Low circulating total adiponectin, especially its non-high-molecular weight fraction, represents a promising risk factor for colorectal cancer: a meta-analysis.

AIM: The principal goal of this meta-analysis is to test the hypothesis that circulating total adiponectin or certain fractions may represent a promising biological candidate in modulating the risk of colorectal cancer.
METHODS: The processes of paper identification, paper selection and data extraction were accomplished independently by two authors. Effect-size estimates were expressed as weighted mean difference (WMD) and 95% confidence interval (95% CI). A total of 31 papers including 48 qualified studies (7,554 patients with colorectal cancer and 9,798 controls) were meta-analyzed.
RESULTS: Pooling all studies found that circulating total adiponectin was significantly lower in patients with colorectal cancer than in controls (WMD: -0.76 µg/mL, 95% CI: -1.20 to -0.32, p=0.001), with significant heterogeneity (I2: 94.2%) and low publication bias (Egger's p=0.336). By adiponectin fractions, the difference in high-molecular weight (HMW) adiponectin was comparable between the two groups (WMD: -0.22 µg/mL, 95% CI: -0.70 to 0.25, p=0.350), while non-HMW adiponectin was significantly lower in patients with colorectal cancer than in controls (WMD: -0.27 µg/mL, 95% CI: -0.35 to -0.19, p<0.001), with marginal heterogeneity (I2: 52.3%). Subgroup analysis revealed that effect-size estimates were heterogeneous when grouping studies by cancer subtype, region, study design, matching status, gender and obesity. Further meta-regression analysis indicated that age and gender were significant potential sources of heterogeneity. The results showed the studied subgroups were not subject to publication bias (Egger's p<0.1).
CONCLUSION: Our data collectively indicate that low circulating total adiponectin, especially its non-HMW fraction, represents a promising risk factor for colorectal cancer. Further studies are needed to explore underlying mechanisms.

Introduction

Adiponectin of molecular mass 28 kDa is a member of the adipocytokines, and it is secreted exclusively by mature adipocytes.1 Biologically speaking, adiponectin possesses a well-defined set of properties, including insulin sensitization, anti-inflammation, anti-atherosclerosis, proapoptosis and antiproliferation.2,3 In circulation, adiponectin exists in forms of trimer, hexamer and high-molecular weight (HMW) complex, and these fractions are proven to have different biological activities.4 For example, HMW adiponectin has a close relationship to insulin sensitivity, while the relevance between non-HMW adiponectin and anti-inflammation is more obvious.5 Mounting evidence from human studies reveal that low total adiponectin in circulation represents a predisposing status for development of colorectal cancer.6–8 By contrast, other researchers failed to confirm this claim,9–11 and some of them even found a higher concentration of circulating total adiponectin in patients with colorectal cancer than in controls.12,13 Several systematic reviews and meta-analyses have summarized published data involving the association between circulating total adiponectin and colorectal cancer risk,14–20 while interpretation of pooled findings in most cases is clouded by the presence of substantial heterogeneity and none considered adiponectin fractions as separate factors. With growing epidemiological data, our understanding on circulating adiponectin and its fractions and colorectal cancer needs replenishment and renewal. Hence, the principal goal of the present study is to test the hypothesis that circulating total adiponectin or certain fractions may represent a promising biological candidate in modulating the risk of colorectal cancer, assessed through a comprehensive meta-analysis. A secondary goal is to explore possible sources of heterogeneity of effect size across studies through both subgroup analysis and meta-regression analysis.

Methods

Guideline

The present study was carried out in strict accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guideline.21

Paper identification

Potentially eligible papers were identified through searching public databases of MEDLINE (PubMed), EMBASE, Web of Knowledge and Scholar (Google). All identified papers from each database were merged together and then duplicates removed. The key items used for paper identification included (“colorectal” or “colon” or “rectal”) in title and (“cancer” or “carcinoma” or “tumor” or “tumour” or “neoplasm” or “adenoma”) in title and (“adiponectin” or “adipocytokine” or “adipokine”) in title/abstract. The process of paper identification was completed independently and in duplicate (Weiqun Lu and Zhiliang Huang) using the same key items mentioned above.

