CONTRIBUTIONS TO DIETARY ASSESSMENT METHODOLOGY

Section:

ChooseTop of pageAbstractCONTRIBUTIONS TO DIETARY ... <<CONTRIBUTIONS TO DIETARY ...CONCLUSIONSREFERENCESCITING ARTICLES

The NHS was initiated in 1976 to assess the long-term effects of oral contraceptive use. Realizing that diet and physical activity might be key modifiable determinants of chronic disease risk, we extended the scope of the study to include dietary intake and physical activity assessment. After a series of pilot studies, a 61-item SFFQ was mailed in 1980 and was returned by approximately 95 000 women in the NHS. An expanded questionnaire containing 130 foods was developed in 1984. With minor modifications to account for changes in the food supply and new hypotheses, this questionnaire has been administered every 4 years in both the NHS since 1986 and the NHS II since 1991 (http://www.channing.harvard.edu/nhs/?page_id=246questionnaires).

Food frequency questionnaires assess usual frequencies of intake using a structured list of foods over a specified period of time in the past; in the NHSs, we have focused on food intake over the previous year.² For the NHS, we developed a semiquantitative questionnaire that incorporates information about usual portion sizes (e.g., asking about “one glass of milk” as opposed to “milk”).³ An SFFQ was the ideal choice for the large-scale NHSs because it was relatively inexpensive, was easy to administer by mail, and had low participant burden.

As an example, the use of repeated SFFQs in the NHSs allowed us to examine how cumulative long-term diet⁴ and changes in diet over time are associated with coronary heart disease (CHD) risk over more than 3 decades,⁵ identifying multiple dietary risk factors for CHD prevention. Periodic diet assessments using SFFQs also enabled us to understand how associations observed with chronic diseases are affected by different ways of modeling dietary exposures. We found cumulatively averaged dietary intake over the entire follow-up to be most strongly associated with CHD risk relative to baseline or the most recent measures of diet,⁶ confirming the long etiologic period involved in the relationship of diet with CHD.

Food frequency questionnaires are invaluable in the study of diet during childhood or adolescence in relation to diseases in later life. For example, in 1998, NHS II participants, then aged 34 to 51 years, were invited to complete a comprehensive questionnaire about their diet while in high school, and 47 355 did so. The validity of this retrospective questionnaire was documented in several ways (including comparison with maternal reporting and with 24-hour recalls and SFFQs collected during high school). We found that adolescent red meat intake was positively associated with and dietary fiber intake was inversely related to the risk of breast cancer. These finding are notable because these relationships were not apparent when we examined diets in midlife and later.

Like all measurements, SFFQs are susceptible to random and systematic error. The evaluation of the reliability and validity of diet assessment methods, particularly SFFQs, have been major NHS contributions. In the first of a series of validation studies,³ we asked a subsample of the original NHS cohort (n = 173) to complete four 7-day weighed diet records (7DDRs) over the year following the initial 1980 SFFQ, after which we readministered the SFFQ. The SFFQ nutrient intakes had reasonable validity relative to the average of the four 7DDRs (for energy-adjusted nutrients; average r = 0.50). Although we cannot rule out correlated errors between the 2 methods, which would overestimate the validity coefficients, there are likely independent errors in 7DDRs that would make these validity coefficients underestimates of true validity. The original NHS has become one of the standards for dietary validation and calibration study design. In 1986, we repeated this design with the revised, expanded SFFQ among 191 NHS participants (of which 92 nurses had participated in the original 1980–1981 validation study).² The average correlation for all nutrients was 0.63 after adjustment for energy intake and variation in 7DDRs.

