NUT602 Research Methods in Nutrition and Food Science - Population-Based Studies PDF

NUT602: Research Methods in Nutrition and Food Science Fall Population- 2024 based Studies Exposures & outcomes 2  3...

NUT602: Research Methods in Nutrition and Food Science Fall Population- 2024 based Studies Exposures & outcomes 2  3 Hill’s Criteria Criteria Rationale Strength of association Strong associations have a higher likelihood of being causal Consistency Associations are more likely to be causal if they are observed repeatedly “by different persons, in different places, circumstances and times.” Specificity An exposure can only cause a single outcome (less relevant in epidemiology) Temporality The cause or exposure must come before the effect or outcome Biologic gradient The ability to demonstrate a dose-response relationship (may not always be present) Plausibility Scientific justification for the cause-effect relationship Experimental evidence Interventions (treatments or risk factor modifications) have predictable effects on the occurrence of outcome Analogy There are similar cause-effect relationships  Classification of research 4 Interventional Longitudinal Prospective vs. vs. vs. Observational Cross‐sectional Retrospective Blinded vs. Not Randomized Single‐blind, Double vs. blind, Unblinded Non‐Randomized  5 Population-Based Observational Studies THE METHODS USED ARE BASED ON EPIDEMIOLOGICAL APPROACHES Hierarchy of evidence 6 Epidemiology 7 ▪ Discipline within public health that looks at the rates of health-related states in different groups of people and why they occur ▪ Connect the dots between exposures and health/disease outcomes ▪ Nutrition epidemiology: subdiscipline of epidemiology ▪ Descriptive epidemiology ▪ Informs about the disease’s prevalence and incidence ▪ Informs about the pattern of the disease (time, place, and personal characteristics) ▪ Analytic epidemiology ▪ Determine the association between an exposure and health-related state ▪ Finding an association does not necessarily make it a causal relationship  Knowledge Check 8 Choose the correct answer from the following choices: A. Experimental B. Observational B. Observational An epidemiologist is doing a study on the sleep patterns of college students but does not provide any intervention. What type of study is this? A. Descriptive B. Analytic B. Analytic 1. A study of heart disease comparing a group who eats healthy foods and exercises regularly with one who does not in an effort to test association A. Descriptive 2. A study to describe the eating habits of adolescents aged 13–18 years in Lebanon  9 Observational studies Explore associations between nutrition and health outcomes (generate hypotheses) Can help to build up evidence to support a suggested effect of a particular dietary factor on a certain outcome, but cannot show cause-and-effect association Types Ecological or Cross- Cohort- Case-control correlational sectional Longitudinal Key consideration: The researcher has no control over the exposure of interest Ecological studies 10 SFA and Serum Cholesterol SFA and CHD deaths Seven Countries Study  Seven Countries Study 11 Hypothesis: Rate of coronary disease would vary in relation to physical characteristic and lifestyle, mainly fat composition of diet and serum cholesterol levels This study was central to the modern recognition, definition, and promotion of the eating pattern found in Italy and Greece in the 1950’s and 60’s now called the “Mediterranean Diet”  Ecological studies 12 ▪ Characterizing population groups rather than linking individuals’ exposures to outcomes ▪ Explore associations between population diet/nutrition indicators and indices of health status ▪ The analysis is not based on data on individuals ▪ Comparing populations’ disease rates with population's per capita consumption of dietary factors ▪ Two population-based measures ▪ One for the exposure of interest (indices of dietary intake) ▪ One for the health outcome (indices of health state)  Methods: Exposure 13 ▪ Indices of dietary intake ▪ Survey data collected for the purpose of the study in a population ▪ Pre-existing dietary data (less costly, but may not sufficiently reflect consumption) ▪ National food supply: internationally available food data (FAO food balance sheets) ▪ Describes the pattern of a country’s food supply for a particular time point → estimated amount of each food available for human consumption ▪ Data used to assess trends in dietary intakes; may overestimate dietary intakes ▪ Household budget surveys: studies that collect data on food availability (household level) ▪ Participants record food purchases and other food coming into the home ▪ Survey data: nutrition and health population-based surveys ▪ For example, used to estimate mean fruit and vegetable intake for the GBD https://www.fao.org/faostat/en/#data/FBS  Methods: Outcome 14 ▪ Indices of health outcomes ▪ Routine measures of mortality and morbidity: at a national level ▪ Usually available through