Polyunsaturated Fatty Acids and Risk of Breast Cancer
Polyunsaturated Fatty Acids and Risk of Breast Cancer
We followed the criteria for conducting and reporting meta-analysis of observational studies. A systematic search was conducted in two databases—PubMed and Embase—up to December 2012. We used the following key words treated as title/abstract for the literature search: ("fat" OR "fatty acid" OR "docosahexaenoic acid" OR "eicosapentaenoic acid" OR "docosapentaenoic acid" OR "alpha-linolenic acid" OR "polyunsaturated fatty acid" OR "omega-3 fatty acid" OR "n-3 fatty acid" OR "fish" OR "fish oil" OR "seafood") AND ("breast cancer" OR "breast neoplasms"). Full details of the search strategy are in Appendix 1 . Our search was restricted to studies in humans and studies published in English. The references of retrieved relevant articles were reviewed to identify potential publications. We did not contact authors for the detailed information of primary studies.
Two investigators (J-SZ and X-JH) independently conducted the literature search, identified potential studies, and extracted detailed information from each included article. Discrepancies were resolved through group discussion with the third investigator (DL). Inclusion criteria were prospective study design (including prospective cohort, nested case-control, and case-cohort studies); the exposure of interest was any type of dietary n-3 PUFA or fish consumption or tissue n-3 PUFA concentrations; the endpoint of interest was incident breast cancer in women; and the risk estimate with corresponding 95% confidence intervals of breast cancer was reported for n-3 PUFA exposure or fish intake. We excluded retrospective or cross sectional studies, studies in animals, non-original research (reviews, editorials, or commentaries), abstracts, unpublished studies, and duplicated studies.
From each identified article, we extracted the first author’s name, study population and region, study design, duration of follow-up, age of participants, number of cases and non-cases, person years for the population and for each exposure category, risk estimates and corresponding 95% confidence intervals for each category of n-3 PUFA or fish intake, menopausal status, method of n-3 PUFA measurement (diet or tissue biomarker), and covariates. We extracted risk estimates with the most adjustment.
Quality assessment was conducted according to the Newcastle-Ottawa criteria for non-randomised studies. A maximum of 9 points was assigned to each study: 4 for selection, 2 for comparability, and 3 for assessment of outcomes (for cohort study) or exposures (for case-control study). We regarded scores of 0-3, 4-6, and 7-9 as low, moderate, and high quality, respectively.
We used relative risk for risk estimates, and hazard ratios in cohort studies and odds ratios in nested case-control studies were treated as relative risks directly. We used log transformed relative risk and its corresponding 95% confidence interval from each eligible study for the meta-analysis. As different studies might use different assessment methods (diet or tissue biomarkers) and report different exposure categories (dichotomous, thirds, quarters, or fifths), we used the study specific relative risk for the highest versus lowest category of fish consumption or n-3 PUFA exposure for the meta-analysis. We then combined the relative risk from each study, weighted by the inverse of their variance, for the meta-analysis with the DerSimonian and Laird random effects model, which takes variation both within and between studies into consideration. Studies that reported relative risk of breast cancer separately for postmenopausal and premenopausal women were considered as independent studies for the meta-analysis. Women in one study were reported as either premenopausal or perimenopausal, and thus we classed them all as premenopausal. One study in which only 10% of women with breast cancer were premenopausal was treated as a postmenopausal study. Most of the women with breast cancer in another study were postmenopausal, and we treated this study as a postmenopausal study.
We conducted meta-analysis for different types of n-3 PUFA separately. Firstly, we estimated the pooled relative risk between the highest versus lowest category of fish intake, total marine n-3 PUFA, alpha linolenic acid (ALA), and total n-3 PUFA, respectively. For studies that did not report a relative risk for total marine n-3 PUFA but that reported risks for eicosapentaenoic acid (EPA), docosahexaenoic acid (DHA), and docosapentaenoic acid (DPA) separately, we pooled these relative risks to represent the relative risk of total marine n-3 PUFA exposure. One study reported relative risks for intake of dried and non-dried fish separately, and we pooled the two relative risks with a fixed effects model to get a summary relative risk for total fish intake in this study for further meta-analysis. Secondly, for marine n-3 PUFA, we conducted meta-analysis for EPA, DHA, and EPA separately.
