| |
A Systematic Review of the Evidence Supporting a Causal Link Between Dietary Factors and Coronary Heart Disease
|
| |
| |
Andrew Mente, PhD; Lawrence de Koning, MSc; Harry S. Shannon, PhD; Sonia S. Anand, MD, PhD, FRCPC
Arch Intern Med. Apr 132009;169(7):659-669.
ABSTRACT
Background- Although a wealth of literature links dietary factors and coronary heart disease (CHD), the strength of the evidence supporting valid associations has not been evaluated systematically in a single investigation.
Methods- We conducted a systematic search of MEDLINE for prospective cohort studies or randomized trials investigating dietary exposures in relation to CHD. We used the Bradford Hill guidelines to derive a causation score based on 4 criteria (strength, consistency, temporality, and coherence) for each dietary exposure in cohort studies and examined for consistency with the findings of randomized trials.
Results- Strong evidence supports valid associations (4 criteria satisfied) of protective factors, including intake of vegetables, nuts, and "Mediterranean" and high-quality dietary patterns with CHD, and associations of harmful factors, including intake of trans-fatty acids and foods with a high glycemic index or load. Among studies of higher methodologic quality, there was also strong evidence for monounsaturated fatty acids and "prudent" and "western" dietary patterns. Moderate evidence (3 criteria) of associations exists for intake of fish, marine {omega}-3 fatty acids, folate, whole grains, dietary vitamins E and C, beta carotene, alcohol, fruit, and fiber. Insufficient evidence (²2 criteria) of association is present for intake of supplementary vitamin E and ascorbic acid (vitamin C); saturated and polyunsaturated fatty acids; total fat; {alpha}-linolenic acid; meat; eggs; and milk. Among the dietary exposures with strong evidence of causation from cohort studies, only a Mediterranean dietary pattern is related to CHD in randomized trials.
Conclusions The evidence supports a valid association of a limited number of dietary factors and dietary patterns with CHD. Future evaluation of dietary patterns, including their nutrient and food components, in cohort studies and randomized trials is recommended.
INTRODUCTION
The relationship between dietary factors and coronary heart disease (CHD) has been a major focus of health research for almost half a century. The pioneering work of Keys and Aravanis1 stimulated many subsequent studies of diet and CHD, which have since evaluated the effects of numerous dietary nutrients, foods, and dietary patterns on CHD risk. More recently, prospective cohort studies and randomized controlled trials (RCTs) have examined these associations in large populations with long periods of follow-up. However, the results of cohort studies and RCTs can be discrepant (eg, for intake of vitamin E and beta carotene),2-13 and the results of some RCTs of dietary supplements paradoxically revealed adverse effects on CHD for certain nutrients that were previously shown to exert protective effects in cohort studies.2, 4, 7-8 This has generated confusion among health care professionals, policy makers, and the population at large who are interested in this information to aid them in CHD prevention strategies.
In his classic study, Hill14 proposed a set of criteria (the Bradford Hill criteria) to evaluate systematically whether a causal link between an exposure of interest and a health outcome exists. These guidelines are used by epidemiologists to test causal hypotheses and have undergone little modification since their original publication.15 Before advocating that specific dietary factors be consumed in large or minimal amounts, it is necessary to base public health recommendations on the best available scientific evidence.16 To address this issue, we conducted a systematic review of the literature examining the association between nutrient intake, dietary components, and dietary patterns (hereafter referred to as dietary exposures) and CHD and its related clinical outcomes. Our specific objectives were (1) to systematically evaluate dietary exposures and CHD using the Bradford Hill criteria; (2) to determine which dietary exposures have been studied sufficiently in RCTs and found to support the findings of prospective cohort studies; and (3) to identify the dietary exposures deemed to have insufficient evidence to be conclusive.
COMMENT
This review is, to our knowledge, the first to systematically assess whether a valid association exists between dietary exposures and CHD using the Bradford Hill guidelines. In applying a predefined algorithm, we identified strong evidence of a causal relationship for protective factors, including intake of vegetables, nuts, and monounsaturated fatty acids and Mediterranean, prudent, and high-quality dietary patterns, and harmful factors, including intake of trans-fatty acids and foods with a high glycemic index or load and a western dietary pattern. Among these dietary exposures, however, only a Mediterranean dietary pattern has been studied in RCTs and significantly associated with CHD. In addition, we found modest evidence to support a causal relationship for intake of fish, marine {omega}-3 fatty acids, folate, whole grains, dietary vitamins E and C and beta carotene, alcohol, fruits, and fiber, and weak evidence of causation for intake of supplementary vitamin E and ascorbic acid, saturated and polyunsaturated fatty acids and total fat, {alpha}-linolenic acid, meat, eggs, and milk. The modest or weak evidence of these dietary exposures is mostly consistent with the findings of RCTs, although RCTs have yet to be conducted for several factors. Taken together, these findings support a causal relationship between only a few dietary exposures and CHD, whereas the evidence for most individual nutrients or foods is too modest to be conclusive.
