Epidemiology Final Exam

selection of controls

Cases and controls should come from the same population-at-risk Controls should have the potential to become a case Controls should be "disease free" of case disease Disease free doesn't always mean healthy Must not share common exposure Use extreme caution when using controls with other diseases! Controls should have similar characteristics as cases with respect to likely confounders (age, sex, etc.) If cases and controls are equal with respect to everything but exposure, then the difference in disease status is ascribed to the difference in exposure status Examples of cases and appropriate controls: Example A: Cases in a sample of the population Controls are non-cases in sample of the same population Example B: Cases from a hospital Controls are a sample of patients without the disease in the same hospital Example C: Cases from the NJ Cancer Registry Controls are NJ residents

relatives and friends as controls

Cases provide names of friends to serve as controls Advantages: Friends might be more likely to participate as controls than people randomly selected from the general population Disadvantages/serious problems: Controls may be too similar to cases in exposures Also, controls may not be from the case source population- this is a big problem! Multiple controls add statistical power to the study Using more than one control group makes the results more believable 1 case : 1 control 1 case : 2 controls 1 case : 3 controls Maximal statistical power with a ratio of 1 case : 4 controls

selecting the comparison group

Comparison groups Are necessary to show a difference between exposed and unexposed Allow for a stronger argument regarding the exposure-disease relationship In a cohort study, the comparison group = unexposed The unexposed comparison group may be: 1. Internal - if from the same population as the exposed External - if from a different population than the exposed Population incidence rates - an unexposed group is not readily available, so population incidence rates are used Option #1: Cohort design using Internal Comparison group Exposed = employees that work with vinyl chloride Unexposed = employees that don't work with vinyl chloride (e.g. secretaries, management staff) Make sure these folks are clearly unexposed Option #2: Cohort design using External Comparison group A good choice when exposure is so widespread that nonexposed in same population hard to find Exposed = employees that work in the vinyl chloride factory Unexposed = employees at another factory (e.g. vinyl chloride not used anywhere in factory) Option #3: Cohort design using Population Incidence Rates group Results in calculation of SMR (standardized mortality ratio) - more on this in future lectures May have limited utility as population rates not adjusted for all likely confounders (e.g. smoking) Exposed = workers at the vinyl chloride factory Unexposed = incidence rate of liver cancer in the general population

confidence intervals

Confidence Intervals (CI) - computed interval of values that with a given probability, e.g. 95%, contains the true value of our estimate Can be calculated for the odds ratio as well as other relative risks (i.e., risk ratio, SMR) calculated with our study designs You'll be learning about these other relative risks in upcoming lectures Most contemporary epidemiologists prefer to focus on Confidence Intervals rather than p-values. Why? Confidence Intervals tell us as much as a p-value, plus quite a bit more : CIs conveys information about the magnitude (how big) and precision (how close) of our point estimate CIs indicate the amount of variability in our data as well as the adequacy of sample size CIs also indicate if our result is statistically significant (on other words, whether or not we can reject the null hypothesis) CIs let us conceptualize epidemiologic research as estimation rather than just decision making A 95% CI is written as a range of numbers: e.g. OR = 2.79 (95% CI: 1.45, 4.78) It is interpreted as meaning, that with 95% confidence, the true value of our OR lies between the 1.45 and 4.78 The width of the CI is related to sample size Large sample sizes produce narrow CIs Small sample sizes provide wide CIs How do we know if this result is statistically significant? Our result is statistically significant (at or below p=0.05) if our 95% CI does not contain the null value for that estimate. Recall that the null value for OR=1 Question: Is this estimate statistically significant (at or below p=0.05)? OR = 1.79 (95% CI: 0.79, 2.38).

experimental research

Considered the "Gold Standard" (i.e., best practice or that which other designs are measured) of study designs Includes intervention designs, such as the Randomized Control Trial (RCT) Used to test effect of new drugs, tools, and programs to see if they prevent, cure, or slow the rate of disease Also includes field trials and community interventions Key factors Investigators manipulate study factor that affects disease Randomization of study subjects

exposure

Contact between a substance or agent in an environmental medium and the boundary of the human body

epi triad- environmental factors

Environment Domain in which the disease causing agent may exist, survive or originate. Sum of total influences that are not part of host May serve to bring agent and host into contact May enhance or diminish survival of disease agents Includes a wide range of physical and climatologic, biological, social and cultural factors Environment may act as reservoir or niche Reservoir - the natural habitat in which an agent lives, grows and multiplies Physical environment E.g., Soil and fungal disease causing San Joaquin Valley fever Human or other animal reservoirs E.g., "Zoonoses" = an infection or infectious disease transmissible under natural conditions from vertebrate animals to humans slide 15 module 11 Example of a Zoonosis: Plague In 1-6 days after an infected flea bite, the subject gets lymphadenopathy The bite gets gangrenous and necrotic - turning black The resultant infection may cause septicemia and death

study designs

Environmental and Occupational epidemiology use a combination of study designs Descriptive studies Identify hazards Formulate hypotheses Set priorities Analytic studies Examine causality Determine the natural history of disease Clinical end-points (disease or injury) assessed by: Self-reported symptoms Morbidity Mortality Exposure assessed by: Self-reported questionnaires Clinical examinations and lab based measures Studies depend upon accurate exposure measurement & assessment

reviewing the literature

Epidemiologists frequently begin by searching biomedical databases, that cover a wide range of health fields PubMed from the National Library of Medicine Same journal coverage as MEDLINE Others include: SCOPUS, Web of Science, Science Direct - access through your school or workplace (subscription based) Databases typically provide author-written abstracts and link to similar studies Often link to full text - sometimes freely available Tip: If you are prompted to pay for an article, be sure to go through Rutgers Libraries instead!

