Colorectal cancer (CRC) is reported to be the fourth most commonly diagnosed cancer and is the second most common cause of cancer deaths in North America( Reference Jemal, Siegel and Ward 1 ). CRC is believed to arise from colorectal adenoma (CRA) through a sequence from adenoma to adenocarcinoma as a consequence of a limited set of molecular events that largely originate with a relatively benign adenoma that progresses to cancer( Reference Jass 2 ). Accumulated data indicate that metachronous lesions occur at a rate of 20 to 30 % per year in post-polypectomy patients. The propensity to develop CRA identifies a sizeable subgroup of the population at an enhanced risk for subsequent adenoma formation and colorectal carcinoma. This indicates that identifying risk factors associated with CRN is essential for the reduction of colorectal carcinoma.
The positive association between obesity and CRA prevalence demonstrates an underlying dose–response relationship according to BMI( Reference Ben, An and Jiang 3 ). Timely screening of obese patients for CRA is thus recommended. Dyslipidaemia is an important component of metabolic syndrome and is demonstrated to contribute to colorectal tumorigenesis through insulin resistance, oxidative stress and inflammatory pathways( Reference van Duijnhoven, Bueno-De-Mesquita and Calligaro 4 ). Alteration of serum lipids (high TAG and low HDL-cholesterol (HDL-C)) has been linked to an increased risk of CRA( Reference Giovannucci 5 , Reference Kim, Lim and Kim 6 ) and several types of malignancy including CRC( Reference Jinjuvadia, Lohia and Jinjuvadia 7 ). A recent large case–control study indicated that a high level of serum TAG was significantly associated with a larger number of adenomas( Reference Otani, Iwasaki and Ikeda 8 ). Although several studies have explored the dyslipidaemia component, e.g. lipid and lipoprotein concentrations individually in relation to CRN risk, there is still inconsistency for this issue. More importantly, the association between components of serum lipids and CRN risk is largely unknown.
An increased understanding between the development of CRN and dyslipidaemia can clarify the mechanistic steps linking components of serum lipids and CRN and may be useful in determining the benefits of early CRN screening. Unfortunately, data on the relationship of serum lipid levels with CRA/CRC are contradictory. Some studies have established a positive association between serum total cholesterol (TC) level and CRN. If this theory were correct, it may contribute to an excess in mortality in individuals with dyslipidaemia compared with those without the disorder. However, other studies showed no significant association between CRN and serum lipids( Reference Park, Joo and Kim 9 , Reference Yang, Rampal and Sung 10 ). Because the association between dyslipidaemia and CRN formation has not yet been systematically assessed, we conducted a systematic review and meta-analysis of all available studies evaluating this issue to investigate the association between components of serum lipids and CRN risk.
Materials and methods
The current review and meta-analysis follows the recommendations of the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) statement( Reference Moher, Liberati and Tetzlaff 11 ).
Search strategy
For the present analysis, a systematic literature search through October 2013 was performed in PubMed, EMBASE and the Science Citation Index to identify relevant studies. Studies investigating the relationship between serum lipids and CRN (including CRA and CRC) were carried out by searching for articles written in English. The search term comprised the following keywords: ‘serum lipids’, ‘triglycerides, TG’, ‘total cholesterol, TC’, ‘high-density lipoprotein-cholesterol, HDL-C, or ‘low-density lipoprotein-cholesterol, LDL-C’ combined with ‘colorectal cancer, CRC’ ‘colorectal adenoma, CRA’ or ‘colorectal neoplasm, CRN’ (see online supplementary material, Supplemental Table 1 for the search strategy in PubMed). All references of the selected articles were checked, including manual searches. Additionally, to find any additional published studies, we interrogated references of all the articles. The titles and abstracts were scanned to exclude any clearly irrelevant studies. The full texts of the remaining articles were read to determine whether they contained information on the topic of interest. All searches were conducted independently by two authors (Y.T. and Y.L.). The results were then compared; any questions or discrepancies were resolved by iteration and consensus.
Study selection
Only publications that fulfilled all of the following criteria were selected for the meta-analysis: (i) the study subjects were adults (older than 18 years); (ii) publication with a case–control, cross sectional, nested case–control or cohort study design; (iii) CRN incidence as the outcome of interest; (iv) reported an estimate of the association of CRN (defined as colorectal cancer or adenoma or both) in individuals with at least one of the selected lipid components (TC, TAG, HDL-C and LDL-cholesterol (LDL-C)); and (v) reported risk ratio (RR; or OR estimates in case–control studies) or hazard ratios (HR) with estimates of their corresponding 95 % CI (or sufficient data to evaluate the above effects).
Data extraction
Information extracted from the extensive review of each publication included: (i) publication data (first author’s name, year of publication and country of the population studied); (ii) type of study design; (iii) study participants’ age range; (iv) sample size (cases and controls or cohort size); (v) type of lesion; (vi) risk estimates with their corresponding CI; (viii) method used to confirm the presence or absence of CRN; and (ix) colonoscopy examination at the time of diagnosis. OR from case–control studies were considered as an estimate of RR( Reference Hogue, Gaylor and Schulz 12 ). If a study provided several risk estimates, the most completely adjusted estimate was extracted and used in the meta-analysis. The information from each study was extracted by two independent researchers (Y.T. and Y.L.), with disagreements resolved with a majority vote by all authors.