government reports or WHO publications ▪ National mortality data and GBD data: http://www.who.int/healthinfo/statistics/en/ Diabetes mellitus (who.int) https://cdn.who.int/media/docs/default-source/gho-documents/global-health-estimates/ghe2019_daly-methods.pdf  Analysis of ecological data 15 r=0.80; p80% ▪ Consider a situation in which the response rate was 90% but 10% of non-responders had very specific and distinct characteristics that made them decide not to respond → Selection bias might be present despite the high response rate ▪ → Obtain some information from non-responders (e.g., short questionnaire or using demographics or other information that might be available for the members of the sample frame) and compare with responders to identify significant differences Cross-Sectional Studies: Hypothesis Testing 23 MeDi diet and AD Cross-Sectional Studies 24 MeDi diet and AD Analysis of Cross-Sectional Data 25 ▪ Statistical methods to evaluate the association of a dietary exposure and a health outcome Cross-Sectional Studies: Causality 26 POMA: Performance-Oriented Mobility Assessment Challenges 27 ▪ Representativeness of the reference population ▪ Dietary assessment methodologies and tools ▪ Use of inappropriate or inaccurate assessment methodologies or tools → biased responses related to dietary exposures, health outcomes, and other studied factors ▪ Bias related to the recall of dietary information from sample members ▪ Potential alterations of long-term dietary habits resulting from the previous presence of a disease or a health condition cannot be thoroughly investigated ▪ Application of data management and statistical analysis methodologies ▪ Failure to account for potential confounding factors ▪ Cause-and-effect relationships between dietary exposures and health outcomes? ▪ Causality: an observed association requires at least that the dietary exposure occurred before the diagnosis of a disease or health condition ▪ Dietary exposure and outcomes measured at the same time point ▪ Causality is not confirmed by a cross-sectional design  Cross-Sectional Studies 28 Advantages Limitations Inexpensive for common disease Unsuitable for rare diseases More representative cases (vs. case series) High refusal rate → inaccurate prevalence estimates Short in duration (less time-consuming) More expensive/time consuming than case control studies Specific population Potential bias in measuring predictors Simultaneous wide variety of measurements (Can study Potential survivor bias (no information about patients who several exposures and outcomes) died) Can yield prevalence (especially for longer-lasting diseases, Unsuitable for disease of short duration (diseases lasting for i.e., chronic diseases) a short period may be underestimated, i.e., Neyman bias) Value for public health planning Not feasible for rare conditions (need large sample size) or quickly emerging outcomes/diseases Can guide further hypothesis testing and studies in the field Most important limitation: impossible to determine whether the exposure and the outcome are causally related (temporal sequence) Time is the best/worst confounder of all Case-Control Studies 29 ▪ Retrospective Study ▪ “To look back” in time to study events that have already occurred TIME Study begins here  30 ▪ 435 classic Hodgkin lymphoma (CHL) cases, 15-79 years, w/o HIV infection at diagnosis ▪ Pathology used to confirm diagnosis ▪ 563 population-based controls matched to cases by age, sex, and state of residence, without a personal history of CHL ▪ Identified through the “Town Books” in the greater Boston area and by random-digit dialing or through the Health Care Financing Administration (Medicare) in Connecticut ▪ Dietary assessment: validated, semiquantitative FFQ, average consumption of 61 food and beverage items, vitamin and mineral supplements, over the year before enrollment ▪ Evaluation of several participants’ characteristics and potential confounders ▪ Potential association between diet and odds for CHL ▪ Diets high in meat or desserts and sweets are associated with higher likelihood of CHL  Methods 31 ▪ Diagnosed (cases) vs. healthy individuals (controls) for past exposure to dietary factors ▪ Significant differences in the dietary exposures in cases vs. controls suggests that the exposure may be associated with decreased or increased likelihood of the disease ▪ Odds ratio (OR) = odds that a health outcome occurs given a dietary exposure, compared with the odds of the outcome occurring in the absence of exposure ▪ Data related to past dietary habits are retrieved by memory recall → Caution when interpretating the results ▪ Matching controls for some confounding factors (e.g., gender, age, SES,…) ▪ Control for confounding of the association between the exposure and outcome ▪ Statistical methodologies allow further controlling for potential confounding ▪ Uses ▪ Studying dietary implications of rare diseases (low incidence, long induction period) ▪ Less time and resources compared with