Intro to Epidemiology Exam 2

Farming and Bladder Cancer Example

"Can you tell if length of farming is associated with bladder cancer?" Given Length of Farming groups, persons examined, cases of bladder cancer, and incidence rate -see pic Process -so how can we show that the longer you farm, the more likely you are to get bladder cancer? Risk of outcomes and Relative Risk So first lets use the incidence rates at the ages -15-29 Years: there 1 out of 123 people w cancer, an 8.1 chance out of 1000 to get bladder cancer -50 Years: theres 14 out of 189 people w cancer, a 74.1 out of 1000 chance to get it What is the strength of association? Find the relative risk: (a/a+b) / (c/c+d) -we get 9.11, or it is 9 times as likely for someone who has 50+ years farming to get cancer than someone who has 15-29 years farming to get cancer

Systematic Error (Bias)

Bias (Systematic Error) -can directly lead to incorrect study result -creates association that is not true -the study result occurs because non-random issue involved in carrying out research -problems in picking study participants, measuring variables, or how subjects participate in follow-p of outcomes -no study is bias free so need to evaluate likely level/importance of bias BIAS -systematic error in design, conduct, or analysis of study that results in mistaken estimate of true exposure-outcome relationship Major Categories 1. Selection Bias 2. Information Bias -non-random errors in process of selecting/analyzing participants -problems in how participants are selected or how subjects participate in follow up of outcomes -results could be biased if study pop. is picked specifically a not representative of general pop. for ex. picking participants more/less likely to develop health issue 2. Information Bias -non-random errors in measuring study variables

Period Prevalence

Number of individuals in a specified population at risk who have the disease of interest over a specified period of time -Denominator is estimated population at mid interval, so subtract any deaths before the midpoint for denom -Numerator is number of people with disease •Must establish a specific time period of interest •The time interval begins at a specified point in time •The time interval ends at a specified point in time Prevalence # of peeps w disease ------------------------ Estimated Population Other Facts -only looking for # of people with disease during a specific point in time, not specific to only NEW cases, just everybody with it -ex: students using ritalin during finals week -only study specific time period, not over years

Incidence Density Rate

Number of new cases of disease occurring over a specified period of time in a population at risk (throughout the interval) -specific to NEW CASES, how many peeps are developing it, a new case, only counting the NEW cases as they come -time interval is over years •the denominator is person-years •person years = time free of disease •e.g., an individual is followed in a study for 4 years. If healthy, they contribute 4 person-years to the denominator. -used for Rate Ratio, Relative Risk (Measure of Association) -over multiple years -represented by person-years (need reoccurring check-in)

Prevalence vs Incidence

Prevalence is the proportion of cases in the population at a given time rather than rate of occurrence of new cases (incidence( Thus, incidence conveys information about the risk of contracting the disease, whereas prevalence indicates how widespread the disease is. -Prevalence refers to proportion of persons who have a condition at or during a particular time period -Incidence refers to the proportion or rate of persons who develop a condition during a particular time period.

Randomized Controlled Trial Study Design

RCT Design -strong support for casual inference -hypothesis testing -experimental manipulation of variables, control -participants assigned level of exposure and/or intervention, done randomly -higher quality evidence, most convincing evidence of E/D -clinical trials are most well-known design -subjects randomly assigned to treatment or comparison groups -cant use RCT to test effects of exposures that can be harmful *ethics Structure -persons w/out outcome recruited as potential participants -participants assigned to either group and given that treatment -rate of development of outcome is determined and compared UNIQUE: peeps don't pick treatment they receive, randomized = balances attributes so lessens outcome being caused by something else Groups Treatment Group -subjects receiving new treatment Comparison Group -subjects receiving no treatment (placebo) or current treatment Advantages -control level of exposure so better knowledge of exposure -higher confidence of E/D relationship -low variation among subjects cu randomization -establish temporality, know exposure before outcome Limitations -very expensive, time consuming -not appropriate for certain hypotheses cuz ethics, cant expose them to harm -length of prospective follow up is 3-5 yrs -interventions that are difficult to maintain compliance - not feasible (intensive exercise), too many stop

Surveillance

The systematic and ongoing collection, analysis, interpretation, and dissemination of health-related data. Gathering information on a health issue as it exists in the community Two Types: Active and Passive Why is it done? - Identify health status of public -Define public health priorities and actions -Provide evidence to move towards disease prevention and control -Provide evidence to evaluate prevention activities What does it do? -Estimates magnitude of problem -Determine geographic distribution of illness -Portray the natural history of disease -Detect epidemics -Generate hypotheses, stimulate research -Evaluate control measures -Monitor changes in infectious agents -Detect changes in health practices -Facilitate planning

Association or No?

There is an Association Correct - outcome really is influenced by the exposure Chance - random error leads to the finding Bias - systematic error leads to the finding Confounding - distortion from a third factor There is NO Association Correct - outcome is not influenced by the exposure Chance - random error leads to the finding Bias - systematic error leads to the finding Confounding - distort

T/F: An advantage of the historical prospective cohort design compared to the case-control design is that the cohort design can directly estimate risk of the outcome.

True

Structure of Historical Prospective Cohort Study

1. Identify hypothesis to investigate and population group to study 2. Determine time frame for the study 3. Identify and plot when disease occurs using existing data 4. Calculate incidence rates for exposed and unexposed group 5. Calculate the Relative Risk first choose time period in past for study, the identify the exposure at that baseline and follow subjects forward in time to classify when outcome happened -follow up periods of 5-30 yrs

Methods Preventing Error

Good Methods -diligence in selecting participants -having a meaningful number of study participants -using proper measurement of study variables -training staff on measurement and cooperation with participants -and using proper statistical assessments

Wk9 -Quiz: Which is not conducive to a successful screening program?

short duration of subclinical disease -a disease with a short time period for the subclinical disease phase would not work well for a population screening program -if its short, too many people would finish the subclinical stage before they could get screened, missing many individuals who could have benefitted from screen Successful Screening Tests -disease is high in prevalence -severe disease -long subclinical phase of natural history -accurate, low cost, safe -treatment available that help subjects with disease

