Week 5: Lecture 24: Module Integration: Choosing A Design for A Research Question

Hello and welcome dear learners to the NPTEL course on nutritional epidemiology, a way towards a healthy life. In the model two, study design and epidemiological methods for the chapter 24, we'll be taking the module entering integration choosing a design for a research question. So, what will be studying over you? We'll be studying from a research question to study design. We'll be taking ahead the types of research question in nutritional epidemiology. Overview of an available study design matching with research questions with study designs. Key factors influencing the choice of the study design, role of measurement in study design, the bias, confounding, validity consideration, causal inference, decision framework, common pitfalls in the study selection, and practical exercise. So, let's begin. Choosing an appropriate study design is one of the most critical decision in nutritional epidemiological research. The importance of choosing the right study design is somewhere you can say the selected design determines validity of findings, strength of evidence, ability to infer causality, cost and feasibility, generalizability and ethical acceptance. So, you may see a lot of importance which is being stated over you. But, if you do something wrong, the poor design choose can lead you to invalid conclusions, which is an accepted waste state of all the resources that have been used, and misleading dietary recommendations, which is somewhere the expectation that you should come out of the study. And no doubt, the entire purpose which goes with a failure is the ineffective public health policies. Thus, the other question remains. Should you drive us to study design? Yes, of course, and why not? So, vice versa, if you can say, choosing a right study design is very, very important. So, from research question to study design, we'll try to take up the entire research process. In a research process, firstly, we should go with step one, that is identifying the problem. If your problem is very clear, you can define it. You can make it more relevant. Especially now, here when we are discussing the nutrition epidemiology, the problem should be clear and relevant to nutrition and health. If I have to give you an example, it is like does frequent consumption of ultra-processed food increase obesity risk among the adolescent? This is like identifying the problem. This is not a hypothetical statement, but it will be clear over the research process. Now, going to next step, that is formulate the research question. So, if you choose the PICO framework, choosing a right design is very important. And when you choose the right design, it must have a clear population, intervention or exposure, comparison, outcome, and time frame. So, if you take for an example, a population between 12 to 18 years, the intervention can be high intake of ultra-processed food. Comparison is like low intake of ultra-processed food versus high intake of ultra-processed food. What is the outcome? You are expecting obesity, increased BMI or body fat. Okay, what is the time framework for this? Follow-up of 2 to 5 years. The research questions will be among the adolescents, does high consumption of ultra-processed food, compared with the low consumption, increase the risk of obesity over 2 to 5 years? So, this is a research question. Okay, now we are getting it clear to the next step, that is determine the objective from this. What are the objectives? It can be described the entire process. Compare. You can predict, you can explain, you can establish the causality, where you can come up with the hypothesis, and evaluate the intervention that is being done. So, whole lot of your population, intervention, comparison, outcome, and time frame can come up into the objectives over here. Which can be more than one. Now, you'll study the design. And you will select a proper design for your study. The selection of study design is a logical structure that accommodates the PICO framework. The elements while minimizing the bias and maximizing the feasibility. So, lesser the bias, and maximizing the feasibility, you'll have a more accurate and more perfect study design. Example of possible design depending on the objectives that you have stated. Okay, there might be a cross-sectional study. You may come across even to case-control study. You may take up a cohort study. You may take randomized control trial. You may take quasi-experimental study. And even ecological studies can also be there. So, any number of possible study designs can come upon your objectives. Okay? Now, there are types of research questions in nutritional epidemiology. These are only the few stated one. You may have much more actually for this. So, very first is like descriptive questions. You may come across multiple studies actually which has used descriptive questions. For example, the purpose of taking it is like you can use it in cross-sectional surveys, you can use it in nutritional surveillance, or ecological study. Okay, what is the average salt intake in adult can be one among the question in descriptive way. Or what proportion