cohort studies (smaller samples, without FU) Selection of cases 32 ▪ Precise, accurate, up-to-date definition of the disease/health condition ▪ Prevalent or incidence cases? ▪ Incident (newly diagnosed individuals): advantageous ▪ Prevalent: ▪ Cases that are present in the population during recruitment (new and old cases) or Fatal cases ▪ More common and easier but has limitations ▪ Identification of cases ▪ Hospital or general practice records ▪ May be more representative of the diseased population ▪ Risk of missing undiagnosed people from the general population ▪ Factors that determined earlier diagnosis might also be linked to differences in diet → risk of spurious associations when studying only newly diagnosed patients ▪ General population ▪ Difficulties in tracing subjects; refusal rate may be greater ▪ Risk of lower representativeness of the diseased population ▪ May be applied in case of a comprehensive registration system (sampling frame) Selection of controls 33 ▪ One of the most difficult aspects of a case-control study (prone to bias) ▪ Controls: Same eligibility criteria, except they do not have the outcome ▪ Selected from the same population as cases → balance confounders between cases and controls and representative in terms of exposure ▪ Selection should be independent of exposure status ▪ Cases selected from screening clinics → controls selected from these clinics ▪ Cases are selected from hospitals → patients with other diseases can be controls ▪ In cases of rare diseases, recruitment of controls is generally easier than for cases ▪ From the close environment of cases (relatives, friends…) ▪ Risk of overmatching: level of the exposure to dietary factors may be similar to that of cases (may be impossible to identify significant differences) ▪ From the general population ▪ Exposure may be more representative of at-risk population ▪ Nonresponse: selection bias due to differences between responders and non-responders ▪ From the same hospitals from which cases have been drawn ▪ Enhances the completeness and accurateness of the collected information  Matching 34 ▪ Matching factors: variables often related to disease and exposure ▪ Confounders of the association between exposure and disease (based on evidence) ▪ Frequently used factors: gender, age, residence, and nationality ▪ Ratio of controls per case: depends on the power calculations ▪ Matching more controls for every case increases the power to detect association, up to 4:1 (4 controls/case) ▪ Individual vs. Group matching ▪ Individual matching: expensive and time-consuming ▪ Group matching (frequency matching)- a form of stratified sampling ▪ Over-matching occurs when a factor is used for matching is not a confounder ▪ Should be avoided: risk of selection bias or reduced efficiency to detect associations Selection of cases and controls- Recap 35 ▪ Recruit cases/controls to ensure a certain level of representativeness of the reference populations and avoid sampling bias ▪ Diversity of exposure to the dietary risk factor(s) being studied in the sample ▪ Cases and controls should be blind to the purpose of the study ▪ Avoid explanations for the disease in the questionnaire ▪ Preferred methodology ▪ Recruiting incident cases ▪ Selecting controls to be representative of the population in which the cases arise ▪ Well defined eligibility criteria (irrespective of exposure status) ▪ Record non-participation, reasons, and other information (age, gender…) if possible Measurement of dietary exposure 36 ▪ Identifying a past diet exposure relevant to the disease process is challenging ▪ Long pre-clinical phase → dietary exposure may have occurred years before diagnosis ▪ Difficult to report past diet accurately ▪ Answers to questions on dietary behavior in the past are strongly influenced by current eating patterns ▪ Changes in the diets among cases as a result of the disease process ▪ Occasions where patients are too ill to respond or to recall information about diet ▪ Main method of collecting dietary information: FFQ ▪ Intake over the past 12 months rather than currently (potentially affected by disease) Analysis of case-control data 37 ▪ Main measure: Odds Ratio (OR) ▪ Odds that an outcome (disease) occurs given a particular exposure (dietary factor), compared with the odds of the outcome occurring in the absence of that exposure ▪ OR = 1: suggests no association ▪ OR > 1: suggests that exposure is associated with increased odds of outcome (positive association) ▪ OR < 1: suggests that exposure is associated with decreased odds of outcome (negative association) ▪ Crude OR (calculated without controlling for confounding variables) vs. Adjusted OR  Analysis of case-control data 38 INTERHEART study ▪ One of the largest case-control studies ▪ 15,152 cases and 14,820 controls from 52 countries from all inhabited continents ▪ Specific objectives ▪ Determine the strength of associations between various risk factors and acute myocardial infarction ▪ Ascertain if this association varied by geographical region, ethnic origin, sex or age ▪ Results ▪ Daily fruit and vegetable intake: OR: 0.70; 95%CI: 0.64, 0.77 ▪ Obesity: OR: 2.24; 95%CI: 2.06, 2.45 Yusuf et al. (2004). Effect of potentially modifiable risk factors associated with myocardial infarction in 52 countries (the INTERHEART study): case-control study. The lancet, 364(9438), 937-952.  