Eligibility criteria

Eligibility criteria consisted of both inclusion criteria and exclusion criteria. Papers were included only when the following conditions were met simultaneously: (i) publication using the English language; (ii) either cross-sectional or nested case–control study design; (iii) involvement of only human beings; and (iv) availability of circulating plasma or serum adiponectin concentration for patients with colorectal or colon or rectal cancer and cancer-free controls. Papers were excluded when any of the following conditions was true: (i) non-original contribution; (ii) narrative or systematic review or meta-analysis; (iii) conference abstract; and (iv) case report or case series.

Paper selection

Two authors (Weiqun Lu and Zhiliang Huang) selected the papers that met the above inclusion criteria through reviewing the title or abstract or the full text, if necessary, of each identified article. When disagreement happened during the selection process, it was resolved by discussion or seeking help from the third author (Haiying Liu).

Data extraction

From each qualified paper, data for analysis including, if available, the first author’s name, publication year, country where the study was conducted, study design, cancer subtype, matching status, source of controls, selection of controls, sample size, assay method for circulating adiponectin, sample type, age, gender, body mass index (BMI), waist–hip ratio (WHR), waist circumference, physical activity, physical inactivity, smoking habit, family history, diabetes mellitus, glucose, insulin, C-reactive protein (CRP) and adiponectin were extracted and entered into databases by two authors (Weiqun Lu and Zhiliang Huang). Extracted data were compared, and disagreement was resolved through consensus and discussion.

Statistical analysis

Data were statistically analyzed by the Stata/SE software version 14.0 (StataCorp LP, College Station, TX, USA). Weighted mean difference (WMD) was adopted to assess the difference in circulating adiponectin concentration between patients and controls in both overall and subgroup analyses, and its 95% confidence interval (95% CI) was estimated accordingly under the random-effects model based on the DerSimonian–Laird method. Heterogeneity was first evaluated with the I2 statistic, and then was explored through both subgroup and meta-regression analyses. The I2 statistic is expressed as a percentage figure, and the higher the figure, the more likely the heterogeneity becomes. It is widely accepted the significance cutoff point of I2 statistic is set at 50%. Publication bias was assessed by the Begg’s funnel plot, filled funnel plot and Egger’s regression asymmetry test. The Egger’s regression asymmetry test records significance if the probability value is <10%.

Results

Qualified studies

A total of 339 papers were identified through searching public databases, and 295 of them were excluded after reading the title and abstract. Further, after full-text reviewing, 13 papers were excluded as they failed to meet our predefined inclusion criteria, leaving 31 eligible papers for meta-analysis in this study.5–13,22–43 Thereof, 16 papers provided data by cancer type, gender or race, and so there were 48 qualified studies.

Study characteristics

Table 1 shows the baseline characteristics of all qualified studies. After removing shared controls, this meta-analysis involved 7,554 patients with colorectal cancer and 9,798 controls. Of 48 qualified studies, only one study focused on HMW adiponectin.39 In addition, eight studies also measured HMW adiponectin, and non-HMW adiponectin was recorded in six studies.

