Diabetes mellitus and the risk of cholangiocarcinoma: an updated meta-analysis.

INTRODUCTION: A number of studies have shown that diabetes mellitus is implicated in susceptibility to several cancers. However, the relationship between diabetes and cholangiocarcinoma remain unclear. AIM: To quantitatively assess the relationship between diabetes and incidence of cholangiocarcinoma in cohort and case-control studies.
MATERIAL AND METHODS: A literature search was performed for entries from 1996 to 2014 using the PubMed and EMBASE databases. Studies were included if they reported odds ratios (OR) and corresponding 95% CI of cholangiocarcinoma with respect to diabetes mellitus.
RESULTS: Twenty studies met the inclusion criteria, which included fifteen case-control studies and five cohort studies from Asia (n = 11), the United States (n = 5), and Europe (n = 4). Compared with individuals without diabetes, the pooled OR of cholangiocarcinoma was 1.74 (95% CI: 1.62-1.87, p = 0.568 for heterogeneity) for patients with diabetes, ICC (summary RR, 1.93; 95% CI: 1.65-2.25; p = 0.037 for heterogeneity), and ECC (summary RR, 1.66; 95% CI: 1.39-1.98; p = 0.001 for heterogeneity). The funnel plot revealed no evidence for publication bias concerning diabetes and the risk of CC (including ICC and ECC).
CONCLUSIONS: The findings from this meta-analysis suggest that diabetes may increase the risk of cholangiocarcinoma. This relationship needs to be confirmed by further follow-up studies.

Introduction

Cholangiocarcinoma (CC) is one of the most lethal human malignant tumours. Anatomically, CC can be classified as either intrahepatic cholangiocarcinoma (ICC) and extrahepatic cholangiocarcinoma (ECC), according to its location [1]. Incidence and mortality rates for CC have risen across the world [2]. In the UK, CC has killed approximately 1500 people annually since the mid-1990s, with approximately equal numbers of men and women [3]. Because of the difficulty in obtaining an early diagnosis, the prognosis is extremely poor [4], even with surgical and chemotherapy intervention.

There are several established risk factors for CC, such as primary sclerosing cholangitis (PSC), which is the commonest known predisposing factor for CC in the western world [5]. Other factors including chronic intraductal gallstones, liver fluke, choledochal (bile duct) cysts and Caroli's disease (intrahepatic biliary cysts), bile duct adenoma and biliary papillomatosis, Thorotrast, inflammatory bowel disease (IBD), and chronic typhoid carriage [6-9]. Many studies have been conducted to explore potential risk factors of CC. Chronic viral hepatitis B or C, obesity, diabetes, fatty liver disease, alcohol, smoking, polymorphisms of genes, inflammation, and biliary transporters may also be risk factors.

Several studies have found that diabetes can increase the risk of cancers, including cancers of the breast, pancreas, liver, and non-Hodgkin's lymphoma [10-12]. Also, some cohort and case-control studies have been conducted to estimate the relationship between diabetes and CC. But the results were controversial. Some studies found that diabetes can increase the risk of CC. Insulin and insulin-like growth factor-1 (IGF-1) are related to CC progress, but some other studies found no relation between them.

Aim

Therefore, a meta-analysis was performed to quantitatively assess the relationship between diabetes and the risk of CC in humans.

Material and methods

Data sources and searches

We screened the relevant studies from the search engines of PubMed and EMBASE (last search update performed on 07/2014), using two investigators independently, with the Medical Subject Heading (MeSH) terms ‘diabetes mellitus’, ‘diabetes’, ‘cholangiocarcinoma’, ‘intrahepatic’, ‘extrahepatic’, ‘bile duct cancer’, and ‘epidemiologic studies’, without language limit. Furthermore, we reviewed reference lists of retrieved articles to search for more studies.

Inclusion and exclusion criteria

Studies were included in the meta-analysis if they fulfilled the following criteria: (1) cohort or case-control design; (2) one of the exposure interests was diabetes mellitus (DM); (3) one of the outcomes of interests was ICC, ECC, or CC; (4) relative risk (RR) in cohort studies or odds ratio (OR) in case-control studies and their 95% confidence intervals (CI) (or data to calculate them) were reported. If data were duplicated in more than one study, the estimated effects controlled for the most appropriate confounders were included.

