Midterm Study bundle
Midterm Study bundle PUBH 3131
Popular in Epidemiology: Measuring Healht/Disease
Popular in Public Health
Test Prep (MCAT, SAT...)
verified elite notetaker
This 35 page Bundle was uploaded by Gabriela Saint-Louis on Saturday February 6, 2016. The Bundle belongs to PUBH 3131 at George Washington University taught by Margaret Ulfers in Spring 2016. Since its upload, it has received 89 views. For similar materials see Epidemiology: Measuring Healht/Disease in Public Health at George Washington University.
Reviews for Midterm Study bundle
Report this Material
What is Karma?
Karma is the currency of StudySoup.
You can buy or earn more Karma at anytime and redeem it for class notes, study guides, flashcards, and more!
Date Created: 02/06/16
01/12/2016→ class lecture ____________________________________________________________________________ _ Learning Objectives: review applicable jargon from public health distinguish between eminence and evidence based medicine what we are doing Epidemiology: Measuring Health & Disease (what)> measuring (how)> (why)> ultimate goal is to prevent disease or at least contol it and have an effect measure health and disease so we can do something about it Goals: learn how Epidemiology fits into public health learn to apply these principles into public health Review>Jargon specific language of academic specialties Morbidity and Mortality:disability and death disease distribution: demographics time, region; classic bellcurve. something of interest in terms of demographics Risk factors: environmental/genetic; markers, indicators, determinants artifactual: artifact of the way things are measured may make things appear to be an association but not real because of misinterpretation case: an individual event; ex. one person with disease case fatality: not just death but from the case incidence and prevalence: risk of it; how much/many Hypothesis Etiology: what causes the disease; background→ how. import for epidemiology to help solve and prevent disease. the more you know about etiology Health “impact””Burden of Disease Economics, mortality, individual vs. group Evidence based vs. eminence based medicine evidence: tested and proven to work Eminence based: respected, well known, subject matter expert ex: observation of respected doctors medicine based on what seemed to subject matter experts that it works example: back to sleep ( babies sleeping positions Levels of Prevention Primary(i.e. education) secondary(i.e. early detection; prevent from going to tertiary) tertiary(limit morbidity and disability/death) Recurring themes: difficulty in determining causality difficulty in studying populations human subjects→ pple are confusing public health vs. personal health 01/14/16→ Class Lecture ____________________________________________________________________________ Populations Epidemiological Perspective adding dimensions changes what we see take home points: epidemiology is scientific foundation of public health jargon goal is to measure data understand epidemiology perspective distinguish between medical and public health application history Epi 101 “ The basic science of public health” “ is a science”> scientific method Hallmarks: data ( empirical & measurable) conclusions ( repeatable & refutable) What is epidemiology? study of health on population level> good, bad, indifferent BTW: off label uses of epidemiology the “epidemiology of…” really refers to the results of epidemiologic studies on… ( distribution, major risk factors and etiology of disease) This course is not on the “epidemiology of…”> but how to evaluate as opposed to incidental examples Epidemiology is a survey course surveying a moving target Epidemiology: Friis points: population focus distribution determinants outcomes quantification control of health problems ( science more with focus on application) Friis on Epidemiology is it a liberal art? why would it matter? has broader applications, wirting aimed at undergrads; push at havign it at lower level classes. “ epidemiology is for everyone, everyday, etc.” self interest Public Health vs. Medicine what distinguishes medicine from public health? the way you describe population differs how you describe an individual where does Epidemiology fit? often only evidence we have to make medical decisions Epidemiology vs. other medical research looking at populations why? Public health interest ( want to know on population level) no other choice Human Populations by definition Populations must be a defined group defined by “some common characteristics” is a population always large? NO Populations vs. subpopulations every population is a “subpopulation” of the population of all humans ever born it’s all relative to who it’s a subpopulation of at least one characteristic differs between subgroups each distinguishing characteristic brings us closer to individual level.. Why does defining the population matter? it’s the denominator! ( who we’ve defined as the population) An “observational science” never a completely experimental science because studying people→ observational mostly look at what happened and observe it ethics/people no choice but to observe Drawbacks? not repeatable; cultural mindset & sometimes can’t Epidemiology vs. Individual Epidemiology uses statistics which are group measures by definition. stat always groups ( ex: average) ex: “ your baby is an individual not a statistic” “ you can’t be having that side effect because only 1/100 do” Life expectancy what doe sit mean for the individual? tells very little about an individual; more so how’s public health doing overall Uses of Epidemiology 