Statistical analysis
All analyses, including publication bias, were performed using the computer program Review Manage version 5·1 (Oxford, UK). Study-specific risk estimates were extracted from each study and log risk estimates were weighted by the inverse of their variances to obtain a pooled risk estimate. The heterogeneity of all publications was evaluated with the Cochran Q test and I 2 statistic( Reference Higgins and Thompson 13 ). An I 2 value of <30 %, 30–50 % and >50 % was considered as little or no heterogeneity, moderate heterogeneity and severe heterogeneity, respectively. For the Q statistic, a P value of <0·1 was considered to have significant heterogeneity. Summaries of RR estimates were evaluated using both fixed-effects and random-effects methods. Random effects are used when heterogeneity is present. Initial analysis, including all studies, was performed to look for an association between serum lipids and CRN. Subgroup analyses were also carried out to estimate the components of serum lipids and the risk of CRC and CRA. For meta-analysis results, the P value of <0·05 indicated statistical significance. A funnel plot for potential publication bias analysis was conducted using the statistical software package Stata 11·0 with Begg( Reference Begg and Mazumdar 14 ) and Egger tests( Reference Egger, Davey Smith and Schneider 15 ).
Results
Study characteristics
Two hundred and eighty-five publications relevant to the words searched were retrieved using the methodology and the search terms described above. Of these, eighty-two duplicated publications were excluded. The remaining 203 studies were selected for further evaluation based on information from abstracts and titles. After screening abstracts and titles, forty-seven studies were considered to be relevant to our study subject. Fourteen articles were excluded as they did not investigate the association of serum lipids with risk of CRN. In total, thirty-three records that met the detailed inclusion criteria were included in the present meta-analysis( Reference van Duijnhoven, Bueno-De-Mesquita and Calligaro 4 , Reference Kim, Lim and Kim 6 , Reference Otani, Iwasaki and Ikeda 8 – Reference Yang, Rampal and Sung 10 , Reference Ahmed, Schmitz and Anderson 16 – Reference Yamada, Araki and Tamura 43 ). All studies reported on at least one of the serum lipid components (TC, TAG, HDL-C and LDL-C) and the risk of CRN. Details of these studies are described in Tables 1 and 2.
Studies contributing to the analysis of serum lipids and colorectal cancer
RR, risk ratio; HDL-C, HDL-cholesterol; TC, total cholesterol; LDL-C, LDL-cholesterol.
Studies contributing to the analysis of serum lipids and colorectal adenoma
RR, risk ratio; TC, total cholesterol; HDL-C, HDL-cholesterol; LDL-C, LDL-cholesterol.
Overall analyses on the association of serum lipids and colorectal neoplasm
The association between serum TC and risk of CRN was analysed first. The sixteen studies on TC (four cohort, five cross-sectional and seven case–control studies) were published between 1990 and 2013 (Tables 1 and 2) and involved a total of 21809 CRN cases( Reference van Duijnhoven, Bueno-De-Mesquita and Calligaro 4 , Reference Park, Joo and Kim 9 , Reference Yang, Rampal and Sung 10 , Reference Bayerdorffer, Mannes and Richter 19 – Reference Chung, Han and Park 22 , Reference Iso, Ikeda and Inoue 25 , Reference Kono, Ikeda and Yanai 28 , Reference Kono, Imanishi and Shinchi 29 , Reference Stocks, Lukanova and Bjorge 34 , Reference Tabuchi, Kitayama and Nagawa 36 , Reference Trichopoulou, Tzonou and Hsieh 38 , Reference Tsilidis, Brancati and Pollak 39 , Reference Wulaningsih, Garmo and Holmberg 42 , Reference Yamada, Araki and Tamura 43 ). Three studies were conducted in the USA( Reference Bird, Ingles and Frankl 20 , Reference Trichopoulou, Tzonou and Hsieh 38 , Reference Tsilidis, Brancati and Pollak 39 ), one in Sweden( Reference Wulaningsih, Garmo and Holmberg 42 ), three in Korea( Reference Park, Joo and Kim 9 , Reference Yang, Rampal and Sung 10 , Reference Chung, Han and Park 22 ), five in Japan( Reference Iso, Ikeda and Inoue 25 , Reference Kono, Ikeda and Yanai 28 , Reference Kono, Imanishi and Shinchi 29 , Reference Tabuchi, Kitayama and Nagawa 36 , Reference Yamada, Araki and Tamura 43 ), one in Finland( Reference Bowers, Albanes and Limburg 21 ), one in Germany( Reference Bayerdorffer, Mannes and Richter 19 ) and two in European countries( Reference van Duijnhoven, Bueno-De-Mesquita and Calligaro 4 , Reference Inoue, Noda and Kurahashi 24 ) (Tables 1 and 2). A random-effects model was considered for a summary RR as there was evidence of heterogeneity for the included sixteen studies (Q=45·28, P value for heterogeneity =0·00001, I 2=65 %). The overall RR for CRN (adenoma or colon cancer) associated with serum TC was 1·00 (95 % CI 0·93, 1·08, P=0·93; Fig. 1(a)), while the funnel plot for potential publication bias was also analysed (Fig. 1(b)). Our result indicated that there was no statistical evidence of publication bias (Egger’s P=0·638, Begg’s P=0·837).