Data extraction

According to the MOOSE (Meta-analysis Of Observational Studies in Epidemiology) guideline for reporting on meta-analyses of observational studies [11, 13], the data were extracted by two investigators independently. The information of each study was extracted as follows: the first author's last name; the region/country where the study was conducted; the year of publication; the number of cases; the mean age of cases and controls; diagnostic criteria; the manner in which the controls were selected; OR, RR, or hazard ratio (HR) of CC and corresponding 95% CI for diabetes; and covariates adjusted in the statistical analysis. Discrepancies were resolved by discussion with a third investigator and a consensus was reached.

Statistical analysis

The study-specific, most adjusted OR, RR, or HR was used to compute a summary OR and its 95% CI. Relative risks and HR were directly considered as OR. The statistical heterogeneity among the studies was estimated by the χ2 test-based Q-statistic, and a significant Q-statistic (p < 0.10) indicated heterogeneity across the studies [14]. The pooled OR was calculated by a fixed effect model (using the Mantel-Haenszel method) or a random effect model (using the DerSimonian-Laird method) according to the heterogeneity among studies [15, 16]. The potential publication bias was evaluated using both Begg's funnel plot and Egger's test, and p < 0.05 was considered statistically significant in publication bias [17]. Analyses were performed by using Stata version 12.0 (StataCorp LP, College Station, TX, USA). A p-value < 0.05 was considered statistically significant, and all the p values were two sided.

Results

Characteristics of literature search and studies

There were 20 studies included in this meta-analysis [18-37]. The continents or countries in which the studies were conducted were as follows: Asia (n = 11), the United States (n = 5), and Europe (n = 4). Characteristics of the studies are shown in Table I.

Table I

Characteristics of case-control studies of diabetes and cholangiocarcinoma

Author/country/year	CC	ECC	ICC	Source	Control (DM, n)	Diabetes assessment	Outcome ascertainment	OR and 95% CI	Adjustments
Yamamoto et al [18]/Japan/2004			50 (11)	Hospital	205 (24)	Hospital records	Pathological	ICC: 1.95 (0.65–5.85)	HCV, hypertension, transfusion, TBi, Alb, Plt count, ALT
Shaib et al. [19]/USA/2005			625 (125)	Population	90834 (14201)	Self-report	Cancer registry	ICC: 2.0 (1.6–2.4)	Age, sex, race, geographic location, medicare/medical enrolment
Welzel et al. [20]/Denmark/2007			764 (15)	Population	3056 (43)	Hospital records	Cancer registry	ICC: 1.43 (0.78–2.63)	NA
Welzel et al. [21]/USA/2007		549 (165)	535 (177)	Population	102782 (22764)	Hospital records	Cancer registry	CC: 1.64 (1.37–1.96)a ICC: 1.8 (1.5–2.1)ECC: 1.5 (1.3–1.8)	Age, sex, race, geographic region
Shaib et al. [22]/USA/2007		163 (19)	83 (12)	Hospital	236 (20)	NA	Histologic diagnosis	CC: 1.55 (0.89–2.69)a ICC: 1.8 (0.7–4.1)ECC: 1.4 (0.7–2.9)	NA
Lee et al. [23]/Korea/2008			685 (96)	Population	124763 (139)	NA	Pathological diagnosis	ICC: 3.2 (2.3–4.3)	Age, sex
Zhou et al. [24]/China/2008			312 (13)	Hospital	438 (11)	Hospital records	Pathological diagnosis	ICC: 1.50 (0.60–3.80)	Age, sex, HCV/HBV markers, heavy drinking
Grainge et al. [25]/UK/2009	372(35)			Population	5760 (342)	NA	Hospital records	CC: 1.48 (1.00–2.17)	Smoking, alcohol use, BMI, NSAID use, gallstone
Shebl et al. [26]/China/2010		191 (20)		Population	959 (78)	Hospital records	Hospital records	ECC: 0.79 (0.30–2.07)	Age, sex, education, aspirin use, BMI, DM duration, waist-to-hip ratio
Tao et al. [27]/China/2010		129 (24)	61 (3)	Hospital	380 (36)	Hospital records	Pathological diagnosis	CC: 1.39 (0.24–8.06)a ICC: 0.53 (0.17–1.65)a ECC: 3.2 (1.7–5.9)	Age, sex, HBV markers, history of cholecystectomy
Peng et al. [28]/China/2011			98 (6)	Hospital	196 (14)	Hospital records	Pathological diagnosis	ICC: 0.85 (0.32–2.28)	Age, sex, HBV infection, liver cirrhosis, hepatolithiasis, liver fluke infestation
Liu et al. [29]/China/2011			87(3)	Hospital	228 (10)	Hospital records	Pathological diagnosis	ICC: 0.779 (0.209–2.899)	Age, sex, smoking, alcohol consumption family history of cancer
Cai et al. [30]/China/2011		313(17)		Hospital	608 (37)	Hospital records	Pathological diagnosis	ECC: 0.851 (0.446–1.626)	Age, sex, hepatolithiasis, cholecystolithiasis and parasitic, biliary ascariasis, liver fluke, liver schistosomiasis
Wu et al. [31]/China/2012		86(19)	102(16)	Hospital	835 (69)	Hospital records	Pathological diagnosis	ICC: 2.065 (1.147–3.718)ECC: 3.148 (1.788–5.544)	Age, sex, HBV infection, cholelithiasis hypertension
Chang et al. [32]/China/2013		2179 (661)	2978 (932)	Hospital	20628 (4027)	Hospital records	Pathological diagnosis	ICC: 2.0 (1.8–2.2)ECC: 1.8 (1.6–2.0)	Age, sex, hepatitis B, hepatitis C, peptic ulcer, inflammatory bowel disease