A wealth of epidemiologic studies have evaluated associations between dietary exposures and CHD. The general consensus from the evidence currently available is that a reduced consumption of saturated and trans-fatty acids and a higher intake of fruits and vegetables, polyunsaturated fatty acids including {omega}-3 fatty acids, and whole grains are likely beneficial.21-23 This is reflected in the revised Dietary Guidelines for Americans 2005 from the US Departments of Health and Human Services and Agriculture.24 However, little direct evidence from RCTs supports these recommendations. In some cases, RCTs have not been conducted, and RCTs that have been conducted have generally not been adequately powered or have evaluated surrogate end points rather than clinical outcomes. Despite this lack of information, evidence-based recommendations derived from cohort studies have been advocated.25 This is cause for concern because dietary advice to limit the intake of a certain nutrient (ie, dietary fat) may result in increased consumption of another (ie, carbohydrates), which can have adverse effects on CHD risk factors.26 Moreover, without large prospective studies in which multiple health outcomes are evaluated, recommendations to modify a dietary component may decrease the likelihood of one chronic disease (ie, CHD) at the cost of increasing another (ie, cancer).16
We found strong evidence that trans-fatty acids are associated with CHD risk, but weak evidence implicating saturated and polyunsaturated fatty acids and total fat intake. Relatively few cohort studies have shown that a higher intake of polyunsaturated fatty acids or a lower intake of saturated fatty acids is related to a reduced risk of CHD.27 Ecological data from the Seven Countries Study showed a strong positive association (r = 0.73) between saturated fatty acid intake and CHD incidence,1 although a much lower correlation was observed for total fat (r = 0.39), suggesting that not all types of fat (ie, polyunsaturated fatty acids) are associated with an increased CHD risk. However, these data were interpreted by some to mean that all fats are associated with increased CHD risk, and subsequent dietary guidelines advocated low-fat diets.21-23 More recently, the lack of benefit of diets of reduced total fat has been established,28 and the evidence supporting the adverse effect of trans-fatty acids on cholesterol levels29 and CHD30-34 has increased, which is reflected in our findings. Single-nutrient RCTs have yet to evaluate whether reducing saturated fatty acid intake lowers the risk of CHD events. For polyunsaturated fatty acid intake, most of the RCTs have not been adequately powered and did not find a significant reduction in CHD outcomes. On the other hand, mechanistic studies have demonstrated that diets low in total fat are associated with increased triglyceride and lower high-density lipoprotein cholesterol levels, whereas diets enriched with unsaturated fatty acids such as olive oil have positive effects on serum lipids.29 Further work is needed to demonstrate the beneficial effects of these fatty acids on clinical outcomes.
Our results support an association between foods with higher glycemic index values and CHD outcomes. Metabolic studies have shown that higher glycemic index scores are associated with coronary risk factors, such as higher fasting triglycerides and lower high-density lipoprotein cholesterol levels.35 Our findings, however, do not imply that every food with a low glycemic index is equally beneficial. Nevertheless, the glycemic index represents 1 functional property of food that can help guide dietary choices and may effectively organize a healthy dietary pattern, if used carefully. This functional index may be supplemented with information about glycemic load, which reflects weighted carbohydrate intake and may provide further information about food choices based on appropriate serving size.
