Nutritional Epidemiology Lecture Notes PDF

Nutritional epidemiology Session 2, October 16, 2024 Basic Lecture Series 1 UN790 Doctoral Program in Epidemiology WS 2024/25 Selma Kronsteiner-Gicevic...

Nutritional epidemiology Session 2, October 16, 2024 Basic Lecture Series 1 UN790 Doctoral Program in Epidemiology WS 2024/25 Selma Kronsteiner-Gicevic, ScD, MSc Postoctoral Researcher [email protected] Nutritional Epidemiology Department of Epidemiology, Center for Public Health 1 Lecture overview What is nutritional epidemiology Why diet? Defining and operationalizing diet: a complex exposure Selected methodological issues in nutritional epidemiology Study design Measuring diet intake Measurement error and its implications Examples of classical achievements in nutritional epidemiology and new directions Department of Epidemiology, Center for Public Health 2 What is nutritional epidemiology? “Nutritional Epidemiology is a one of the younger branches of epidemiology that focuses on the study of the relationship between diet, nutrition, and health outcomes in populations. This field integrates nutritional science with epidemiological methods to understand how dietary factors contribute to the occurrence and prevention of disease.” Scholar GPT 3 Goals of nutritional epidemiology Monitor food consumption, nutrient Generate new hypotheses about intake and nutritional status of a relationships of diet and disease population over time Goals Contribute to the prevention of nutrition-related disease and Improve dietary assessment methods improvement of public health (inform policy) Goals of nutritional epidemiology Monitor food consumption, nutrient 1. Evaluate literature intake and nutritional status of a Generate new hypotheses about relationships of diet and disease 2. Generate hypothesis, e.g.:time population over “High dietary salt intake is associated with stomach cancer Goals risk” 3. Test hypothesis using data Contribute to the prevention of nutrition-related disease and Improve dietary assessment methods improvement of public health (inform policy) Goals of nutritional epidemiology Monitor food consumption, nutrient Generate new hypotheses about intake and nutritional status of a relationships of diet and disease population over time Goals -Develop and validate a Contribute to the prevention of food frequency nutrition-related disease and Improve dietary assessment methods questionnaire improvement (FFQ) of public health policy) (inform Goals of nutritional epidemiology Monitor food consumption, nutrient Generate new hypotheses about intake and nutritional status of a relationships of diet and disease population over time Goals -”Soda” tax Contribute to the prevention of -Fortifying salt with nutrition-related disease and Improve dietary assessment methods iodine improvement of public health (inform policy)  Why focus on diet in epidemiological research? Diet and burden Diet and m orbidity 250 million deaths in 2017 of disease Global Panel on Agriculture and Food Systems for Nutrition, 2016  Based on the results from a GDB Diet and burden of disease study, data from 200 countries Ezzati et al. Global Burden of Disease Study, 2013 Diet and premature mortality 11 million deaths in 2017 GBD 2017 Diet Collaborators. Health effects of dietary risks in 195 countries, 1990- 2017: a systematic analysis for the Global Burden of Disease Study 2017, 2019 Diet vs. other Global Burden of Disease risk factors T h e s ig n ific a n c e o f d ie ta r y a s s e s s m e n t H o w d ie t ra n k s a m o n g 2 5 G B D S t u d y ris k fa c to rs in te rm s o f a tt rib u ta b le m o rta lity Rank GBD Risk factors: 1 ▪ Environmental 2 ▪ Occupational ▪ Behavioral 3 ▪ Metabolic 4 (2 0 1 0 , b o th s e xe s , a g e d -sta n d a rd ize d ) G lo b a l B u rd e n o f D ise a s e S tu d y 2 0 1 0 , IH M E 12 Historical development Goldberger observations of corn, from nutrient deficiencies… milk, meat role> Scurvy (severe lack of vitamin C) epidemiology was key Beriberi (severe lack of thiamine/vitamin B1) to solving the mystery of pellagra Pellagra (severe lack of vitamin niacin/vitamin B3) Back in 1700´s From 1980´s… …to chronic diseases/NCDs -multiple causes -long latent periods -relatively low frequency -under or over-consumption of dietary components Double burden of malnutrition In developing countries: overnutrition and undernutrition co-exist, sometimes within a single household IFPRI, 2017 14 Defining and operationalizing diet in in epidemiological reseach Role of diet and nutrition in epidemiological research mediator exposure outcome effect modifier covariates/confounders