Effectiveness

the extent to which an intervention, when deployed in the population, does what it is intended to do (i.e. produces a beneficial result)

Scientific Method

-used to find determinants of disease in analytical epidemiology Method 1. Identify a research idea/question to investigate 2. Frame a hypothesis related to the question 3. Design a study to answer the hypothesis 4. Collect data in the study 5. Determine results and assess their validity 6. Determine if results support or refute your hypothesis -Begin by formulation research idea then question then framing a hypothesis from the question so its testable. Ideas come from -any suspicion that a factor may influence occurrence of disease -observations in clinical practice or lab research -examination of disease patterns -ideas from evidence in descriptive epidemiology such as patterns of disease and what patterns mean for possible factors

Screening Accuracy

-depends on method and tech -accurate if both reliable and valid Reliability -produce the same results when given under varied conditions -if the test is conducted multiple times on the same sample, it will show the same results. Validity -ability to discriminate b/w disease and no disease -assess validity by comparing result of screening test to result of diagnostic test. --Did it test positive when there was a disease? How many tested negative when they actually were positive? -high sensitivity and high specificity Good Screening Test -high in sensitivity, high in specificity, high in predictive value positive, high in predictive value negative -high benefits (improves health to most) and low risks (low potential harm) -can have different accuracies between populations (breast tissue affects accuracy of mammography)

Registries

-type of surveillance -incidence -commonly focused on chronic diseases or injuries, issues not captured through notifiable disease approaches -tries to identify all cases of injury or disease (aim is 100% certainty) -high accuracy = better description of frequency and key health problems -enhances decisions on potential hypotheses and prevention -data concerning all cases of a disease in a defined population so cases can be related to population -require collection of information from multiple locations -slow procedure, expensive, but VERY high certainty -EX: Trauma registries for injury data -helps identify trauma needs, injury prevention efforts

Measures of Association

-calculations to portray numerical relationship b/w a potential determinant/cause and disease -quantify if level of disease/outcome is different between subjects exposed to potential cause or not Absolute Risk Measures -risk difference/attributable risk = exposed - unexposed -indicate magnitude of risk in group of persons exposed -important for clinical and public health policy -evidence already shows exposure is important, identifies interventions to reduce exposure Relative Risk Measures -relative risk ratio -odds ratio = exposed / unexposed -indicate whether exposure is associated w/ increased risk of disease -important for identifying potential causes of disease -communicates degree E and D are related and if hypothesis is correct -how likely is subject to devleop disease if exposed? Risk -likelihood that health outcome will develop where it did not exist before -describes incidence of disease -characterizes relationship between exposure and devleopment of disease

Relative Risk

-characterize incidence of disease (the chances of something happening)/ (the chances of all things happening) -Denom: population at risk, all in pop at risk Relative Risk Incidence Rate of Disease in Exposed / Incidence Rate of Disease in Non-Exposed -identifies risk of outcome, uses prospective timeframe like cohort and trial Types Risk Ratio Cumulative Incidence of Exposed / Cumulative Incidence of Unexposed -best for cumulative incidence -encompasses entire timeline in study Rate Ratio Incidence Density of Exposed / Incidence Density of Unexposed -best for incidence density -over multiple years -represented by person-years in study so need reoccurring check ins Table Measurement (R.R): (a/(a+b)) / (c/(c+d)) -Num: incidence in exposed group -Denom: incidence in non exposed group -The comparison group is typically what investigator believes is the level of exposure in the population that has nothing to do with the disease

Odds Ratio

-characterize prevalence of disease -seen w cross-sectional and case control -frequency of exposure in the cases and in the controls for ex. Formulas (the chances of something happening)/ (the chances of it not happening) -Denom: population that did not have event Odds of Exposure in Disease Group / Odds of Exposure in No Disease Group -usually applied prevalence, not incidence Measurement of Odds (a/b) / (c/d) or (a*d) / (c*b) Chart Parts a = number of persons exposed and with disease b = number of persons exposed but without disease c = number of persons unexposed but with disease d = number of persons unexposed: and without disease a+c = total number of persons with disease (case-patients) b+d = total number of persons without disease (controls) Interpretation of Results 1 = No Association b/w E/D -prevalence rates are identical b/w groups >1 = Positive Association -disease group has higher prevalence of exposure -exposed groups have higher prevalence of disease <1 = Negative Association, Protective Effect -disease group has lower prevalence of exposure -exposed group has lower prevalence of disease

Casual Inference

-complex health scenarios -guide decision making -process of making a general statement about causality on basis of existing evidence -involves judging quality or strength of evidence in each criteria (Hill) -strengths and weaknesses weighed against each other -decision should be open to reevaluation as new information is available Pyramid of Associations -largest proportions of finding support null hypothesis where exposure cannot lead to the outcome -other associations may be statistically significant but affected by error so that it is doubted -top of pyramid, smallest amount, are causal relationships -Evidence is evaluated from perspective that exposure is likely to be causal rather than exposure is casual -info needed to support view that exposure is likely to be casual, evidence from multiple research studies on it -Result from just one study alone can't take association to causation Hill's Criteria for Casual Inference -guidelines to evaluate probabilistic and mechanistic evidence for causation from studies -trial has strongest support

Ecologic Study Design

-based on analysis of aggregated data across POPULATIONS rather than data specific to an individual -analyzes information from within groups of individuals rather than information from specific individuals -examines correlations b/w exposure and outcome in many populations, not just one pop. -multiply groups are examined to identify if a relationship exists within these groups. -Hypothesis Testing Advantages -provides quick way of exploring associations (large number, amount of data) -tests hypothesis at group level -simple, quick, inexpensive Disadvantages -does not identify E/D relationship at individual level, may mask effects that exist w/n individuals -based upon average exposures -limited ability to control for confounding since it relies on pre-existing information from varied sources -low quality evidence Ecological Fallacy -key limitation -incorrect interpretation of meaning -error in conclusion that may occur because of a bias where associations at a group level do not represent associations at individual level -researcher concludes results seen at group level as what is happening for individuals, not necessarily true -goal of research is to find exposure or intervention for individuals, not for groups -erroneous inference