of children are vitamin D deficient? We are taking all these more from Indian context. What are the dietary pattern among urban and urban women particularly? What are the dietary pattern among urban women? You can even have a comparison also. Urban versus rural. So, this gives you a best design actually if your descriptive questions are coming. Okay, then comes association questions. These are also very important and many times you have seen in lot of nutritional studies association question comes. The purpose of putting association question is to determine whether exposure and the outcome are related or not. Okay, now you will have a association very clearly stated in this example. Is red meat intake associated with colorectal cancer? So, you can find this. There is an association which is being questioned. Also, is fruit intake associated with lowering the hypertensive risk. Very important. This can be seen as some way the questions can be further stated as hypothesis. What are the best designs that can be used for associated question? Cross-sectional, case-control, and cohort studies. Okay. Then comes etiological or causal questions. Causal are very important. You can see it in studies also. What is the purpose of etiological causal questions? It is to determine whether exposure caused the outcome. So, exposure is linked over here. So, you may find exposure. Now, soda beverages linked with obesity, trans fat related or associated with cardiovascular risk. Example can be does sugar-sweetened beverage increase or can be a cause of obesity? Does trans fat intake increase cardiovascular disease risk? Okay. The etiological causal question can be used in randomized controlled trial, prospective cohort, Mendelian randomization, or natural experiments. Okay, you have a variety to select as a best design. Then comes intervention questions. What is the purpose of taking the intervention question? It is to assess the effectiveness of dietary intervention. So, you may have a lot of dietary interventions which are done, and when you have to assess it, you can use the intervention questions. So, example can be does iron supplementation reduce anemia? Or you can see giving more nutrition fortified food can reduce the condition of malnutrition. Or does DASH diet lowers blood pressure? Okay, there's a DASH diet and how it is effective in blood pressure. So, what are the best designs being stated over here? RCT cluster RCT or community trial. So, intervention questions are very very important. So, you can select among these also. Okay, prognostic questions. Prognostic you may see it is when you have the effect of that. So, purpose of is predict future outcomes. We have nutrition epidemiology and the dietary intervention to give us future outcomes. So, if you have to predict this, you can use prognostic question. Can dietary pattern predict diabetic risk? This can be one among an example. Or do biomarkers predict the cardiovascular disease? Yes, it is and it has been proven that they can do it. What can be the best designs? Cohort long assessment longitudinal studies. Okay. The last one is the mechanistic questions. What is the purpose of putting the mechanistic questions? Understand the biological pathways. Okay, the example can be how does omega-3 affect the inflammation? Truly speaking, omega-3 really affects the inflammation. How does vitamin D affect the immune function? It increases, no doubt. So, mechanistic questions can also be put and it can be studied. What is the best way to study as in design? Feeding studies. If you're feeding omega-3 you can study the effect of inflammation. If you are a feeding vitamin D, dairy products and all that, you can affect you can see the effect of immune function getting increased. Control metabolic studies. This is very important diabetes patient, thyroid patients and all those. Experimental research of an individual person where you are giving a particular type of diet and you are seeing the effect of this mechanistic question. So, among the descriptive, associated, etiological, intervention, prognostic, mechanistic question, you can choose one or more of the research questions in your nutritional epidemiological studies. Okay. Overview of the available study designs. We have multiple study designs and out of those, whatever available you can get, we will try to understand. Observational studies. In the observational studies, you will get much a picture of ecological studies. Population versus a naive population. So, you can see a national fat intake and the risk of breast cancer. What are the limitations of ecological studies? Fallacy. Ecological fallacy. Cannot control individual confounders who may say that they are not able to meet it. So, ecological studies are benefiting, but among the observationals, you may have some sort of fallacy which is coming as a limitation for this. In cross-sectional study, what it is? So, timing, math examination, dietitian, asking of questions at the same time, same moment. So, quick, inexpensive at prevalence. So, you may say cross-sectional may have some level of limitation. Cross-sectional studies here cannot establish temporality. They have limited causal