We found moderate evidence of valid associations involving fish intake and heterogeneity in the effects of {omega}-3 fatty acids. Metabolic studies have shown that these factors exert beneficial effects on surrogate measures of CHD such as levels of serum triglycerides and thrombotic factors, markers of endothelial dysfunction, and prevention of cardiac arrhythmias.36-37 The findings from cohort studies, however, have been inconsistent, and some have reported a detrimental effect.31-32,38-39 Our summary estimate (RR, 0.81) for fish intake and CHD risk was similar to that reported in a meta-analysis by Whelton et al,40 who also reported similar estimates for fatal vs nonfatal CHD. A previous systematic review reported a stronger protective effect for fish intake in populations at higher risk of CHD than initially healthy populations,41 which we also observed. Similarly, our analysis showed that the effects of marine {omega}-3 fatty acids on CHD events are context driven because benefits are observed predominantly in female cohorts and secondary prevention studies, and the individual RCTs corroborated these benefits in subjects with CHD. However, a meta-analysis by Bucher et al42 suggested that an equal benefit from dietary and supplemental sources existed. Our findings did not support an association between intake of plant {omega}-3 fatty acids (eg, {alpha}-linolenic acid) and CHD in cohort studies, and evidence from RCTs is still lacking. These findings are consistent with recent information showing that {alpha}-linolenic acid supplementation does not affect cardiovascular risk markers.43 Collectively, the evidence suggests a benefit of increased marine {omega}-3 fatty-acid intake against CHD in certain population subsets, although more studies are needed before widespread recommendations are made for the general population.
The discrepant findings of cohort studies vs RCTs often draw the attention of investigators. For instance, high-impact clinical trials such as Gruppo Italiano per lo Studio della Sopravvivenza nell'Infarcto Miocardico, Heart Outcomes Prevention Evaluation, Heart Outcomes Prevention Evaluation-The Ongoing Outcomes, and the Physician's Health Study3, 7, 13, 44 did not support the findings of large cohort studies that supported a protective association between vitamin E and beta carotene intake and CHD. In our analysis, differentiating between dietary and supplemental intake and implementation of the Bradford Hill guidelines helped to demonstrate that discrepant results involving the 2 designs are minimal. In particular, our findings of modest or weak evidence of a causal link between CHD and intake of polyunsaturated fatty acids and total fat and vitamin E and ascorbic acid supplements are compatible with the results from RCTs. Similarly, the strong evidence of causation involving a Mediterranean dietary pattern is compatible with the evidence from the Lyon Diet Heart Study trial. These findings lend support to the usefulness of the Bradford Hill guidelines in gauging the evidence of causation and emphasize the importance of examining the evidence from observational studies, given some of the limitations of RCTs (eg, subject compliance, disease latency, and duration of exposure).
We observed strong evidence of a causal link between CHD and dietary patterns. Population-based cohort studies have demonstrated the protective effect of a quality diet against CHD and all-cause mortality,45-49 and these benefits are additive with other lifestyle activities aimed at improving health.47 The Lyon Diet Heart Study showed that a Mediterranean dietary pattern reduces cause-specific and all-cause mortality in patients with CHD.50 Dietary patterns have the advantage of taking into account the complex interactions and cumulative effects of multiple nutrients within the entire diet because these effects may be larger and easier to detect than the effect of a single nutrient or food.51-52 Finally, CHD is a complex condition involving numerous physiologic systems, which makes it unlikely that modifying the intake of a few nutrients would alter these systems and influence clinical outcomes. Given the advantages of evaluating dietary patterns vs single nutrient components, we recommend that future RCTs test various dietary patterns in sufficiently large populations and determine the effects of these patterns on multiple important health outcomes, including cardiovascular disease and cancer.
Our study has a number of strengths because we undertook several measures to minimize bias, including restricting our review to studies with the strongest causal inference (eg, cohort studies and RCTs), conducting an independent assessment of study eligibility by 2 of the authors, using predefined criteria to evaluate the evidence of causation, and performing stratification analyses for a number of extraneous variables. In addition, we examined high-quality or larger studies with sufficient outcome events, evaluated the potential for heterogeneity of effects across cohort studies, and assessed publication bias.
We may be criticized for creating arbitrary definitions of strong, moderate, and weak evidence, although these classifications have face validity and similar scoring systems have been used to assess the evidence of causation from observational studies.53-54 Second, we had to derive the RR cutoff points to define a strong association from the distribution of RR values in cohort studies because the true cutoff points for defining clinically meaningful effects are not known. Third, the heterogeneity of cohort studies may have influenced our results. However, our scatterplots of RR values against the difference in mean and median intake between quantile extremes showed no relationship for each dietary predictor, suggesting that differences in mean intake across studies do not explain the variation in RR values. Last, the evidence of causation may depend on the prevention strategy (primary vs secondary) or dietary assessment tool used in studies or may vary across populations. However, our subanalysis showed that, in general, the summary estimates are consistent across the strata of potential effect modifiers.
