Week 5: Lecture 21: Biases in Nutritional Epidemiology

Hello and welcome dear learners to the NPDL course on nutrition epidemology of way towards a healthy life. In the module 2 study design and epidemological methods for the chapter 21, we'll be talking about the biases in nutritional epidemology. Why biases matters and biases versus random errors. The major type of bias and special biases in nutritional epidemology. Also, we will be talking about minimizing the bias and case discussion on bias in nutritional epidemology will be taken higher. So why bias matters in nutritional epidemology? Establishing a robust causal link between dietary exposures and disease outcome is notoriously difficult. Diet is a complex continuous and highly correlated exposure matrix. Bias matters because it represents a systematic error that distorts the true association between dietary factor and a health outcome. Unrecognized or unmititigated bias leads to speurious associations. treating internal validity and consequently misguiding public health nutrition policies, dietary guidelines and clinical recommendations. So you can understand why bias needs to be taken very seriously over here. Recognizing the systematic errors is critical step in rigorously study design and interpretation within the hierarchy of evidence. So you can understand over here what is the difference between bias and random error. On the other side you have the random error that is precision error and bias that is called as systematic error. So systematic deviation from the true value is what bias is and fluctuation around the true value is what the random error is. Consistence direction example always under reporting which is being seen in systematic error that is bias and in random error it is oscillates around the truth. It is around the truth that comes as a random error. Effects on the estimate is invalid in bias that points the estimate where versus in the random error the estimate effect on estimate is less precise but not average valid. Effect of sample size in bias does not appear with large samples and effect of sample size in random error shrinks with larger samples. So you can understand how it comes with the precision versus with the systematic error. Example in nutrition if I have to give you about the bias is under reporting of total energy or social desiraability and selective reporting for the random error. It is daily variation in food intake with person fluctuation. So this is a very clearcut differentiation between bias and random error. Now we'll try to understand what are the major types of bias. There comes the selection bias as first followed by information bias that is measurement bias and confounding bias. What is selection bias as such? Selection bias occurs when the systematic criteria used to select the subjects for a study or factors influencing the retention are associated with both exposure and outcome. You can understand with the example of the prospective cohort study where the cohort is recruited. Healthy diet is in green given as low risk. Average diet which is being shown in brown which is an average diet and in red that is poor that is called as high risk. After 20 years you can see loss to follow poorer diet and higher cardiovascular risk. The impact is the remaining sample appears healthier leading to an underestimate of the true diet disease association. So you can see there can be overweight, obesity, high risk participants, poor dietary habits, higher risk of CBD or more likely to drop out. So this is a selection bias. So for example, loss to followup in a 20 years prospective cohort study where the participants with the poorest diet and highest cardiovascular risk systematically drop out skewing the remaining samples. So the exit may be likely to be drop out. Now we can see the information bias that is called as measurement bias. This bias arises from systematic errors in measuring the exposure outcome or co-variants. Non-ifferial mclassification versus differential misclassification. So in non-ifferial misclassification, it occurs when the error is measuring diet is independent of DC status. Often pulling results towards the null. So you may have a diet assessment same type of error in both the groups cases and controls and the estimate effect is towards null where versus in the differential mclassification it occurs when the error depend on the disease status. Example recall bias in case control studies where the cases remember their past diet differently than the healthy controls. You can find a difference where always from the null over the underestimate. So the common sources of information bias is recall bias, reporting bias, under or over reporting and misclassification bias, interview bias. So what is impact? Impact can either exaggerate or underestimate the true diet disease association. Now the third as an confounding bias. Confounding bias while technically distinct from bias as it represents a true though non-causal association. Confounding acts similarly by distorting exposure disease relationship. A confounder must be associated with the dietary exposure an independent risk factor for the disease not lie on causal pathway. So the confounder for example physical activity, smoking, socio economical status can affect the dietary exposure. Example high rate meat intake or disease outcome might be cardiovascular disease. Examples of the confounder can be physical activity, smoking, socioeconomic status, other dietary factors or alcohol consumption. What is the impact? The impact of the confounder is associated with both the dietary exposure and the disease leading to speurious or exaggerated association. So this way you can understand the major types of three biases. Now what is special biases in nutritional epidemology? So there are four types of biases in nutritional epidemology. the measurement error in diet, reverse causality, social desiraabilities bias and energy misreporting. So in the measurement error in diet, inaccurate estimation of dietary intake due to limitations of dietary assessment methods. It can be FFQ, 24 hours recall or dietary record. The example is underestimation of portion sizes in a food frequency questionnaire where there can be a measurement error which respondent is difficult