Together, these 2 studies provided invaluable information about the validity of SFFQs. They helped to establish that a relatively simple instrument, such as an SFFQ, can give researchers approximately as much information as a far more expensive and time-consuming 7DDR. We also found that correlation coefficients between the SFFQ and 7DDR intakes can be improved through energy adjustment and by correcting for within-person variation in the reference method. Correlations of dietary intakes assessed by the average of the 1980, 1984, and 1986 SFFQs with intakes from the average of the 1980 and 1986 7DDRs were higher than were those reported within the individual validation studies, indicating that there is considerable improvement in measuring long-term diet using multiple SFFQs (deattenuated correlation coefficients were 0.71 for total fat, 0.80 for saturated fat, 0.79 for dietary cholesterol, and 0.55 for protein).² Lastly, we found that SFFQs can measure change in dietary intake over a 6-year period when compared with the 6-year change assessed by 7DDRs. The low cost, ease of administration, and demonstrated reproducibility and validity of the SFFQ have made it one of the most widely used diet assessment methods in large-scale, long-term observational studies. Similar questionnaires have been developed to fit the dietary and cultural contexts of many different study populations.⁷

In addition to using diet records or multiple 24-hour recalls as reference methods in validation studies of SFFQs, we have examined correlations with nutrient biomarkers. Although these provide only a qualitative assessment of validity because biomarker levels are affected by many factors in addition to intake, documentation of an association is valuable because the errors are usually independent. In the NHS, intakes of linoleic acid, trans fatty acids, and long chain N-3 fatty acids correlated well with levels in adipose tissue or red blood cells.^8–10 Providing objective evidence for the validity of total fat intake has been elusive because it has no specific biochemical indicator. However, fasting plasma triglyceride levels are known to increase with replacement of carbohydrate by fat; so these can provide an indirect indicator of dietary fat. After adjustment for several determinants of triglyceride levels, we found that triglyceride levels were nearly twice as high in the highest compared with the lowest fat intake.¹¹ These findings provide evidence that the SFFQs used in these cohorts can assess both total dietary fat and specific fatty acids with reasonable validity. We also have documented that intakes of specific carotenoids, vitamin E, folate, and vitamin D are correlated with their corresponding plasma levels.

We recently completed the Women’s Lifestyle Validation Study, a far more extensive validation study of our dietary and physical activity assessment methods that includes more than 700 women, about half each from the NHS and NHS II. Participants provided multiple measures of our SFFQs and physical activity questionnaires, four 24-hour recalls of diet and activity, two 1-week diet records, two 1-week accelerometer measurements, four 24-hour urine samples (for protein, sodium, and potassium), and 2 blood samples (for fatty acids, carotenoids, and B vitamins). Energy expenditure was assessed by doubly labeled water.

Despite changes in the food supply and in eating behaviors over time, the findings comparing nutrients calculated from the SFFQ with those from weighed diet records almost exactly replicated the findings reported from our 1986 validation study.² In analyses using plasma biomarkers of diet as the comparison method, the SFFQ had validity similar to a 1-week weighed diet record and superior to the average of three 24-hour recalls.¹² In 2003, using absolute and energy-adjusted protein intakes as examples, the authors of the Observing Protein and Energy Nutrition study suggested that validation studies employing diet records or recalls as the comparison method might seriously overstate validity stemming from correlated errors.¹³ However, in the larger and more comprehensive Women’s Lifestyle Validation Study, we found that the validity of energy-adjusted protein intake assessed by our SFFQ was virtually identical whether compared with intake estimated by weighed diet records, multiple 24-hour recalls, or 24-hour urinary nitrogen and doubly labeled water.^12,14 The Women’s Lifestyle Validation Study provides a uniquely detailed examination of sources of error in diet and activity questionnaires. In conjunction with earlier validation studies, these permit more precise adjustments for measurement errors, a potentially important source of bias in diet assessment.

Over the past few decades, we have developed and documented methods for correcting random within-person errors in correlation coefficients and errors in measures of association in the presence of differential and nondifferential measurement error while adjusting for multiple confounding factors.² These techniques have allowed more valid estimates from large cohort studies with the use of SFFQ data.