We carried out a dose-response analysis for the trend estimation using generalised least squares regression (two stage GLST in Stata). For a study without information on the number of cases, number of healthy controls, or person years for exposure categories, we used variance weighted least squares regression for the dose-response estimation. Studies with fewer than three exposure categories were excluded from trend estimation. Dose-response analysis was conducted only in studies that reported dietary intake of n-3 PUFA or fish intake because the results of tissue n-3 PUFA compositions varied according to different tissues (serum, erythrocytes, or adipose) and units used and were not appropriate for standardisation. For fish intake, all the different units were transformed to g/day as described. One study reported the unit for the fish intake category as g/1000 kcal, which we transformed to g/day assuming an average energy intake of 2000 kcal/day in this population. For dietary n-3 PUFA, we carried out dose-response analysis among studies with exposure units as % energy/day and g/day, separately. We did not do trend estimation for EPA, DHA, DPA, or total n-3 PUFA because there were limited studies with dietary information for these exposures. To estimate a potential curve linear association between fish, dietary n-3 PUFA, and breast cancer, we used a restricted cubic spline model (three knots).
Study heterogeneity was estimated with I statistic, with values of 25%, 50%, and 75% representing low, moderate, and high degrees of heterogeneity. Subgroup analysis was conducted to examine sources of study heterogeneity and the influence of potential residual confounding factors, such as age, body mass index (BMI), total energy intake, and education. Univariate meta-regression was performed to examine the significance of the difference in relative risks by different subgroups, including study region (Asian countries v western countries), duration of follow-up (lower v higher than mean duration), exposure measurement (diet v tissue biomarker), study type (cohort v nested case-control), menopausal status (premenopausal v postmenopausal), study quality (score ≤7 v >7), risk expression (hazard/rate ratio, relative risk, or odds ratio), and adjustment of covariates (including age, BMI, total energy intake, and education). Sensitivity analysis was conducted by omitting one study at a time and examining the influence of each individual study on the overall relative risk. Publication bias was evaluated by visual inspection of a funnel plot and Egger’s regression test (significant at P<0.1). We used a trim and fill algorithm if possible publication bias was detected to identify and correct for the asymmetry of funnel plot from publication bias and provide an adjusted summary relative risk based on all the studies, including the estimated missing studies. Stata version 12 (StataCorp LP, College Station, TX, USA) was used for all the statistical analyses.
Methods
Search Strategy and Selection Criteria
We followed the criteria for conducting and reporting meta-analysis of observational studies. A systematic search was conducted in two databases—PubMed and Embase—up to December 2012. We used the following key words treated as title/abstract for the literature search: ("fat" OR "fatty acid" OR "docosahexaenoic acid" OR "eicosapentaenoic acid" OR "docosapentaenoic acid" OR "alpha-linolenic acid" OR "polyunsaturated fatty acid" OR "omega-3 fatty acid" OR "n-3 fatty acid" OR "fish" OR "fish oil" OR "seafood") AND ("breast cancer" OR "breast neoplasms"). Full details of the search strategy are in Appendix 1 . Our search was restricted to studies in humans and studies published in English. The references of retrieved relevant articles were reviewed to identify potential publications. We did not contact authors for the detailed information of primary studies.
Two investigators (J-SZ and X-JH) independently conducted the literature search, identified potential studies, and extracted detailed information from each included article. Discrepancies were resolved through group discussion with the third investigator (DL). Inclusion criteria were prospective study design (including prospective cohort, nested case-control, and case-cohort studies); the exposure of interest was any type of dietary n-3 PUFA or fish consumption or tissue n-3 PUFA concentrations; the endpoint of interest was incident breast cancer in women; and the risk estimate with corresponding 95% confidence intervals of breast cancer was reported for n-3 PUFA exposure or fish intake. We excluded retrospective or cross sectional studies, studies in animals, non-original research (reviews, editorials, or commentaries), abstracts, unpublished studies, and duplicated studies.
Data Extraction
From each identified article, we extracted the first author’s name, study population and region, study design, duration of follow-up, age of participants, number of cases and non-cases, person years for the population and for each exposure category, risk estimates and corresponding 95% confidence intervals for each category of n-3 PUFA or fish intake, menopausal status, method of n-3 PUFA measurement (diet or tissue biomarker), and covariates. We extracted risk estimates with the most adjustment.
Quality assessment was conducted according to the Newcastle-Ottawa criteria for non-randomised studies. A maximum of 9 points was assigned to each study: 4 for selection, 2 for comparability, and 3 for assessment of outcomes (for cohort study) or exposures (for case-control study). We regarded scores of 0-3, 4-6, and 7-9 as low, moderate, and high quality, respectively.