1. Historical→ understanding past 2. community health→ understanding/planning 3. Health services→ Best practices for cost ( now) 4. Risk assessment→ from population to individual 5. Disease causality→ determine which associates; real vs. artifiactual History of Epidemiology term epidemiology didn’t come into use until 1800’s “epidemiologist” even later only more recently really a separate field of study pretty young as a science 1600’s: Bills of mortality records in local courthouses of when people lived and died (for tax purposes) males to fight, etc. not out of public health concern John Graunt statistician true populations tudy considered age when making comparison ( quantified it) standardizing of age to make comparisons analyzed patterns empirical data 1700’s: Scurvy James Lind of groups of sailors ( 12, in 6 groups) compared different treatments experimental approach fan early idea not an immediate success ( took 5060 years before the British Navy started using it) 1700’s: Chimney sweeps SirPercival Potts surgeon observed chimney sweeps higher rates of scrotal cancer to to the soot exposure note importance of comparisons even though not formally done, still know others had lower rates. 1800’s: Data collection william Farr was an “epidemiologist” who studied medicine medical doctor by training collected data, comp 1800’s: Snow and Cholera cholera plague of the times: could kill you in 8 hours;misunderstood; untreatable and deadly john snow doctor in london in mid 1800s identified cause of epidemic through descriptive and quantitative stats methods ( i.e. epidemiology) Known as father of Epidemiology removed pumps so people couldn’t get water; made pa with lines for every case of cholera( no cholera near brewery because they mostly drank what was brewed) able to make convincing argument. spoke to places to ask them where they got their water from. 1800’s: Snow and Ricketts Snow published paper on Rickets caused by Vitamin D ignored and forgotten the true dietary nature of rickets was not elucidated and accepted until 1920’s Snow & Cholera ( pump handle put back on ) 1800’s: Ignaz Philipp Semmelweis puerperal fever ( 1646 paris) while doctor, 2030% mortality rate for women giving birth in hospitals Hungarian OB working in Vienna Austria: compared 2 clinics and showed deaths due to “cadaverous materials” on student doctors ; other clinic was with midwife mortality rate reduced to 1% with hand washing increased following his departure huge controversy in medical community. 1900’s: Noninfectious disease 1948: FHS began large studies where began following populations for diseases started studying chronic diseases and doing studies following people Biostatistics: many mentioned were staticians if epi is science of public health, biostats is math of science. 1930’s: RAFisher developed concept of null hypothesis and statistics test 1948: FHS began 1950’s: BradfordHill and Richard don used statistical methods to quantify certain risks from smoking 197320005: Bogalusa heart study ***History is English Centric Version illustrates>not standard scientific progression multidisciplinary relatively young tends to be build by glacial accretion as a science epidemiology more about the tools than the results biostatistics is important to oland a lot of what we use is very new 01.19.2016 ____________________________________________________________________________ ___ A transition in Epidemiology judging by this history of epidemiology: what was the focus in the past? ( infections communicable diseases) what did it shift to? ( chronic diseases ) is that shift permanent? (no) What are we measuring in Epi? health and disease what is necessary in order to measure something? some type of scale; quantifiable units; gather statistics an agreed on and clear definition that allows for consistent quantification Definitions Health how has it been defined for you so far? From 1948 WHO statement “a state of complete physical, mental, social wellbeing and not merely the absence of disease of infirmity” original root word just meant to be whole ( looking for what’s missing) The basics: vital statistics refers to birth and death ( includes marriage/divorce records most common public records historically ( how long someone lived) generally gives you the where, who ( age, gender, race, when and how can be used to calculate group descriptors average life span→ this is where we get life expectancy from birth and death rates statistics by subgroup ( sex, age, places, etc. ) Measuring health can you measure complete well being? what can we measure life expectancy ( may be specific to subgroups) or what the average lifespan was pros: minimum requirement for health is to be alive fairly easy to compute cons: maybe too minimal only an average so “ expected” part is iffy ( based on past experience) Health adjusted life expectancy subtracts expected days with disabilities/illness pros: closer to defnition of health cons: expectancy on average subjectivity aspect various quality of life scales (QoL) Happiness indexes Hale and QoL mostly physical, the happiness indexes try to look at level of functioning as well but are you happy or not. this is a new area of measuring health new scales? > all scales have some issues of subjectivity Measurement issue: subjectivity=inconsistency (easily result in inconsistencies) Measuring the complement of health limited in trying to completely measure health so we often measure the complement of health scientists love this trick: ( the easier thing to measure) for example: if you can’t count life, count deaths if we can’t measure health can we measure