The associations we have documented between diet and various diseases in prospective analyses provide further evidence that our SFFQs are able to detect important biological relationships. For example, we have found associations between red meat and adenomas and cancer of the colon and CHD; low cereal fiber and CHD and T2D; regular consumption of nuts and lower risk of CHD and mortality; red and processed meat and T2D; glycemic index and load and T2D; soda consumption and weight gain and T2D and cardiovascular disease; coffee consumption and low risk of T2D and cardiovascular mortality; low dietary fiber and diverticular disease of the colon and constipation; low fruit fiber and hypertension; saturated fat (vs unsaturated fat) and CHD; trans fat (vs cis unsaturated fat or carbohydrate) and CHD and T2D; low dietary lutein or zeaxanthin and cataracts; calcium and potassium and kidney stones (inverse); low folic acid and colon adenomas and colon cancer; and low vitamin D and hip fractures.

We have also documented associations between biomarkers of dietary components that are difficult to assess using self-report measures and various disease endpoints. For instance, the NHS was the first study to objectively measure the intake of selenium in toenail samples collected from 68 000 participants between 1982 and 1984, and it found no association with breast cancer incidence. We also performed one of the first epidemiological investigations of a biomarker of acrylamide (hemoglobin adduct of acrylamide), a putative carcinogen, and found no association with ovarian cancer.

CONTRIBUTIONS TO DIETARY GUIDELINES AND NUTRITIONAL POLICIES

Section:

ChooseTop of pageAbstractCONTRIBUTIONS TO DIETARY ...CONTRIBUTIONS TO DIETARY ... <<CONCLUSIONSREFERENCESCITING ARTICLES

When formulating dietary guidelines, randomized controlled trials are typically considered the highest level of evidence. However, they are usually not the most appropriate or feasible study design for answering questions regarding long-term effects of diet on chronic disease risk because the drug trial paradigm does not neatly apply to dietary hypotheses. Diet randomized controlled trials can be difficult to blind, are rarely placebo controlled, often have high drop-out rates, tend to have poor compliance, can suffer from obsolescence, and often do not consider substitution effects or interactions.¹ Also, in contrast to drug trials, the control arm does not have zero exposure.

Assessment of effects that occur over many years or decades—highly plausible for cancer and some other outcomes—would be particularly difficult in a randomized controlled trial. Thus, prospective cohort studies are invaluable for making evidence-based recommendations, because they are often the strongest study design for causal inference in the absence of large randomized controlled trials on hard endpoints. When possible, data from small randomized trials with intermediate endpoints and findings from long-term large prospective cohort studies with disease endpoints should be used as complementary evidence; together these can provide a high level of certainty regarding causality to inform dietary guidelines.

Every 5 years, the US Departments of Health and Human Services and Agriculture jointly convene the Dietary Guidelines Advisory Committee to systematically review evidence on diet and health, and the Dietary Guidelines Advisory Committee’s findings are used to update the US Dietary Guidelines.¹⁵ Over the years, the findings from the NHSs have contributed in a major way to these guidelines as well as other policies at the federal, state, and local levels. A prime example is that of trans fatty acids (TFAs) resulting from the partial hydrogenation of vegetable oils. Developed to stabilize vegetable oils and to create solid fat that mimics lard and butter, TFAs were extensively used in food manufacturing starting in the early 20th century. Because partial hydrogenation alters the structure of essential fatty acids, we became concerned about the long-term health consequences of TFAs and have incorporated details on their food sources in our SFFQs from 1980. Simultaneously, we created a unique food composition database for TFAs with direct analyses of foods that have been updated every 4 years. Using 8-year follow-up data from the NHS, we published the first large prospective study documenting a positive association between trans fat intake and incident CHD, with similar results in updated analyses.