Association between TC and CRN (adenoma and cancer combined): (a) forest plot; (b) funnel plot. In (a), the study-specific RR and 95 % CI are represented by the black square and horizontal line, respectively; the area of the black square is proportional to the specific-study weight to the overall meta-analysis. The centre of the diamond presents the pooled RR risk and its width represents the pooled 95 % CI. In (b), Begg’s funnel plot with pseudo 95 % CI is presented (TC, total cholesterol; CRN, colorectal neoplasm; RR, risk ratio; IV, fixed-effects model; HR, hazard ratio)
There were twenty-nine studies reported on TAG (twelve cohort studies, four cross-sectional and thirteen case–control studies) published between 1990 and 2013 and involved a total of 31546 CRN cases (Tables 1 and 2)( Reference van Duijnhoven, Bueno-De-Mesquita and Calligaro 4 , Reference Kim, Lim and Kim 6 , Reference Otani, Iwasaki and Ikeda 8 – Reference Yang, Rampal and Sung 10 , Reference Ahmed, Schmitz and Anderson 16 – Reference Ashbeck, Jacobs and Martinez 18 , Reference Bird, Ingles and Frankl 20 , Reference Chung, Han and Park 22 – Reference Inoue, Noda and Kurahashi 24 , Reference Kaneko, Sato and An 26 – Reference Saydah, Platz and Rifai 32 ). Nine studies were conducted in the USA( Reference Ahmed, Schmitz and Anderson 16 , Reference Ashbeck, Jacobs and Martinez 18 , Reference Bird, Ingles and Frankl 20 , Reference Saydah, Platz and Rifai 32 , Reference Sturmer, Buring and Lee 35 , Reference Trevisan, Liu and Muti 37 – Reference Tsushima, Nomura and Lee 40 ), one in Sweden( Reference Wulaningsih, Garmo and Holmberg 42 ), five in Korea( Reference Kim, Lim and Kim 6 , Reference Park, Joo and Kim 9 , Reference Yang, Rampal and Sung 10 , Reference Chung, Han and Park 22 , Reference Kim, Shin and Hong 27 ), two in China( Reference Hu, Chen and Lin 23 , Reference Liu, Hsu and Li 30 ), eight in Japan( Reference Otani, Iwasaki and Ikeda 8 , Reference Inoue, Noda and Kurahashi 24 , Reference Kaneko, Sato and An 26 , Reference Kono, Ikeda and Yanai 28 , Reference Kono, Imanishi and Shinchi 29 , Reference Morita, Tabata and Mineshita 31 , Reference Tabuchi, Kitayama and Nagawa 36 , Reference Yamada, Araki and Tamura 43 ), one in Austria( Reference Ulmer, Borena and Rapp 41 ) and three in European countries( Reference van Duijnhoven, Bueno-De-Mesquita and Calligaro 4 , Reference Aleksandrova, Boeing and Jenab 17 , Reference Stocks, Lukanova and Bjorge 34 ) (Tables 1 and 2). Pooled analysis showed a significant association between serum TAG and CRN (n 29 studies; summary RR=1·08; 95 % CI 1·05, 1·12, P<0·00001; Fig. 2(a)). A random-effects model was considered for a summary RR because of the significant heterogeneity of the included twenty-nine studies (Q=110·44, P value for heterogeneity<0·00001, I 2=68 %). The funnel plot for potential publication bias was also implemented (Fig. 2(b)); the results of the statistical analysis showed a potential publication bias (Egger’s P=0·006, Begg’s P=0·225).
Association between TAG and CRN (adenoma and cancer combined): (a) forest plot; (b) funnel plot. In (a), the study-specific RR and 95 % CI are represented by the black square and horizontal line, respectively; the area of the black square is proportional to the specific-study weight to the overall meta-analysis. The centre of the diamond presents the pooled RR risk and its width represents the pooled 95 % CI. In (b), Begg’s funnel plot with pseudo 95 % CI is presented (CRN, colorectal neoplasm; RR, risk ratio; IV, fixed-effects model; HR, hazard ratio)
Next, we performed an analysis to evaluate the association between serum LDL-C and risk of CRN. There were a total of eight studies carried out on LDL-C (three cohort studies, three cross-sectional and two case–control studies) published between 1993 and 2013 which included 12 473 CRN cases (Tables 1 and 2)( Reference van Duijnhoven, Bueno-De-Mesquita and Calligaro 4 , Reference Park, Joo and Kim 9 , Reference Yang, Rampal and Sung 10 , Reference Bayerdorffer, Mannes and Richter 19 , Reference Kaneko, Sato and An 26 , Reference Kono, Imanishi and Shinchi 29 , Reference Schoen, Tangen and Kuller 33 , Reference Wulaningsih, Garmo and Holmberg 42 ). One study was conducted in the USA( Reference Schoen, Tangen and Kuller 33 ), one in Sweden( Reference Wulaningsih, Garmo and Holmberg 42 ), one in Germany( Reference Bayerdorffer, Mannes and Richter 19 ), two in Korea( Reference Park, Joo and Kim 9 , Reference Yang, Rampal and Sung 10 ), two in Japan( Reference Kaneko, Sato and An 26 , Reference Kono, Imanishi and Shinchi 29 ) and one in European countries( Reference van Duijnhoven, Bueno-De-Mesquita and Calligaro 4 ) (Tables 1 and 2). Data provided evidence that LDL-C is an increased risk factor of CRN development (summary RR=1·07; 95 % CI 1·00, 1·14, P=0·04; Fig. 3(a)). A random-effects model was considered for summary RR due to the existence of heterogeneity of the included eight studies (Q=79·32, P value for heterogeneity <0·00001, I 2=90 %), while the funnel plot for potential publication bias was also conducted (Fig. 3(b)). Results of the statistical analysis showed no significant publication bias (Egger’s P=0·573, Begg’s P=0·835).