Case-Series Study Design

-Descriptive design, not hypothesis testing (proposes one) -hypothesis generating -identify/examine characteristics of series of individuals w. a common trait like health problem or exposure -identifies experience of series of persons w similar outcome -large number of individuals examined -one reporting source -proposing hypothesis -prevalence (freq. of unique event) is characterized -observed over set period of time (eg. one year) -identified from one reporting source (clinical practice, population surveys, etc.) Clinical vs Population source -individuals in case identified as either -most based on clinical sources (consecutive number of patient in hospital/clinical setting) -population source looks at all subjects meeting case definition from existing surveillance system like registry or pop. survey -pop sources are better representative of overall population Basic Structure -After a source is chosen, all persons who meet case definition are identified from source. The design analyzes cases together to find patterns important to use of disease or outcome Advantages -informative for making hypothesis Disadvantages -cannot study cause/prognosis of disease since there is no comparison group w/out exposure or outcome -case series only has individuals that have common exposure or health concern, no control -cannot assess disease frequency in clinical case series

Measure of Association Finding + Research

-Goals of Research: use evidence from research to develop interventions to decrease exposure in population or strategies to increase use of treatments -better the quality of results, greater the success of intervention MoAF -indicates a numerical relationship b/w exposure and outcome variables -Outcome variable is ALWAYS a health issue (not education) -Goal of MoAF: characterizes if finding is an accurate representation of the relationship -evaluates quality and accuracy of research study results Research is about a health issues that relates to: 1. number of individuals it affects 2. types of populations it affects 3. its casual pathway or meaning for prognosis Significant Association - a finding that is likely to exist and the level of random error affecting the finding is viewed to be acceptable

Cumulative Incidence Rates

-Number of new cases of disease occurring over a specified period of time in a population at risk (at the beginning of the interval) -specific to NEW CASES, how many peeps are developing it, a new case, only counting the NEW cases as they come -new cases at a specific period of time, beginning of time period looking at -Used for Risk Ratio -good for evaluating incidence rates in population at risk at beginning of interval

Case-Control Study Design

-ODDS RATIO, do not do relative risk ratio -identification of persons w existing disease (cases) and those w/out disease (control) -hypothesis testing -prevalence, odd ratio, examine existing exposures and outcomes -shorter time to carryout since based upon existing cases of disease -great for rare diseases cuz can investigate case group w preexisting rare disease that would take years prospectively -stronger evidence than cross-sectional since it includes temporality (no chick/egg) -follows retrospective timeframe -current outcome status is known and examined to past exposure status, shows exposure existed before outcome Structure -classify study population by basis of outcome status, first persons w outcome and then persons w.out -exposure status in two outcome groups is measured retrospectively to reflect exposure variable before outcome existed -proportion of cases that experience the exposure are compared to proportion of controls that experienced exposure Basic Rules for Cases -subjects meet case definition for outcome -controls are free of outcome -controls come from same source population as cases -prevalence of exposure in controls is same as prevalence of exposure in source population (if 10% of source pop was exposed, 10% of controls need to have been exposed) Goal -assess relationship b/w E/D where only difference is outcome and exposure -gives stronger evidence that relationship is due to exposure and not a different variable Strengths -fast and inexpensive -efficient for studying rare diseases Limitations -exposure measurements are taken after disease states is known, may affect measurement of exposure -accuracy of exposure, measuring today what values for exposure were in past may not be accurate -disease status can influence selection of subjects -investigators know outcome status and can influence which cases and control they select.

Successful Surveillance Systems

-Simple (practical and clear case definition) -Stable (ongoing in operation) -Acceptable (to data providers) -Standardized, uniform collection of data -Timely (in reporting events) -Representative (of the population) -Sensitive (to changes over time) -Flexible (to changing surveillance needs) -Accuracy vs Timeliness Chart

Questions to ask to determine if you should surveil

-What level of resources are needed and available to carry out disease monitoring? -What outcomes from the surveillance are desired? -Can the counting methods used in surveillance be readily accepted into the community?

Study Design

-a specific method of how a research is conducted to collect info of whether a statement (from the research hypothesis) is true or not -outlines plan on how to carry out study, methods of data collection, data analysis, and interpretation of results Types (Ranked by Quality of Research Evidence) 1. Randomized Controlled Trial 2. Cohort 3. Case-Control 4. Case Series 5. Cross-Sectional 6. Ecologic Categories of Designs 1. Intent Descriptive -to describe the patterns related to individuals who have experienced a similar exposure or outcome -used to GENERATE a hypothesis, don't have hypo. -Does the project intend to describe patterns of outcome or patterns of exposure? = Case-Series Analytic -analyze determinants of health events -assess association b/w Exposure vs Outcome -TEST a hypothesis, already have hypo -Does the project intend to test a hypothesis of the relationship between an exposure and an outcome? =Group Data: Ecologic =Individual Data: Cross-Sectional, Case-Control, Cohort, Randomized Controlled Trials Hypothesis Testing: -will have exposure and outcome variable -exposure and outcome levels will have different levels (yes/no, low/medium/high) 2. Focus Observational -non-controlled -method is to observe and record what happens regarding exposure or intervention -no experimental control over exposure/intervention, but just an observation of what treatments the participants individually use -less used cuz less precise than experimental =Group Data: Ecologic =Individual Data: Cross-Sectional, Case-Control, Cohort Experimental -level of exposure or type of intervention is controlled by investigator -common in health care research where a specific intervention is given to one group but not to another = Randomized Controlled Trials 3. Time Perspective Prospective - examine future outcome events -looking forward into the future to characterize the relationship between E/D -exposure and outcome variable in different periods of time -exposure variable taken at beginning of study and then outcome is assessed as it occurs at later points in time Retrospective -examines past exposure events -looking backward in time to characterize the relationship b/w E/D -outcome variable is taken at beginning of study and level of exposure is assessed at some point in past before outcome happened Cross-Sectional -examines exposure and outcome events happening at the same time -looking at one point in time to characterize relationship -no difference in time be exposure and outcome -identifies a point in time and measures values for E/D at that point as they exist