Alb – albumin, ALT – alanine aminotransferase, AORs – adjusted odds ratios, CC – cholangiocarcinoma, CI – confidence interval, DM – diabetes mellitus, ECC – extrahepatic cholangiocarcinoma, ICC – intrahepatic cholangiocarcinoma, NA – not available, NSAID – nonsteroidal anti-inflammatory drug, Plt – platelet, Tbi – total bilirubin

The AOR and 95% confidence intervals were derived by pooling the site-specific AORs.

The 15 case-controlled studies reported a total of 3610 cases with ECC and 6380 cases with ICC. And the report from Grainge presented results for 372 cases with CC [25]. Among these 10,362 cases, 2399 cases with diabetes were reported (Table I), whereas, among 351,908 controls, 41,815 patients had diabetes. The control individuals included originated from a population-based [19–21, 23, 25, 26] or hospital-based setting [18, 22, 24, 27–32]. Diabetes status was ascertained by a self-reported history [19] of DM or hospital records [18, 20, 21, 24, 26–32], with the exception of three studies in which the methods of DM ascertainment were not available [22, 23, 25]. Ascertainment of ECC or ICC was based on histological methods or a review of medical records in 12 studies, and the remaining three studies were based on diagnostic codes [19-21]. Adjustments were made for potential confounders of tor more factors in studies 12 of 13, with the exception of 2 studies in which only the univariate OR was available [20, 22].

We identified five cohort studies that showed an association between DM and the risk of ICC or ECC (Table II). Among these five studies, the standardised incidence ratio as a measure of RR was used in two diabetic cohorts, and the other three studies used rate ratio as the measure of RR [33, 37]. These five cohort studies comprised between 56,881 and 836,283 persons with a median follow-up period of 6.7 years, reporting a total of 878 incident cases of ICC or ECC. The methods of DM ascertainment were based on medical records in three studies, and in the other two studies they were based on self-reported history or registry of disease [34, 36]. The ascertainment of outcome was based on cancer registry in all studies. Potential confounders (at least for age) were controlled in all studies. The Newcastle-Ottawa scale was adopted in our quality assessment. The maximum score was 9, and all studies scored 7 or higher.

Table II

Characteristics of cohort studies of diabetes and cholangiocarcinoma

Author/country/year	Number of patients	Demographics of all patients (age in years)	Diabetes assessment	Cancer ascertainment	Follow-up [years]	ICC/ECC in DM	Adjusted RR (95% CI)	Adjustments
Adami et al. [33]/Sweden/1996	153 852	74 male: 64%	Hospital discharge diagnosis	Cancer registry	6.7	272	ECC: 1.4 (1.1–1.8)	Alcohol use, hepatitis, cirrhosis, jaundice, etc
Khan et al. [34]/Japan/2006	56 881	40–70	NA	Cancer registry	18–20	40	ECC: 0.30 (0.04–2.22)	Age, sex, race, geographic location, medicare/medical enrolment
El-Serag et al. [35]/USA/2009	718 687	52 male: 97%	Registry	Cancer registry	2.3	NA	CC: 1.60 (0.67–3.83)aICC: 2.54 (1.31–4.94)ECC: 1.04 (0.59–1.83)	Age, gender, baseline visit date, type of visit
Jamal et al. [36]/USA/2009	836 283	65 male: 98%	Hospital discharge diagnosis	Cancer registry	NA	NA	ECC: 2.1 (1.6–2.5)	Age, sex, race, geographic region
Hemminki et al. [37]/Sweden/2010	125 126	> 39 male: NA	Medical records	Cancer registry	15	566	ECC: 2.53 (1.44–4.11)	Age