Association between LDL-C and CRN (adenoma and cancer combined): (a) forest plot; (b) funnel plot. In (a), the study-specific RR and 95 % CI are represented by the black square and horizontal line, respectively; the area of the black square is proportional to the specific-study weight to the overall meta-analysis. The centre of the diamond presents the pooled RR risk and its width represents the pooled 95 % CI. In (b), Begg’s funnel plot with pseudo 95 % CI is presented (LDL-C, LDL-cholesterol; CRN, colorectal neoplasm; RR, risk ratio; IV, fixed-effects model; HR, hazard ratio)
Finally, we performed analysis of studies to evaluate the association between HDL-C and CRN. The included twenty-three studies on HDL-C (ten cohort studies, four cross-sectional and nine case–control studies) were published between 1990 and 2013 (Tables 1 and 2) and involved a total of 21426 CRN cases( Reference van Duijnhoven, Bueno-De-Mesquita and Calligaro 4 , Reference Kim, Lim and Kim 6 , Reference Park, Joo and Kim 9 , Reference Yang, Rampal and Sung 10 , Reference Ahmed, Schmitz and Anderson 16 – Reference Bowers, Albanes and Limburg 21 , Reference Hu, Chen and Lin 23 , Reference Inoue, Noda and Kurahashi 24 , Reference Kaneko, Sato and An 26 – Reference Morita, Tabata and Mineshita 31 , Reference Schoen, Tangen and Kuller 33 , Reference Trevisan, Liu and Muti 37 – Reference Tsilidis, Brancati and Pollak 39 , Reference Wulaningsih, Garmo and Holmberg 42 ). Seven studies were conducted in the USA( Reference Ahmed, Schmitz and Anderson 16 , Reference Ashbeck, Jacobs and Martinez 18 , Reference Bird, Ingles and Frankl 20 , Reference Schoen, Tangen and Kuller 33 , Reference Trevisan, Liu and Muti 37 – Reference Tsilidis, Brancati and Pollak 39 ), one in Sweden( Reference Wulaningsih, Garmo and Holmberg 42 ), one in Germany( Reference Bayerdorffer, Mannes and Richter 19 ), two in China( Reference Hu, Chen and Lin 23 , Reference Liu, Hsu and Li 30 ), four in Korea( Reference Kim, Lim and Kim 6 , Reference Park, Joo and Kim 9 , Reference Yang, Rampal and Sung 10 , Reference Kim, Shin and Hong 27 ), five in Japan( Reference Inoue, Noda and Kurahashi 24 , Reference Kaneko, Sato and An 26 , Reference Kono, Ikeda and Yanai 28 , Reference Kono, Imanishi and Shinchi 29 , Reference Morita, Tabata and Mineshita 31 ), one in Finland( Reference Bowers, Albanes and Limburg 21 ) and two in European countries( Reference van Duijnhoven, Bueno-De-Mesquita and Calligaro 4 , Reference Aleksandrova, Boeing and Jenab 17 ) (Tables 1 and 2). A random-effects model was considered for a summary RR as there was significant heterogeneity of the included twenty-three studies (Q=63·10, P value for heterogeneity=0·0001, I 2=57 %). Analysis of the twenty-three studies indicated that HDL-C was not significantly associated with CRN (RR=1·03; 95 % CI 0·98, 1·07, P=0·25; Fig. 4(a)), while the funnel plot for potential publication bias was also analysed (Fig. 4(b)). Our result indicated that there was no statistical evidence of publication bias (Egger’s P=0·983, Begg’s P=0·252).
Association between HDL-C and CRN (adenoma and cancer combined): (a) forest plot; (b) funnel plot. In (a), the study-specific RR and 95 % CI are represented by the black square and horizontal line, respectively; the area of the black square is proportional to the specific-study weight to the overall meta-analysis. The centre of the diamond presents the pooled RR risk and its width represents the pooled 95 % CI. In (b), Begg’s funnel plot with pseudo 95 % CI is presented (HDL-C, HDL- cholesterol; CRN, colorectal neoplasm; RR, risk ratio; IV, fixed-effects model; HR, hazard ratio)
Overall analyses on the association of serum lipids and colorectal cancer
Eight publications in the present meta-analysis on TC (four cohort and four case–control studies) and risk of CRC were published between 1992 and 2012 and involved a total of 10 979 CRC cases (Table 1)( Reference van Duijnhoven, Bueno-De-Mesquita and Calligaro 4 , Reference Bowers, Albanes and Limburg 21 , Reference Chung, Han and Park 22 , Reference Iso, Ikeda and Inoue 25 , Reference Stocks, Lukanova and Bjorge 34 , Reference Trichopoulou, Tzonou and Hsieh 38 , Reference Wulaningsih, Garmo and Holmberg 42 , Reference Yamada, Araki and Tamura 43 ). One study was conducted in the USA( Reference Trichopoulou, Tzonou and Hsieh 38 ), one in Sweden( Reference Wulaningsih, Garmo and Holmberg 42 ), one in Korea( Reference Chung, Han and Park 22 ), two in Japan( Reference Iso, Ikeda and Inoue 25 , Reference Yamada, Araki and Tamura 43 ), one in Finland( Reference Bowers, Albanes and Limburg 21 ) and two in European countries( Reference van Duijnhoven, Bueno-De-Mesquita and Calligaro 4 , Reference Stocks, Lukanova and Bjorge 34 ) (Table 1). Pooled data did not support the association between TC and CRC (RR=0·95; 95 % CI 0·85, 1·06, P=0·38). There was a statistically significant heterogeneity among studies (Q=40·30, P value for heterogeneity <0·00001, I 2=80 %; Table 3).