Case-Control Studies 39 Selection of cases Selection of controls Incident or prevalent Independent of exposure status Identified from hospital, general practice or the - Source population from which cases were drawn ➔ balance general population confounders - Representative of the source population in terms of exposure Specificity of diagnosis - More than one control can be matched to each case - Avoid overmatching Advantages Limitations Less expensive and time-consuming - Obtaining unbiased measure of previous dietary exposure - Recall bias can compromise internal validity Efficient for studying rare diseases/conditions Do not estimate incidence or prevalence Provide a potentially weaker causal investigation of an outcome Often are not generalizable  Cohort Studies 40 ▪ Participants are observed over time ▪ Free of the disease at baseline ▪ Baseline data on exposures of interest are collected for all individuals ▪ Incidence of outcome (new cases) under investigation is recorded during follow-up ▪ Suggests an effect of the dietary exposure on the outcome ▪ Time relationship between exposure and outcome can be determined Example: Framingham Heart Study 41 ▪ In 1948, the Framingham Heart Study (prospective cohort study) started to study the impact of various exposures on CVD. For each exposure, researchers identified the “exposed” and “not exposed” groups, then followed these participants for the development of CVD: ▪ Exposures: Diet, smoking, body weight, diabetes, exercise ▪ Outcomes: Blood pressure, coronary heart disease, stroke ▪ The first cohort included more than 5,000 men and women (ages 30 to 62) from the town of Framingham, Massachusetts. After an initial physical exam and gathering of lifestyle data, they were asked to return every 2 years for another assessment to evaluate risk factors and the development of CVD ▪ In 1971, another cohort was started; this time it included children and spouses of the first cohort ▪ In 2002, the grandchildren of the original cohort were enrolled in a new study ▪ Through the FHS, scientists have learned the risk factors for CVD, including high blood pressure, high cholesterol, unhealthy eating patterns, smoking, physical inactivity, or unhealthy weight. TYPES OF COHORT STUDIES 42 Selection of study population 43 ▪ Sampling frame compiled or sourced (ideally) → healthy (i.e., outcome-free) individuals ▪ If exposure is common ▪ Sample from the general population ▪ Electoral registers, school registers, list of patients in general practices… ▪ Random recruitment of participants after the application of inclusion and exclusion criteria ▪ Advantages: allows studying multiple dietary exposures in relation to a wide range of outcomes ▪ Disadvantages: challenging follow-up depending on the scale of the area selected for recruitment ▪ Participants from a specific group ▪ Army servants or workers belonging to a union ▪ Advantages: facilitates a more comprehensive follow-up of participants because they are already organized around a well-defined system; degree of exposure to dietary risk factors may be higher in this group compared with the general population → assist the study of the associations ▪ Disadvantage: owing to this difference in risk factors present in the study group, the estimated incidence cannot be representative of the general population ▪ If exposure not common ➔ select sample based on exposure (sufficient individuals) ▪ Power calculations are needed to determine the size of cohort required ▪ Resources and practices available may affect the choice of the study population Measurement of dietary exposure 44 ▪ Baseline ➔ assumption that eating habits remain relatively stable ▪ May not always be the case due to changes in dietary fashion and advice ▪ Additional wave collections at different follow-ups (additional resource requirements, losses to follow-up, and complexities of analysis) ▪ 24-hour dietary recalls or diaries (four to seven days) ▪ More current and detailed ▪ Short term ▪ More time and resources ▪ FFQs (estimated average intake relating to the previous 12 months) and diet histories ▪ Elements of retrospective recall of exposure data Analysis of cohort data 45 ▪ Measure disease incidence (rate of new disease development) ▪ Relative risk: ratio of disease incidence in the exposed group and that in the nonexposed group ▪ RR expresses how much more (or less) likely it is for the exposed person to develop an outcome (relative to an unexposed person) ▪ RR=1.0 → no difference in risk between groups (exposure did not increase/decrease risk of outcome) ▪ RR>1 → increased risk of that outcome in the exposed group ▪ RR