to gauge. What is the impact? It can weaken or in some cases it can distort the true diet disease association what actually the purpose of the estimate is. So how to minimize this measurement error? You can use a validated tool which is pre-ested or multiple dietary assessment actually to curate or to minimize the error. [snorts] Portion size aids and training to the interviewer or someone who is trying to assist actually for the questionnaire to reduce this. Adjust for total energy intake to take accumulative result of the assessment. Use biomarkers which are doubly labeled water se serum kerotenoids and other so you can get a clear or closer estimate of the diet assessment. Then comes reverse causality. What is reverse causality? the disease or preclinical condition influences dietary intake rather than diet causing the disease. So there is a early disease symptoms and change in diet. Okay. So you can understand where as an example if I have to quote it for relating people with early signs of hyper cholesterolmia reduce fat intake before the diagnosis. So what will happen? It will be closer to the disease condition but not the exact disease condition. The impact may lead to false inverse association or obscure the real effect of the diet. So the diet should come after the disease occurrence. So how to minimize this reverse causality pas? You should use a prospective study design which is after exclude early years of followup lag analysis which is very important to reduce this bias. Also you can repeat dietary assessment which is very oftenly done to improve the assessment. Also careful interpretation of the finding is very important for the researcher to do it. If you overcome or you try to overcome this bias, then comes social desiraability bias. This bias the participant report what is socially acceptable rather than their true dietary intake. So there is a difference which is being put as in subjective or under the pressure of social disability. Okay. There is one thinking that I should eat healthy which is being stated over here for a person actually over here who is in the dilemma where versus with the social disability the person will say I will report only healthy foods. Okay. So what is the effect? There can be over reporting fruits, vegetable, whole grains. The person may be in that social disability where versus there will be under reporting for sugar, food, fried food, alcohol, beverages and so much of things. You may say example of over reporting of veggies, fruits, intake of under reporting of sweets, candies, snacks, cakes, alcohol and beverages. What is impact? The impact leads to overestimation of healthy diet which will make a very different picture and underestimation of unhealthy diet which is not being told. So how to minimize this bias? So you can assure confidentiality to the respondent and try to take the original information. You can also use neutral or non-judgmental questions which will make it very clear to overcome this bias. Also, you can use biomarkers and objective measures to estimate accurately the desired information. Also, you can validate self- reports where the respondents can also confirm the given answer. Then comes energy misreporting. So this is a bias where the systematic under reporting or over reporting of total energy intake is done. In under reporting it is which is very common in all diet assessment. The reported intake is less than the actual intake. Well versus there might be in proportion, measurement, timing, volume, type and so many other under reporting can be there which is very useful where versus for the over reporting part which is lesser common where reported intake is more than actual one which doesn't seems to be there might be in some cases we can see this example of this energy misreporting can be overweight. Eight individuals under report might be because of the desire, shame and usual habit. Calorie dense food and total calories also can be seen to be shown as under reporting. What is the impact of this? The impact is distorts the nutritional intake. Estimates can bias diet disase association which is where the impact is already seen on a lower side. How to minimize this bias? You can use multiple 24-hour recalls which can reduce maximum of this energy intake possibility checks like Goldberg's cut offs also the use of biomarkers is advised where you can have a doubly labeled water so you may understand I have taken it again you are taking it so you may get a sort of an also So statistical methods to adjust for misreporting. So you can overcome with the help of the statistical method if there is misreporting done. Now we will understand minimizing bias. How to take ahead minimizing the bias. So in all the four different phases particularly we can minimize the bias step by step in the design phase strategies where for the randomization for the trials suppose you have a group A and group B which eliminates the confounding bias by design. Randomization where random assignment of participants can be done prospectively on the basis of the cohort design where versus on the prospective cohort design that tracks forward reduces the recall bias. Also in careful sampling you can ensure the representiveness document and nonresponse participant by ensuring this and you can include versus exclude for those participants very carefully. So this way in the design phase you can do randomization prospective cohort design and careful sampling to reduce or to minimize the bias in the exposure measurement. What can be done? There are also three types over here. Multiple past 24-hour recalls or food records. These are the two ways. Out of the two ways in the very first the use multiple past 24-hour recall or food records which reduces the memory or portion size error and you can calibrate the self-report with the biomarkers which will help you to give the exact measurement. In the second case, what you can do calibrate self-report with biomarkers which is also very physible calibration a regression calibration which will give you a very exact information. Also there can be a third way that is harmonized dietary instrument for comparability across the studies. It is very important that you can harmonize the dietary instrument which will give you a very closer