Directed Acyclic Graph (DAG) (simplified) Diet vs. cigarette smoking as exposure? Why is there so much more research on adverse effects of cigarette smoking vs. diet? Why are not that many controversial research findings of cigarette smoking (e.g.: “smoking is good for you”)  Diet vs. cigarette smoking as exposure Diet Cigarette smoking Measuring: Measuring: - Difficult to measure Simple to measure - Inter-correlated aspects of nutrients Accurate reporting – brand, number and foods per day - Misreporting- not aware what is in Self or spouse reporting food Study design: Study design: We can be non-smokers/ quit smoking -Isocaloric (what we use to replace (y/n exposure) removed food is important) Diet as a complex exposure Nutrient Disease outcome Food Disease outcome Dietary Disease pattern outcome Exposure of interest: single nutrient Example: Folic acid and neural tube defect RR=0.27 Folic acid (400 mcg daily) Spina bifida risk Dietary pattern Spina bifida risk diluted association MRC Vitamin Study, 1991 Exposure of interest: food/food group Example: sugar-sweetened beverages and type 2 diabetes RR=1.41 1 or more SSB per day T2DM risk -SSBs main source of added sugar in US diet -contain high fructose corn syrup similar to sucrose -using food as exposure can be beneficial for public health messaging -Diet pattern can serve as a covariate in models Schulze et al., JAMA, 2004 Exposure of interest: diet pattern PREDIMED trial, Spain, 7447 participants, 3 diets, 4.8 years follow-up, outcome: combined heart attack, stroke or death Estruch et al, 2018, NEJM Why study dietary patterns? Complexity of the diet: we eat foods, not nutrients Dietary factors are correlated: interactions and synergies Effects of a single nutrient might be too small More easily translated into dietary guidelines and policy In general, research on dietary patterns tends to be more consistent across studies compared to single nutrient or food studies Clinical trials show positive health outcomes with changes in the “whole diet” for example Mediterranean diet. Does not allow for studying: Mechanisms individual effects of nutrients of foods Role of diet and nutrition in epidemiological research mediator exposure outcome effect modifier covariates/confounders Directed Acyclic Graph (DAG) (simplified) 24 Consumption of Olive Oil and Diet Quality and Risk of Dementia-Related Death In joint analyses, participants with the highest olive oil intake had a lower risk for dementia-related mortality, irrespective of their AMED score (28% to 34% lower risk compared with participants in the combined low olive oil and high AMED) and of their AHEI (27% to 38% lower risk compared with participants with low olive oil and high AHEI) JAMA Network Open 2024 Confounding by Dietary Patterns of the Inverse Association Between Alcohol Consumption and Type 2 Diabetes Risk Alchohol consumption and 7-year risk of type 2 diabetes mellitus in 2,879 healthy adults enrolled in the Framingham Offspring Study (1991–2001) After adjustment for standard risk factors, consumers of ≥9.0 drinks/week had a significantly lower risk of type 2 diabetes mellitus compared with abstainers (hazard ratio = 0.47, 95% confidence interval (CI): 0.27, 0.81) Adjustment for selected nutrients had little effect on the hazard ratio, whereas adjustment for dietary pattern variables by factor analysis significantly shifted the hazard ratio away from null (hazard ratio = 0.33, 95% CI: 0.17, 0.64) The data suggest that alcohol intake, not dietary patterns associated with alcohol intake, is responsible for the observed inverse association with type 2 diabetes mellitus risk Imamura et al. 2009, AJE Diet as an outcome Diet quality index AHEI-2010 by socioeconomic status U.S. NHANES data Evaluating “upstream” predictors of diet quality Other examples: education level, age, gender, marital status, deprivation index, migration status, geographic location, etc. Wang et al. 2014, JAMA Study design issues in nutritional epidemiology A quick reminder: study types in epidemiology 29 The hierarchy of evidence (Cochrane)  Some reasons Inadequate study design – e.g. cross sectional vs. prospective Measurement error – e.g. inadequate data collection method Unmeasured confounding – e.g. not adjusting for important covariate …Findings may always be due to chance (due to random error) Let´s have a more detailed look! 33 Ecological studies/correlation studies in diet Useful, but: Aggregated data Correlation on group level do not always exist on individual level (“ecological fallacy”) Unmeasured confounding Temporal relation Cannot be reproduced