AORs – adjusted odds ratios, CI – confidence interval, DM – diabetes mellitus, CC – cholangiocarcinoma, ECC – extrahepatic cholangiocarcinoma, ICC – intrahepatic cholangiocarcinoma, NA – not available, RR – relative risk

The AOR and 95% confidence intervals were derived by pooling the site-specific RRs

Diabetes mellitus and risk of cholangiocarcinoma

Six case-controlled studies and one cohort study reported results on DM and risk of CC (or risk of ICC and ECC, respectively) (Figure 1). Of these, four studies found statistically significant positive relationships [21, 25, 31, 32], and the other three studies did not find a significantly increased risk of CC in patients with diabetes [22, 27, 35]. In the analysis of all five studies that reported RR of DM and CC, the summary RR and corresponding 95% CI were 1.74 (95% CI: 1.62–1.87) in a random-effects model for those with diabetes compared with those without diabetes. There was no statistically significant heterogeneity among studies (p = 0.568; I2 = 0%).

Figure 1

Forest plot of the relationships between diabetes and CC risk

Diabetes mellitus and risk of extrahepatic cholangiocarcinoma

We identified 12 studies (7 case-controlled and five cohort studies) that presented results on diabetes and risk of ECC (Figure 2). Of these, 7 studies found an increased risk of ECC in patients with diabetes [21, 27, 33, 36, 37], and in another five studies positive relationships were not found [22, 26, 30, 34, 35]. In the analysis of all studies, the summary RR of ECC were 1.66 (95% CI: 1.39–1.98; p = 0.001 for heterogeneity) in a random-effects model for those with diabetes compared with those without diabetes, with evidence of significant heterogeneity among studies (Q = 31.42, p = 0.001, I2 = 65%). We then conducted subgroup meta-analyses by geographic area and study design (Table III). The summary RR of the associations between diabetes and ECC risk were similar for cohort studies and case-controlled studies (summary RR (95% CI) 1.61 (1.14–2.29), in cohort studies and 1.66 (1.32–2.1) in case-controlled studies, respectively). We conducted subgroup analysis by geographic area (Figure 3). A significant association between DM and ECC risk was found in studies conducted in non-Asian regions (the USA and Europe) (summary RR, 1.62; 95% CI: 1.32–2.00) and in Asia (summary RR, 1.60; 95% CI: 1.01–2.54).

Figure 2

Forest plot of the relationship between DM and ECC risk in case-controlled studies and cohort studies

Table III

Summarised relative risks for the association between diabetes and ECC and ICC by study characteristics

Subgroup		Number of studies	Relative risk (95% CI)	Tests for heterogeneity
				Q	p	I2 (%)
ECC	Geographical region:
	Asia	6	1.60 (1.01–2.54)	18.19	0.003	72.5
	Non-Asian	6	1.62 (1.32–2.00)	12.70	0.026	60.6
	Study design:
	Case-control study	7	1.66 (1.32–2.10)	17.38	0.008	65.5
	Cohort studies	5	1.61 (1.14–2.29)	13.78	0.008	71.0
ICC	Geographical region:
	Asia	7	1.82 (1.27–2.60)	18.50	0.005	67.6
	Non-Asian	5	1.88 (1.66–2.12)	2.19	0.701	0.0
	Study design:
	Case-control study	11	1.90 (1.62–2.23)	19.91	0.030	49.8
	Cohort studies	1	2.54 (1.31–4.93)	–	–	–

CI – confidence interval, ECC – extrahepatic cholangiocarcinoma, ICC – intrahepatic cholangiocarcinoma.

Figure 3

Forest plot of the relationship between DM and ECC risk for twelve studies by geographic region

Diabetes mellitus and intrahepatic cholangiocarcinoma risk

We identified 11 case-controlled and one cohort study that presented results for the association of diabetes and ICC risk (Figure 4) [18–24, 27–29, 31, 32, 35]. Five of these 12 studies found a statistically significant positive association (range of individual RR, 0.53–3.2; summary RR for all 12 studies, 1.93; 95% CI: 1.65–2.25). There was significant heterogeneity among studies (p = 0.037, I2 = 46.9%). Subgroup meta-analyses by study design indicated that the positive association was significant not only among case-controlled studies (summary RR, 1.90; 95% CI: 1.62–2.23; p = 0.030 for heterogeneity) but also among cohort studies (summary RR, 2.54; 95% CI: 1.31–4.93) (Table III). In addition, we also conducted subgroup analysis by geographic area (Figure 5). The association between diabetes and ICC risk was significantly positive in studies conducted in both Asian (summary RR, 1.82; 95% CI: 1.27–2.60; p = 0.005 for heterogeneity) and in non-Asian regions (summary RR, 1.88; 95% CI: 1.66–2.12; p = 0.701 for heterogeneity).