Summary risk estimates of the association between serum lipids and colorectal cancer and adenoma risk
RR, risk ratio; TC, total cholesterol; HDL-C, HDL-cholesterol; LDL-C, LDL-cholesterol.
The association between serum TAG and risk of CRC was analysed. Fourteen studies on TAG (eight cohort and six case–control studies) and risk of CRC were published between 1992 and 2012 and involved a total of 13 785 CRC cases (Table 1)( Reference van Duijnhoven, Bueno-De-Mesquita and Calligaro 4 , Reference Ahmed, Schmitz and Anderson 16 , Reference Aleksandrova, Boeing and Jenab 17 , Reference Chung, Han and Park 22 , Reference Inoue, Noda and Kurahashi 24 , Reference Saydah, Platz and Rifai 32 , Reference Stocks, Lukanova and Bjorge 34 , Reference Sturmer, Buring and Lee 35 , Reference Trevisan, Liu and Muti 37 , Reference Trichopoulou, Tzonou and Hsieh 38 , Reference Tsushima, Nomura and Lee 40 – Reference Yamada, Araki and Tamura 43 ). Six studies were conducted in the USA( Reference Ahmed, Schmitz and Anderson 16 , Reference Saydah, Platz and Rifai 32 , Reference Sturmer, Buring and Lee 35 , Reference Trevisan, Liu and Muti 37 , Reference Trichopoulou, Tzonou and Hsieh 38 , Reference Tsushima, Nomura and Lee 40 ), one in Sweden( Reference Wulaningsih, Garmo and Holmberg 42 ), one in Korea( Reference Chung, Han and Park 22 ), one in Austria( Reference Ulmer, Borena and Rapp 41 ) two in Japan( Reference Inoue, Noda and Kurahashi 24 , Reference Yamada, Araki and Tamura 43 ) and three in European countries( Reference van Duijnhoven, Bueno-De-Mesquita and Calligaro 4 , Reference Aleksandrova, Boeing and Jenab 17 , Reference Stocks, Lukanova and Bjorge 34 ) (Table 1). The association of serum TAG with CRC risk was not observed (RR=1·07; 95 % CI 0·99, 1·15; P=0·10). A random-effects model was constructed as there was statistically significant heterogeneity among studies (Q=31·81, P value for heterogeneity=0·01, I 2=50 %; Table 3).
The association between serum HDL-C and risk of CRC was then analysed in our study. Nine studies on HDL-C (six cohort and three case–control studies) and risk of CRC were published between 1992 and 2012 and involved a total of 7328 CRC cases (Table 1)( Reference van Duijnhoven, Bueno-De-Mesquita and Calligaro 4 , Reference Ahmed, Schmitz and Anderson 16 , Reference Aleksandrova, Boeing and Jenab 17 , Reference Bowers, Albanes and Limburg 21 , Reference Inoue, Noda and Kurahashi 24 , Reference Schoen, Tangen and Kuller 33 , Reference Trevisan, Liu and Muti 37 , Reference Trichopoulou, Tzonou and Hsieh 38 , Reference Wulaningsih, Garmo and Holmberg 42 ). Four studies were conducted in the USA( Reference Ahmed, Schmitz and Anderson 16 , Reference Schoen, Tangen and Kuller 33 , Reference Trevisan, Liu and Muti 37 , Reference Trichopoulou, Tzonou and Hsieh 38 ), one in Sweden( Reference Wulaningsih, Garmo and Holmberg 42 ), one in Japan( Reference Inoue, Noda and Kurahashi 24 ), one in Finland( Reference Bowers, Albanes and Limburg 21 ) and two in European countries( Reference van Duijnhoven, Bueno-De-Mesquita and Calligaro 4 , Reference Aleksandrova, Boeing and Jenab 17 ) (Table 1). The pooled RR of serum HDL-C for CRC was 0·97 (95 % CI 0·80, 1·18, P=0·77), suggesting no significant relevance. Statistically significant heterogeneity existed among studies (Q=33·01, P value for heterogeneity=0·0001, I 2=73 %; Table 3).
There were three studies regarding LDL-C (two cohort and one case–control study) and risk of CRC that were published between 1990 and 2013 and involved a total of 5322 CRC cases (Table 1)( Reference van Duijnhoven, Bueno-De-Mesquita and Calligaro 4 , Reference Schoen, Tangen and Kuller 33 , Reference Wulaningsih, Garmo and Holmberg 42 ). One study was conducted in the USA( Reference Schoen, Tangen and Kuller 33 ), one in Sweden( Reference Wulaningsih, Garmo and Holmberg 42 ) and one in European countries( Reference van Duijnhoven, Bueno-De-Mesquita and Calligaro 4 ) (Table 1). The association between serum LDL-C and CRC risk was not observed (RR=0·88; 95 % CI 0·77, 1·01, P=0·07). A fixed-effects model was used as there was no statistically significant heterogeneity among studies (Q=5·34, P value for heterogeneity=0·07, I 2=63 %; Table 3).