Information Bias - Types

-error in study result due to systematic differences b/w study groups in way data was obtained -problem during data collection, how main variables were measured, inaccurate result Recall Bias Limited Recall -inability for research participants to remember what the correct response to specific variable (especially in retrospective studies) -limited recall increases probability of inaccuracy of exposure variable -longer period of recall, higher recall bias potential -self reported measures Differential Recall -individuals w/ disease tend to think about possible causes of outcome Reporting Bias -inaccurate variable because participants are intentionally reporting inaccurate info -intentional misreporting of values -releant w/ socially sensitive variables -pay attention when researching culturally unacceptable or illegal variables Interviewer Bias -bias arising from actions of interviewer -possible for interviewer to approach asking questions to participants differently instead of the exact same approach w. all participants. -leads to inaccurate measurement of variable -may be intention or unintentioal -training interviewers can reduce bias -blinding care reduce tendencies to ask questions differently (no knowledge of participant/which group) Control of Bias -can be prevented or controlled during design and conduct of study -not possible to reduce bias after data has been collected Use good/unbiased methods to: -select participants -measure exposure and outcome variables -in prospective studies, try to limit loss to follow-up (incentives) Ways to Reduce Bias/Improve -increase participation (monetary incentive), decrease loss to follow up (rapport), good training of staff, valid and reliable measurement tools, blinding procedures when appropriate, shorter recall periods

Cohort Study Design

-examination of a group of subjects and experiences of group over given time frame -hypothesis testing -can be years -identify subjects from their exposure status and identify events/outcomes as they occur in future -prospective -incidence -can directly estimate risk of outcome Structure -identify group w/out outcome -classify into two groups: if they had exposure and if they didnt -follow them forward in time to determine if outcome develops -structure allows for a more accurate assessment of exposure sine its the first focus, pop picked based on it -structure examines E/D temporally -identifies risk of new outcome, since following forward Types 1) Prospective -most common -begins at present data and measures level of exposure levels and then follows subjects on regular basis 3-5 yrs -both exposure and future outcome are measured -starts at exposure 2) Historical (Retrospective) -historical data on E/D are used to evaluate previous exposure and outcome events preceding date of study -used when record of health/exposure is available (health, work, insurance) -study begins when outcomes occurred, but uses the past to trace from exposure to outcome -faster than prospective since all data is available -can directly estimate risk of the outcome Strengths -exposure status determining before disease detection so measure of exposure is better -subjects selected before disease detection -measures new disease/outcomes Limitations -expensive and time-consuming (3-15 yrs) -inefficient for rare diseases, diseases with long latency -subjects lost to follow up, long periods of follow up, higher opportunity of drop outs

Analytical Epidemiology + Risk/Prognostic Factors

-gathering evidence to support evaluation of causes of diseases in community -provide evidence to support hypothesis of link between E/D or refute it Risk Factor -exposure that evidence shows is a finding b/w relationship of E/D -exposure that increases risk for disease or outcome in population -ex: environmental factors, social behaviors, genetics factors, cultural factors, etc. Prognostic Factors - exposure/intervention that evidence shows is related to health outcome in population with disease -changes prognosis of disease course, decreases future adverse health event -intervention affects probability of outcome -ex: behavior, treatment, environmental, social Measure of Disease Frequency -measures of disease risk in study sample -Prevalence: period prevalence -Incidence: incidence density and cumulative incidence

Hill's Criteria for Casual Inference

-guidelines to evaluate probabilistic and mechanistic evidence for causation from studies -all 9 elements do not have to be met to support a casual role for an exposure -ex: chronic diseases focus on temporality, strength of association, consistency of findings, plausibility, and biologic gradient Includes 1. Consistency of Findings -the study results that are consistent show stronger evidence of causation for exposure -results of multiple studies on same hypothesis are reviewed -uses replication to support causality because results are consistent across different populations, different study designs, and different investigators during different times 2. Strength of Association -larger the association, more likely the exposure is causing the disease -the associations with the greater evidence of causation (strong associations) shows that the relationship is less likely due to error, bias, or confounding 2. Biological Gradient -results that show gradient changes in exposure lead to gradient changes in outcome have strong evidence of causation -dose-response relationship -changes in dose = changes in outcome -positive and negative associations work here 4. Temporal Sequence -cause-effect relationship where exposure as cause produces disease as effect. -exposure must exist before the outcome to shows causation 5. Biological Plausibility -evidence of a biologic mechanism of how exposure can lead to disease increases proof -evidence for this may not exist yet since we sometimes find the mechanism after we make associations -relationship may be indirect 6. Coherence -body of research evidence should be make sense given current knowledge about disease (natural history, biology, epidemiology) -proposed mechanism should not contradict what is known about disease 7. Specificity of Association -an exposure leads to a single (specific) effect, or affects persons with a specific susceptibility -exposure produces this one, very specific, outcome. -rarely met for chronic diseases 8. Experiment -evidence from study shows intervention on exposure is successful in reducing freq. of outcome 9. Analogy -does a similar type of relationship exist for another disease process, or with another exposure disease relationship? -can argue danger of motorcycles w danger of cars

What study design is represented?

-look at first step of study design -What is the first step being done in this situation? Ex: Carrying out a survey -so events would happen at one point in time -data collected at one specific point in time -narrow it down from there

Vital Statistics

-type of surveillance -state law requires all vital events be registered to government authorities -events: birth, death, fetal death, marriage, divorce -National center of health statistics maintains vital data, works to set standards of registration of data Birth Certificate -identifies citizenship, age, birthplace, parentage -Useful Because: gestation, weight, prenatal care, birth outcomes Death Certificate -used in burial docs and settlement of estate and insurance claims -Useful Because: cause of death, age of death, person, place, time -Part I: health conditions that directly caused the death --start with most recent conditions on top line and oldest or original condition at bottom line --Immediate COD: the condition that directly led to death, without this condition, death would not have occurred at that moment --Underlying Cause of Death: the condition that started the course of events leading to death -Part II: lists conditions and other factors contributing to death, includes medical factors and risk factors not directly leading to death --smoking, obesity, surgery long ago, drug use Variation of CoD cuz: -limited training on how to complete CoD statement -bias in recording CoD because physicians have differing opinions, physicians not aware of underlying condition

