E&O_lecture1_2021.pptx

Exposure and outcome measurement in epidemiology M A S T E R E P I D E M I O LO GY UNIVERSITY OF ANTWERP A N N E L I E S VA N R I E OVERVIEW Measurement of exposure and outcome variables: how, when, what to measure ? Directed acyclic graphs (DAGs): which covariates should be measured?...

Exposure and outcome measurement in epidemiology M A S T E R E P I D E M I O LO GY UNIVERSITY OF ANTWERP A N N E L I E S VA N R I E OVERVIEW Measurement of exposure and outcome variables: how, when, what to measure ? Directed acyclic graphs (DAGs): which covariates should be measured? Measurement error: understanding their causes and effects Measurement instruments: questionnaires and scales Environmental and human specimens, biomarkers and surrogate markers Validity and reliability studies Overview MEASUREMENT Measurement is at the core of epidemiology “One must go and seek more facts, paying less attention to techniques of handling the data and far more to the development and perfection of methods obtaining them” Hill, 1953 Class 1 – slide 1 Measurement is at the core of epidemiology “The most elegant design of a clinical study will not overcome the damage caused by unreliable or imprecise measurement” Fleiss , 1999 Class 1 – slide 2 What is measurement? Measurement is the assignment of numbers or labels to objects and events according to rules The use of rules is what determines scientific rigor This requires the use of a protocol, training and the existence of standard operating procedures A method of measurement is considered operational if two requirements are satisfied Instructions for the method must exist and be understandable to other investigators who may want to use them There must be a demonstration (at least a pilot study) that measurements resulting from this method are reproducible. Stevens, Goude, Anderson and Mantle Class 1 – slide 3  Scales of measurement Continuous Nominal Ordered Dichotomous Infinite number Only two values possible no ranking ranking order of values Class 1 – slide 4 Scales of measurement Best approach: always collect variables in the greatest detail possible in order to have maximal flexibility in the analysis Exemption: ethical constraints Class 1 – slide 5 Types of exposures collected in epidemiology  Exposures that cause or are associated with an event or outcome (growth, disease, health,..)  Exposures that confound the association between the exposure and outcome of interest  Exposures that modify the association between the exposure and outcome of interest  Exposures that influence disease diagnosis American College of Obstetricians and Gynecologists, 2013 Class 1 – slide 6 Common exposures in epidemiology  Demographic (age, sex, race,..)  Psychosocial (stress, stigma, discrimination…)  Diet  Clinical (anthropometrics, comorbidities, surgery, medication, hemoglobine, lipid profile,…)  Substance abuse (alcohol, smoking, drugs)  Occupation  Environmental exposures  Genetics (host or bug) Class 1 – slide 7 Common difficulties in measuring exposure We are often interested in things are not straightforward to measure Example: Does environmental lead exposure affect cognitive achievements? Does exposure only matter in childhood? Class 1 – slide 8 Exposure can be difficult to define Example: alcohol use in different cultures Example: socio-economic status different cultures Class 1 – slide 9 Exposures fluctuate over time, even within the same person Class 1 – slide 15 The true exposure is often not measurable, resulting in the need to choose a surrogate When choosing the surrogate, one needs to balance sensitivity and specificity accuracy with patient burden accuracy with cost Class 1 – slide 10 Exposure can occur through many sources Example: Does exposure to aluminum welding fumes can cause pulmonary fibrosis? Class 1 – slide 11 Conclusion True exposure – which is what we want to measure- is seldom measured The first step is to acknowledge this. Then try to create the operational definition of the best possible exposure that can be measured Class 1 – slide 12 BIOLOGICAL AGENT OF INTEREST Isolating the biological agent of interest By specifying the active agent, the researcher can make the measurement of the exposure - specific: isolate the putative causal agent ◦ If caffeine is the exposure of interest, do not ask about coffee in general but exclude decaffeinated coffee - sensitive: include all sources of exposure ◦ While coffee is the main exposure of caffeine, some other drinks contain caffeine, some sports gels contain caffeine and some medication contain caffeine Class 1 – slide 13  Does vaccination cause autism due to exposure