inference. So, you are not able to meet it up as a requirement. Can be a some of sort of a limitation coming over here. Also, case-control study. Case-control study has a direct to move back forward from past to the past history. So, disease person, healthy person. Colorectal cancer as an example. So, you can say good for the rare disease, efficient for healthy people, but relatively quick for understanding the risk. But, there are few limitations for this. Recall bias as if the person needs to have even selection bias whom you are selecting. So, there is a temporality issue coming over here. So, for every study, you need to understand the limitation and then and then you can move uh to have a selection of a proper design. Cohort studies. What is a cohort study coming in at? Like a direction. It is establishing the temporality. You can study multiple outcomes. So, cohorts are something actually where you have, for example, for 20 years down the line, assess the diet and brain health relationship. Actually, it is there. But, these are very expensive study. Cohort are being taken something very time-consuming and cohort has lost to the follow-up. Sometimes, you may not be able to take a proper follow-up. So, limitations over here are more versus the benefit something for taking such a very directional study. It is very good. Okay, there is one more. Nested case-control or case-cohort. The case-control designs or cohort design over here is being seen nested to a case-control. A particular group of a people selected and they're reducing the cost over here. Limitations can be efficient use of biospecimens. Requires existing cohort infrastructure, which is already being seen something as an also great limitation to be taking this as. Now we move on on the other side of the story that is interventional studies. Interventional studies, I'm not stating that they are comparatively to the observational studies, they are much higher, but yes, with the benefits you can understand. Randomized controlled trial. Where you can see the gold standard for establishing causality. Randomization is like Mediterranean diet versus usual diet. So randomization is giving you a pluses actually to be selecting this. Community or field trials, unit of randomization versus a control group. So you can see salt reduction programs in a village versus a controlled group. So strong causal inference is coming and in a very short duration you are able to given outcome of the study. Community or field trials. Unit of randomization, salt reduction program in a village. And you may find in a cluster contaminated actually which requires a larger sample. So for public health intervention, it is very important that you should go for community or field trials. Then comes the quasi-experimental design. Before and after. So like midday meal, school lunch programs. Portion of yarn. You can have seen lot of this coming after the study and you can read it. You may come up with an example like no randomization but includes intervention. Uh which is done in the previous community trials. Where the practical when RCTs are not feasible, you can do this quasi-experimental. There might be a selection bias is equal to randomized group, but the limitation is like weaker causal inference. So, cause is where you cannot really establish over here. Now, matching research questions with the study design. So, it is very important like you can take it as research question on the one hand and appropriate design on the other hand. Okay, you can take it as an benefit for your own projects. You can also develop project on basis of what are the stated and very proper appropriate designs which are being stated over here. So, prevalence of anemia, if you have to understand as a research question, you should go ahead with appropriate design called as cross-sectional study design. Dietary habits of adults and multiple dietary habits you can study. You can also go with cross-sectional over here. Diet and cancer risk diet and cancer risk is a different type of cancer risk and even diet, whatever diet. So, you can have multiple and different studies. It can be done beautifully with cohort. Risk factors for rare cancer case control. Case control is a very uh proper way to deal with this type of studies. Impact of vitamin supplementation, suppose vitamin B12, vitamin D, vitamin C, vitamin A, RCT. RCT are the best. RCT is the appropriate. Effectiveness of school nutrition program, as I stated, midday meal. Okay. Portion Abhiyan. Okay, or even the nutrition rehabilitation centers. You can see the cluster RCT to be taken as a benefit. Salt intake before and after. What are the policy? You can say natural experiments can be done at individual level also. Biomarkers predicting disease, little costly affair, but you can take it with the help of the cohort. And dietary trends across the countries global level one country comparing with other countries smaller with a larger, so ecological study. You may have few limitations, but yes, you can do it with the help of the ecological study design. Now, the key factors influencing the choice of study design. So, before you go and finalize a study design, there