Table 1

The baseline characteristics of all qualified studies

Study	Year	Region	Cancer subtype	Control features	Study design	Matching status	ADI assay (ELISA)	Sample size		Sample source	Adiponectin (µg/mL)		Age (years)		Male gender		BMI (kg/m2)
								Patients	Controls		Patients	Controls	Patients	Controls	Patients	Controls	Patients	Controls
Saetang et al25 (CRC)	2016	East Asia	CRC	W/o colonic polyps	Retrospective	NA	Thermo Fisher	82	30	Serum	17.56	19.49	62.4	59.1	NA	NA	NA	NA
Saetang et al25 (colon)	2016	East Asia	Colon	W/o colonic polyps	Retrospective	NA	Thermo Fisher	52	30	Serum	15.65	19.49	62.4	59.1	NA	NA	NA	NA
Saetang et al25 (rectum)	2016	East Asia	Rectum	W/o colonic polyps	Retrospective	NA	Thermo Fisher	30	30	Serum	20.49	19.49	62.4	59.1	NA	NA	NA	NA
Inamura et al29 (HPFS)	2016	USA	CRC	W/o CRC	Nested	Yes	ALPCO	155	296	Plasma	5.00	5.60	66.2	66.2	1.000	1.000	26.1	25.3
Inamura et al29 (NHS)	2016	USA	CRC	W/o CRC	Nested	Yes	ALPCO	152	297	Plasma	7.90	8.20	58.5	66.2	0.000	0.000	25.2	24.7
Zekri et al13	2015	Middle East	CRC	W/o CRC	Retrospective	NA	RayBiotech	34	29	Serum	11.28	6.59	41.3	66.2	0.588	0.552	NA	NA
Demir et al37	2015	Middle East	CRA	W/o CRC	Retrospective	NA	eBioscience	32	30	Serum	8.57	9.70	53.5	66.2	0.625	0.500	29.1	26.1
Chandler et al11	2015	USA	CRC	W/o CRC	Nested	Yes	ALPCO	275	275	Plasma	6.00	6.24	58.8	66.2	0.000	0.000	26.8	26.3
Ochs-Balcom et al26 (white)	2014	USA	CRA	W/o CRA	Nested	NA	R&D	217	650	Plasma	10.58	11.64	57.4	66.2	0.521	0.369	29.1	32.3
Ochs-Balcom et al26 (AA)	2014	USA	CRA	W/o CRA	Nested	NA	R&D	175	378	Plasma	7.76	8.02	58.2	66.2	0.406	0.241	31.7	32.3
Aleksandrova et al5 (colon)	2014	Europe	Colon	W/o CRC	Nested	Yes	ALPCO	794	794	Serum	6.70	6.80	58.6	66.2	0.530	0.530	26.8	26.3
Aleksandrova et al5 (rectum)	2014	Europe	Rectum	W/o CRC	Nested	Yes	ALPCO	466	466	Serum	6.50	6.80	58.0	66.2	0.457	0.457	26.6	26.4
Aleksandrova et al40 (men)	2014	Europe	Colon	W/o CRC	Nested	Yes	ALPCO	292	292	Serum	5.40	5.30	59.3	66.2	1.000	1.000	27.3	26.5
Aleksandrova et al40 (women)	2014	Europe	Colon	W/o CRC	Nested	Yes	ALPCO	370	370	Serum	8.30	8.40	58.6	66.2	0.000	0.000	26.4	26.1
Abe Vicente et al43	2014	USA	CRC	Healthy	Retrospective	NA	R&D	39	40	Serum	4.60	3.60	61.0	66.2	0.590	0.650	23.9	23.9
Song et al24 (men)	2013	USA	CRC	W/o CRC	Nested	Yes	Linco	270	519	Plasma	4.99	5.32	65.8	66.2	1.000	1.000	26.2	25.4
Song et al24 (women)	2013	USA	CRC	W/o CRC	Nested	Yes	Linco	346	686	Plasma	8.04	8.19	59.0	66.2	0.000	0.000	26.0	25.5
Danese et al38 (men)	2013	Europe	CRA	W/o CRC	Retrospective	Yes	Mediagnost	21	26	Plasma	7.86	5.83	63.0	66.2	0.525	0.650	26.0	26.0
Danese et al38 (women)	2013	Europe	CRA	W/o CRC	Retrospective	Yes	Mediagnost	19	14	Plasma	11.81	9.89	NA	NA	NA	NA	NA	NA
Kemik et al8 (colon)	2012	Middle East	Colon	Healthy	Retrospective	NA	Linco	32	40	Serum	4.80	6.50	49.5	40.4	0.563	0.550	15.9	21.5
Kemik et al8 (rectum)	2012	Middle East	Rectum	Healthy	Retrospective	NA	Linco	35	40	Serum	4.70	6.50	44.8	40.4	0.543	0.550	16.2	21.5
Ho et al30	2012	USA	CRC	W/o CRC	Nested	NA	Millipore	457	834	Plasma	27.20	29.10	NA	NA	NA	NA	NA	NA
Hillenbrand et al31 (men)	2012	Europe	CRC	Healthy	Retrospective	NA	Millipore	42	30	Serum	14.20	11.80	65.5	46.0	1.000	1.000	27.2	NA
