Lecture 48: Evidence Synthesis in Nutritional Epidemiology: From Research to Recommendations

Hello and welcome dear learners to the NPTEL course on nutritional epidemiology a way towards a healthy life. For the model four diet nutrients dietary patterns and disease epidemiology for the chapter 48 we'll be discussing about evidence synthesis in nutritional epidemiology from research to recommendation. This chapter particularly deals with hierarchy of evidence in nutritional research systematic reviews and meta-analysis sources of heterogeneity in nutrient studies the great framework for evidence assessment the Bradford Hill consideration for causality translating evidence into dietary recommendations and giving you the future directions. For the hierarchy of evidence in nutritional research in classical biomedical science the hierarchy of evidence places a randomized controlled trials that is RCT as the gold standard. So with the observational cohorts ranked lower the nutritional epidemiology however the rigid pyramid faces unique operational and biological limitations. So there is a nutritional evidence matrix. What is this matrix we'll try to understand. The prospective cohort studies and the randomized controlled trials. Here in the PCTs we have a decadal exposure tracking mirrors freely living populations and capturing the chronic incident points. In the RCT we have a high compliance for short terms vulnerable for to dropouts and crossover and limited to surrogate biomarkers. While the RCT excel at testing isolated chemical compounds over a period they are often unfeasible for evaluating the whole dietary patterns over decades. Forcing human participants to comply with an artificial diet for 20 years to observe a clinical outcomes. The outcomes can be more for cardiovascular mortality or a colon cancer and logistically and ethically impossible. Therefore, the evidence synthesis in nutrient relies on a complementary matrix. That is the prospective cohort studies. That is PCS. And where we come across to see that there's a vital data on long-term real-world exposure of hard clinical endpoints while short-term feeding trials contribute to a mechanistic support by tracking intermediate and circulatory biomarkers. Circulatory biomarkers are glycemic tracking, lipid fractions, and inflammatory cytokines. The systematic reviews and meta-analysis. In the systematic review, in a structured process used to identify, a critically appraise, and extract data from all relevant primary studies. Which are more specifying towards the nutritional questions. In the meta-analysis other side, we see the mathematically combining the individual data points to calculate a single high-powered pulled risk estimate. Where This is very important. Pulled risk estimate, which is giving you something which is called as a pulled relative risk, RR, and a hazardous ratio, that is a HR. In the primary study, secondary study, third study, you may come across a meta-variance pooling engine. So, meta-variance pooling engine is like a randomized effect model, which is giving at a first level, which is followed by a pulled risk estimate. Like from accounting for between a study variance to a summary diamond on a forest plot. So, because the nutritional habits vary across a global study population, meta-analysis typically apply a random effect model. So, you will find that a fixed effect model, which assumes every study shares on identical underlying truth, a random effect model assumes the truth effect of sizes, which vary naturally between cohorts due to a geography, demographics, and a culture. Which incorporates the between study variance alongside with a study sampling error, ensuring the final summary estimate is not disproportionately skewed by a single massive cohort. Okay? So, you can understand the difference between a systemic review and a meta-analysis. Now, the concept of meta-variation pooling engine. So, when an epidemiologist conducts a meta-analysis, they are trying to take multiple separate studies from around the world and combine them into a one definitive headline result. Okay? This is like pulling from different sources. Okay? And trying to take all the study and putting some sort of a headline result, which is giving them a meta-analysis's diversion. So, like you can say a study one, or a study two, or study three, and so on. Study one can be conducted in Japan, the study two can be conducted in America, and a study three can be conducted in Sweden, and so on. Which finds that the study one is 15% reduction in risk 5%, followed by 25%, and so on. So, there is a basic statistical model, the software will assume that all the three countries, four countries, and all are identical, and that the differences in their numbers are just not random, but a statistical flux. The random effect model is much smarter. It acts like a meta-variation pooling engine. It looks that the data says that weight people in Japan, the US and the Sweden have completely different lifestyle. Genetics and background diet. Of course their numbers are different. Okay? So this is like meta-analysis is very important in understanding meta variation pulling into. What's the illustration now that I will try to make you explain? The engine mathematically measures the between study disagreement. Okay? So this is something which you need to score it on. Where the statisticians call the chi squared. Okay? It folds that the real world disagreement right into the calculation alongside a standard margin of error testing. Which gives you the random effect engine where you can find that the study which is done in Japan versus America and Sweden which calculates the accounts for the difference into the total where the final summary is more like perfectly giving you the reflection towards the global complexity. So the final outcome that actually we are expecting that is a summary diamond at the bottom of the meta-analysis chart will be something like an extracting an engine to give a global differences. You will find that it is more not skewed up but a balance between a high power average that represents the general human health without getting skewed. So skewedness is avoided over here with a single unusual study. Now for the heterogeneity in the nutrition study. What does heterogeneity gives