in Europe [15,21,23] and the remaining four in Asia [13,14,16,24]. Three studies reported results separately for males and females, but not combined [18,22,23]. Almost all studies adjusted for age and sex, but most did not adjusted for alcohol consumption, diet style, family history of CRC and physical activity (baseline data and other details are shown in Table 1).

Figure 1.

Flow diagram of screened, excluded and analysed publications

Table 1.

Characteristics of included studies that investigated the association between insulin therapy and colorectal cancer risk

Study	Year of publication	Study design	Country	Data ascertainment methods	All subjects	Colorectal cancer cases	Study period	Duration of insulin use	Sex	Adjusted risk estimate (95% CI)	Confounders for adjustment
Onitilo et al. [18]	2013	Retrospective cohort	USA	Marshfield Clinic electronic medical record and cancer registry	9486	106	1995–2009	NR	M/F	Male HR: 1.37 (0.74−2.5) Female HR: 2.22 (1.18−4.17)	BMI, age, date of diagnosis, insurance status, comorbidities, smoking history and location of residence
Gu et al. [24]	2013	Retrospective cohort	China	Shanghai Diabetes Registry	8774	31	2001–2010	NR	M/F	RR: 1.25 (0.50–3.12)	Age, sex, smoking status, diabetes duration, macrovascular complications (coronary artery disease, cerebrovascular disease, and peripheral artery disease) glycosylated haemoglobin, concomitant oral glucose-lowering agents
Chang et al. [13]	2012	Case–control study	China	Taiwan National Health Insurance claims database	108 920	468	2000–2007	≥2 years	M/F	OR: 2.67 (2.29−3.12)	Use of sulfonylurea, glinides, metformin, thiazolidinediones, α-glucosidase inhibitors, chronic liver disease, nephropathy, statins, β-blocker, calcium channel blockers, cerebrovascular disease, angiotensin-converting enzyme inhibitors, chronic kidney disease and aspirin
Wong et al. [20]	2012	Case–control study	USA	Two electronic medical records systems within the University of Pennsylvania Health System	1168	196	1998–2007	≥3 years	M/F	OR: 2.0 (1.2–3.4)	Sex and age
Hsieh et al. [16]	2012	Case–control study	China	Taiwan's National Health Research Institutes database	61 777	1739	2000–2008	NR	M/F	OR: 2.14 (1.23–3.72)	Sex and age
Carstensen et al. [23]	2012	Retrospective cohort	Denmark	National diabetes register data	22 826	320	1995–2009	NR	M/F	Male RR: 0.99 (0.86–1.14) Female RR: 0.90 (0.76–1.09)	Age, current date of follow-up, date of birth
Campbell et al. [22]	2010	Prospective cohort	USA	Cancer Prevention Study II (CPS-II) Nutrition Cohort	184 194	2474	1992–2007	≥4 years	M/F	Male RR: 1.11 (0.82−1.51) Female RR: 0.94 (0.60−1.48)	Age, education, BMI, physical activity, NSAID use, alcohol use, family history of colorectal cancer and endoscopy history
Vinikoor et al. [19]	2009	Case–control study	USA	North Carolina Central Cancer Register	3752	1688	1996–2006	≥1 year	M/F	OR: 2.31 (1.23–4.32)	Age, sex, NSAID use, calcium intake, family history of colorectal cancer, education and offset (Offset terms to account for the randomized recruitment strategy)
Currie et al. [15]	2009	Retrospective cohort	UK	The Health Information Network	62 809	292	2000–2009	NR	M/F	HR: 1.69 (1.23–2.33)	Age, sex, smoking status, diagnosis of a previous cancer
Chung et al. [14]	2008	Case–control study	Korea	Medical records in the Hallym University Sacred Heart Hospital	325	100	2003–2006	NR	M/F	OR: 3.0 (1.1–8.9)	Sex and age
Koro et al. [17]	2007	Case–control study	USA	Integrated Healthcare Information Services (IHCIS) managed care database	2435	408	1997–2004	NR	M/F	OR: 4.46 (1.05–19.0)	Age, sex, calendar time, length of follow-up, years of recorded history in database before index date
Yang et al. [21]	2004	Case–control study	UK	General Practice Research Database of the UK	24 918	125	1987–2002	≥5 years	M/F	OR: 1.2 (1.03–1.4)	Sex, history of cholecystectomy, smoking, duration of type 2 diabetes mellitus, BMI, metformin use, sulfonylurea use and NSAID/aspirin use

Abbreviations are as follows: BMI, body mass index; CI, confidence interval; F, female; HR, hazard ratio; M, male; NR, not reported; NSAID, nonsteroidal anti-inflammatory drug; OR, odds ratio; RR, risk ratio.