are few factors that determines actually or shape the impact of that design and the choice of that. If you're seeing the disease frequency, rare versus common disease, you may see rare disease you should have a case control to be taken, and for the common disease you can say cohort. Okay? For tuberculosis as an example, go for a cohort. Rare disease if you're taking any, you can say case control. Okay? Latency period, which is very important, cohort are the best and short latency cross-sectional. Exposure frequency, rare versus very common frequency of taking. Suppose for meat or for cake, candy versus very common as fruits or even porridge, okay, rolls. So, you can say if you're taking common exposures, case control. Rare exposure, cohort. Coming as previous one. Temporality needs, again, strong temporality cohort RCT for the tail and weak for cross-section. Causal influence strength, cause and effect, strong causality, you should go with RCT. Moderate, you should go with cohort. Ethical consideration. There is very, very important part whenever you are taking the choice of your study design, you need to understand this. Ethics plays a very important role. Harmful exposure, smoking, alcohol, trans fat intake being seen as very important ethical consideration and you need to make it clear. And must use observational study designs when you are trying to study such ethical considerations. How ethics comes that time observational study designs are very important. Resources, no doubt, this is a very important factor. Money, time, and the infrastructure. So, limited resources, low budget, time, staff, infrastructure, cross-sectional and case-control are the best to deal with this as a challenge. Okay. So, if you take cost control and case control, they can help you to overcome your resource constraints. What is the role of measurement in study design? No doubt, we can understand with through the different methods. Dietary assessment method. So, if you take 24-hour recall, strength are short-term accurate for group means. Limitation, day-to-day variability. And best design context, you can say cross-sectional cohort, multiple recalls can be there. You can do it more faster if you have to take this as a method. And no doubt, 24-hour recalls comes as one of the first choices. If you take food records particularly, you can be detailed real time. You can get a more strength. Strength are more over you. Limitation, burden reactivity. Uh the best design context is like validation studies and short-term trials. This can be coming as an choice for you. Okay. Then comes the food frequency questionnaire, FFQs. Food frequency questionnaire captures usual intake and low button. But they have a limitation of recall bias and limited food list. This is very important. And then you have to choose a design that is large cohort studies and case control. You may have a diet history as being seen as a method which is very comprehensive usual intake done by nutritionist by clinical nutritionist. But it is a very time intensive interviewer bias and all that. Lot of inclusivity. It is like more of qualitative. Lot of subjectivity deals with this. Lot of validation requires and so where I can say or advise take small studies validation. So uh as in hierarchy, we can say 24-hour recall, food records, food frequency, and then diet history. Also, it depend on what is the strength that you have to acquire and to reduce maximum limitations actually. And to make the study more best you should try to do the dietary assessment in that way. What are the nutritional biomarkers? Nutritional biomarkers The objective is like biological indicators that reflect the nutrition intake status, metabolism, or effect. So biomarkers serve multiple purposes. Objective measures, validation tools, mechanistic insights, monitoring and intervention. So you may say like a blood example vitamin C and folate. Urine, nitrogen and sodium. Stool, gut and metabolites. Hairs, long-term exposure of particular nutrition. Adipose tissue, fatty acids, and vitamin D. You have seen a lot of association in our examples. Saliva, cortisol, and vitamin C. nutritional biomarkers over here. Okay. What are the key consideration? Does it reflect only the nutrient of interest? Avoid cross-reactivity with the other factors. The specificity comes as a very consideration. To take it as a proper biomarker. Then you have to see the sensitivity part. Can it detect meaningful difference in intake or status? So, if you are taking any biomarker, nutritional biomarker, you need to also see the sensitivity part. Okay. Timing is also very important. Just like the biomarkers have multiple purposes, you also have a key consideration while selecting for your study design. The particular biomarker, you have to see the timing. Is it a short-term or a long-term biomarker that matches with your study objectives? Your study objectives are important while versus your timing is also somewhere actually which is making it more clear for you as in part. Also additional as a key consider, you need to consider the cost. Cost is very important. Like for every study, the cost is very important. Available resources and burden on