Hillenbrand et al31 (women)	2012	Europe	CRC	Healthy	Retrospective	NA	Millipore	25	30	Serum	21.90	22.50	66.0	44.0	0.000	0.000	26.4	NA
Gulcelik et al32	2012	Middle East	Colon	W/o cancer	Retrospective	NA	B-Bridge	27	40	Serum	9.51	13.91	52.1	52.4	0.593	0.250	26.8	26.8
Chen et al39 (E)	2012	East Asia	CRC	W/o colonic polyps	Retrospective	NA	Adlitteram	71	102	Plasma	10.10	13.20	62.1	58.3	1.000	1.000	23.5	22.9
Chen et al39 (A)	2012	East Asia	CRC	W/o colonic polyps	Retrospective	NA	Adlitteram	94	102	Plasma	9.30	13.20	61.8	58.3	1.000	1.000	22.4	22.9
Aleksandrova et al41 (colon)	2012	Europe	Colon	W/o CRC	Nested	Yes	ALPCO	755	755	Serum	6.71	6.84	58.6	58.6	0.542	0.542	26.9	26.3
Aleksandrova et al41 (rectum)	2012	Europe	Rectum	W/o CRC	Nested	Yes	ALPCO	451	451	Serum	6.38	6.79	58.0	58.0	0.452	0.452	26.5	26.4
Al-Harithy and Al-Zahrani42 (men)	2012	Middle East	CRC	Healthy	Retrospective	Yes	ALPCO	31	30	Serum	5.59	6.38	55.0	49.0	1.000	1.000	24.8	29.1
Al-Harithy and Al-Zahrani42 (women)	2012	Middle East	CRC	Healthy	Retrospective	Yes	ALPCO	29	30	Serum	4.47	8.89	53.5	52.0	0.000	0.000	26.8	28.8
Gialamas et al34	2011	Europe	CRC	Healthy	Retrospective	Yes	RI	104	208	Serum	9.45	10.38	69.8	69.1	0.625	0.625	NA	NA
Chronis et al10 (NOAA)	2011	Europe	CRA	W/o CRA	Retrospective	NA	Linco	22	138	Serum	10.90	10.20	65.1	63.5	0.545	0.587	26.4	27.4
Chronis et al10 (AA)	2011	Europe	CRA	W/o CRA	Retrospective	NA	Linco	46	138	Serum	11.20	10.20	66.2	63.5	0.565	0.587	27.1	27.4
Catalan et al7	2011	Europe	Colon	Healthy	Retrospective	NA	R&D	11	18	Plasma	4.07	8.09	66.0	44.0	NA	NA	26.9	29.3
Al Khaldi et al12	2011	Middle East	Colon	Healthy	Retrospective	Yes	Linco	58	68	Plasma	8.60	4.10	53.0	60.0	0.500	0.500	27.9	26.0
Yamaji et al22 (men)	2010	East Asia	CRA	W/o CRA	Nested	NA	NA	523	480	Plasma	3.98	4.37	NA	NA	1.000	1.000	NA	NA
Yamaji et al22 (women)	2010	East Asia	CRA	W/o CRA	Nested	NA	NA	255	255	Plasma	6.81	7.36	NA	NA	0.000	0.000	NA	NA
Nakajima et al27	2010	East Asia	CRA	W/o CRC	Retrospective	Yes	Otsuka	115	115	Serum	8.90	8.90	63.7	63.5	0.600	0.600	22.9	23.1
Kemik et al28	2010	Middle East	Colon	Healthy	Retrospective	NA	Linco	126	36	Serum	4.30	6.50	43.5	40.4	0.421	0.444	NA	NA
Gonullu et al33	2010	Middle East	CRC	Healthy	Retrospective	NA	Bio-Source	36	37	Serum	5.50	6.20	56.6	51.0	0.500	0.541	27.2	27.0
Erarslan et al36 (CRA)	2009	Middle East	CRA	W/o colonic polyps	Retrospective	NA	RayBiotech	31	50	Serum	7.40	9.20	63.0	59.0	0.548	0.540	26.0	29.2
Erarslan et al36 (CRC)	2009	Middle East	CRC	W/o colonic polyps	Retrospective	NA	RayBiotech	23	50	Serum	7.10	9.20	57.0	59.0	0.609	0.540	24.6	29.2
Stocks et al9 (men)	2008	Europe	CRC	W/o cancer	Nested	Yes	R&D	125	245	Plasma	7.00	6.60	59.8	NA	1.000	1.000	NA	NA
Stocks et al9 (women)	2008	Europe	CRC	W/o cancer	Nested	Yes	R&D	181	350	Plasma	11.40	11.40	59.7	NA	0.000	0.000	NA	NA
Ferroni et al35	2007	Europe	CRC	W/o CRC	Retrospective	NA	BioVendor	60	35	Serum	8.30	13.10	64.0	63.0	0.517	0.514	NA	NA
Wei et al23	2005	USA	CRC	W/o CRC	Nested	Yes	Linco	179	177	Plasma	7.40	7.80	66.6	66.5	1.000	1.000	25.9	25.4
Otake et al6	2005	East Asia	CRA	W/o colonic polyps	Retrospective	NA	Otsuka	51	52	Plasma	7.00	10.60	59.0	58.0	0.686	0.654	23.6	22.8