Main analysis

Given that significant heterogeneity (P < 0.001, I² = 91.8%) was observed, a random-effects model was chosen over a fixed-effects model, and we found that insulin therapy (compared with non-use) was associated with a statistically significant 69% higher risk of CRC in patients with type 2 DM [RR = 1.69, 95% CI (1.25, 2.27)]. Both multivariable adjusted RR estimates with 95% CIs of each study and combined RR are shown in Figure 2.

Figure 2.

Forest plot of the overall meta-analysis of insulin use and colorectal cancer risk. Squares indicated study-specific risk estimates (size of square reflects the study statistical weight, i.e. inverse of variance); horizontal lines indicate 95% CI and diamonds indicate the summary relative risk estimate with its corresponding 95% CI. CI, confidence intervals; ES, effect estimate

Subgroup analyses and sensitivity analysis

We found that insulin therapy was associated with a statistically significant 115% higher risk of CRC in patients with type 2 DM among case–control studies [RR = 2.15, 95% CI (1.41, 3.26)]; however, no significant association was found between insulin use and risk of CRC among cohort studies [RR = 1.25, 95% CI (0.95, 1.65); presented in Table 2]. When the various studies were stratified by geographical location, significant association was noted among studies conducted in USA [RR = 1.73, 95% CI (1.15, 2.60)], and Asia [RR = 2.553, 95% CI (2.14, 3.04)], however, no significant association was found among studies conducted in Europe [RR = 1.20, 95% CI (0.92, 1.57)]. No significant association was observed in either the male population [RR = 1.02, 95% CI (0.90, 1.16)] or the female population [RR = 1.14, 95% CI (0.73, 1.77)]. When we examined whether the associations differed by adjustment for BMI or smoking status, the associations did not vary by these factors. Furthermore, it was observed that studies with higher control for potential confounders (n ≥ 7) as well as studies with lower control (n ≤ 6) presented a significant association between insulin use and the risk of CRC [RR = 1.61, 95% CI (1.06, 2.44) and RR =1.80, 95% CI (1.15, 2.81), respectively; shown in Table 2]. To test the robustness of association, sensitivity analysis was carried out by excluding studies one by one. Sensitivity analysis indicated that there was no significant variation in combined RR from exclusion of any of the studies, confirming the stability of the present results.

Table 2.

Summary risk estimates of the association between insulin therapy and colorectal cancer risk

	No. of studies	Pooled estimate		Tests of heterogeneity
		RR	95% CI	P value	I2 (%)
All studies	12	1.69	1.25–2.27	<0.001	91.8
Study design
Cohort	5	1.25	0.95–1.65	0.002	75.7
Case–control	7	2.15	1.41–3.26	<0.001	89.1
Geographical location
Europe	3	1.20	0.92–1.57	0.001	85.8
USA	5	1.73	1.15–2.60	0.014	67.8
Asia	4	2.55	2.14–3.04	0.373	3.9
Gender
Male	3	1.02	0.90–1.16	0.512	0.0
Female	3	1.14	0.73–1.77	0.026	72.5
Adjusted for confounders
Number of adjustment factors
n ≥ 7 confounders	6	1.61	1.06–2.44	<0.001	92.6
n ≤ 6 confounders	6	1.80	1.15–2.81	<0.001	83.8
Smoking status adjusted
Yes	4	1.42	1.13–1.79	0.153	43.0
No	8	1.89	1.19–2.98	<0.001	94.5
Body mass index adjusted
Yes	3	1.22	1.00–1.49	0.158	45.8
No	9	1.94	1.25–3.03	<0.001	93.6

Abbreviations are as follows: CI, confidence interval; RR, relative risk.

Meta-regression analysis

In order to investigate the possible sources of between-study heterogeneity better, a meta-regression analysis was performed. Study design (cohort or case–control study), geographical location (Europe, USA, Asia), year of publication, number of adjustment factors and the major confounders adjusted (BMI, smoking status), which may be potential sources of heterogeneity, were tested by a meta-regression method. Meta-regression analysis revealed that study design (P = 0.046) and geographical location (P = 0.039) were the sources of heterogeneity.

Publication bias

In the present meta-analysis, no publication bias was observed among studies using Begg's P value (P = 0.08) or Egger's test (P = 0.24), which suggested that there was no evidence of publication bias (Figure 3).

Figure 3.