the participants. So, participants burden, resources, cost, these all makes very invasiveness actually which is a need that you have to take this cost and invasiveness as very important key consideration while you are taking this nutritional biomarkers in your selection patterns. Although otherwise, they are very very important, but if you see the overall is like more of a cost coming over here. Bias, confounding, and validity consideration. So, if I'm seeing the bias, recall bias, selection bias, measurement bias, confounding, and reverse causality. I may say the recall bias, you come across cases remember past diet differently compared to the controls. There is a prevention strategy that can be used. Validate with the records. Use biomarkers. And preferential design will be cohort or case control. By the bias for selection bias, you can say study sample is not representative of the target population. So, you have to use a prevention strategy like population-based sampling or a high participation response rate. And the perfect design will be population cohort. There can be measurement by systematic error in dietary assessment leading to misclassification of exposure. You may come up with a prevention strategy of validation sub-duty or use a biomarkers again. And preferential design for this, any with validation. Validation is very important over here. Confounding. Confounding, a third variable is associated with it both exposure and outcome distorting the true association. There, you can use a prevention strategy as randomization, matching, and statistical adjustment. No doubt, the preferable design over here is RCT towards cohort. And the bias, like reverse causality, the disease process influences the dietary behavior and not the other way around. So, this is where you can come across the reverse causality. But, you have very strategical prevention. Use a prospective design over here. Exclude early cases and lag analysis. So, you have only cohort versus cross-sectional over here. Now, understanding and controlling confounding in nutritional research is a confused relationship that is common diet confounders and mitigation strategies. You may come across common confounders in diet relationship like lifestyle as a factor or socioeconomic factor, demographic factor, and health status. So, you may have dietary intake which can influence the disease outcome, no doubt. A healthy eating index may prevent a lot of NCDs, non-communicable diseases. Okay, lifestyle factors are many. Smoking, physical activity, alcohol, socioeconomic factors, our education, income, occupation, demographic factors as age, sex, ethnicity, and health status might be coming with comorbidities, BMI, and so many of these. So, your control strategies may come up in a design phase or can come in analysis phase. You have to do a lot of proactive planning or you have to do a lot of reactive adjustment over here. And you have to make it more of match actually versus a lot of multi-variable adjustment can also be done as a control strategies. Epidemiological research validity ensuring accuracy and applicability. So, you may have an internal validity coming with the study correctly measures what it intends to or external validity which results generalizing to the other populations. >> [snorts] >> Or you can even have a construct validity which measures the theoretical construct. And a statistical conclusion validity which corrects the statistical inferences which is also being used to ensure like how the appropriate test are being stated to be used. So, you should have a epidemiological research validity and it is it is ensuring the accuracy and applicability part, which is no doubt very important. Then comes the Bradford Hill criteria. So, you have criteria which comes with nine associations that is strength, stronger association more likely causal, consistency, specificity, temporality, dose response, and possibility, coherence, experiment, and analogy. So, what I can say over here that is there's no single criteria which is essential. Multiple criteria is something like you have to fulfill. And this helps you to guide like nutrition and public health decision. So, it is very important. Like if I have to give you an example over here, high processed food intake association with heart disease. So, this is a Bradford Hill criteria that you can take it to understand the causal inference. Causal inference hierarchy, already we have studied it previously. That highest is something as when you have to take it, the systematic review and meta-analysis will always be very very important to be taken followed by RCT, then prospective cohort, nested case control, case control, and cross-sectional, and lastly ecological, which is weakest. But yes, it can be a choice also sometimes. According to the stars, the strength is there according to limitation which are stated on the either side. Counterfactual framework, we have already studied over here where we can see the causal inference ask you what would have happened if the exposure were different. So, you may have an observed reality versus a counterfactual reality. You may come up with key assumptions for causal inference