those things that impinge upon it? morbidity mortality Outcomes: adverse health events ( negative parts) Broad Categories Direct disease injury Mental health/QoL Indirect Behaviors/conditions that are known disease or injury risks Reality Check what is disease? (this is the one we tend to focus on the most) generally thinking of a process ( mostly use this term in a narrow sense where there is some sort of physiological dysfunction that’s part of a process) General disease categories: Infections disease: disease resulting from the presence of a biological agent reproducing in the host ex: parasites, bacteria, viruses, prions may or may not be communicable must produce disease symptoms eventually generally acute, short latency What’s the difference ( http://emergency.cdc.gov/preparedness/quarantine/) infections communicable* contagious ( directly transferable; very communicable and directly from someone) *where epidemiologists start to get excited Review: communicable disease transmission direct transmission( person to person) indirect transmission vehicle transmission ( inanimate on something)>toilets, doorknobs vector transmission (animate)> mosquitoes, birds, fleas,ticks aka: vector borne Airborne bacteria virus can float in air Either! What’s fomite? other matter General disease categories: Chronic diseases generally not thought to be infectious long lasting condition >3 months according to NCHS often with slowly worsening conditions ex: cancers, diabetes, CVD, osteoporosis, asthma often unknown etiology particularly in individual less of a strict category than infectious Communicable vs. Chronic: Diagnosis communicable: generally straightforward diagnosis requires accurate knowledge of pathogen Chronic complicated,often must rule out infectious requires agreed on standards/definitions Communicable vs. Chronic: mortality communicable generally highest in young/very old ( immune system strength) higher when little access to medical care ( communicable tends to be treatable but no access makes it deadly) may be quickly fatal chronic higher rates with age seen more in absence of other causes of death not as fatal Communicable vs. chronic: risk factors communicable exposure to infectious agent immunity factors suppressed/weakened immune system unvaccinated not previously exposed populations chronic often lifestyle/behavior factors genetics various and multiple exposures communicable vs. Chronic: Epidemiology research focus communicable: prevention treatment chronic understanding causation pathways ( mostly for prevention purposes) early diagnosis ( if not prevention, catch it early) Which is it? cervical cancer lyme disease ( it is infectious food poisoning ( doesn’t fit in either) generally not communicable but not infectious AIDS Blindness there are things that result in chronic disease but caused by infection. diseases don’t always fit in categories We could categorize as: chronic vs. Acute communicable vs. noncommunicable 1.21.2016 Kkoch and causation 1. organism present in every disease case 2. it must be isolated and grown in pure culture 3. inoculation with culture 4. etc. Koch and causation...today original postulates: based on infectious disease ) no role of chance cannot be applied to organisms that only infect humans need to find causes of noninfectious disease need to identify “causes” that don’t always result in disease leeds to importance of concepts: risk and risk factors Defining risk risk implies the role of chance ( i something increases or diminishes your risk it’s like tossing a dice) it is the probability that something (probably bad) will happen Risk factors are those things that affect risk Determinants of health another word for cause, and a lot of times they are risk factors biological lifestyle/behavior environment access to health care ***according to Friis*** Causal Models there are different ways to picture causality keep in mind individual vs. populations issues consider how they address the role of chance there are models, models and then models Triangles also known as “epidemiological Triad” Host> environment> agent ( generally the pathogen) Wheels of causation figure 3.5 in a person,all of these things generally affect whether they have this disease or not Web of Causation how do the risks interact? Direct vs. Indirect Causal “Models” Direct: A causes B Indirect: C influences A which then cases B subsequent models tend to look like clear path ways. A→ B → C etc. Practical examples? Or...Necessary, Sufficient and the Component Cause Model A is Necessary if: Must have A toget B ( example infectious Diseases; you don’t have Aids without having HIV) Sufficient if: A alone results in B may apply to a set of causes ( risk factors) A and C are component causes i: in causal in combination → may be different combinations that are causal → may be combined with direct/indirect terms Rothman’s “Wheels” Sufficient case Sufficient cause I II some set of things ; if all are present you will get the disease. if you remove any one of them you don’t get disease or a specific disease there are several different combinational factors. these models illustrate sufficient component causes. What is necessary ( if only I and II lead to disease)? there is almost always a factor that is unknown does not show relationship between components each separate piece can be called a contributory cause Contributory causes in component cause model contributory causes need to be neither necessary nor sufficient ex: smoking has a causal relationship with cancer ( smoking is neither sufficient nor necessary) Most risk factors we study are contributory if