Funnel plot for publication bias in the studies investigating relative risk (RR) for colorectal cancer associated with the use of insulin

Discussion

Nearly all patients with type 2 DM will eventually need exogenous insulin therapy for adequate glycaemic control as pancreatic β-cell function deteriorates. With increasingly stringent guidelines for glycaemic control put forth by the American Diabetes Association [34] and growing evidence for a reduction in microvascular complications with introduction of insulin therapy earlier in the disease course [35,36], insulin users represent a substantial and growing proportion of the entire type 2 DM population. Mechanistically, insulin stimulates cell proliferation through two pathways [37]. The minor pathway involves direct binding of insulin to either insulin or insulin-like growth factor-1 (IGF-1) receptors, while the major pathway is via inhibition of IGF-binding proteins and the resultant increase in IGF-1 availability to the IGF-1 receptor [37]. The IGF system is a potent growth regulator closely linked with carcinogenesis [38]. Furthermore, insulin has been demonstrated to be associated with the development of colonic neoplasia in animal models [39,40].

In this comprehensive meta-analysis of 12 studies analysing the effect of insulin use on modifying the risk of CRC in 491 384 patients with type 2 DM, we found that, relative to non-use, use of insulin was associated with a 69% increase in the risk of CRC. In the present meta-analysis, significant heterogeneity was observed among all studies; therefore, a random-effects model, which provided a more conservative standard error and a larger confidence interval, was chosen over a fixed-effects model to determine the pooled RR estimates in our meta-analysis. Meta-regression analysis suggested that study design and geographical location were the sources of between-study heterogeneity.

In our subgroup analyses, the results were substantially affected by study design. The results from case–control studies showed a significant association between insulin therapy and CRC risk; however, the results from the cohort studies showed that there was no significant association between insulin therapy and CRC risk. This may be partly related to the study design, because case–control studies are susceptible to potential recall and selection biases (especially dietary recall bias) due to their nature. Furthermore, it is noteworthy that there were only five cohort studies (four retrospective cohort studies and one prospective cohort study) investigating the association between insulin use and the risk of CRC, which is too low a number to enable firm conclusions to be drawn. More prospective cohort studies are needed to confirm the association between insulin use and the risk of CRC in the future.

When we carried out subgroup analysis by geographical location, we found a significant risk increase in CRC in American and Asian populations but not in European populations. The differences in genetic susceptibility, culture and lifestyles may explain part of the inconsistency in the results. No significant association between insulin therapy and the risk of CRC was observed in both male and female populations; however, only three studies reported results separately for males and females, and that number was rather low to enable firm conclusions to be drawn. Future studies should therefore report results separately for males and females. When we examined whether the associations differed with adjustment for smoking status or BMI, we did not find any substantial differences, indicating that the influence of those adjusted confounders on the results might be small.

The strength of the present meta-analysis lies in inclusion of 12 studies, reporting data from 491 384 participants and 7947 CRC cases. Publication bias resulting from the tendency not to publish small studies with null results was not found in our meta-analysis. Furthermore, our findings were stable and robust in sensitivity analysis.

There were several limitations in our meta-analysis. First, most of the included studies did not reported results separately for males and females. There were only three studies studying the association between insulin therapy and the risk of CRC among males and females separately. That number was rather low to enable firm conclusions to be drawn; therefore, future studies should report results separately for males and females. Second, there were five studies conducted in the USA; however, only the study by Vinikoor et al. [19] examined the association between insulin use and CRC risk by race. Among Caucasians with diabetes, insulin use was positively associated with CRC [RR = 2.53, 95% CI (1.21, 5.28)]; however, no significant association was observed among African American diabetics using insulin [RR = 1.81, 95% CI (0.55, 6.00)]. Furthermore, none of the studies conducted in Asia and Europe examined the association by race; therefore, studies in the future should examine the association by race. Third, about half of the included studies did not report the duration of insulin treatment. Only two studies reported the results of different durations of insulin use. Wong et al. [20] found that there was evidence of a duration–response effect, with a 20% incremental increase in odds of CRC for each additional year of insulin use on average [RR = 1.20, 95% CI (1.00, 1.30)]. Likewise, Yang et al. [21] demonstrated that the risk for CRC increased with duration of use [RR = 1.20, 95% CI (1.03, 1.40)]. Given that a short duration of insulin use may not provide a long enough period of insulin exposure to increase the risk of developing CRC, more studies with longer duration of insulin therapy are needed in the future. Fourth, most studies did not adjust for alcohol consumption, diet style, family history of CRC, physical activity or duration of DM, which are risk factors for CRC.

In summary, this meta-analysis suggests that insulin therapy may increase the risk of CRC. More prospective cohort studies with longer follow-up durations are warranted to confirm this association. Future studies should report stratified results by gender and race, and should adjust the results by more confounders.

Competing Interests

All authors have completed the Unified Competing Interest form at http://www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and declare: no support from any organization for the submitted work; no financial relationships with any organizations that might have an interest in the submitted work in the previous 3 years; no other relationships or activities that could appear to have influenced the submitted work.