Abbreviations: ADI, adiponectin; ELISA, enzyme-linked immunosorbent assay; CRC, colorectal cancer; HPFS, Health Professionals Follow-up Study; NHS, Nurses’ Health Study; CRA, colorectal adenoma; AA, advanced adenoma; NOAA, without advanced adenoma; W/o, without; BMI, body mass index; WHR, waist–hip ratio; WC, waist circumference; DM, diabetes mellitus; CRP, C-reactive protein; NA, not available.

Overall analysis

Overall analysis was based on 41 qualified studies, and the funnel plot illustrating the difference in total adiponectin concentrations between patients with colorectal cancer and controls is presented in Figure 1. Pooling effect size of all studies found that total adiponectin was significantly lower in patients with colorectal cancer than in controls (WMD: −0.76 µg/mL, 95% CI: −1.20 to −0.32, p=0.001). However, this significant finding was clouded by strong evidence of statistical heterogeneity (I2: 94.2%, p<0.001).

Figure 1

Forest plot for overall difference in total adiponectin between patients with colorectal cancer and controls.

Note: Weights are from random-effects analysis.

Abbreviations: WMD, weighted mean difference; CI, confidence interval.

Figure 2 shows another funnel plot quantifying the differences in adiponectin fractions – HMW adiponectin and non-HMW adiponectin – between patients with colorectal cancer and controls. The difference in HMW adiponectin was comparable between the two groups (WMD: −0.22 µg/mL, 95% CI: −0.70 to 0.25, p=0.350), with significant heterogeneity (I2: 97.3%, p<0.001). By contrast, non-HMW adiponectin was significantly lower in patients with colorectal cancer than in controls (WMD: −0.27 µg/mL, 95% CI: −0.35 to −0.19, p<0.001), with only marginal heterogeneity (I2: 52.3%, p=0.063).

Figure 2

Forest plot for differences in two adiponectin fractions (HMW and non-HMW) between patients with colorectal cancer and controls.

Note: Weights are from random-effects analysis.

Abbreviations: HMW, high molecular weight; WMD, weighted mean difference; CI, confidence interval.

Subgroup analysis

Table 2 summarizes the differences in total adiponectin between patients and controls by grouping studies according to cancer subtype, region, study design, matching status, sample source, gender, BMI, WHR and diabetes mellitus, respectively.