Cross-Sectional Study Design

-observational design -Large sample size -identifies E/D at a single point in time -prevalence, odd ratio, examine existing exposures and outcomes -follow a cross sectional time perspective -preliminary design, results provide lower quality evidence about E/D -based upon info from surveillance mechanisms like population survey and registries -identifies prevalence of exposure at given time and prevalence of outcome at given time -best for chronic diseases since they have a prolonged course of disease and will be captured in most cross sections -not good for rare, highly fatal, or short duration disease -hypothesis testing Structure -based upon first identifying a cross section of time for given population -identify values for outcome and exposure variable at that time -categorize if exposure was present or absent at time and if outcome was present or absent at time -examine if categories of exposure differ for categories for outcome Strengths -quick and low cost answer to study hypothesis -evidence obtained in shorter time frame -often first design for testing hypothesis -good for testing more than one exposure for diseases -estimate prevalence of condition in population -study many associations at once Limitations -weak observational design, measure prevalence not incidence (prevalent cases are survivors), won't include exposure data from people who had disease but died quickly -unable to show that exposure preceded development of outcome -timeline of exposure and effect may be impossible to determine -not good for rare or quickly emerging disease Chick or Egg Dilemma -because were measuring both variables at same time, if both are present on June 5th for ex., which came first? -unable to determine if outcome lead to exposure or if exposure lead to outcome -lack temporality, time line

Odds Ratio Estimating Risk Ratio

-odds ratio is a good estimate of risk ratio when: 1. The outcome is RARE OR 2. The effect size is small/modest Not a good estimate when outcome is common AND the effect size is large; Effect Size: measures the strength of the relationship between two variables

Passive Surveillance

-outreach BY persons in community to health officials agency -health department personnel do NOT go out into community -develop instruments that community can use to collect the info and send to health department -most surveillance systems

Active Surveillance

-outreach TO community BY health agency officials -one place reaching out to everyone -regular monitoring of surveillance sites by health agency officials -gather data themselves, may involve visits to clinical sites -the information collected MOST OFTEN involves a new case of disease or the health issue of concern -good for identifying incidence of a chronic disease -quality of data is higher in active approach than passive -more expensive than passive

Point Estimate of Study Design

-point estimate translates to the quality of research obtained -Point Estimate is the result of the measure of association in study (relative risk odds ratio, etc.) -a point in some parameter which tells "best guess" or "best estimate" of an unknown population parameter Ex: Why is the point estimate from a cohort study of higher quality than the point estimate from case-control study? -cuz bias is lower in cohort since you're capable of measuring the exposure and outcome variables more accurately. Subjects are selected before disease selection so measure of exposure is better.

Population Surveys

-prevalence of disease -deals with chronic, not sure about infectious Two Types: -Health Survey: focused on identifying the health of the US population -Health Care Use Survey: focused on identifying the reasons why individuals use health care services -PROS: representative data for population, often standard assessments or interview methods -CONS: self report surveys may have bias of recall of events or willingness to report info, info limited to setting of surveys, expensive to conduct -Slow, expensive, estimated certainty -NCHS main place for survey info -NHIS: cross-sectional household interview survey info on chronic diseases and disability --self report survey relies on answers by participants to questions

Probabilistic Evidence

-probabilistic evidence and mechanistic evidence should be assessed together to judge likelihood exposure may be causal Principles that increase probability of relationship -findings that indicate exposure may lead to new cases of outcome -finding indicate that removing exposure should result in no outcomes related to exposure -findings indicating strong association -findings indicating dose-response relationship Question to Ask -Does the exposure increase the probability of the outcome? -What is the magnitude of the association? =Stronger associations increase probability. -Does a reduction in exposure lead to less disease? -Does a dose-response relationship exist? -Does a biologic mechanism exist?

Efficacy

-the extent to which an intervention produces a beneficial result under ideal conditions. -high efficacy: drug is very beneficial -extent to which an intervention does more good than harm under ideal circumstances

Health Interview Survey

-type of population survey (which is a type of surveillance) -principal source of info on health of civilian pop -used to monitor trends in illness/disability and track progress toward meeting health objectives -ex: household interview survey -self report on illness and disease -information represents prevalence of health issues -survey based upon representative sample of US households NHANES: health survey of pop -National Health and Nutrition Examination Survey -Uses two components -Interview: assesses demographic, socioeconomic, dietary, and health-related behaviors -Clinical Examination: medical/dental examinations, physiological measurements, lab tests BRFSS -Behavioral Risk Factor Surveillance System -collect state data about U.S. residents regarding their health-related risk behaviors, chronic health conditions, and use of preventive services -cross sectional questionaire NHIS -based upon self reported data -National Health Interview Survey -mainly interview type NHCS: collects large array of health care surveys that identify patterns of health care use YRBSS: Youth Risk Behavior Surveillance System -CDC population survey -monitors six categories of priority health-risk behaviors like sex, tobacco, alcohol, diets -learn more about adolescent drug use in the community NAMCS/NHAMCS -The (hospital) ambulatory medical care survey includes reasons for visits to emergency departments in its administration.

Notifiable Disease Surveillance

-type of surveillance -understands frequency of infectious diseases and aims to limit spread of infectious diseases, measures disease trends -detects sudden changes in disease occurrence and identifies epidemics -monitor infectious diseases -notifiable diseases: diseases required by law to be reported to public health authorities. -have high rate of communicability or a high rate of death -such as malaria, influenza, west nile, lyme disease, HIC, anthrax, pertussis -fast, inexpensive, low certainty