like exchangeability, positivity, consistency, and no inference. So, you may see that there's a relationship with observed reality versus a counterfactual reality, and you have to come across with the causal inference over here. Direct acyclic graphs, that is DAGs, that we have taken examples in the previous discussions where how the DAG helps you. They helps you in identifying the confounders. They detect the colliders. They visualize the causal effects pathway, and they plan analysis before data collection. >> [snorts] >> There are rules that you can take it as a quick guide for DAG rules. And you can use the DAGs to identify the confounders, adjusting the colliders, see causal pathway, and plan analysis actually. Decision framework, while choosing, you have to take all the nine steps. Define your research question. Identify your question type particularly based on your etiology, intervention, validation, prevalence. Assess the disease exposure characteristic. Evaluate the practical constraints. Consider measurement issues. Assess the bias confounding risk. Select a primary design. You can take it at the first level. Don't be in rush actually to take that. Consider hybrid or alternative designs also to be nested in, and then finally you check against the causal inference need what you have to do it next. When you take the conceptual errors, you need to come up with the pitfall. And you need to understand through this that there are common pitfalls in study design and and and their consequences also. Okay. You may come up with lot of consequences. So, you need to understand it in the whole, and then only you should take it. Is So, these are a few examples you can take a glance actually to understand how it is really important to come up with the wrong design question, ignoring temporality, overlooking variety, and mismatch causality needs. There are some measurement errors which are very very common over here where you can say a dietary assessment, no validation, ignoring biomarkers or wrong biomarkers timing. So, you may have a lot of measurement error attenuation, unknown measurement properties, or you may come up with systematic bias and misclassification. Practical errors are also there. Underpowered study, ignoring loss to follow-up, overambitious scope and budget miscalculation. Very important to take a glance and understand like how practical errors are there when you are not taking it in consideration. This is the exercise you can take it. Where you can take uh that you want to investigate whether high processed meat consumption increases colorectal risk in Indian women age 40 to 65. Etiological factor is very important. What type of research question is required? What study design you should choose? Whether you should stick prospective cohort or case-control? What are the key factors? Dietary assessment recall bias and long latency. What confounders may must control? Age, BMI, physical activity, family history of CRC, screening, and socioeconomic status. What practical constraints must have come? Large sample size is needed. Long follow-up 10 years, not only in fuel day. High cost and loss to the follow-up is also very important. This is also one of very important from the PICO formulation. You can take it. Uh the question is does vitamin D supplementation reduce the fracture risk in middle elderly Indian women? No doubt it is giving you in the PICO question is very right. You can understand like how the population intervention comparison and outcome is very important over here for this study. Ex- exercise three. Bias identification. I advise you to take this exercises and uh whatever we have taken in the previous sections, you should try to apply and solve this as an question. In this exercise, a case-control study found the breast cancer cases reported lower past vegetable intake in than the controls. It's very direct association. It's very important over here. So, in the summary, uh choosing an appropriate study design is important for the most important decision in nutritional epidemiology because it determines the validity, feasibility, and causal strength of the finding. The research objective, temporality, disease frequency, and ethical consideration and available resources should guide design selection. You have a lot of different questions which may give you and land up to you to the different designs. The designs might be cross-sectional, case control, cohort, and RCTs. A researcher must also consider measurement methods, potential biases, and confounding factors and causal inference whenever they have to select a proper study design. So, a structured design decision framework helps ensure to choose the right design. And it helps the researcher with the questions like maximizing the specific scientific rigor they require to minimize the error. So, ultimately, an appropriate study design is essential for generating reliable evidence that can inform nutrition practice, policy, and public health decision-making. So, with this, we come to an end to this chapter. You can take the help of these references to meet up your requirements to fulfill it. Thank you. Thank you very much. >> [music and bell] [music]