something is necessary, easy to study and not requires a lot of studies to prove that Directed Acyclic Graphs: graphical way to describe causal associations. direction of causal chain indicated by arrows. can not be a loop ( acyclic). arrows do not indicate sole causes U> S D> Y extended to looking t things on a group level; at an individual you may not have all of these things. POpulation vs. Individual models do not always make clear distinctions between population and individual may be multiple causes in population but not individual may be multiple causes ( component cause) for individual case epi studies look at causation on population level. What studies find: Associations how much do we know that something is truly causal association does not imply a direction says nothing about direct, indirect,necessary or sufficient weasel word? epidemiologists like it actually most accurate description of results How do we decide an association is a cause if we can’t use Koch’s postulates? The scales of epidemiology! criteria that researchers have come up with, if we meet enough of them we say it’s causal and if not we say it wasn’t Dr. Ulfers Quick Guide to Causal Criteria Caveats 1. STRENGTH OF ASSOCIATION – Interpreted here as a large effect àproblematic if artifactual, or by chance (from small sample), how large is strong enough? (Sometimes people interpret this as “statistical strength” or a small pvalue but this may be due to a small effect found in a very large sample so we won’t …this will make sense later J) 2. CONSISTENCY – Multiple studies addressing the same question give same results à Problematic if same artifactual associations found in all studies à Best if same results under different conditions and/or with different designs 3. SPECIFICITY – A risk factor produces one particular effect that is generally not an effect seen from other risk factors example: vinyl chloride (rare exposure) is risk factor for hepatic angiosarcoma (rare outcome) à Problematic in that it rarely applies in noncommunicable diseases, few causal associations, aside from pathogens leading to infectious disease, are truly specific 4. TEMPORALITY – Exposure precedes disease – Will disprove causal association if refuted (i.e. if the “cause” did NOT occur/exist before the “effect” it can’t be the cause!) à Problematic because a lot of things came before the disease!!! à The most abused of the criteria. 5. BIOLOGICAL GRADIENT – aka Doseresponse relationship, but response seen on a group level à Problematic if artifactual association and false factor closely associated with true cause à Can’t see this if only yes/no exposure status 6. PLAUSIBILITY – Makes biological sense àProblematic because sometimes accepted “knowledge” is incomplete or even wrong. 7. COHERENCE – Fits in with other research findings (ex. Smoking causing lung cancer fits with: more men die from lung cancer, more men smoke, carcinogens in smoke, etc., etc.) 8. ANALOGY – Other disease models have similar mechanisms à Both 7 and 8 similar to plausibility with similar caveats Some comments on repeatability scientific principle that a study’s results should be confirmed by repetition yielding the same results rarely ever done in medical studies consistency is as close as we generally get and it’s not quite the same thing 01.23.2016 ______________________________________________________________________ Review/Questions How do you measure association? ex: smoking & Lung Cancer Samples vs. Population N ( Population); N could be infinite n ( Sample); sample size is limited but has to represent all of N Estimates we can only measure samples ( usually) only estimate true population value unbiased estimates will be right on average due to random error ( chance) if population is measurable, no estimation needed How good is estimate depends on sample size ( could be relative to group you are referring to) closer to ‘n’ is to N, the closer estimate will be to true value depend on variability of possible measurements ( how representative it is) More on Confidence wider range= less confidence in estimate X% confidence means what exactlY? 100% confidence would include all possible values almost always a 95% confidence interval though Measure of association estimates afeected by chance measured association can be due to chance and not ture association if no true association→ relative risk=1; OR=1, RD=0 if risks are eaual, odds same, RD=0 Value for Null Hypothesis null hyp. states no association therefore RR=1, OR=1, RD0 ( if those are appropirate measures Alternative hypothesis is complement therefore RR does not equal 1, OR does not equal 1, and RD does not equal 0 In statistical testing assumption is that there is no bias consider random error [handouts] 01.26.2016 ____________________________________________________________________________ __ Friis pg. 98100 Causality in Epidemiologic Studies: issue of causality includes several criteria that must be satisfied: criteria of causality causal and noncausal associations non causal: the association could be merely a one time observation, due to chance and random factors, or due to errors in methods and procedures Causal: Criteria of causality A.B. Hill proposed a situation in which there is a clear association between two variables and in which statistical tests have suggested that this association is not due to chance Hill’s Criteria of causality strength, consistency specificity temporality biological gradient plausibility coherence experiment analogy Types of causality: multifactorial ( multiple causality) Figure 511: the declaration of a causal association involves a