Dietary Fiber Intake and Risk of Cardiovascular Disease
Dietary Fiber Intake and Risk of Cardiovascular Disease
We conducted literature searches for articles published from 1 January 1990 to 6 August 2013. Initial literature searching was part of a larger systematic review concerning studies reporting associations between any form of carbohydrate intake and cardiometabolic health outcomes (between 1 January 1990 to 17 November 2009). Searches were conducted by an information specialist in addition to other members of the research team (VJB, DET, CLC). We used six electronic databases: the Cochrane Library, Medline, Medline in-process, Embase, CAB Abstracts, ISI Web of Science, and BIOSIS. This search was extended from 2009 using Medline, Medline in-process, and Embase, which were the main sources of included articles in the initial search. This update searching was carried out only for articles reporting dietary fibre and CVD or CHD. We used the same key search terms including, among others, "dietary fibre," "cellulose,” "lignin," various other subfractions and subtypes of fibre, "cardiovascular diseases," "coronary diseases," and "myocardial ischaemia" (Web Fig 1). Hand searching of selected journals and cross checking of bibliographies from other published review articles was also done to supplement the electronic searches. Standard guidelines for conducting and reporting meta-analyses of observational studies were followed.
(Enlarge Image)
Web Figure 1.
Medline search strategy for update searching adapted from the original search strategy (for all carbohydrate intake and all cardiometabolic health outcomes)
After removal of duplicate references, initial screening of article titles and abstracts was undertaken by members of the review team (DET, CLC, CELE, VJB). This process removed those articles that were clearly not relevant—such as editorials, single case study reports, and therapeutic approach articles. Reviewers used prespecified guidelines to ensure a consistent approach. Potentially relevant articles were obtained in full text and read independently by two review team members (DET, CLC, CELE, CN, VJB) using a structured flow chart and detailed guidelines to determine eligibility for inclusion. Any disagreements were settled by a third reviewer. We obtained copies of all articles identified as being of potential importance, including contacting authors as necessary.
Articles eligible for inclusion were reported in English language since 1990 and were original research from prospective observational studies, with at least three years’ follow-up. Studies had to report incident or fatal events of primary CHD or CVD in relation to dietary fibre intake.
Because primary prevention of CVD was the main focus for this work (rather than secondary prevention), we excluded studies if participants had been specifically recruited because of ill health or history of disease and if the study did not recruit from a generally healthy population.
Multiple reports from the same cohort study were reviewed, and papers with the longest follow-up for identical outcomes were included. If insufficient data were presented at the longer follow-up for inclusion in meta-analysis, we used the shorter follow-up data.
Methodological quality of studies was evaluated using the Newcastle-Ottawa Scale for assessing cohort studies used in meta-analyses. Studies were not excluded based on these scores, but aspects of study quality—such as follow-up duration, case ascertainment, and adjustment for various important confounders—were investigated through meta-regression.
Data from articles were extracted directly into a Microsoft Access database, which included predefined fields set up to capture aspects of study design and quality as well as all results (relative risks and confidence intervals). We also extracted details on exposure type and quantity, case numbers, the definition of outcome, and adjustments used within analyses. This method of data extraction was based on the approach used for the World Cancer Research Fund second expert report. Multiple risk estimates were extracted from most publications, because risk of CHD or CVD was provided for a range of fibre subtypes and sources as well as for total dietary fibre. Risk ratios plus 95% confidence intervals for continuous (linear) exposures were extracted preferentially or, where these data were not available, risk estimates across increasing fibre intake categories were extracted.
Included studies use a range of different definitions for their exposure categories. We therefore derived an estimated dose-response trend for each study, using the method recommended by Greenland and Longnecker. These dose-response trends were then combined using random effects meta-analysis. This method generates study specific slopes (with 95% confidence intervals), based on the results presented for each category of fibre intake. To derive a linear dose-response curve for each study, the distribution of cases and person years, or cases and non-cases, with risk ratios and estimates of uncertainty (for example, confidence intervals) for at least three categories of quantified fibre intake needed to be presented in the reviewed publication. If the total number of cases or person years was presented, but not the distribution, we estimated the distribution on the basis of definitions of the quantiles. The median or mean level of fibre intake was then assigned to the corresponding risk ratio for each study. When medians and means were not presented, we used the category midpoint. Where the highest or lower category was unbounded, we assumed the width of the category to be the same as the next adjacent category, so that we could assign a midpoint. If studies already reported a linear dose-response trend, with confidence intervals or standard error, this was used directly.