Table 2

Subgroup analysis of the difference in circulating adiponectin between patients with colorectal cancer and controls

Characteristics	Subgroups	Studies (number)	WMD	95% CI	pD	I2	pH	pE
Cancer subtype	CRC	20	−0.736	−1.519 to 0.047	0.065	94.0%	<0.001	0.514
	CRA	10	−0.892	−1.573 to −0.211	0.010	80.1%	<0.001	0.473
	Colon	10	−1.073	−2.095 to −0.050	0.040	97.3%	<0.001	0.403
	Rectum	4	−0.681	−1.358 to −0.005	0.048	75.4%	0.007	0.778
Region	East Asia	6	−1.707	−2.979 to −0.436	0.008	95.1%	<0.001	0.213
	America	10	−0.343	−0.599 to −0.087	0.009	36.7%	0.115	0.683
	Europe	15	−0.749	−1.451 to −0.047	0.037	94.3%	<0.001	0.541
	Middle East	10	−0.787	−3.015 to 1.442	0.489	97.1%	<0.001	0.787
Study design	Retrospective	22	−1.179	−2.536 to 0.177	0.088	96.2%	<0.001	0.489
	Nested	19	−0.266	−0.380 to −0.152	<0.001	12.4%	0.303	0.409
Matching status	NA	21	−1.472	−2.452 to −0.491	0.003	96.1%	<0.001	0.411
	Yes	20	−0.099	−0.389 to 0.191	0.504	79.0%	<0.001	0.693
Sample source	Plasma	18	−0.714	−1.358 to −0.070	0.030	95.1%	<0.001	0.783
	Serum	23	−0.803	−1.430 to −0.176	0.012	93.3%	<0.001	0.303
Sample size (median)	≤230	21	−1.189	−2.609 to 0.231	0.101	96.4%	<0.001	0.511
	>230	20	−0.271	−0.385 to −0.157	<0.001	12.1%	0.304	0.262
Gender	Men	9	−0.671	−1.255 to −0.086	0.025	90.4%	<0.001	0.536
	Women	8	−0.329	−0.693 to 0.035	0.077	49.6%	<0.001	0.206
BMI	<25 kg/m2	7	−1.627	−3.092 to −0.163	0.029	92.1%	<0.001	0.434
	≥25 kg/m2	24	−0.527	−1.016 to −0.039	0.034	92.4%	<0.001	0.611
WHR	<0.9 m	5	−0.878	−1.798 to 0.043	0.062	94.7%	<0.001	0.112
	≥0.9 m	9	−0.522	−0.939 to −0.106	0.014	60.6%	0.009	0.496
DM	Without	12	−1.385	−3.003 to 0.233	0.093	96.1%	<0.001	0.349
TNM stage	I–II	9	−1.280	−2.115 to −0.446	0.003	90.1%	<0.001	0.103
	III–IV	9	−1.196	−2.033 to −0.358	0.005	92.6%	<0.001	0.219

Notes: pD, the p-value for WMD; I2, inconsistency index; pH, the p-value for heterogeneity; pE, the p-value for Egger’s regression asymmetry test.

Abbreviations: WMD, weighted mean difference; CI, confidence interval; CRC, colorectal cancer; CRA, colorectal adenoma; BMI, body mass index; WHR, waist–hip ratio; DM, diabetes mellitus; TNM, tumor node metastasis; NA, not available.

By cancer subtype, significant difference in total adiponectin was observed in patients with colorectal adenoma (CRA), colon cancer and rectum cancer, respectively, relative to controls (WMD: −0.892, −1.073 and −0.681 µg/mL, p=0.010), with significant heterogeneity (I2: 80.1%, 97.3% and 75.4%).

By geographic region, besides Middle East, total adiponectin was significantly lower in patients than in controls from America, Europe and especially East Asia (WMD: −1.707 µg/mL, p=0.008), and only in America, there was no evidence of heterogeneity (I2: 36.7%).

By study design, significant difference was observed only in nested case–control studies (WMD: −0.266 µg/mL, p<0.001), without heterogeneity (I2: 12.4%).

By matching status, studies with matched patients and controls failed to identify any significance (WMD: −0.099 µg/mL, p=0.504) (I2: 79.0%), while in studies with unclear matching status, there was significant difference in total adiponectin between patients and controls (I2: 96.1%).

By sample source, total adiponectin was significantly and comparably reduced in studies collecting plasma and serum samples for adiponectin measurement (WMD: −0.714 and −0.803 µg/mL, p=0.030 and 0.012, I2: 95.1% and 93.3%, respectively).

By sample size, grouping studies by median total sample size showed that the reduction of total adiponectin was obvious in studies with total sample size >230 (WMD: −0.271 µg/mL, p<0.001), and the likelihood of heterogeneity was low (I2: 12.1%).