CONCLUSIONS
Our implementation of the Bradford Hill criteria identified strong evidence that a causal association exists between CHD and intake of vegetables, nuts, monounsaturated fatty acids, foods with a high glycemic index, trans-fatty acids, and overall diet quality or dietary patterns. Among these factors, only a Mediterranean dietary pattern was related to CHD in RCTs. Although investigations of dietary components may help to shed light on mechanisms behind the benefits of dietary patterns, it is unlikely that modifying the intake of a few nutrients or foods would substantially influence coronary outcomes. Our findings support the strategy of investigating dietary patterns in cohort studies and RCTs for common and complex chronic diseases such as CHD.
RESULTS
The Figure displays the number of studies evaluated and excluded through the stages of the literature review. The search of the MEDLINE database yielded 146 prospective cohort studies describing 361 subcohorts and 43 RCTs involving 51 subgroups. Most of the cohort studies (125 [86%]) were primary prevention studies, whereas 32 RCTs (74%) were secondary prevention trials.
Supplementary appendix 4 summarizes the characteristics and results of the included prospective cohort studies by dietary exposure. Among the 361 subcohorts included in the review, 201 were from the United States, 130 were from Europe, and 12 were from Asia. For each dietary factor, an average of 29 209 individuals were included, and the median length of follow-up was 11 (range, 2.8-28) years; mean age was 53 years; and 41% were women. Most of the studies (89%) used a food frequency questionnaire to assess dietary intake..
Supplementary appendix 5 summarizes the characteristics and results of the included RCTs by dietary exposure. For each dietary variable, there was an average total of 7204 individuals, and the median length of follow-up was 3.7 (range, 1-12) years; mean age was 58 years; and 36% were women.
POOLED ESTIMATES FROM COHORT STUDIES
The pooled analyses of cohort studies showed that an increased consumption of alcohol (RR [95% CI] for heavy consumption, 0.69 [0.64-0.75]; for light/moderate consumption, 0.71 [0.67-0.75]), dietary beta carotene (0.73 [0.65-0.82]), fiber (0.78 [0.72-0.84]), fish (0.81 [0.70-0.92]), total folate (0.68 [CI, 0.57-0.79]), dietary folate (0.62 [0.50-0.79]), fruits (0.80 [0.66-0.93]), marine {omega}-3 fatty acids (0.86 [0.75-0.97]), monounsaturated fatty acids (0.80 [0.67-0.93]), nuts (0.70 [0.57-0.82]), vegetables (0.77 [0.68-0.87]), total vitamin C (0.82 [0.71-0.92]), dietary vitamin C (0.80 [0.68-0.91]), total vitamin E (0.77 [0.66-0.89]), dietary vitamin E (0.77 [0.55-0.99]), and whole grains (0.81 [0.75-0.86]) and increased Mediterranean (0.63 [0.53-0.72]) and high-quality diet patterns (0.63 [0.45-0.81]) were each associated with a significantly lower risk of CHD. Conversely, an increased consumption of trans-fatty acids (RR, 1.32 [95% CI, 1.16-1.48]) and foods with a high glycemic index (1.32 [1.10-1.54]) were associated with a significantly higher risk of CHD (Table 2).
Higher intake of {alpha}-linolenic acid, eggs, meat, milk, polyunsaturated fatty acids, saturated fatty acids, total fat, and ascorbic acid (vitamin C) and vitamin E supplements and prudent and western diet patterns were not significantly associated with CHD (Table 2). Among studies of higher methodologic quality, prudent (RR, 0.73 [95% CI, 0.62-0.83]) and western (1.55 [1.27-1.83]) diet patterns were each associated with CHD. In keeping with previous information, fish intake was protective against fatal CHD (RR, 0.83 [95% CI, 0.71-0.94]), but marine {omega}-3 fatty acids (0.88 [0.66-1.11]) and {alpha}-linolenic acid (0.98 [0.60-1.36]) were not.
Table 3 shows the summary RRs for each dietary exposure, stratified by dietary assessment tool, sex, geographic region, and type of prevention strategy. Primary prevention studies indicate favorable associations for vitamin E, beta carotene, vitamin C, and fiber intake, whereas secondary prevention studies show beneficial associations for fish and {omega}-3 fatty acid intake. The intake of marine {omega}-3 fatty acids was beneficial in women. The summary estimates are generally consistent across the strata of other potential effect modifiers.
POOLED ESTIMATES FROM RCTs
The pooled analysis of RCTs showed that increased consumption of marine {omega}-3 fatty acids (RR, 0.77 [95% CI, 0.62-0.91]) and a Mediterranean dietary pattern (0.32 [0.15-0.48]) were each associated with a significantly lower risk of CHD. Higher intake of beta carotene supplements, fiber, fish, folate supplements, fruits and vegetables, and polyunsaturated fatty acids relative to saturated fatty acids, total fat, and ascorbic acid and vitamin E supplements were not significantly associated with CHD. The summary RRs were generally consistent across the strata of methodologic quality (Table 4).