to mercury? o Thimerosal, a compound that contains ethylmercury, was used as a preservative in multi-dose vials of vaccines o Mercury is a metal found in air, water and soil as it is released into the environment from volcanic activity, weathering of rocks and as a result of human activity (batteries, lamps, dental fillings,...) o Exposure to small amounts of mercury can cause health problems in utero and early in life o People are mainly exposed to methylmercury, an organic compound, when they eat fish and shellfis14 Class 1 – slide 14 Doe eating meat cause colon cancer? Red meat may affect cancer risk because naturally occurring chemicals formed during digestion have been found to damage the cells that line the bowel. Processed meat is often made from red meat and it also contains added nitrates and nitrites which are also broken-down during digestion to form chemicals that can cause cancer. Barbecues and charred meat may increase the risk of cancer. Substances called heterocyclic amines are formed in foods that are cooked at high temperatures and blackened or charred. In animal studies, heterocyclic amines are carcinogenic (cancer causing). However, the evidence in human studies is not clear. Another factor could be that those who eat a lot of meat may miss out on eating other protective foods such as fruit and vegetables or wholegrain cereals. Class 1 – slide 15 Does smoking cause prostate cancer? Different types of smoking (cigarettes, cigars, pipe, weed, E- cigarettes, …) Different toxic components in different types (nicotine, tar, acetone, …) Quantifying the dose: how much do you smoke? Class 1 – slide 16 CONCLUSION: considerations when selecting exposure measurement 1. How does exposure relate causally to the outcome of interest? ◦ What is the exposure-disease pathway ◦ What is the biological agent of interest? 2. What is the best way to represent the amount or dose of exposure? ◦ Dose-response relationships are an important base for inferences about etiology 3. What is the most critical time period during which exposure is likely to be causally associated with the outcome? 4. How can one translate all the above into one variable that can be used in the analysis? Class 1 – slide 17 THE EXPOSURE DISEASE PATHWAY Exposure-disease pathway: M. tuberculosis Example 1: Exposure to MTB and latent TB Example 2: Exposure to MTB and active TB disease Example 3: Tuberculosis treatment for active TB Cobat et al 2013 Class 1 – slide 18 Exposure-disease pathway: vitamin D Osteoporosis is a long-term effect of calcium and vitamin D insufficiency Vitamin D is a fat-soluble vitamin that is naturally present in some foods, added to others, and available as a dietary supplement. It is also produced endogenously when UV rays from sunlight strike the skin Vitamin D obtained from sun exposure, food, and supplements is biologically inert and must undergo two hydroxylations in the body for activation. Serum 25(OH)D is a biomarker of exposure to vitamin D The extent to which serum 25(OH)D serve as a biomarker of effect (i.e., health outcomes) has not been clearly established Class 1 – slide 19 The exposure-disease pathway: some thoughts Exposure to disease is not a single step but a pathway of multiple steps, a sequence of causal events or states Exposures can be measured at multiple steps/states Disease can be measured at multiple steps/states As you move down the exposure-disease pathway, what one measures is closer to the causal agent, which makes it more likely you can show a non- biased association between exposure and disease. As you move up on the exposure-disease pathway, what one measures is further removed from the causal association between exposure and disease As you move along the exposure-disease pathway, the perspective of the study shifts from a public health to a basic science. Both perspectives are valid, but one needs to be clear what the goal of the study is. Class 1 – slide 20 EXPOSURE DOSE Defining the dose of exposure Binary exposures are easy as there is no dose ◦ Is exposure to secondary smoke a binary exposure? ◦ Is the presence of a mutation in the human DNA a binary exposure? Most exposures are more complex and have a dose component with variations in dose ◦ Duration (how long, when start…) ◦ Frequency ◦ Intensity (depen on season,…) ◦ Example All complex exposures can be simplified, but many simple measures result in substantial measurement error Class 1 – slide 21 Cumulative dose  Cumulative dose = the level or quantity of exposure over a time period Cumulative dose = ∑i (yearsi x frequencyi x intensity)  Sometimes the equation can be simplified if there is little