By gender, total adiponectin was reduced significantly in male patients relative to male controls (WMD: −0.671 µg/mL, p=0.025) (I2: 90.4%), and there was no observable significance in females.

By obesity, studies were grouped by mean or median BMI at 25 kg/m2, and significant reduction in total adiponectin was noted in both groups, especially in studies with BMI <25 kg/m2 (WMD: −1.627 µg/mL, p=0.029) (I2: 92.1%). In addition, when studies were grouped by mean or median WHR at 0.9 m, there was significant reduction in studies with WHR ≥0.9 m (WMD: −0.522 µg/mL, p=0.014).

After restricting to studies involving subjects free of diabetes mellitus, the reduction in total adiponectin was not statistically significant, and there was significant heterogeneity.

By tumor node metastasis stage, the reduction in magnitude of total adiponectin was slightly stronger in patients with stage I–II colorectal cancer (WMD: −1.280 µg/mL, p=0.003) than in patients with stage III–IV cancer (WMD: −1.196 µg/mL, p=0.005), and significance was detected in both subgroups with significant heterogeneity but a low probability of publication bias.

Meta-regression analysis

Table 3 lists the results of meta-regression analysis by incorporating all characteristics to assess whether they can explain the heterogeneity between total adiponectin and colorectal cancer. Age and smoking in both patients and controls were significant potential sources of heterogeneity, and BMI, waist circumference and CRP can account for heterogeneity only in patients (p<0.05).

Table 3

The meta-regression analysis of all characteristics in both patients with colorectal cancer and controls

Characteristics	Studies (number)	Patients		Controls
		Coefficient	p-value	Coefficient	p-value
Age	36	−0.21	0.012	0.18	0.019
Gender	38	−8.45	0.076	8.37	0.073
BMI	29	0.41	0.039	−0.35	0.091
WHR	14	5.04	0.387	−10.47	0.282
WC	19	0.14	0.008	−0.09	0.177
DM	15	−3.71	0.913	8.86	0.840
Physical activity	6	−0.05	0.785	0.04	0.842
Physical inactivity	9	1.46	0.836	−7.74	0.315
Smoking	20	−10.59	0.032	10.59	0.043
Family history	9	−0.66	0.842	0.34	0.909
Glucose	10	0.93	0.640	2.37	0.609
Insulin	10	0.28	0.316	0.03	0.930
CRP	12	−0.37	0.015	0.20	0.698

Abbreviations: BMI, body mass index; WHR, waist–hip ratio; WC, waist circumference; DM, diabetes mellitus; CRP, C-reactive protein.

Publication bias

Overall comparison of total adiponectin between patients with colorectal cancer and controls was not subject to publication bias, which was illustrated by both Begg’s and filled funnel plots in Figure 3, as well as by the Egger’s regression asymmetry test (p=0.336). For comparisons in subgroup analysis, there was consistently low likelihood of publication bias across all subgroups (Egger’s test, p<0.1), as presented in Table 2.

Figure 3

Begg’s and filled funnel plots for difference in circulating total adiponectin between patients with colorectal cancer and controls.

Abbreviations: mdadi, mean value of total adiponectin; SE, standard error.

Discussion

The most noteworthy finding of this present study is that in line with the conclusions of most previous studies,14–20 our data indicated that low total adiponectin in circulation was a significant risk factor for colorectal cancer. Moreover, for the first time, we pooled data according to adiponectin fractions and found that only non-HMW adiponectin differed significantly between patients with colorectal cancer and controls. We further identified age, gender, smoking, obesity, study design, matching status and sample size as potential sources of heterogeneity. To our knowledge, this is thus far the largest meta-analysis dealing with the relationship of circulating adiponectin and its fractions with colorectal cancer.