Accuracy of Research Results

Accuracy -results shows either an association b/w exposure and outcome or no association and that choice is accurate (ex. there is no associate) or inaccurate (ex. there actually was an association) How to evaluate it? -assessment checking 1. how random error affected results 2. how systematic error affects results 3. how confounding affected results *confounding: causing surprise or confusion by acting against expectations Types of Errors -If the conclusion is results are accurate, then evidece can be used for future healthcare decisions -If inaccurate, theres an error in correct relationship b/w outcome and exposure. *Null Hypothesis: there is no connection between variables Type I (Alpha) Error: study result shows there is a connection when there really isnt one; result rejects null hypothesis when it should have confirmed it Protecting Against Type I -significant association identified via p-value and confidence interval protect against errors due to random error -P-value/CF does not protect against bias/confounding. only random error -statistical findings confirm association but do not identify casual relationships. Type II (Beta) Error: study result shows there is NO connection when there really is one; study results does not reject null hypothesis when it should have rejected it. -if related to random error, could happen cuz study pop is small not large Ex: If study does not reject Null Hypothesis, then it is saying "there is no connection between the variables". If there actually is no association, it is the correct decision. If there is a connection, then its Beta Error Type II Protecting Against Type II -exists whens study finds no association but should have found one -having a large number of study subjects reduces error -reduced by hypothesizing and testing large difference, not small -may exist cuz of bias/confounding What causes Errors? -Random Error -Systematic Error (Bias) -Confounding *confounding: causing surprise or confusion by acting against expectations

Population Attributable Risk (PAR)

Assess risk of disease in total population attributable to exposure -identifies the reduction in outcomes that could be achieved if the population were entirely unexposed -Helps in determining which exposures are most relevant to public health in community -incidence levels in the non-exposed are assumed to represent the level of the outcome that has nothing to do with the exposure. PAR = Ipopulation- Iunexposed Calculate Population Attributable Risk Percent PAR% = (Ipopulation- Iunexposed) / Ipopulation *100 PAR% = incidence rate in population - incidence rate in unexposed divided by the incidence rate in the population

Association vs Causation

Association -relationship b/w exposure and a disease or health outcome -suggests that exposure may be cause/determinant for disease, but DOES NOT prove it -indicates relationship b/w E and D exists -DOES NOT demonstrate a mechanism for how exposure leads to outcome/disease Causation -implies that there is a true mechanism that leads from exposure to disease -vital to have MECHANISM -difficult to pin down for chronic diseases because they can have many causes, multiple exposures, and long time frames

Attributable Risk (AR)

Attributable Risk -# of cases among the exposed that could be eliminated if the exposure were removed -measures the amount of the outcome in the exposed group related to exposure -only includes outcomes that develop because of exposure by subtracting incidence in unexposed -incidence levels in the non-exposed are assumed to represent the level of the outcome that has nothing to do with the exposure. Equation = Incidence in Exposed - Incidence in UnExposed Attributable Risk Percent -proportion of disease in exposed pop. that could be eliminated if the exposure were removed Equation = (Incidence in Exposed - Incidence in UnExposed) / Incidence in Exposed

Wk9 - Why would an epidemiologist be interested in increasing the positive predictive value of a screening test?

Because the current PVP values are extremely low and argue against the implementation of a public health screening program. In the current analysis, the benefits of the program would likely be outweighed by the risks and costs. Another manner to increase PPV is to change the target population. You can increase PPV by focusing on a high-risk population (a population at high risk for bladder cancer). This will increase the pre-clinical prevalence of disease in the population, and, thus, increase the PPV.

Confounding

Confounding *confounding: causing surprise or confusion by acting against expectations -creates a situation when interpretation of relationship is incorrect -describes association that is true but misleading due to third variable and not exposure variable More Facts -a third variable (not outcome or exposure) that distorts observed association between outcome/exposure -confusion of the effect of an exposure on an outcome -effect of confounding factor is mistakingly attributed to exposure -you think its the exposure leading to the outcome -its actually the outcome resulting from a third factor, not exposure -clouds proper interpretation of how exposure and outcome are related Example Variables -family history of disease -socio-economic status -exposure to other "radiation" -cigarette smoking Rules -3 Conditions for a variable to be a confounder (all 3) 1. Factor MUST be associated/related w disease 2. Factor MUST be associated/related w exposure 3. Factor CAN NOT be an intermediate link in pathway from exposure to outcome -Exposure DOES NOT cause factor which then causes outcome Examples of Confounding: Ex: Alcohol -> Lung Cancer -are there other factors that impact result? -smoking for instance -it IS CONFOUNDING because 1. smoking and drinking go hand in hand (related to consumption) 2. smoking is related to lung cancer 3. Alcohol DOES NOT cause smoking, so its NOT an intermediate link -so its unclear if lung cancer is from alc or smoking Place of Delivery vs C-Section Exposure: hospital vs med center Disease: C-section Confounder: high-risk pregnancy State of Residence and ABC Mortality Exposure: Alaska vs Arizona Disease: mortality due to ABC Confounder: Age Controlling Confounders During Design: -restriction: limiting eligibility for participation to only individuals who fall w/n specific categories of cofounders (no people who smoke, who are high risk preggos), eliminates cofounder -matching: used in case-control, individuals are selected so confounding variable is equally distributed among study outcome groups, confounding present but controlled -for every person who si high risk, there is a low risk, every smoker has a non smoker -randomization: allocated subjects to intervention groups, equal balance in frequency of confounding variable in each treatment group (= # of smokers) During Data Analysis: -stratified analysis separate analysis' for different values of confounder, adjust for it ex: one analysis for exposure/outcome for smokers, one analysis for nonsmokers -multivariate analysis statistical modeling, uses model to account for influence of exposure, outcome and confounding variables simultaneously.