you? We'll try to see that later multiple researchers, multiple studies which are frequently encountered with different individual papers. So there's a high heterogeneity which is being with the singles that the primary studies are not telling you the same story. And in the nutritional meta-analysis the variation terms of the three main sources. The first is the exposure measurement variance. What is exposure variance give you? These are the different cohorts using tools to capture dietary habits. One study might use a ultra detailed which have of 150 items and validated FFQs which are administered every 4 years. While another is more with the 24-hour dietary call, so you can understand the difference. The difference is more producing towards you to introduce waiting levels of uh recall errors. Okay? And with the mass classification which flattens the distorted risk estimate. And the second is coming with the compositional differences in the food groups. Which gives you more about the geographic regions. For instances, you have an East Asian cohort stating that a seafood intake. Okay? And from the other part that is the lean white fish with an iodine rich uh marine seaweeds. And then the western cohort, you find that the same seaweed feeding take might be dominated by deep-fried whitefish. Or a processed seafood items which is consumed along with the refined carbohydrate. So, can you understand over here? Here it is like a combining this distant food matrix understanding the single generic heading introducing a structural heterogeneity. So, heterogeneity is more to be concentrated over here to understand these two differences which are coming from an East Asian cohort and a western cohort. Now, the background diet and the substitution matrix. Is it something that we will be trying to understand more different with the previous one? The answer is stated over here that health effects of adding a protective food group to a diet dependent heavily on what the food grouping is replacing. The replacing is very important. So, if a participant intakes their intake of whole grains by replacing a refined sugar sweetened pastries, their metabolic markers will improve dramatically. There is no way it is bounded actually that you can say that another participant which adds whole grains to already hyperbo and a plan for our diet, the observed benefit will be minimal. It will be like reducing. Okay. Is there expected failure? The failure is more towards the substitution matrix creating and conflicting statistical results. Like you may have across the different global populations. So, might be a variation which may come up. And the grade framework for evidence assessment. Grade framework like a grading which gives a recommendation coming with a development evaluation assessment which is something a tool which is being developed with the standard efforts call it as a standard tool by WHO. As the agency is very predominantly talking about the quality of body of evidence, it is grading towards an automatically downgraded observational data. Low quality stating a trial which frequently penalizes the solid nutritional data. And we can find the grade nutrient criteria where downgrading risk factors and upgrading strength factors are given. Either on the sides tries to understand where the downgrading risk factors give a serious risk of a selective bias, okay? Followed by unexplained high hydrogenicity. And procedure in confidence bands and publication funding bias. Wherein the upgrading strength vectors, you may come across a large magnitude effect followed by a cleaner uh clear linear dose response curve. Clear linear dose response curve made is more all plausible confounding control or acquired for. So, you can say that upgrading strength vectors versus to the downgrading risk factors giving you more the body of evidence in the downgraded if there is a high risk of residual confounding with a severe self-reporting bias. And explains statistical heterogeneity or high dependency on the industry funded trials. Upgrading triggers, which is from the observational evidence, can be upgraded to moderate or high. Okay, the clarity is more on the plausible residual confounders where you may come up across with a weakened and the observed effect. So, the triggering may be seen with respect to the downgrading risk factors and the upgrading strength factors. Okay. The Bradford Hill consideration for causality. You might have seen this and we have explained it previously also that how a Bradford Hill criteria is evaluating which is being observed to a statistician association reflecting in a true underlying biological causes. And the relationship comes with a consistency and strength association. Temporality and biological gradient. Where the association at the first level gives you more and uh consistently replicated across an independent cohort using differentiating methodologies. Where the methodologies across distinct culture, geographic, and ethnic demographics. And you may come up with temporality which is being more clearly indicating to you a proceed development of clinical endpoints. Prospective cohort which is achieving by assessing the dietary patterns at the baseline among individuals who are completely free at the target disease. And then they are tracking them to forward over time. Biological gradient. Here you have to see the hallmark of a true biological causality uh which is a clear dose-response relationship. Okay, it is it is like a thinner to understand like a for instance that you are documenting that every 50 g daily increase in a processed meat consumption predicts a corresponding stepwise 80% increase in colorectal cancer. There's a sort of an hypothetical but a assumption which is based more on the facts which is giving you the risk strongly to support a causal pathway. The causal pathway is more clearly understand in this hypothesis. The plausibility and coherence. The epidemiological observation must align with established basic science and clinical pathophysiology. Here an example is cited over here. The example is like association between trans fat intake and ischemic heart disease. That is called as IHD. It's very complicated but it's very important to understand that this disease condition how it is connected with trans fat. Where you may find