The connection between obesity and colorectal cancer is well established,44,45 and a great deal of supporting evidence from clinical and epidemiological data has suggested that this connection is probably mediated by abnormal adiponectin in circulation.46 Adiponectin is the most abundant hormone secreted by adipose tissue, and its anti-inflammatory and insulin-sensitizing properties may, at least in part, expound on the etiologic link between obesity and colorectal cancer.47,48 Moreover, accumulating evidence has highlighted a contributory role of adiponectin in anti-carcinogenesis.49,50 In vitro studies have showed that adiponectin can promote endothelial apoptosis, repress the maturation and proliferation of colorectal cancer cells and control colony formation, likely through the activation of AMPK-mTOR signaling pathway.39,51,52 Based on the above evidence, it is more reasonable to presume that circulating adiponectin may be involved in the pathophysiological process of colorectal cancer.

Consistent with the findings of major published studies,14–20 we confirmed in this present study that low total adiponectin in circulation was a significant risk factor for colorectal cancer. As with most previous meta-analyses, heterogeneity is a serious issue that limits interpretation of effect estimates and requires careful exploration.53 As a secondary goal of this study, we employed both subgroup analysis and meta-regression analysis to seek possible causes of heterogeneity between studies. In particular, the reduction in total adiponectin was more obvious for CRA and colon cancer, and in studies enrolling subjects of male gender, from East Asia and with normal weight. In addition, the association of total adiponectin with colorectal cancer was markedly significant in studies with a nested case–control design and a large sample size, relative to studies with a retrospective case–control design and a small sample size, indicating the robustness of our observation. Colorectal cancer is a highly heterogeneous disease to which environmental exposure, germ-line susceptibility determinants and accumulated genetic and epigenetic changes contribute interactively.54 For example, cigarette smoking and alcohol drinking, which are more prevalent in men than in women, are established risk factors for the development and progression of colorectal gender-specific difference in adiponectin observed in the present study. Nevertheless, even though we have made great endeavors to seek causes of heterogeneity, there is still strong evidence of heterogeneity in some subgroups, indicating that residual confounding from other sources of heterogeneity in colorectal cancer risk is likely. We agree that further explorations are necessary, and especially analysis of individual participant data could yield further insights.

However, a growing body of research has reported diverse biological activities of different adiponectin fractions, mainly focusing on HMW (insulin sensitivity) and non-HMW (inflammation response) forms.5,24 In this context, the sufficient number of eligible studies in this meta-analysis enables us to investigate different adiponectin fractions, and our findings revealed that low non-HMW adiponectin in circulation was a significant risk factor for colorectal cancer and there was no observable significant difference for HMW adiponectin. As previously discussed, non-HMW adiponectin plays a major role in the inflammatory process.24 Systemic inflammation is a key manifestation of cancer progression and metastasis in many types of cancer including colorectal cancer.57 We thus develop a further presumption that non-HMW adiponectin may be involved in colorectal carcinogenesis by regulating inflammatory responses. Addressing this presumption is beyond the scope of this meta-analysis, and further experimental studies are required.

Finally, the present study needs to be interpreted cautiously, bearing in mind the following limitations. First, we only retrieved the literature for the papers published in English language, and selection bias cannot be ruled out. Second, all involved studies in this meta-analysis were observational in nature, either cross-sectional or nested, which hindered further causality inference. Third, as discussed above, sources of heterogeneity for a majority of comparisons were not fully accounted for, and additional considerations were necessary. For example, low circulating adiponectin was found to be associated with KRAS-mutant colorectal cancer risk but not with KRAS-wild-type cancer risk.29 Fourth, only data on circulating plasma/serum adiponectin were summa-rized, and it is expected that tissue adiponectin concentration in tumor environment may be more relevant to the evaluation of colorectal cancer risk than its circulating concentration. However, data on tissue adiponectin are rarely reported. Fifth, circulating adiponectin concentration was affected by medical treatment or drug intervention, which cannot be taken into account because of lack of data.

In conclusion, through a comprehensive meta-analysis of 7,554 patients with colorectal cancer and 9,798 controls, our data indicate that low total adiponectin, especially its non-HMW fraction, represents a promising risk factor for colorectal cancer. Although there is still residual confounding unaccounted for, we believe that this study can aid in better understanding cancer heterogeneity, highlighting the importance of anti-inflammation therapies to prevent or delay the occurrence of colorectal cancer and thereby providing new insight in its physiology. In addition, further studies are needed to explore underlying mechanisms.