process that is similar to a jury weighing the evidence in a trial. Figure 512: the web of causation ____________________________________________________________________________ Hill’s list of causal criteria not the only list, some authors list up to 14 for instance… Reversibility: ( removing exposure you remove disease) but this requires: a. ability to remove the exposure and b. specificity ( sometimes reversibility is proof of specificity) Communicable vs. Chronic: etiology communicable diseases usually have a necessary cause ( the particular pathogen) → chronic generally multiple causes ; multiple pathways generally multiple risk factors in an individual or in a population there is multiple risk factors ( some people get it from x and some people get it from Y) Major role of genetics Lecture 5: Descriptive epidemiologyPerson, Place and TIme Learning objectives; Know what descriptive epidemiology is know what are the most common person, place and time variables understand the importance of variables in epi consider ole of estimation in sampling Where to start Descriptive describing everybody in some population at some place and time ( whole group or one category) of diseased only of healthy only of exposed only of unexposed only descriptive study asks questions what is everyone like, etc. table 1 describes distribution of those variables how something, any variable is distributed along the lines of another group. who’s got the disease ( with age) how is it distributed by age, place, etc. geographically also Generate hypotheses I think A is related to B or A is causing B descriptive studies aren’t answering that they are just observing it; you might have an idea of what you're gonna find you want to confirm it? describing findings not testing hypotheses Analytical Tests hypothesis Variables in descriptive study mostly demographics ( gender race age, etc) common to many studies, not just a particular study exposures ( risk factors) Health outcomes Personal characteristics ( that go beyond general demographics) particular to specific study How do we quantify the variables? Discrete or categorical variables in health ( ex: mortality) Frequencies/Counts Continuous average/means **Note difference between data for individual and group!!! “Simple” Statistics Counts ( a group measure) ( discrete variables) Percentages/proportions and rates ( discrete variables) means/medians ( for continuous variables) → all may be given for subgroups of interest, and sub-sub groups, i.e. stratified Demographics What is included? Person, Place, and Time Descriptors of interest in population What isn’t included? outcomes or risk factors….in genral PERSON age most important health determinant?? overlaps with time ( the older you are the longer you’ve been exposed….biological significance to it and just risk) average age isn’t always enough ( different population group might have a different distribution) in any particular group or sample you will have a different distribution of age → distribution is important!!! Sex 2nd most important determinant biological makeup and lifestyle associated with different risk factors Biological differences/behavioral → can’t assume results will apply to both sexes Race Differ in Health “determinants” Politically/culturally important Ethics Genetics may be studied as risk factors diagnostic markers to define subgroups → not as commonly used ( more analytical than descriptive study) SES related to many health determinants no one definition ( ex: income levels); multiple variables used to measure this in different ways Education indirect health effects a determinant of other determinants) often a surrogate for SES ( ex: to measure income without asking directly if you can’t get that information) or separate from SES **** this is nont an exhaustive list of what might be considered person variables Place: Geographic Location Provides information on: Environment: political/social Environment: physical Population density *** all this from just the name of the place Time time from exposure to disease time from diagnosis to death time from treatment to ….. Time: calendar period provides information on: social/political environment physical environment ( changes based on time of year and time period) timing of exposures ( in time period) Age, Period, Cohort cohort: everyone who was born around the same time ex: person got cancer at age 70 in 1950 ( all born in 1920’s shared similar experience) an event may be related to: age( biological importance) period ( calendar time) cohort ( born in same period) → age at event + cohort = period of event Person, Place and Time provides the context or any study so important they're usually in the title ( if not, it means it’s current) Are person, place, and time variables risk factors/ depends on your perspective may be associated with, but not a risk factor ( is it a risk factor if it is a characteristic that can’t be changed?) Generally not included in descriptive section of analytical study if risk factor of interest have to wait or the results potential risk factors of interest included in a descriptive study if not yet analyzed or tested Defining Populations Describing groups by selected variables Reality check: Target Population ( interest) vs. Study population ( availability) vs. Sample ( actual data) Sample vs. Census sometimes we count everyone ( census) vs. who we could get ( sample) usually we can’t get a sensus so ( sample) Estimation Statistics calculated from a sample are only an estimate of the true numbers all estimates are affected by chance and bias bias is systematic error and not just