Overall analyses on the association of serum lipids and colorectal adenoma
Nine studies reporting associations between TC (five cross-sectional and four case–control studies) and risk of CRA were published between 1990 and 2013 and involved a total of 10 935 CRA cases (Table 2)( Reference Park, Joo and Kim 9 , Reference Yang, Rampal and Sung 10 , Reference Bayerdorffer, Mannes and Richter 19 , Reference Bird, Ingles and Frankl 20 , Reference Chung, Han and Park 22 , Reference Kono, Ikeda and Yanai 28 , Reference Kono, Imanishi and Shinchi 29 , Reference Tabuchi, Kitayama and Nagawa 36 , Reference Tsilidis, Brancati and Pollak 39 ). Two studies were conducted in the USA( Reference Bird, Ingles and Frankl 20 , Reference Tsilidis, Brancati and Pollak 39 ), one in Germany( Reference Bayerdorffer, Mannes and Richter 19 ), three in Korea( Reference Park, Joo and Kim 9 , Reference Yang, Rampal and Sung 10 , Reference Chung, Han and Park 22 ) and three in Japan( Reference Kono, Ikeda and Yanai 28 , Reference Kono, Imanishi and Shinchi 29 , Reference Tabuchi, Kitayama and Nagawa 36 ) (Table 2). The results from nine studies showed that serum TC had a significant association with risk of CRA (RR=1·04; 95 % CI 1·01, 1·09, P=0·02) and there was no significant heterogeneity among these studies (Q=5·26, P value for heterogeneity=0·73, I 2=0 %; Table 3).
We next performed a meta-analysis specifically for TAG and risk of CRA. The sixteen studies on TAG (four cohort studies, four cross-sectional and eight case–control studies) and risk of CRA were published between 1990 and 2013 and involved a total of 17 830 CRA cases (Table 2)( Reference Kim, Lim and Kim 6 , Reference Otani, Iwasaki and Ikeda 8 – Reference Yang, Rampal and Sung 10 , Reference Ashbeck, Jacobs and Martinez 18 , Reference Bird, Ingles and Frankl 20 , Reference Chung, Han and Park 22 , Reference Hu, Chen and Lin 23 , Reference Kaneko, Sato and An 26 – Reference Morita, Tabata and Mineshita 31 , Reference Tabuchi, Kitayama and Nagawa 36 , Reference Tsilidis, Brancati and Pollak 39 ). Three studies were conducted in the USA( Reference Ashbeck, Jacobs and Martinez 18 , Reference Bird, Ingles and Frankl 20 , Reference Tsilidis, Brancati and Pollak 39 ), five in Korea( Reference Kim, Lim and Kim 6 , Reference Park, Joo and Kim 9 , Reference Yang, Rampal and Sung 10 , Reference Chung, Han and Park 22 , Reference Kim, Shin and Hong 27 ), two in China( Reference Hu, Chen and Lin 23 , Reference Liu, Hsu and Li 30 ) and six in Japan( Reference Otani, Iwasaki and Ikeda 8 , Reference Kaneko, Sato and An 26 , Reference Kono, Ikeda and Yanai 28 , Reference Kono, Imanishi and Shinchi 29 , Reference Morita, Tabata and Mineshita 31 , Reference Tabuchi, Kitayama and Nagawa 36 ) (Table 2). The pooled RR for CRA was 1·06 (95 % CI 1·03, 1·10, P=0·0009) which indicated that serum TAG was significantly associated with CRA development. Significant between-study heterogeneity was found in this analysis (Q=62·15, P value for heterogeneity <0·00001, I 2=69 %; Table 3).
There were fourteen studies on HDL-C (four cohort studies, four cross-sectional and six case–control studies) and risk of CRA published between 1990 and 2013 and involved a total of 14 098 CRA cases (Table 2)( Reference Kim, Lim and Kim 6 , Reference Park, Joo and Kim 9 , Reference Yang, Rampal and Sung 10 , Reference Ashbeck, Jacobs and Martinez 18 – Reference Bird, Ingles and Frankl 20 , Reference Hu, Chen and Lin 23 , Reference Kaneko, Sato and An 26 – Reference Morita, Tabata and Mineshita 31 , Reference Tsilidis, Brancati and Pollak 39 ). Three studies were conducted in the USA( Reference Ashbeck, Jacobs and Martinez 18 , Reference Bird, Ingles and Frankl 20 , Reference Tsilidis, Brancati and Pollak 39 ), one in Germany( Reference Bayerdorffer, Mannes and Richter 19 ), two in China( Reference Hu, Chen and Lin 23 , Reference Liu, Hsu and Li 30 ), four in Korea( Reference Kim, Lim and Kim 6 , Reference Park, Joo and Kim 9 , Reference Yang, Rampal and Sung 10 , Reference Kim, Shin and Hong 27 ) and four in Japan( Reference Kaneko, Sato and An 26 , Reference Kono, Ikeda and Yanai 28 , Reference Kono, Imanishi and Shinchi 29 , Reference Morita, Tabata and Mineshita 31 ) (Table 2). A pooled analysis of these fourteen studies showed that the RR for CRA with serum HDL-C was 1·03 (95 % CI 0·99, 1·06, P=0·12) with a significant heterogeneity indicated (Q=29·88, P value for heterogeneity=0·03, I 2=43 %; Table 3).