variation in one of the three components between subjects (although this is uncommon)  Sometimes the equation needs to be more complex: if frequency changes over time (example physical activity in summer and winter) if multiple exposures need to be taken into account (example vitamin D) - Class 1 – slide 22 Concepts in dose of exposure The cumulative dose can obscure the relationship between exposure and outcome if - one component is more important than another. example: smoking relates to cancer by the second power of intensity (number of cigarettes smoked per day) and the fourth power with duration of smoking - there is a dose threshold, where only exposure above a certain dose is associated with the outcome example: alcohol use or binge drinking - Class 1 – slide 23 Example Molina PE, Nelson S. Binge Drinking’s Effects on the Body. Alcohol Research, 2018 “A limitation to our understanding of the consequences of binge alcohol consumption on organ injury is the lack of information on the time period, duration, and number of binge occurrences that describe the long-term practice of binge drinking. Preclinical studies conducted under controlled conditions provide opportunities to examine quantity and frequency variables in the investigation of the effects of alcohol consumption on organ injuries. However, interpreting, comparing, and integrating the patterns of alcohol consumption described in clinical reports is difficult because of the different types of data collected across studies. This difficulty underscores the need for researchers to perform more rigorous comprehensive and systematic data collection on alcohol use patterns Class 1 – slide 24 Operationalizing exposure measurement in an epidemiological study Class 1 – slide 25 EXPOSURE TIMING Timing of exposure Critical or etiological window of exposure Not all exposures matter… only “being at the wrong time at the wrong place” Class 1 – slide 26 Etiological window The critical exposure window is defined as the time period during which the exposure has the greatest causal effect on the outcomes. Inclusion of exposures outside of the “etiological window” will result in exposure measurement error. Examples  prenatal exposure: different effect of exposures in 1st, 2nd and 3rd trimester  exposures during different phases of childhood  effects of cannabis exposure during the peripubertal period is a risk factors for the emergence of schizophrenia-like deficits  time in relation to menopause  Postmenopausal women have a higher risk for cardiovascular disease. Only looking at age of women will result in measurement error as not all women start menopause at the same age. Class 1 – slide 27 Induction period The induction period for an exposure is the period of time from when the causal action of exposure is complete until disease initiation Exposures early in the sequential ‘component causes’ of disease have a longer induction period then causes occurring late in the sequential component causes of disease By definition, the induction period for the causal exposure that acts last is zero The induction period is thus not for the outcome (disease) but for the exposure Class 1 – slide 28 Induction period Any factor that postpones the onset of an event after exposure to a causal factor will increasing the induction period for another cause and is a preventive exposure. If we prevent death = we postpone death Even for outcomes such as cancer, which are assumed to have a long induction period, both exposures that occurred a long time ago and exposures that occurred more proximal to the event can be important. Class 1 – slide 29 Latent period  The latent period is the interval from irreversible effect of exposure to the occurrence of symptoms ◦ Example: time from one cell or clone of prostate cells that has irreversibly gained a cancerogenous mutation to start of difficulties with urination ◦ Example: time from infection with HIV to the first symptoms of AIDS Class 1 – slide 30 Latent period  The length of the latent period can be modified by factors influencing disease progression  The latent period can be reduced with improved diagnostics ◦ In contrast: the induction period is not changed with improved diagnostics because disease occurrence marks the end of the induction period Class 1 – slide 31 Exposure in the latent period  Exposure of the etiological agent during the latent period do not contribute to the etiology. Including them may thus lead to misclassification of exposure  Exposures during the latent period can contribute to o speed of disease progression o clinical presentation  Exposures during the latent period can be