that a considerable causal effect of a short-term metabolic feeding trial prove that trans fatty acid directly elevate the LDL particles. Nowadays we are very much worried about the cholesterol and the reduction in the HDL counts and the active vascular endothelial inflammation which is a point of concern when you particularly visit your cardio logist. And translating the evidence into dietary recommendation we come up with a final stage critical to understand the evidence synthesis in turn understanding the translating complex statistical data into actionable public health policies. The policies diverge towards giving you a guideline. The committee is to balance the pure nutritional science against a real world humanistic behavior. Where evidence to the policy comes with a raw multi-omic and cohort data. Where the omic data of the part which is more towards a gradient which is being stated in the Brad Hole Hill review is more for the recommendations. The recommendation first comes more with the nutrient based that is extracting for the target figures for clinical settings and uh food based dietary guidelines which is coming with the the swapping of meat for legumes or actionable behaviorally realistic patterns. Might be seen that we may not see this all in India but yes this is something which is more based on the food based dietary guidelines, which is a translating evidence. Now, for the nutrient-based versus food-based guidelines, you may come across with a guideline by a nutrient-based matrix, which is giving you more of ranges. What are those ranges? The ranges are coming from RDAs, that targeting a sodium to less than 2,300 mg per day will essentially come with a clinical dietitian's formula that are difficult for the general public to apply, where they don't measure when they eat or when they are really not directed to take the proportions. The FBDGs, which is translating the chemical matrix into clear practical eating habits, like consume three proportions of whole grains daily, or replace your butter with liquid vegetable oils. Uh, so it is coming with a purchasing of the whole foods, not isolated chemical nutrients, and a behavioral pattern, which is more focused on shifting towards the health outcomes. Whether uh things should be in mind to understand the health outcomes. In terms of purchasing, so there should be a guideline which you should be directing towards this. Now, managing the stakeholder conflict. So, is there any stakeholder conflict which is coming with an economic pressure? Truly speaking, formulating the nutritional guideline often comes with a substantial economic and industrial pressure. Okay. The multi-billion dollar agriculture sector economy, directly they're speaking about the lobby which guidelines the committees to soften the warnings. The warnings can be lobbying to change the reduced red and processed meat consumption to a more ambiguous directive, like choose lean protein sources. Okay. This is one, and there are multiple such examples for this. To protect the public health integrity, modern guideline panels require a stricter financial disclosure protocols, interpretation of scientific chairs, and completely transparent systematic review pipelines. So, this is something which is directing you towards going for a policy recommendation. For the future directions, the nutritional epidemiologist uh should get involved in the past of the limits of the traditional self-reported paper questionnaires. And you should have some sort of integration of the advanced digital technologies and molecular biology. Digital biomarkers and real-time trackers, which gives you more from the GGMs and smartphones. Photo-based dietary tracking. Digital kitchen scales is replacing retrospective memory-based food recalls. These tools capture food intake data in a real-time, dramatically reducing the recall errors and social disability, which is sometimes very important to be reduced. And then there is a metabolomic profiling and objective exposure markers, which are coming as a rise of a nutritional metabolism, which is instead of relying on the sole reported questionnaires, which is more verifying with the citrus food, where you can come across a citric food with the whole grains, which is giving more by researcher which analyzes the blood plasma or urine samples throughout the whole mass spectrometer. The process isolates the distinct objective chemical fingerprint. Metabolomic biomarkers left behind the specific food matrix. Combining the objective exposure markers with a deep metagenomic and genomic tracking allows. Okay. So, this is our future directive for systematic reviews, which is more relying to achieve whether you have a traditional pharmaceutical trials or the either other way. Now, finally we have to come up with a key takeaway to take this then a complementary evidence comes with a nutritionist to take or a nutritional epidemiologist to understand there is a balances of long-term prospective cohorts versus with the short-term clinical trials. Okay. And systematic sorting, which is a meta-analysis which is used to understand a random effect models to pull the relative estimates across the global cohort. The global cohort which is advancing with a grade two screen. Which is more based on the residual confounding and evaluating a data quality. How to see this heterogeneity? To see the heterogeneity, we have to see the unlocking. So, the variation may come up with the study findings like you have a survey design. You may have a local composition local food composition or a specific food substitution. There are three ways. These three ways are something which are unlocking. Okay, and making the participants more relying on you to understand for this part. Lastly, the policy realism. Where you as a policy person for public health were turning to see the evidences into a successful public policy which require shifting from abstract chemical matrix to a practical. Okay, and food based guidelines that are transparently protected from commercial industry influence which can be also applicable in country like India. Okay, with this we come in end to this chapter and thanking to you and stating that this references might be very very important for you to understand the grade matrix in the grade guidelines. Thank you very much.