correlation≠causation! r=0.85(m), 0.89(w) Armstrong and Doll, IJC, 1975 Cross-sectional studies Prone to “reverse causation” Persons change their eating habits due to condition E.g. CVD patients tend to eat healthier compared to general population BECAUSE of diagnosis Obese patients change their diet pattern IN ORDER to lose weight Result of cross-sectional design? wrong “Obesity is protective” “Healthy diet pattern is associated with a higher CVD risk” Case-control studies Recall bias: cases report their past eating habits differently from controls (e.g., obese vs. lean participants) Difficulties in recalling eating habits decades ago Temporal ambiguity: difficulty distangling whether diet was changed as a result of disease (reverse causation) Selection bias: may occur depending on how controls are selected (from which population) Not suitable for long-term exposures occurring decades before the disease (e.g., diet and cancer) 36 Experimental studies/RCTs Practical for studies of vitamins, trace elements (can be fed as pills) Limitations: Differential dropout Often not feasible leads to bias as study arms Long latent periods no longer exchangeable Long follow-up/ insufficient duration Poor adherence & Dropout Blinding sometimes impossible (food are not pills) Ethical issues when adverse effect suspected 38 Prospective studies (e.g. cohorts) Often optimal design in nutritional epidemiology Yet: prone to (unmeasured) confounding and bias (e.g. selection bias) Association ≠ causation Meta-analyses gone wrong ‘study-of-studies’ At best: a useful summary of a large pool of data if high-quality studies with similar groups of people and study methods are used. At worst: a mish-mash of findings from studies that differ significantly, essentially comparing apples with oranges that offer meaningless—or even misleading—conclusions. Making headlines and stirring up controversies: “This paper is bound to cause confusion. A central issue is what replaces saturated fat if someone reduces the amount of saturated fat in their diet. If it is replaced with refined starch or sugar, which are the largest sources of calories in the U.S. diet, then the risk of heart disease remains the same. However, if saturated fat is replaced with polyunsaturated fat or monounsaturated fat in the form of olive oil, nuts and probably other plant oils, we have much evidence that risk will be reduced.” Walter Willett, Former Chair of the Department of Nutrition at the Harvard TH Chan School of Public Health, 2014 41 “...no single study can provide a definitive answer and no study design is without limitations” (Hu & Willett, 2018) Frank Hu Walter Willett 42 To conclude In nutritional epidemiology research: RCTs are not always feasible Cross-sectional studies prone to reverse causation Case-control studies may be subject to recal and selection biases Prospective cohort studies (especially „pooled cohorts“) with a large number of covariates (i.e minimizing unmeasured confounding) often remain the best evidence for many disease outcomes Meta-analyses and systematic reviews‘ quality depends on the „what does in?“ Measuring diet in epidemiological research Diet assessment methods 44 How can we Objective measures measure diet? Observational methods New technology- supported methods Objective measures Subjective measures Biomarkers Blood (e.g., serum, plasma Food diary/ diet record folate) 24-Hour dietary recall Urine (e.g., 24h urinary Diet sodium) (24HR) assessment Toenails (e.g., selenium) Food frequency methods questionnaire (FFQ) Hair (e.g., zinc) Diet history Adipose tissue (e.g., fatty acids) Diet screener More recently: Metabolomics 45  Traditional diet assessment methods: questionnaires Types Characteristics Food diary/ diet record: often called “gold Self-report! Not objective standard”, prospective 24h diet recall: Day-to-day variation in intakes Retrospective: (if intake recorded on a single day) 24-Hour dietary recall (24HR): multiple ones FFQ: population specific, may be prone to adjusted for random error can also work! systematic error if important food not included Food frequency questionnaire (FFQ): measures Recall bias and respondent memory usual diet which is conceptually more relevant Diet records: Reactivity bias exposure than diet on day x Error in estimating amount of food eaten Diet screener (typically short): can work well for Food composition database (converting food screening purposes, adherence to diet in RCTs to nutrient data) 46 Traditional diet assessment methods: dietary biomarkers Types Characteristics Objective Recovery biomarkers: e.g., 24h „The real gold standard“ urinary nitrogen (protein intake), BUT: 24h urinary sodium But do not exist for all nutrients and Concentration biomarkers: e.g., foods! serum folate, carotenoids, High participant burden/invasive, specific fatty acids expensive to analyse, lab errors 47 Novel diet assessment tools: examples Smart glasses for diet intake and physical activity (Chung et al., 2018) Deep Neural Networks for Image- Based Dietary Assessment (Mezgec et al. 2021) Bromage et al. Video repository of novel tools https://app.jove.com/de/methods-collections/284/innovative-methods-of-dietary-assessment-and-analysis Identification of new biomarkers (using omics technologies) Metabolomics, measuring a large number of low-weight metabolites including essential amino acids and molecules not synthesized in the human body (e.g. trimethylamine N-oxide, TMAO as a marker of fish/animal-based foods). But what about TMAO from fish? 49 Choice of method: considerations Usually a mix of methods is chosen in large studies Relevant Diet on any specific day, or dietary exposures over longer time periods? exposure E.g. diet and colon cancer, diet and stroke Level of detail How much precision is needed in absolute intake? Or do we need to be able to rank individuals by their intake, i.e. would relative intakes be sufficient for evaluating associations between diet and disease? Cost How costly (e.g. personnel, equipment, data analysis etc.) are diet assessment techniques? What can we afford? Respondent How burdensome is the technique of choice for participants? burden Intrusion and risk How intrusive/risky is it for participants? Measurement How prone to measurement error? Is measurement error associated with error the method of acceptable nature and size? 50 To conclude Traditional measures: each method comes with a set of issues, there is no perfect method Novel methods: promising, but still not ready for prime time! Dietary biomarkers: while objective, they do not exist for each nutrient/food The best solution, for now, is using a mix of data collection methods in studies – choosing those with uncorrelated errors! 20. Januar 2023 51 Measurement issues in nutritional epidemiology Measurement error, its implications, and correcting for measurement error 52 Types of measurement error Diet assessment methods - FFQ, 24h diet recall, diet record are always biased! How to measure/quantify bias in order to minimize it? Measurement error Random Systematic ‘noise’ ‘bias’ e.g. day-to-day variation e.g. over-reporting “good foods” in food intake and under-reporting “bad foods” (social desirability bias) Type of measurement error Willet, W. 2013 “Nutritional Epidemiology” textbook Nature of variation in diet e.g. interviewer effect, seasonal effect 𝑁𝑢𝑡𝑟𝑖𝑒𝑛𝑡 𝑌 = ᶙ+ 𝑠𝑢𝑏𝑗𝑒𝑐𝑡𝑖 + 𝑓𝑎𝑐𝑡𝑜𝑟 𝑋 + 𝑑𝑎𝑦 𝑜𝑓 𝑤𝑒𝑒𝑘 + ԑ Day-to-day variation in dietary intake Day-to-day variation in intake Day of week, weekend vs. weekday Seasonal variation Simplified model (e.g. summer vs. winter) Nutrient Y= ᶙ+𝑠𝑢𝑏𝑗𝑒𝑐𝑡𝑖 + ԑ Interviewer effect Between-person variation Within-person variation Willet, W. 2013 “Nutritional Epidemiology” textbook Day-today variation of macro vs. micronutrients Micronutrients, macronutrients and foods vary differently Fat variation constrained by total energy intake Vitamin A concentrated in small number of foods (e.g. carrot) with a high day-to-day variation Foods: episodically consumed foods like fish or nuts require large number of 24h diet recalls/diet records (or a FFQ!) 194 women, four 7-day diet records Willet, W. 2013 “Nutritional Epidemiology” textbook Estimating “true” intake (i.e. usual intake) for individual Willet, W. 2013 “Nutritional Epidemiology” textbook Within-person variation in intakes in a group Example NHANES survey: distribution of intakes using 1 or 2 days of data Exposure of interest: true intake in a group/population A single day data used: true mean for a group, but large SD („thick tails“) Solution: corrected 24h recall using 2nd day data and posthoc statistical methods Effects of random within-person variation on measures of association in epidemiological studies (attenuation bias) Hypothetical example Exposure of interest: Nutrient intake (no error in true intake of individual Sw2/Sb2=3 measurement) Measures of association High Low OR=(0.31/0.69) / (0.16/0.84) = Cases a=0.31 b=0.69 are attenuated: 2.36 Noncases c=0.16 d=0.84 Pearson correlation coefficient Nutrient intake (random error in Misclassification measurement) Regression coefficient High Low (beta) Cases a’=0.40 b’=0.60 Risk ratio, odds ratio OR=(0.40/0.60) / (0.31/0.69) = Noncases c’=0.31 d’=0.69 1.53 Willet, W. 2013 “Nutritional Epidemiology” textbook Implications of and correcting for measurement error Within-person Between persons Random Day-to-day variation in dietary intake Using single 24h diet recall data Result: true mean, large SD Result: true mean (law of large numbers), exaggerated SD, attenuated measures of Solution: multiple replicates association (attenuation bias) (reproducibility); correct for error using post-hoc methods Solution: use at least 2 days and posthoc methods to correct for the error (e.g. ANOVA to partition the variance to between- and within- components) Systematic Person repeatedly underreporting Some are over-reporters and others are unhealthy foods/social desirability bias under-reporters due to omitting food from FFQ, social desirability/recall bias Result: Bias, incorrect mean, loss of Result and solution: If all subjects power affected the same / measures of association not affected. If differential Solution: validation and calibration using between cases and noncases, bias is not ‘gold standard’ amenable to correction (e.g., case-control studies with diet) To conclude Day-to-day variation in nutrient intake is a major source of random measurement error A single 24h diet recall or diet record may provide an accurate estimate of the mean for a group, but standard deviation is highly overestimated Random error attenuates measurements of association (e.g. relative risk, correlation coefficient) in epidemiologic studies to the point of being undetectable. Systematic error leads to incorrect means but measures of association (e.g. relative risk) are not affected if all subjects equally affected (i.e. if it is not differential in terms of exposure) Systematic error leads to uncorrectable bias if differential in terms of exposure (e.g. classical case-control studies in diet) From evidence to public health policy Selected nutrition epidemiology findings and their wider implications 62 Can you think of the examples of nutritional epidemiology research that “made it” into policy/interventions? Discuss and give examples. Example 1: folate and spina bifida Spina bifida: a developmental disorder caused by maternal folate deficiency Early epidemiologic studies prior to the knowledge of etiology of disease in 1970‘s suggested that insufficient folate intake during pregnancy was linked to an increased risk of neural tube defects like spina bifida and anencephaly One of remarkable successes of epidemiology Followed by the famous MRC Vitamin study (RCT) in 1991 Folic acid supplementation and neural tube defect RCT MRC Vitamin study, 1991. Neural tube defects prevention, RCT, The Lancet From evidence to public health measure Folic acid fortification Folic acid supplementation The U.S. was first to fortify white flour routinely 69 countries today have Women of mandatory, 47 voluntary reproductive age fortification of staple food and pregnant (2023) women 77 countries still not fortifying despite clear evidence (including the EU!) Quinn et al, 2023, The Lancet Example: margarine (trans-fatty acids/TFAs) Industrial TFA vs. Ruminant TFA Partial dehydrogenation: coverting cis to trans bond Developed by German chemist Wilhelm Normann, who invented the process of hydrogenation Example: it is the type of fat (not total fat!) 1960‘s evidence on SFA and LDL cholesterol Second half of 20th century: total fat reduction guidelines in the U.S. /low- fat diet guidance based on poor evidence Implication: replacing fat with carbohydrates (often refined!) and carbometabolic outcomes (followed by a steep rise in obesity and T2D rates in the U.S.) Omniheart RCT (feeding study) Appel et al., 2005, JAMA Effect of Trans and Saturated Fat (10% E) on Blood Lipids (vs. Monounsaturated fat) Monounsaturated fat) Trans fat Saturated fat Total cholesterol +6% +12% LDL cholesterol +14% +18% HDL cholesterol -12% 0% LDL/HDL ratio +29% +18% Feeding study, 34 women and 25 men on mixed diets (TFA, SFA, MFA/ref), 3 weeks (Mensink & Katan, 1990) 9.110 Type of Dietary Fat and Risk of Coronary Heart Disease The Nurses' Health Study 14-Year Follow-up 100 Y =SFA/E+MFA/E+PFA/E+protein/E+TEI+other covariates Trans (Effects of each type of fat can be observed in relation 80 to the same % energy from carbohydrates) % Change in CHD 60 40 20 Sat 0 1%E 2%E 3%E 4%E 5%E -20 -40 Mono Poly Y=SFA/E+MFA/E+PFA/E+protein/E+TEI+other covariates (Effects of each type of fat can be observed in relation to the same % energy from carbohydrates) Hu FB, et al. N Engl J Med 1997;337:1491-9 9.131 Age-Adjusted Plasma CRP by Quintiles of Trans Fatty Acid Intake in the Nurses’ Health Study (P, trend = 165 165-143