influences by the disease “reverse causality”  Example: In an observational study, researchers may observe that people who earn higher annual incomes report being happier overall in life. Thus, they may simply assume that higher income leads to more happiness. However, in reality it may be that people who are happier are better workers and earn higher incomes. Thus, higher income might not cause more happiness. More happiness might be the cause for higher income Class 1 – slide 32 CONCLUSION 1. Including exposures during the latent period can result in misclassification as they do not contribute to causality 2. Including exposures during the pre-clinical period, when they could be influenced by the disease process can lead to reverse causality ◦ Behavioral exposures should only be included before any symptoms occur. ◦ This is usually done by establishing a reference date which is then applied to both cases and controls. ◦ Example: 2 years for colon cancer 3. Timing of exposure is especially tricky to determine in case-control studies, as definition measurements are done after disease onset ◦ In case control study, ‘only’ reliable measure of exposures are those that are fixed over time(e.g. genetic info) ◦ Class 1 – slide 33 Unfortunately, the timing of the etiological window, the induction period and latent period are often not known Class 1 – slide 34 Relating timing of exposure to disease risk 1. Latency analysis or exposure lagging ◦ When computing cumulative dose: exclude exposures that occurred in the induction or latent period ◦ Example: a cumulative dose of asbestos excluding the most recent 10 years of exposure showed the strongest association with lung cancer 2. Exposure window analysis ◦ Only counts exposures that occurred in the etiological window by excluding exposures before and after ◦ Example: a cumulative dose of asbestos for the time window 10 to 25 years before diagnosis/reference date in a case-control study had the greatest power to detect an association with lung cancer Class 1 – slide 35 Relating timing of exposure to disease risk 3. Time since first exposure or age at first exposure ◦ Important for causes that are early on the sequential cause chain ◦ Example: Age at immigration is important for risk of melanoma 4. Time since last exposure ◦ Important for causes that occur late in the sequential cause chain as those who were recently exposed will be at greatest risk ◦ Example: current smokers have a threefold risk of myocardial infarction whereas those who quit smoking two years before had risks close to never smokers 5. Regency analysis ◦ Considers only recent exposure ◦ Works for exposures that are causal in the later part of the sequential cause chain ◦ Example: smoking in the past year is the best exposure window for the association smoking-myocardial infarction Class 1 – slide 36 Statistical approaches to relate timing of exposure to disease risk 1. Exam cumulative exposure dose for different non-overlapping time periods in the same model and determine which has the greatest strength of association 2. Exam multiple overlapping periods in separate models and compare the models for goodness of fit 3. weighted cumulative exposure, with each time window carrying different weights Caution: - Exploratory modeling tends to overestimate the effect size - Components of dose and time are correlated, so need to adjust for other exposure windows Class 1 – slide 37 Conclusion: measuring exposure step 1 1. What is your research question? 2. What is the biological agent of interest? 3. Where in the exposure-disease pathway do you want to measure exposure? Now you have identified what you want to measure. Class 1 – slide 38 Conclusion: measuring exposure step 2 1. What is the ideal and most feasible method of measurement? 2. How can I ensure my measurement method is both sensitive and specific? 3. What is the timing of exposure I want to include? - When Now you have identified what, how and when to measure. Class 1 – slide 39 Conclusion: measuring exposure step 3 1. What is the duration of exposure? 2. What is the frequency of exposure? Is there important within person variability of frequency? 3. What is the intensity of exposure? Is there important within-person variability of intensity? Now you have determined how to operationalize your measurement All information above should result in the instrument and SOP for measurement and later be applied to the development of the analysis plan Class 1 – slide 40 An excellent epidemiologists avoids….  Next class 1. Possibility for asking questions relating to the material presented 2. Exercise to implement what you just learned