During the early 1990s, controlled feeding studies also documented adverse effects on blood lipids and other CHD risk factors.¹⁶ On the basis of this evidence, in 2003 Denmark became the first country to restrict the availability of TFAs in the food supply,¹⁷ followed by several other countries, including Austria, Hungary, Norway, Switzerland, Iceland, Sweden, Canada, Brazil, Chile, Argentina, and South Africa.^18,19 In 2005, on the basis of a review of observational cohorts and controlled feeding studies, the US Dietary Guidelines added a recommendation to limit the intake of trans fats.²⁰ In 2006, the Food and Drug Administration required manufacturers to list TFAs in the nutrition facts panel,²¹ and in 2015, they required that TFAs be eliminated by 2018 because of their determination that partially hydrogenated oils are not “generally recognized as safe.”²² Observational findings from the NHS cohorts provided key evidence of harmful effects of this dietary exposure on disease risk, resulting in a wide impact on global nutrition policy and in improved health.

“Negative” findings can also be important. Beginning in the 1980s, national and international dietary guidelines focused on reducing the consumption of total fat with the belief that this would substantially reduce risks of cancers of the breast and other sites. The NHS was the first prospective study to examine the relation between dietary fat and breast cancer risk in detail. No relation was seen in this initial report, follow-up analyses, or a pooling study of prospective studies, even after correction for measurement error using the methods and validation data we have described. This and null findings for many outcomes from the NHSs and from other cohorts, as well as limited evidence from randomized controlled trials, led to the removal of the restriction on total fat in the 2015 US Dietary Guidelines.²³ Reports from the NHS on CHD risk also highlighted the importance of types of fats. Saturated and trans fats were associated with increased risk, and mono- and polyunsaturated fats with lower risk. These findings have provided strong support for dietary guidelines that have emphasized specific types of fat rather than total fat.²³

More recently, several reports from the NHSs have documented significant positive associations between a higher consumption of sugar-sweetened beverages (SSBs) and an increased risk of weight gain, obesity, type 2 diabetes (T2D), CHD, and stroke. These findings, together with randomized controlled trials showing that a decreasing consumption of SSBs reduces excess weight in children and adolescents,^24,25 have provided a strong scientific foundation for developing polices to reduce the consumption of SSBs. Such policies have included increasing taxation in Mexico,²⁶ removing SSBs from most schools in the United States,²⁷ and banning the sales of SSBs in public buildings in some cities, such as Boston, Massachusetts.²⁸ The 2015 Dietary Guidelines Advisory Committee report²⁹ recommends limiting SSB consumption and substituting healthy beverages such as water and suggests that the Food and Drug Administration include the amount of added sugars and percentage daily value in the revised nutrition facts label.³⁰ Following the advice of the Dietary Guidelines Advisory Committee, the 2015 US Dietary Guidelines recommend that no more than 10% of daily caloric intake come from added sugar,²³ consistent with the upper limit recommended by the World Health Organization.³¹

Analyses using the NHSs have also helped resolve key controversies in the dietary etiology of chronic diseases. For instance, longstanding recommendations to limit saturated fat intake to 10% of energy intake³² were questioned by a recent meta-analysis concluding that saturated fat intake has no significant relationship with CHD risk.³³ However, the meta-analysis failed to specify the replacement macronutrient for saturated fat. In 1997, we showed that replacing 5% of energy from saturated fat with the same amount of energy from carbohydrates was not associated with CHD risk, but replacing the same amount of energy with polyunsaturated fat was associated with a significantly reduced risk of CHD. Subsequent analyses with updated follow-up and using different energy-adjustment models⁶ confirmed these results.

In a more recent NHS analysis, saturated fat intake was not associated with CHD risk when compared with refined starch and added sugars but was positively associated with CHD risk when compared with intakes of unsaturated fats or whole grains. These analyses, consistent with the effects on blood lipids, underscore the importance of choosing healthy types of fats and carbohydrates in reducing cardiovascular risk. Of note, the energy-adjustment methodology that was pioneered by the NHS investigators³⁴ has played a key role in these analyses. Adjusting for energy intake is now standard practice in nutritional epidemiology, not only to understand substitution effects but also to control for potential confounding by determinants of energy intake (e.g., physical activity and body size), remove extraneous sources of variation, and understand how dietary composition relates to the risk of disease, which is more amenable to modification by individuals than is absolute intake.²