happening randomly) all in one direction….doesn’t matter how big study, missing same things over and over again. can take care of random errors chance is a random error ( could be in any direction) ….cancel out A little confidence we can account for random errors with statistical techniques ( like confidence intervals) These statistics all assume no BIAS Bias is a study/measurement issue ( not that researcher is biased but something they are doing is always causing a problem.) results are wrong but every time you do the study it is going to be wrong in same direction WHy we love biostatistics describe a group by person, place, time etc. biostats help us decide if estimates are accurate and true representation probably 01.28.2016 ____________________________________________________________________________ __ know standard measures used in epi studies rates, ratios, risk, probability and proportions how used in epi prevalence and incidence importance of denominator how time is included in epi measures Common Language strict mathematical definitions vs. epi usage vs. sloppy usage Basic Terms: Ratios Basic Terms: Proportions a ration of a part to the whole, or subset to entre set ( always same unit of measure) in common uses in epidemiology we use ration for comparisons that are not proportions by this definition Basic Terms: Probabilities chance of an event is expressed s number of events over the number of times event could have happened Percentages vs.Percentiles percentage just a proportion expressed out of 100 percentile: mostly with distribution not same as percentage ( below or above a certain point) Basic terms: random sample idea that everyone had an equal probability of being selected or observed not to mean each possible value has an equal chance Probability and risk risk is the probability of a future event true risks are therefore always estimates ( can’t have actually measured risk because it is in future tense) probabilities of events that we observe in health is often interpreted as the risk of the event since is the ris is what we are usually interested in, we have a tendency to call the probability of what we observe the risk more about risks risks are usually applied to the individual but it has always been computed from group data risk not used for characteristics of an individual Reminder various factors can be used to define either the events or the observation Basic Terms: Frequency a count of events most useful if reporting Prevalence Incidence/Incident Cases Incidence is something new; a measure of new cases or events relative to the population at risk over time can be measured as a count ( incident cases) or a proportion ( incidence) but always in a defined population over a period of time incidence of disease and rate of disease used interchangeably [SLIDES POSTED] 02.02.2016 ___________________________________________________________________________ missed few minutes!!! True rates and person years Crude vs. specific vs adjusted measures crude rates are those observed over entire population. no adjustment is made variable specific rates are rates given a particular value of some other variable or combination of variables aka “stratified” most commonly by age, gender, and/or race adjusted is a summary of strata specific measures that removes effect of stratifying variable on comparisons standardization is a common method of adjustment Standardizations concern is that it would be misleading to compare populations that differ by age, gender or race differences in distribution are adjusted for instead of making separate comparisons Direct standardizations give us comparison ready rates apply age ( race, gender)specific rates to age ( race, gender) distribution of standard population for new summary measure ……………. ………. Public Health General Knowledge framework for interpreting various rates ( and knowing what to expect) age and mortality racial differences gender differences SES effects BMI effects exercise effects smoking ***general core knowledge **exceptions matter! did you miss something? was underlying assumption wrong? lecture 7 most simply graphical display: “simple” tables Histogram ( continuous) vs. Bar graph histogram is a typical way to depict the distribution of variables the more people in study, the finer the bars will be pie charts proportions PMR (proportionate mortality ratio) easy to show in a pie chart limited line graph example, showing mortality by cohort period effect, age effect, cohort effect scatterplot dots can represent individuals or a group relationship between two continuous relationship Why are descriptive studies important? public health applications for planning understanding etent, burden of disease ( knowing how bad it is not just or planning but for prioritizing) want to know person, place and time generates hypotheses regarding: risk factors prevention approaches relationships in causal pathways Epidemiological studies what are we looking or ultimately? ways to prevent disease identify the links Epidemiological hypotheses links suggested by descriptive data not tested by descriptive data motivates analytical studies What sort o descriptive data suggest links”? Epidemiological hypotheses exposure/risk factor/treatment group has a different outcome than non exposure/risk factor/treatment group exposure associated with outcome at group level 02.04.2016 ____________________________________________________________________________ __ Normal Probability density function ( standard normal mu and sigma= 0,1) LEcture 8: Making comparisons EWhy make comparisons? a true causeeffect means absence of cause