If results were only presented separately for men and women, we derived separate dose-response curves. These curves were then combined into a single estimate for the study using a fixed effects meta-analysis, before combining with the other studies using a random effects meta-analysis. This method maintained the correct degrees of freedom for tests of heterogeneity.
Heterogeneity between studies was tested using Cochran's Q statistic, alongside the proportion of total variation in study estimates that is due to heterogeneity (I). Pooled estimates are only presented if I did not exceed our prespecified cut point of 75% and if studies had included appropriate adjustments. For comparability, fibre increments presented in the dose-response figures were chosen to be approximately one standard deviation of the mean, based on European population intakes, from multiple sources. Increments are equivalent to Association of Official Analytical Chemist (AOAC) fibre values, because most contributing studies assessed fibre in this way.
We explored other methodological features through predefined meta-regression. These included fibre intake assessment method (AOAC or non-AOAC); inclusion of non-fatal events; follow-up duration (<10 years or ≥10 years); geographical location (Americas, Europe, or other); and whether the results were adjusted for age, alcohol, anthropometry, energy intake, physical activity, or sex. All analyses were performed in Stata 12.1.
The potential for small study effects such as publication bias was explored using funnel plots with Egger's test of asymmetry where there were sufficient studies to allow the test.
We plotted non-linear dose-response curves using restricted cubic splines for each study, with knots fixed at percentiles 10%, 50%, and 90% through the distribution, and the median intake as the reference. These curves were combined using multivariate meta-analysis, and the summary curve was plotted, with 95% confidence limits, using the median over all studies as the reference. Studies with three or fewer categories could not be included in this analysis.
Methods
Search Strategy
We conducted literature searches for articles published from 1 January 1990 to 6 August 2013. Initial literature searching was part of a larger systematic review concerning studies reporting associations between any form of carbohydrate intake and cardiometabolic health outcomes (between 1 January 1990 to 17 November 2009). Searches were conducted by an information specialist in addition to other members of the research team (VJB, DET, CLC). We used six electronic databases: the Cochrane Library, Medline, Medline in-process, Embase, CAB Abstracts, ISI Web of Science, and BIOSIS. This search was extended from 2009 using Medline, Medline in-process, and Embase, which were the main sources of included articles in the initial search. This update searching was carried out only for articles reporting dietary fibre and CVD or CHD. We used the same key search terms including, among others, "dietary fibre," "cellulose,” "lignin," various other subfractions and subtypes of fibre, "cardiovascular diseases," "coronary diseases," and "myocardial ischaemia" (Web Fig 1). Hand searching of selected journals and cross checking of bibliographies from other published review articles was also done to supplement the electronic searches. Standard guidelines for conducting and reporting meta-analyses of observational studies were followed.
(Enlarge Image)
Web Figure 1.
Medline search strategy for update searching adapted from the original search strategy (for all carbohydrate intake and all cardiometabolic health outcomes)
Study Selection
After removal of duplicate references, initial screening of article titles and abstracts was undertaken by members of the review team (DET, CLC, CELE, VJB). This process removed those articles that were clearly not relevant—such as editorials, single case study reports, and therapeutic approach articles. Reviewers used prespecified guidelines to ensure a consistent approach. Potentially relevant articles were obtained in full text and read independently by two review team members (DET, CLC, CELE, CN, VJB) using a structured flow chart and detailed guidelines to determine eligibility for inclusion. Any disagreements were settled by a third reviewer. We obtained copies of all articles identified as being of potential importance, including contacting authors as necessary.
Articles eligible for inclusion were reported in English language since 1990 and were original research from prospective observational studies, with at least three years’ follow-up. Studies had to report incident or fatal events of primary CHD or CVD in relation to dietary fibre intake.