Data Synthesis
We used relative risk for risk estimates, and hazard ratios in cohort studies and odds ratios in nested case-control studies were treated as relative risks directly. We used log transformed relative risk and its corresponding 95% confidence interval from each eligible study for the meta-analysis. As different studies might use different assessment methods (diet or tissue biomarkers) and report different exposure categories (dichotomous, thirds, quarters, or fifths), we used the study specific relative risk for the highest versus lowest category of fish consumption or n-3 PUFA exposure for the meta-analysis. We then combined the relative risk from each study, weighted by the inverse of their variance, for the meta-analysis with the DerSimonian and Laird random effects model, which takes variation both within and between studies into consideration. Studies that reported relative risk of breast cancer separately for postmenopausal and premenopausal women were considered as independent studies for the meta-analysis. Women in one study were reported as either premenopausal or perimenopausal, and thus we classed them all as premenopausal. One study in which only 10% of women with breast cancer were premenopausal was treated as a postmenopausal study. Most of the women with breast cancer in another study were postmenopausal, and we treated this study as a postmenopausal study.
We conducted meta-analysis for different types of n-3 PUFA separately. Firstly, we estimated the pooled relative risk between the highest versus lowest category of fish intake, total marine n-3 PUFA, alpha linolenic acid (ALA), and total n-3 PUFA, respectively. For studies that did not report a relative risk for total marine n-3 PUFA but that reported risks for eicosapentaenoic acid (EPA), docosahexaenoic acid (DHA), and docosapentaenoic acid (DPA) separately, we pooled these relative risks to represent the relative risk of total marine n-3 PUFA exposure. One study reported relative risks for intake of dried and non-dried fish separately, and we pooled the two relative risks with a fixed effects model to get a summary relative risk for total fish intake in this study for further meta-analysis. Secondly, for marine n-3 PUFA, we conducted meta-analysis for EPA, DHA, and EPA separately.
We carried out a dose-response analysis for the trend estimation using generalised least squares regression (two stage GLST in Stata). For a study without information on the number of cases, number of healthy controls, or person years for exposure categories, we used variance weighted least squares regression for the dose-response estimation. Studies with fewer than three exposure categories were excluded from trend estimation. Dose-response analysis was conducted only in studies that reported dietary intake of n-3 PUFA or fish intake because the results of tissue n-3 PUFA compositions varied according to different tissues (serum, erythrocytes, or adipose) and units used and were not appropriate for standardisation. For fish intake, all the different units were transformed to g/day as described. One study reported the unit for the fish intake category as g/1000 kcal, which we transformed to g/day assuming an average energy intake of 2000 kcal/day in this population. For dietary n-3 PUFA, we carried out dose-response analysis among studies with exposure units as % energy/day and g/day, separately. We did not do trend estimation for EPA, DHA, DPA, or total n-3 PUFA because there were limited studies with dietary information for these exposures. To estimate a potential curve linear association between fish, dietary n-3 PUFA, and breast cancer, we used a restricted cubic spline model (three knots).
Study heterogeneity was estimated with I statistic, with values of 25%, 50%, and 75% representing low, moderate, and high degrees of heterogeneity. Subgroup analysis was conducted to examine sources of study heterogeneity and the influence of potential residual confounding factors, such as age, body mass index (BMI), total energy intake, and education. Univariate meta-regression was performed to examine the significance of the difference in relative risks by different subgroups, including study region (Asian countries v western countries), duration of follow-up (lower v higher than mean duration), exposure measurement (diet v tissue biomarker), study type (cohort v nested case-control), menopausal status (premenopausal v postmenopausal), study quality (score ≤7 v >7), risk expression (hazard/rate ratio, relative risk, or odds ratio), and adjustment of covariates (including age, BMI, total energy intake, and education). Sensitivity analysis was conducted by omitting one study at a time and examining the influence of each individual study on the overall relative risk. Publication bias was evaluated by visual inspection of a funnel plot and Egger’s regression test (significant at P<0.1). We used a trim and fill algorithm if possible publication bias was detected to identify and correct for the asymmetry of funnel plot from publication bias and provide an adjusted summary relative risk based on all the studies, including the estimated missing studies. Stata version 12 (StataCorp LP, College Station, TX, USA) was used for all the statistical analyses.