equals absence of effect presence of the caus makes a differences to measure a difference there must be at least 2 numbers to compare ( quantitative values to measure a difference) Implied comparison groups might be no FORMAL comparison group if: outcome in unexposed well known clear expectations Formal comparison groups necessary in epidemiology to test hypotheses May be part of the study ( internal) investigator collects all data may be general population ( external)...we know exposure group is so different than general population we can compare the measurements in our study population to measurements that have already been made in general population need data to be available ( and if you're looking at anything tricky want to make sure they measured it the same way you measured it) Can you ever look for causation without a comparison group? NO you can never really talk about causation without a comparison group 33 people took an experimental drug (A) and 29 had immediate hair loss. 33 healthy people took a different experimental drug (B) and none had hair loss but 3 reported headaches the following week. there is no formal comparison group but is there evidence for a causal association between either drug and the reported effects. you would need a comparison group for drug B but not A the point is that when we make decisions there has to be a comparison groups but sometimes it's okay if it’s not a formal one and an implied one. who are you really comparing them to? ( expectations of healthy people) WHen we measure a difference… what then? Have to test to determine if it is real Test hypotheses: what we think will happen..PLUS Testable ( is it real or is it something that could have happened by chance available test require: measurable variables, quantitative value to test or quantitative hypothesis value no difference between comparison groups gives a testabel and interpretable quantity Null hypothesis if any 2 group measures are the same, the difference is zero or the ratio is 1 so we test if there is NO effect, aka the NULL when comparing groups the null hypothesis predicts groups will have same outcome mathematically test to see if what you got was due to chance *null* or reject Null and alternative alternative hypothesis: what we’re left with if we reject the null they are complementary: together they encompass all possibilities they are mutually exclusive EX: Null: group A has the same incidence of disease as group B alternative: group A and B have different disease incidence if we reject the null we can conclude the alternative Does smoking cause lung cancer? nUll: smokers have the same rate of lung cancer as nonsmokers alternative: they have different rates conclude there is an association You are interested in the health effects of exercise. need measurable comparison. Narrow down to measurable outcome and measurable “ exposure” null: the mortality rate of people who exercise more than 30 min. twice a week is the same as for those who exercise less. Association if only difference between A and B is exposure status the exposure and disease are associated what if there are other differences? Scientific method ( again) requires empirical evidence requires measurable differences to have quantitative comparison requires quantitative measures concept of constants scientific ideal is to hold all but variable of interest constant too many variables with humans to keep constant term not used in epi too far from reality control of differences at group level instead Ideal comparison groups: almost exchangeable ( everything esle is same except for risk factor) except for the factor we are investigating will need same units of measure should be comparable Top 3 issues in student designed studies: 3.not having any comparison group 2. not having proper comparison groups 1. not having measurable variables Confounding ( not ideal) comparing groups that differ by some other factor that changes the outcome can’t know if results are due to difference in exposure or difference in confounder ** confounded results are artifactual Examples of confounders anything and smoking location nd average age of groups occupational exposures and healthy workers SS and location ( or education or race) apparent effect of studied factor due to confounding confounding can hide real associations and make associations that aren’t there. changing comparison of the groups Which is the confounder the one with the rue association no association with outcome = no confounding therefore: confounders must have been associated with the exposure status/level affect the outcome Most popular example death rates are much higher in florida than in Alaska proving that living close to disney world is dangerous problems? other differences age highly associated with mortality the apparent difference in death rates changed when controlling for age Consider age as a confounder can stratify analyses results age specific rates Mathematically adjust comparison standardize rates include age variable in math models standardized mortality ratios ( indirect..) **these approaches are part of analyses Match subjects between groups by age compare groups with same age distributions compare groups that are age “restricted” ***part of study design Any confounder can stratify analyses results specific rates by variable level mathematically adjust comparison aka “c
Are you sure you want to buy this material for
You're already Subscribed!
Looks like you've already subscribed to StudySoup, you won't need to purchase another subscription to get this material. To access this material simply click 'View Full Document'