Figure 4

Forest plot of the relationship between DM and ICC risk in case-controlled studies and cohort studies

Figure 5

Forest plot of the relationship between DM and ICC risk for twelve studies by geographic region

Publication bias

The funnel plot revealed no evidence for publication bias concerning diabetes and the risk of CC, either in Egger's and Begg's tests or visualisation of the Begg's funnel plot (Egger's test: ECC, p = 0.661, ICC, p = 0.735; Begg's tests: ECC, p = 0.15, ICC, p = 0.373). A sensitivity analysis, which was performed to evaluate the stability, revealed that there was no significant impact on the overall results with removal of any of the studies.

Discussion

In this meta-analysis, we reviewed the case-control studies and cohort studies with information of diabetes and CC in two regions. The overall result suggests a positive association between diabetes and the risk of CC. Even though among diabetic individuals absolute risks of CC are low, our results have important clinical and public health significance. Diabetes may be a pathogenic factor for the development of CC. Sub-group analysis showed that this increased risk was largely attributed to the summary risk estimates from case-controlled studies and cohort studies. Although some meta-analyses were conducted with respect to the relationship between diabetes and CC, some other meta-analyses were conducted with respect to the relationship between diabetes and ICC or extrahepatic cholangiocarcinoma ECC. We screened more studies in this meta-analysis to quantitatively assess the relation between diabetes and CC, including CC, ICC, ECC and a significant association between DM and ECC risk was found in non-Asian regions and Asian regions. This result is different from other previous meta-analysis.

Diabetes is a common disease all over the world, and its incidence and mortality is increasing. In many countries diabetes and its comorbidities have become a major public health concern. Epidemiological studies have provided strong evidence that diabetes can increase the incidence of many types of cancer (including cancers of the breast, endometrium, non-Hodgkin's lymphoma, pancreas, and the liver) [38, 39]. Many studies found that pancreatic gland disease (pancreatic cancer, pancreatitis) is associated with DM. The results from the present meta-analysis suggest that diabetes may be a pathogenic factor for the development of cancer. Insulin-like growth factor-1 and insulin play an important role in the course of malignant transformation in the tissues of the colon, breast, lung, bladder, and prostate. Studies have confirmed that insulin can stimulate growth of many malignant tumour cell lines, and thus up-regulate the level of IGFs [40]. Some studies have suggested that diabetes increases the incidence of CC and level of IGF-1 associated with the development and progression of CC [19, 21, 25, 41]. In some studies, DM has been considered to be an independent risk factor for cholelithiasis [42, 43], which is one of the primary risk factors for CC. Inflammatory cytokines produced by adipose tissues, such as interleukin-6, monocyte chemoattractant protein, and plasminogen activator inhibitor-1, may play important roles in carcinogenesis, cancer progression, and poor prognosis. These may be the possible mechanisms whereby diabetes causes CC.

Heterogeneity across studies is often a concern in a meta-analysis. It was not surprising that a certain degree of heterogeneity was observed given the between-study variation, such as race, study design, and sample source. The degree of heterogeneity was somewhat attenuated among the studies conducted in Asian countries, suggesting that race may be a potential source of heterogeneity. However, meta-regression was adopted and no variables were identified as potential contributors to heterogeneity.

Some potential limitations should be considered in the present meta-analysis. First, since we only searched papers in English and Chinese, the completeness of evidence is impeded by language bias. Second, most of the studies did not distinguish between type 1 and type 2 diabetes, which might attenuate any true relationship between diabetes and CC risk. Third, confusion is also likely to be present because these two diseases share several risk factors, such as aging, smoking, alcohol consumption, and obesity. However, the relationship between these two diseases’ risk factors was only marginally attenuated after adjustment for a wide range of potential confounders. Finally, as in any meta-analysis, the possibility of publication bias is of concern, because small studies with null results tend not to be published. Publication bias may have resulted in an overestimate of the relationship between DM and risk of CC. However, the results obtained from funnel plot analysis and formal statistical tests did not provide evidence for such a bias. Last but not least is the number of included studies. More studies, especially cohort studies, about the association of diabetes and CC risk are needed to update the results.

Conclusions

The current meta-analysis showed that diabetes may increase the risk of CC. Well-designed cohort studies are warranted to confirm this association.