Cohort studies provide abundant evidence of an association with total mortality for many dietary exposures (supplementary appendix 6). Randomized controlled trials corroborate these associations for the consumption of {omega}-3 fatty acids and a Mediterranean diet because most of the other dietary factors have not been evaluated to date.
APPLICATION OF BRADFORD HILL GUIDELINES FOR CAUSALITY
Agreement of observed data from cohort studies with Bradford Hill criteria for each dietary exposure is presented in Table 2. Strong evidence supports a causal link with CHD (³4 criteria) for several protective factors, including intake of vegetables and nuts and Mediterranean and high-quality dietary patterns, and for harmful factors, including consumption of trans-fatty acids and foods with a high glycemic index or load. There is also strong evidence supporting a valid association for monounsaturated fatty acid intake and a prudent diet (protective factors) as well as a western diet (harmful factor) among studies of high methodologic quality. Moderate evidence (3 criteria) for associations exists for intake of fish, marine {omega}-3 fatty acids, folate, whole grains, dietary vitamins E and C, beta carotene, alcohol, fruit, and fiber. Weak evidence (²2 criteria) is present for protective factors, including intake of supplementary vitamin E and ascorbic acid, polyunsaturated fatty acids, {alpha}-linolenic acid, eggs, and milk, and for harmful factors, including intake of meat, saturated fatty acids, and total fat. In separate analyses of the biological gradient (supplementary appendix 7), evidence of a linear dose-response with CHD is found for most of the dietary exposures except for monounsaturated fatty acids and total fat, folate, fruits, {omega}-3 fatty acids, fish, vitamin C, eggs, and milk.
As shown in Table 4, sufficient support from RCTs to satisfy the criterion for experimental evidence is observed only for marine or total {omega}-3 fatty acid intake and a Mediterranean dietary pattern. Little or weak evidence from RCTs is found for consumption of fruits, vegetables, fish, fiber, polyunsaturated fatty acids, and total fat and supplemental intake of beta carotene, vitamin E, ascorbic acid, and folate. Other factors have not been evaluated singly in clinical trials (Table 4). The evidence from RCTs agrees with the Bradford Hill results from cohort studies for intake of ascorbic acid and vitamin E supplements, polyunsaturated fatty acids, and total fats and a Mediterranean dietary pattern, but disagree for fish consumption, which shows moderate evidence of a causal link with CHD in cohort studies but virtually no effect in RCTs. Among the factors with strong evidence of causation, only overall healthy dietary patterns are significantly associated with CHD in RCTs (Table 5).
METHODS
SEARCH STRATEGY AND STUDY SELECTION
We searched the MEDLINE database for prospective cohort studies and RCTs from 1950 through June 2007. The bibliographies of retrieved articles were scanned for additional cohort studies and RCTs. Two of us (A.M. and L.D..) independently assessed study eligibility. Excluded studies and reasons for exclusion were listed, and disagreement was resolved by discussion and consensus. We included original English-language articles pertaining to the effect of diet on the following primary outcomes: coronary or ischemic heart disease and fatal or nonfatal myocardial infarction. These articles were also evaluated for the following secondary outcomes: angina pectoris, sudden death, cardiovascular disease, and total mortality. Relative risks (RRs) of outcomes are presented with their 95% confidence intervals (CIs) after adjusting for potential confounders. We only considered studies that followed up subjects for at least 1 year. Cohort studies had to include estimates of dietary intake using conventional dietary assessment tools (eg, food frequency questionnaires, food records, or 24-hour diet recall). Clinical trials had to be randomized and to compare dietary exposure with a control diet or a placebo. Crossover trials were excluded if plasma biomarkers or atherosclerotic indicators were not evaluated because coronary outcomes occurring after a crossover would be difficult to interpret.
DATA EXTRACTION
The following data were extracted from the selected studies: (1) study design (cohort study or RCT, ie, parallel or factorial); (2) country of origin; (3) number of participants; (4) characteristics of participants (ie, age and sex); (5) dietary assessment tool; (6) intake dosage; (7) length of follow-up; (8) description of the interventions (RCTs); and (9) cardiovascular disease outcomes.