and reduce the exposure med. In the third case of minimizing the bias, the confounding can effect estimation adjustment can be done. There also there are three ways to do that. Prespecify confounders like age, sex, BMI, PA, education, income, smoking. Collect carefully. Multiple regression propensity score methods structural models this can be done to reduce the residual confounding which is pre-specifying confounders especially we use age sex BMI education income and smoking and in multiple regression we can use the propensity score or structural modeling then the Third is transparently report adjustment sets where it can avoid the over adjustment for mediators. Reporting and analytical transparency is also one of the way of minimizing the bias. There also you can do it by pre-register study protocols and analysis plans. This can be done on public registers. registers all analysis and avoid selective reporting which can report both positive and null findings. Risk of bias tools in systematic reviews use risk of bias tools. So this can be adapted for nutrition separately quality from bias in systematic reviews. Now we will have case discussions. There are two sets of case discussion that we will be discussing about. One is of a global studies and second is of Indian studies. In the global studies we will have dietary fat and heart disease. Historically in the western countries it has been seen and the observations that are made are suggesting that higher fat increases cardiovascular risk. The confounding bias over here can be smoking, physical activity and socioeconomic status. What were the initial results of most of the global studies? They were overestimated harm of fat due to an unadjusted confounders. So confounding bias was seen something to be adjusted over here. What are the evidence inside? Nutritional epidemology criticized for difficulty accurately measuring diet and reliance on observational design leading to misleading confusions if bias are not addressed. There is also an association of betaarotin and lung cancer. Observational versus randomized control trials. observation where high beta carotene lowers lung cancer. Healthy user bias confounding overall and healthy lifestyle is something where you can use as a confounding bias. In RCTs, beta carotene supplement increase lung cancer risk in smokers. So interpretation is observational finding confounded by healthy lifestyle patterns. So you can have a differentiation to be seen over here. Sugar intake and obesity. The global surveillance data states that there is energy misreporting by most of the people who take more in sugar versus social disability will also become as in diet actually where the reporting is under. So overweight individuals under reportport energy intake especially sugar and fat weakening the association. So global dietary surveillance which is done in multiple country give the comparison that there is measurement bias and selection bias non-standard data and misleading to international comparison and policy direction versus in the Indian studies you may have diet and diabetes in urban India where you may understand that there is a role to be seen that India is having more number of diabetic patient particularly in urban setups. The context is the participants under reportport oil, sugar, processed food and beverages, underestimation of diet, diabetes association where the bias is mostly being coming from social deservity bias or sometimes from recall bias in some areas. The National Nutrition Surveys which are supported by NFHS and NSSO which usually gives us a ecologic policy household versus individual intake where you can take a intrahold food distribution measurement error or gender age differences can be coming as some sort of differences from this studies where you can get a very important view of intrahousehold food and distribution of food which leads to miscalcification. Nutrition transition studies in India where you can find a rapid dietary change due to urbanization is one away and in the other way you may find transitional to processed food. So there is a rapid dietary change in urban setups and a traditional to processed food which is also being seen as an habit. There comes a confounding soioeconomic status and urbanization which is putting is this as a bias where the urban samples are over presented. So what is the evidence? The food system inequalities and changing environment influence patterns difficult to isolate diet effect from income, lifestyle and access to healthare. So these are very important to understand the nutrition transition studies in India. micronutrient deficiencies such as iron, vitamin A, vitamin D where you can come across a measurement error, misclassification bias and the issue comes with poor recalls of proportion sizes. Seasonal variation in diet was not captured also leading to inaccurate estimation of deficiency prevalence is also being seen. So you can understand there are variations in diet and diabetes in urban India. National nutrition surveys which are giving us a greater information towards adding for the nutritional epidemology and nutritional transition studies in India which are back also micronutrient deficiency studies are also done. So with this we come an end to this chapter where bias in nutritional epidemology is being referred to systematic errors that distort the true relationship between diet and health outcomes. The bias is commonly arising from issues like inaccurate dietary measurement confounding by lifestyle factors and selective reporting which is very being common. This is because the diet is difficult to measure precisely. These biases can lead to over or underestimation of association. So there are some special challenges that include energy misreporting, social desiraability bias and reverse causality that can affect it. If not addressed, this bias can mislead research findings and public health recommendations. Therefore, carefully study design, accurate measurement and transparent analysis are essential to minimize this bias and ensure valid conclusion. So with this we come an end to this chapter. You can take this references as a guidance for this and you can understand more in detail about the nutritional assessment. Thank you. >> [music] [music]