Finally, we performed a meta-analysis specifically for the association between LDL-C and risk of CRA. Five studies on LDL-C (one cohort study, three cross-sectional and one case-control study) and risk of CRA were published between 1990 and 2013 and involved a total of 7151 CRA cases (Table 2)( Reference Park, Joo and Kim 9 , Reference Yang, Rampal and Sung 10 , Reference Bayerdorffer, Mannes and Richter 19 , Reference Kaneko, Sato and An 26 , Reference Kono, Imanishi and Shinchi 29 ). One study was conducted in Germany( Reference Bayerdorffer, Mannes and Richter 19 ), two in Korea( Reference Park, Joo and Kim 9 , Reference Yang, Rampal and Sung 10 ) and two in Japan( Reference Kaneko, Sato and An 26 , Reference Kono, Imanishi and Shinchi 29 ) (Table 2). Because there was statistical heterogeneity among studies (Q=70·43, P value for heterogeneity<0·00001, I 2=93 %), the random-effects model was applied. The pooled RR for CRA was 1·11 (95 % CI 1·04, 1·19, P=0·003) which presented an increased risk for CRA (Table 3).
Subgroup and sensitivity analyses
The association of lipid components with risk for CRN was performed by subsets within the study design, geographic area, age, gender or the number of cases. Table 4 presents detailed results of subgroup analyses.
Results of subgroup analysis of serum lipids and colorectal neoplasm risk
RR, risk ratio; TC, total cholesterol; HDL-C, HDL-cholesterol; LDL-C, LDL-cholesterol.
We performed a sensitivity analyses to explore the heterogeneity among studies of serum lipids and CRN. By using a stepwise process, we determined that most of the heterogeneity was accounted for in two studies( Reference Bowers, Albanes and Limburg 21 , Reference Trichopoulou, Tzonou and Hsieh 38 ) reporting the association between TC and CRN. After excluding the two studies, there was no study heterogeneity (P=0·15, I 2=27 %) and the RR for CRN was 1·04 (95 % CI 1·01, 1·07). Most of the heterogeneity was accounted for in eight studies( Reference Otani, Iwasaki and Ikeda 8 – Reference Yang, Rampal and Sung 10 , Reference Saydah, Platz and Rifai 32 , Reference Stocks, Lukanova and Bjorge 34 , Reference Tabuchi, Kitayama and Nagawa 36 , Reference Wulaningsih, Garmo and Holmberg 42 , Reference Yamada, Araki and Tamura 43 ) reporting the association between TAG and CRN. After excluding these studies, there was no study heterogeneity (P=0·11, I 2=26 %) and the RR for CRN was 1·01 (95 % CI 1·00, 1·01). We found that most of the heterogeneity was accounted for in two studies( Reference Yang, Rampal and Sung 10 , Reference Bayerdorffer, Mannes and Richter 19 ) reporting the association between HDL-C and CRN. After excluding the two studies, there was no study heterogeneity (P=0·14, I 2=38 %) and the RR for CRN was 1·00 (95 % CI 0·99, 1·01). Most of the heterogeneity was accounted for in four studies( Reference Aleksandrova, Boeing and Jenab 17 , Reference Bayerdorffer, Mannes and Richter 19 , Reference Liu, Hsu and Li 30 , Reference Trichopoulou, Tzonou and Hsieh 38 ) reporting the association between LDL-C and CRN. After excluding these studies, there was no study heterogeneity (P=0·16, I 2=23 %) and the RR for CRN was 1·01 (95 % CI 1·00, 1·02).
Discussion
Our data clearly indicated the positive association between serum TAG and LDL-C and the increased risk of CRN. Subgroup analysis indicated that serum TAG was associated with an increased risk of CRA, but not CRC. It remains uncertain whether TAG is the causal factor responsible for increasing the risk of CRA but not for CRC, although several mechanisms have tried to explain the association between increasing TAG levels and CRN risk. Serum TAG plays an important role in insulin-like growth factor-1 levels( Reference Ibrahim and Yee 44 ), a hormone with proliferative and anti-apoptotic effects. It is demonstrated that the insulin/insulin-like growth factor-1 pathway increases the activity of the ras protein occurring in abnormal colonocytes and stimulates the progression of adenomas into cancers through mediating mitogenicity by activation of K-ras( Reference Siddiqui 45 ). Additionally, hypertriacylglycerolaemia also increases pro-inflammatory cytokines, including IL-6 and TNF-α, while decreasing anti-inflammatory cytokines such as IL-10. The increased inflammatory response has been linked to DNA damage and to effects on the growth, apoptosis and proliferation of colorectal tumour cells. Furthermore, high levels of serum TAG may also result in oxidative stress and in the development of reactive oxygen species( Reference Cowey and Hardy 46 ), which could damage DNA and affect carcinogenesis by affecting gene expression, mutation and chromosomal rearrangement( Reference Valko, Izakovic and Mazur 47 ). Serum TAG may also affect colorectal tumorigenesis by mechanisms involving modification of bile acid excretion, circulating hormones and energy supply to neoplastic cells( Reference McKeown-Eyssen 48 ). Experiments with animal models indicated that Apc-deficient mice showed age-dependent hypertriacylglycerolaemia and a number of intestinal polyp formations, which could be suppressed by anti-hyperlipidaemic medicines. This was confirmed by another report that azoxymethane injection in obese rats with hypertriacylglycerolaemia resulted in an increased number of advanced colon aberrant crypt foci, which are putative precursors of colon cancer( Reference Raju and Bird 49 ). In Apc-deficient FAP model mice, because of the reduced activity of the lipoprotein lipase, serum TAG and intestinal polyp formation decreased significantly after systemic administration of a PPAR ligand( Reference Niho, Mutoh and Takahashi 50 ). Although the biological or molecular mechanism is unclear so far, experiments with animal models most likely showed a clear association between TAG and intestinal neoplasms. Further laboratory and epidemiological studies are still necessary to shed light on the association between serum TAG and CRN development.