Measures of Potential Impact

Definition -aid in picking which prevention program to use to gain max health benefit -provide numerical estimates of degree of prevention an intervention can give -indicate how prevention efforts may affect disease levels in population Important: All require that a cause-effect relationship exists b/w E and D -causation must come before outcome Types Attributable Risk (AR), Rate -Attributable Risk Percent (AR%), percentage -fraction of exposed cases due to exposure -measures potential impact in exposed cases -outlines value of prevention programs where target audience is only those who have been exposed in the population -ONLY EXPOSED PEEPS Population Attributable Risk (PAR), Rate -Population Attributable Risk Percent - (PAR%), percentage -measures potential impact in population as a whole -fraction of population cases due to exposure -outlines value of prevention programs where target audience is the entire population, as if the entire population would receive the intervention (covid vaccine) -BOTH EXPOSED AND NONEXPOSED PEEPS

Influenza Ex. of Risk vs Odd

EX: Among 100 people at baseline, 20 develop influenza Risk: 20/100 (number who have it / how many at risk) Odd: 20/80 (number who have it / number who don't have it)

Attributable Risk Contingency Table

Ex: car crash vs speed Outcome l No Out Exposure A l B No Exposure C l D -Attributable Risk Measure is calculated by: a/a+b minus c/c+d -Attributable Risk Percent is calculated by: (a/(a+b) - c/(c+d))/a/a+b Rate/Risk -Population Attributable Risk Measure is calculated by: (Total Outcome/Total Participants) - (Outcome by No Exposure (c)/Total unexposed) -Population Attributable Risk Percent is calculated by: PAR% = incidence rate in population (a+c) - incidence rate in unexposed divided by the incidence rate in the population Specific Numbers 100/2000 = 0.050 80/8000 = 0.010 180/10000 = 0.018 Attributable Risk Measure (100/2000) = 0.05 (80/8000) = 0.01 0.05 - 0.01 = 0.04 Attributable Risk % 0.04/0.05 = 0.8 Population Attributable Risk 180/10000 = 0.018 80/8000 = 0.010 0.018-0.010 = 0.008 Population Attributable % 0.018-0.010/0.018 *100 =44%

Screenings - Purpose

Goal: Use safe and inexpensive test to reduce morbidity and mortality from disease among persons being screened -reduce the impact of existing disease in a population --separate peeps into two groups: probably diseased, probably disease free Purpose: Detect disease or body dysfunction before an individual would normally seek care --individuals in subclinical stage How? -Earlier detection and treatment of disease slows progression of natural history of disease. -using a simple test to find individuals who have early stages of disease (in the subclinical disease stage) and introducing treatment early Does it Reduce Disease (Incidence)? -does not reduce the disease, but rather, screening improves the prognosis of the disease by reducing future morbidity and mortality in the natural history of the disease What to Consider? -Severity, prevalence, natural history, diagnosis and treatment, cost, efficacy, and safety When is it beneficial? -Serious disease -Early treatment is valuable -Prevalence is high When is it not appropriate? -Early detection has no clinical significance (don't show symptoms, non-operable affliction) -Prevalence is low --exception when its easy, low cost, simple, and highly accurate (at birth)

Which type of surveillance would be used?

Notifiable Disease Surveillance -infectious and reportable diseases -good for short duration, infectious diseases Vital Events Surveillance -high mortality rate, generally involves death -use death certificate to identify frequency -only involves info on births, deaths, marriages, and divorces Registries -focus largely on chronic diseases -possibly long term infectious disease -good for identifying incidence of long term disease, not prevalence -require collection of info from multiple locations Population Surveys -natural history has to be long enough -identify prevalence of disease, NOT INCIDENCE -good for common conditions because they are expensive to undertake

Wk 8 Quiz Q: The systematic and continuous gathering of information about the occurrence of pertussis is best characterized as:

Notifiable disease surveillance -Pertussis is an infectious and reportable disease. -doesn't usually cause deaths so death certificate are unhelpful -not registry cuz not a chronic disease or injury or a long term infectious disease -natural history is too short to make a survey for

Interpreting Relative Risk Measurement

Null Hypothesis: no association 1 = No Association b/w Exposure and Disease -incidence rates are identical -null hypothesis was supported >1 = Positive Association -exposed group has higher incidence than non-exposed group -exposed group has higher risk of disease <1 = Negative Association or Protective Effect -exposed group has lower incidence -exposed group has lower risk of disease -ex: 0.5 -exposure is protective for outcome: means that in presence of exposure, level of disease is lower -the farther the point is from null (1.0), the stronger the negative association Rylander Classification R.R. >/= 10 -people easily recognie risk -very strong assocaition R.R. between 2-9 -comfort zone -strong to moderate assocation R.R. < 2 -uncertainty zone -weak association

Random Error (p-value, confidence interval)

Random Error -can directly lead to incorrect study result -creates association that is not true -caused by random factors out of the control of researcher -occurred at random where study population, by chance, is not similar in traits to overall population -Key issue: one sample pop. may not have same results as another sample pop due to random variation in samples -sample may not be representative of population by chance (red and white balls, pick three (only red by chance)) -since there's always a chance of this error, all studies should evaluate extent of random error effect HOW? -statistical testing: use statistics to estimate probability that observed results are due to chance -will the relationship found exist over and above a given level of random error? how much fo result is actually whats happening and how much is random error? -Answer using p-value and confidence interval p-Value (0.05) -probability that an effect occurred by chance alone, if there is no relationship between exposure and disease -protect against Type I (alpha) Errors α = level of type I error p = 0.05 (1/20) -if p is less than or equal to 0.05, the result is not likely due to chance (valid, statistically significant) -if p is greater than 0.05, then it is most likely chance (invalid, not significant, chance not excluded) -Limitation: still chance of chance Confidence Interval (95%) -considers an interval estimate with probability statement that true value for population lies in the interval -range of values for a variable constructed so that this range has a specified probability of including true value of variable -measure of study's precision -has upper and lower limit, w/n range is true value of relationship be outcome and exposure -95% CI: means true estimate of effect lies within 2 standard errors of population effect 95 times/100 More Info on Confidence Intervals -A study has a relative risk that the results came up with. This is the p value determined. To take into account that this value is a "give or take", the confidence interval adds a range that pretty much says, "It might not be the p-value, but it is definitely within this range". If 1.0 is part of the range of true value, the finding is not statistically significant. Why? **** -if the true relative risk/main finding has values greater than 1.0, it is a positive association and statistically significant. -if the true relative risk/main finding has values less than 1.0, it is a negative association and not statistically significant.