The NHSs have also been instrumental in going beyond the single nutrient approach to examine the total diet holistically with respect to disease risk. This allows dietary recommendations to be more translatable, because it may be easier for people to understand and adopt recommendations regarding dietary patterns, as opposed to optimizing many different nutrients.³⁵ There are 2 main approaches to dietary pattern analysis.³⁶ In the empirical approach, such as principal component analysis, statistical methods are used to derive predominant dietary patterns from existing dietary data. Using this method, we identified 2 dominant dietary patterns in the NHS: a “prudent” diet that is high in the intake of whole grains, fruits, vegetables, low-fat dairy, poultry, and fish, and a “Western” pattern that is high in the intake of red and processed meat, butter, high-fat dairy, French fries, desserts, and refined grains. These patterns have been robustly associated with a wide range of health outcomes, including CHD, stroke, T2D, obesity, and premature deaths.

The second a priori approach to developing a dietary pattern uses existing knowledge of a hypothesized healthy diet to create a dietary index. These include the Dietary Approaches to Stop Hypertension,³⁷ the Mediterranean diet,³⁸ and the Healthy Eating Index (HEI), which was developed by the US Departments of Health and Human Services and Agriculture to measure adherence to the 1995 US Dietary Guidelines.³⁹ The HEI-1995 did not predict a composite of major health outcomes, raising doubts about these guidelines, which focused on a reduction of total fat intake. To better capture adherence to a diet that significantly predicts chronic disease risk reduction, we created a modified version of the HEI, called the Alternate HEI (aHEI); this was significantly better at predicting the risk of major chronic diseases than was the original HEI.⁴⁰ The US Departments of Health and Human Services and Agriculture then created the HEI-2005 to reflect updated dietary guidelines,⁴¹ which also performed better than did the original HEI.

Following the release of the 2010 dietary guidelines, the US Departments of Health and Human Services and Agriculture created HEI-2010, which had most of its elements in common with aHEI-2010.⁴² This most updated version was able to predict chronic disease mortality risk almost as well as did the aHEI-2010.⁴³ The aHEI-2010 has also been used to evaluate US trends in diet quality using data from the nationally representative National Health and Nutrition Examination Survey. It shows steady, modest improvements since 2000, mainly because of an almost complete elimination of TFAs and an important reduction in SSB consumption.⁴⁴ We combined these results with data from our cohorts and estimated the improvements to have prevented more than 1 million premature deaths and to have substantially reduced the number of incident cases of T2D (by 12.6%) and cardiovascular disease (by 8.6%). This is in line with an earlier study documenting an improvement in blood lipid levels from 1988 to 2010 in the same National Health and Nutrition Examination Survey populations, which is thought to be largely because of decreases in trans fat intake.⁴⁵ We have also demonstrated inverse associations between the Dietary Approaches to Stop Hypertension and Mediterranean dietary patterns and various health outcomes in the NHSs.

Because of the relevance of dietary patterns to overall health, the 2015 Dietary Guidelines Advisory Committee report²⁹ and the 2015 US Dietary Guidelines²³ have put much greater emphasis on recommending healthy dietary patterns. On the basis of a systematic review of evidence from randomized controlled trials and observational cohorts, which included many reports published on the basis of the NHSs, the committee identified 3 dietary patterns that are associated with a reduced risk of chronic diseases and improved diet quality: the healthy US-style pattern, the healthy Mediterranean-style pattern, and the healthy plant-based dietary pattern. These 3 patterns have several elements in common: a higher consumption of fruits, vegetables, whole grains, nuts and seeds, legumes, low-fat dairy, and seafood and a lower consumption of red and processed meats, sugar-sweetened foods and beverages, and refined grains. The 2015 Dietary Guidelines indicate that the core features of a healthy diet can be attained through multiple approaches by incorporating varying personal preferences in differing cultural contexts.