Because primary prevention of CVD was the main focus for this work (rather than secondary prevention), we excluded studies if participants had been specifically recruited because of ill health or history of disease and if the study did not recruit from a generally healthy population.
Multiple reports from the same cohort study were reviewed, and papers with the longest follow-up for identical outcomes were included. If insufficient data were presented at the longer follow-up for inclusion in meta-analysis, we used the shorter follow-up data.
Methodological quality of studies was evaluated using the Newcastle-Ottawa Scale for assessing cohort studies used in meta-analyses. Studies were not excluded based on these scores, but aspects of study quality—such as follow-up duration, case ascertainment, and adjustment for various important confounders—were investigated through meta-regression.
Data Extraction
Data from articles were extracted directly into a Microsoft Access database, which included predefined fields set up to capture aspects of study design and quality as well as all results (relative risks and confidence intervals). We also extracted details on exposure type and quantity, case numbers, the definition of outcome, and adjustments used within analyses. This method of data extraction was based on the approach used for the World Cancer Research Fund second expert report. Multiple risk estimates were extracted from most publications, because risk of CHD or CVD was provided for a range of fibre subtypes and sources as well as for total dietary fibre. Risk ratios plus 95% confidence intervals for continuous (linear) exposures were extracted preferentially or, where these data were not available, risk estimates across increasing fibre intake categories were extracted.
Statistical Methods
Included studies use a range of different definitions for their exposure categories. We therefore derived an estimated dose-response trend for each study, using the method recommended by Greenland and Longnecker. These dose-response trends were then combined using random effects meta-analysis. This method generates study specific slopes (with 95% confidence intervals), based on the results presented for each category of fibre intake. To derive a linear dose-response curve for each study, the distribution of cases and person years, or cases and non-cases, with risk ratios and estimates of uncertainty (for example, confidence intervals) for at least three categories of quantified fibre intake needed to be presented in the reviewed publication. If the total number of cases or person years was presented, but not the distribution, we estimated the distribution on the basis of definitions of the quantiles. The median or mean level of fibre intake was then assigned to the corresponding risk ratio for each study. When medians and means were not presented, we used the category midpoint. Where the highest or lower category was unbounded, we assumed the width of the category to be the same as the next adjacent category, so that we could assign a midpoint. If studies already reported a linear dose-response trend, with confidence intervals or standard error, this was used directly.
If results were only presented separately for men and women, we derived separate dose-response curves. These curves were then combined into a single estimate for the study using a fixed effects meta-analysis, before combining with the other studies using a random effects meta-analysis. This method maintained the correct degrees of freedom for tests of heterogeneity.
Heterogeneity between studies was tested using Cochran's Q statistic, alongside the proportion of total variation in study estimates that is due to heterogeneity (I). Pooled estimates are only presented if I did not exceed our prespecified cut point of 75% and if studies had included appropriate adjustments. For comparability, fibre increments presented in the dose-response figures were chosen to be approximately one standard deviation of the mean, based on European population intakes, from multiple sources. Increments are equivalent to Association of Official Analytical Chemist (AOAC) fibre values, because most contributing studies assessed fibre in this way.
We explored other methodological features through predefined meta-regression. These included fibre intake assessment method (AOAC or non-AOAC); inclusion of non-fatal events; follow-up duration (<10 years or ≥10 years); geographical location (Americas, Europe, or other); and whether the results were adjusted for age, alcohol, anthropometry, energy intake, physical activity, or sex. All analyses were performed in Stata 12.1.
The potential for small study effects such as publication bias was explored using funnel plots with Egger's test of asymmetry where there were sufficient studies to allow the test.
We plotted non-linear dose-response curves using restricted cubic splines for each study, with knots fixed at percentiles 10%, 50%, and 90% through the distribution, and the median intake as the reference. These curves were combined using multivariate meta-analysis, and the summary curve was plotted, with 95% confidence limits, using the median over all studies as the reference. Studies with three or fewer categories could not be included in this analysis.
Source...