Wk 8 Quiz Q: Monitoring cancer incidence in the US population over time is a major application of which of the following forms of surveillance:

Registries -chronic disease so not notifiable disease system -incidence is registries, prevalence is population survey -not all new cases = death so not vital event

Relative vs Attributable Risk

Relative Risk -measure of strength of association -assess possibility of casual relationship -understand etiology Attributable Risk -Measure of the potential for prevention of disease if the exposure could be eliminated (assuming a causal relationship) -used in policy decisions

Selection Bias - Types

Selection Bias -non-random errors in process of selecting/analyzing participants -results could be biased if study pop. is picked specifically a not representative of general pop. for ex. picking participants more/less likely to develop health issue Exists Because 1. Manner of selection affected likelihood of having exposure or having outcome 2. Factors influencing continuing participation in study affect likelihood of having exposure/outcome --(prospective studies), bias if those who remain in the study are different (in terms of the exposure and outcome) from those who left the study. TYPES 1. Self-Selection Bias, Non-Response Bias, Volunteer Bias -characteristics of persons who choose to participate may be different than characteristics of those who decline -if these differences are related to exposure/outcome, findings may be inaccurate 2. Berkson's Bias -using participants from healthcare settings (WPH) -their health status is different than persons not hospitalized (general pop.) -persons w/ more than one disease are more likely to be hospitalized than just one disease -exposure values of hospitalized subjects may not be = to exposure values for general population -hospitalized subjects have higher exposure values than pop -hospitalized subjects have different levels of severity of outcome than nonhospitalized peeps. -important in case-control 3. Healthy Worker Effect -comparing persons w a job and individuals in general pop -workers are healthier than gen-pop people -arises when persons may be selected for employment based on good health, persons which continue to work may be "selected" because they are still in good health compared to those who arent working/worse health 4. Loss to Follow-Up Bias -characteristics of those who remain in study differ from characteristics of those who left, those who left may have different outcomes

Screening Process

Successful Screening -Exit Strategy --facilities for treatment/diagnosis should be readily available when someone tests positive --unethical to offer screening if you aren't prepared for a positive Test: -feasible for large populations -relative sensitive and specific (valid and reliable) -simple and inexpensive -very safe -acceptable to subjects and providers

True Disease/Test Status

True Positive (a): predicted to have disease (screen positive) and actually had disease -Risk: labeling effect, overdiagnosis True Negative (d): Predicted to be disease free (screen negative) and actually are disease free False Positive (b): predicted to have disease (screen positive) but are actually disease free -Risk: anxiety, fear of future tests, expensive False Negative (c): predicted to be disease free (screen negative) but actually have disease (missed case of disease) -Risk: delayed intervention, delayed diagnosis cuz false sense of security Sensitivity Sensitivity %: True Positive / All Cases Sensitivity % = a / (a+c) -affects False Negatives: disease could spread, higher mortality -important for infectious diseases, not chronic diseases (80% is fine) Specificity Specificity %: True Negative / All Negative Cases Specificity %: d / (b+d) -affects False Positives, involves unnecessary diagnostic test (cost, safety), unnecessary anxiety -important for chronic diseases Predictive Value Positive PV+ %: True Positive/ All who test positive PV+ %: a / (a+b) -specificity changes this, higher specificity = higher PPV -increase PPV by changing target population to a high-risk pop which increases prevalence increases PPV -want higher PPV cuz more likely health program to be implemented, benefits outweigh risk Predictive Value Negative PV- %: True Negative/ All who test negative PV- %: d / (c+d) Values of Predictive Values are affected by... -sensitivity of screening -specificity of screening -prevalence of disease *same screening test can have different PVs in a different pop. (heart disease in men vs women) Implications of False Positive: -cost of unnecessary procedures -invasive diagnostic procedures -possible complications/risk for diagnostic test -emotional stress -reluctance to be screened again (distrust) Implications of False Negative: -false reassurance of no disease -communicable disease can be spread -social cost from inaccurate test -health care implications of missed treatment -reluctance to be screened again (distrust)

Wk9 -Quiz: How many true positive test results would be found from a screening antibody test? * Test can't tell if infection is still present or went away, just that they have antibodies

Unable to be determined when given how many people have active infection numbers, Note: can be told when given how many people have antibodies because then they definitely had the infection and are true positives -the antibody test does not identify active infection...this is only known after the second test. Therefore, it is not possible with the information available to identify how many were positive from the first test.

Measures of Association - Break Down

What are they? -quantifies relationship b/w E/D -Exposure: age, race, waist size, occupation, socioeconomic status, STD, food, etc. -Examples: risk ratio (relative risk), rate ratio, odds ratio, and proportionate mortality ratio Risk Ratio = (Risk of Disease in Main Group) / (Risk of Disease in Comparison Group) =Exposed/Unexposed -use cumulative incidence Rate Ratio = (Rate of Disease in Main Group) / (Rate of Disease in Comparison Group) =Exposed/Unexposed -use incidence density Odds Ratio -ad/bc Odds of Exposure in Disease Group / Odds of Exposure in No Disease Group -usually applied prevalence, not incidence

Wk 8 Quiz Q: Surveillance based on a process where public health officials identify a specific respiratory illness from hospital emergency department records is best characterized as:

a form of active surveillance -public health official going out to hospital ED to get records on disease -registries would require collection of information from multiple locations so not it -passive surv. would involve ED going to Health not Health to ED -vital only involves births, deaths, etc. not respiratory illness

Efficiency

a measure of the economy (or cost in resources) in which a procedure is carried out [benefits relative to costs] -about "whether a drug is worth its cost to individuals or society. -The most efficacious treatment, based on the best evidence, may not be the most cost-effective option. It may not be acceptable to patients

Wk 8 Quiz Q: Which of the following surveillance approaches best represents a passive form of surveillance?

a system where health agency officials receive information on SARS CoV-2 from the laboratories in the region -information from community coming to health department

Analytic Epidemiology

examines casual hypotheses of association b/w exposure and outcomes -finds link b/w E and D(isease) Exposure -> Disease Intervention -> Outcomes A possible cause for disease is often characterized as an exposure. A possible treatment for disease is listed as an intervention.

Mortality (Death Rate)

major measure of disease in a population -info from death certificates -use death certificate data to assess disease frequency in different populations. -Death Rate: (Number of deaths in a time period/ Number at risk of dying) -denom: population at risk Death Rate elements -specifically defined population group (denom.) -time period -number of deaths occurring in that population group during that time period (num.)