Our meta-analysis showed no positive association between TC level and the prevalence of CRN or CRC, although a suggestive association was observed in CRA. A prospective study showed a significant association between serum TC and CRC, which also indicated the risk was higher in patients with colon cancer than with rectal cancer( Reference Tornberg, Holm and Carstensen 51 ). The possible association between TC and colorectal tumorigenesis may be at least partly caused by genetic factors, such as an apoE phenotype which affects both cholesterol metabolism and susceptibility to carcinoma( Reference Kervinen, Sodervik and Makela 52 ). However, the pooled results of our study did not provide evidence for the association between TC and CRN. In our study, we found different associations of serum TC with CRA and CRC. Further research including a large number of studies is necessary to clarify this issue.
In our meta-analysis, higher levels of serum LDL-C were significantly associated with an increasing prevalence of CRN, while serum HDL-C levels were not significantly associated with a decreasing prevalence of CRN. Subgroup analysis demonstrated the positive association between serum LDL-C and CRA, although it was not associated with CRC. HDL-C, however, did not associate with either CRA or CRC. Previous studies have demonstrated that lipids and lipoproteins have been associated with neoplastic processes such as inflammation( Reference Esteve, Ricart and Fernandez-Real 53 ), insulin resistance and oxidative stress. Although there are several possible mechanisms whereby serum lipoproteins influence CRC, little is known regarding the mechanisms by which LDL-C and HDL-C participate in colorectal carcinogenesis. Further mechanistic studies are needed to understand the deferent roles of LDL-C and HDL-C in the development of CRA as well as advanced carcinoma.
The other important finding is that none of the components of serum lipids included in the present meta-analysis (TAG, TC, HDL-C and LDL-C) was significantly correlated with the development of CRC, although positive associations between serum TAG, TC and LDL-C and CRA were found. It was previously reported the significant association of LDL-C with low-grade but not with high-grade CRA( Reference Kaneko, Sato and An 26 ), a stronger correlation between high levels of serum TAG and the number of adenomas, and different associations of serum lipids and adenomas according to histological examination( Reference Otani, Iwasaki and Ikeda 8 ). With the accumulated data, it is difficult to interpret the different effects of serum lipids on CRA and CRC because of the limited mechanisms that have demonstrated a clear biological plausibility for differential effects of lipids on the development of adenoma and advanced carcinoma. Because the potential effects of lipid parameters on different stages of colorectal carcinogenesis have not yet been established, this provides a potential chance for further study.
Our meta-analysis of studies with large numbers of incident cases provides high statistical power for estimating the relationship between components of serum lipids and prevalence of CRN. Despite the strength of the meta-analysis, our study also has several limitations. First, several studies we included are observational. Second, a meta-analysis is not able to solve problems with confounding factors because it did not take into account other possible confounding variables such as dietary patterns, family history of colon cancer and alcohol use, which might be associated with the risk of CRN. However, most studies in the meta-analysis adjusted for other known and potential risk factors for CRN development. Third, heterogeneity may be introduced because of methodological and demographic differences among studies, although appropriate well-motivated inclusion criteria were used. Fourth, the RR values of the baseline serum lipids were endorsed by different panels/organizations in the original studies. Finally, the individual studies may adjust for different covariates including diets, which may affect our results. It is reported that several attributes of diet (such as alcohol and high intake of saturated fats) appear to alter levels of individual lipids( Reference Hegsted and Kritchevsky 54 , Reference Rietman, Schwarz and Blokker 55 ). Many of these diets, which determine serum lipid levels, are established risk factors for CRN( Reference Vargas and Thompson 56 , Reference Vrieling and Kampman 57 ). Therefore, dietary factors might be an important confounding factor in our study.
Conclusion
In conclusion, the present systematic review and meta-analysis demonstrated that high levels of serum TAG and LDL-C are positively associated with the prevalence of CRN. In addition, persons with high levels of the serum lipid components TC, TAG and LDL-C have a greater risk of suffering from CRA but not CRC. Neither CRA nor CRC is linked with serum HDL-C levels. Given the rise in the epidemic of dyslipidaemia worldwide, health-care providers should be more vigilant and adhere with colorectal carcinogenesis screening guidelines in subjects with dyslipidaemia, especially those with abnormal serum TAG, TC and LDL-C levels.
Acknowledgements
Sources of funding: This study was supported in part by funding from the National Natural Science Foundation of China (Y.T., grant number 81302110) (Y.L., grant number 81200263). The National Natural Science Foundation of China had no role in the design, analysis or writing of this article. Conflict of interest: The authors have declared no conflicts of interest. Authorship: Y.T., K.W. and J.L. contributed equally to this work. Y.L. and Y.T. conceived and designed the study, and were involved in drafting of the manuscript; K.W. and J.L. were involved in study concept and design; J.W., Z.W. and Y.F. were involved analysis and interpretation of data; Y.Y. was involved in acquisition of data; G.J. and Y.L. were involved critical revision of the manuscript for important intellectual content. Ethics of human subject participation: Ethical approval was not required.
Supplementary material
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/S1368980015000646