For the prospective cohort studies, data were extracted on the estimates of the association between the dietary exposure and disease outcome. The cohort studies typically report estimated intake as quantiles (usually quartiles or quintiles). The higher intake level is compared with the lowest intake level and the results are reported as odds ratios or RRs for each clinical outcome. The P values for trend, where available, were used to evaluate dose-response relationships. For RCTs, we compared the RRs of disease outcomes between the dietary intervention and control groups. The cohort studies and RCTs were stratified according to risk of bias (methodologic quality), as determined using the criteria in supplementary appendix 1 (this and the other appendices are available on request from the authors).17-18 The different dietary patterns evaluated in studies were noted. The "Mediterranean" dietary pattern emphasizes a higher intake of vegetables, legumes, fruits, nuts, whole grains, cheese or yogurt, fish, and monounsaturated relative to saturated fatty acids. The "prudent" dietary pattern is characterized by a high intake of vegetables, fruit, legumes, whole grains, and fish and other seafood. The "western" pattern is characterized by a high intake of processed meat, red meat, butter, high-fat dairy products, eggs, and refined grains.
APPLICATION OF BRADFORD HILL GUIDELINES
A modified algorithm of the Bradford Hill criteria was used to systematically evaluate the evidence of a causal relationship between each dietary exposure and CHD.14 As summarized in Table 1, the following 4 criteria were used in the review of cohort studies: strength, consistency, temporality, and coherence. A fifth criterion, biological gradient, was not included in the algorithm because dietary exposures may not exhibit dose-response relationships with CHD owing to possible threshold effects or "j-shaped" relationships (Table 1). Nevertheless, we explored evidence of this criterion separately.
The 4 Bradford Hill criteria were used to derive a causation score for each dietary exposure. The score was computed as the unweighted sum of the number of criteria that were met, for a possible range of 0 to 4. A score of 4 was considered strong evidence of a cause-and-effect relationship between the dietary exposure and disease. A score of 3 was deemed to indicate moderate evidence of causation. A score of 2 or less was considered a reflection of weak evidence of causation. As a final step, a sixth criterion, experiment, was used to examine whether the evidence from cohort studies was consistent with that from RCTs. Three criteria were omitted, including plausibility, specificity, and analogy because these factors were already satisfied by default (plausibility), considered nonspecific to CHD (specificity), or deemed to be subjective (analogy) (Table 1).
STATISTICAL ANALYSES
We used commercially available statistical software (Comprehensive Meta Analysis software, version 2.2 [Biostat, Englewood, New Jersey] and SAS, version 9.1 [SAS Institute Inc, Cary, North Carolina]). Statistical heterogeneity across studies was assessed using the Q statistic,19 with significant heterogeneity for all of the dietary predictors except for beta carotene, eggs, monounsaturated fatty acids, trans-fatty acids, nuts, whole grains, and a "Mediterranean" dietary pattern. Summary estimates were calculated using a general variance-based method (random-effects model) with 95% CIs.19 Because the potential confounders considered in multivariate analyses vary across studies, we used the parameter estimates in the most complex model, which typically include demographic, lifestyle, and dietary factors. Stratified analyses were also conducted by methodologic quality, sex, type of dietary assessment tool (food frequency questionnaires vs food records and 24-hour recall), continental region (North America, Europe, and Asia), and type of prevention strategy (primary vs secondary) to assess the influence of these factors on the observed associations between diet and CHD.
INVESTIGATION OF HETEROGENEITY
The comparison of quantile extremes to compute RRs in cohort studies may add heterogeneity to summary RR estimates because the mean and median levels of intake in the quantile groups varied from study to study. To examine this, we created a scatterplot and linear regression for each dietary exposure, with the RR value for a given study on the y-axis and the difference in mean or median intake between quantile extremes on the x-axis.. A slope of greater than or less than 0 would suggest that the differences in mean intake across studies influenced the summary RR value, whereas a slope of 0 would suggest that differences in mean intake across studies probably have a negligible effect on the summary estimates. As shown in supplementary appendix 2, no relationship was observed for {omega}-3 fatty acids (as well as other dietary exposures; data not shown), suggesting that different quantile values across studies likely cannot explain the variation in RR values. Finally, to assess whether the results might be explained by certain dietary factors being studied more frequently or the possible reporting of mostly positive results, we examined the relationship between the summary RR values derived from the cohort studies and the sample size from all cohort studies pooled together for each particular dietary factor, with separate plots for beneficial and harmful dietary exposures. A slope of 0 would suggest that a propensity to study certain dietary factors more than others likely does not account for the variation in the effects across dietary exposures. No associations were observed (supplementary appendix 3).
|
|
| |
| |
|
|
|