significance testing and confidence intervals

Epidemiologists frequently employ statistics to help us interpret our observed associations Significance tests and P-Values Confidence Intervals ("CI") Our goal is to understand what each is and compare their relative value In general, anything that you can get from a P-Value, you can get from a CI Plus, a CI tells us much more! Significance tests - a statistical procedure that, when applied to a set of observations, results in a p-value relative to some hypothesis Examples: Student's t-test, Chi-square test of independence Significance level - criterion we will use to rejecting the null hypothesis Examples: 0.05 or 5%, or 0.01 or 1% Acceptable level varies by academic field and by study design In general, the lower the significance level, the more the data must diverge from the null to be significant. P-value - indicates the probability of obtaining the observed result or one more extreme if no real effect exists (if the null hypothesis is correct) Examples: p= 0.045 or 4.5% ; p=0.001 or 0.1%

exposure assessment

Exposure is an estimate We rarely have exact exposure measurements, particularly if the exposure happened long ago Assumption: exposure already happened by the time disease process began Exposure measurements are subject to recall bias and interviewer bias: Recall bias: When assessment relies on subjects' memories, cases may search for an explanation for their disease and may be more likely to report (or misreport) information than controls. Interviewer bias: When interviewer has knowledge of subjects' disease status, he/she may ask certain questions differentially of cases than of controls.

interpreting findings

Finding no statistically significant difference... in a small study tells us nothing Could lack statistical power (sample size) Lacks ability to demonstrate an association if one exists in a large study tells us treatments or outcomes are essentially equivalent Could be other factors that make one a better choice for patients in a clinical setting Finding a statistically significant difference... in a small study may not be able to be replicated because of sampling variation in a large study reveals a true difference, but the findings may not be clinically important Rule of thumb: If the values indicated by your CI is smaller than a clinically relevant difference, you should not change your practice no matter what the statistical significance

descriptive designs

Focus on Person, place, time Used to describe the distribution of disease May be used to examine correlations between exposure and disease Look at the large-scale picture May use group or individual data depending on design Do not account for temporality Cannot infer causal association Typically used to generate hypotheses

individual matching

For every case, a control is selected with the same factors of interest Age Race Smoking status Examples: For every case that is 21 years old, select a 21-year-old control For every Asian case, select an Asian control For every case that smokes, select a control that smokes

ionizing radiation

From particle or light energy Can cause DNA damage in cells, mutagenic effects Cancers with known sensitivity to radioactivity: bronchus, trachea and lung; breast; some lymphatic and hematopoietic tissue, organs); thyroid Latency a problem Nuclear accidents 1979 Three Mile Island - increased incidence of breast cancer initially observed, within 5 yrs, but no apparent increase in long term mortality 13 yrs later 1986 Chernobyl & increased thyroid cancer in young adults 2011 Fukushima Daiichi - future cancer concerns (radioactive milk) Effects of nuclear weapon testing - heavily influenced by age, diet, sex and location Atomic (bomb) Exposure at ages 10-19 years increases the risk of breast cancer in Japanese Radon Inert gas produced by decay of radium and uranium found in earth's crust May cause up to 20% of U.S. lung cancer

hypotheses

Generating hypotheses are very important! For any public health problem, first step is to formulate a reasonable and testable hypothesis. Hypothesis = suppositions that are tested by collecting facts that lead to their acceptance or rejection They may also be thought of as "refutable predictions" Three common ways of stating hypotheses: 1. Positive declaration (research hypothesis) Example: "The infant mortality rate is higher in one region than another" 2. Negative declaration (null hypothesis) Example: "There is no difference between the infant mortality rates of two study regions" 3. Implicit question Example: "To study the association between infant mortality and geographic region of disease."

characteristics of persons- nativity and migration

Geographical movement of people can affect health From place of birth to new land Geographic movement in general Examples Honolulu Heart Study - effects of migration on Japanese men Over time, increased risk for heart disease in new country Associated with change in diet, environment, stress and/or other behaviors 1930 study of mental hospital patients in NY Higher levels of psychological stress among foreign born than in native-born people Importation of formerly controlled/eradicated diseases Malaria, intestinal parasites, bed bugs, etc.

evaluating the exposure-disease relationship

Goal of undertaking a study is to evaluate the exposure-disease relationship Descriptive studies: describe populations Analytic studies: causal associations To assess, we must compare at least two groups with vs. without exposure with vs. without disease Epidemiologic studies usually organize data in tables 2x2 contingency tables offer a simplified view Disease status goes in columns: yes/no in top row Exposure status goes in rows: yes/no in side column What does the 2x2 table tell us? Number of people with disease/other health outcome (a+c) Number of people without disease/other health outcome (b+d) Number of people with exposure (a+b) Number of people without exposure (c+d) Total number of people in the study (N) Using 2x2 tables in analytic designs allows us to calculate measures of association Used to make inferences using Hill's Criteria, e.g., Strength of Association We can only infer causality, however, if temporality is also satisfied - otherwise, the association is correlation Required for a "good" study design A clear case definition Defining what constitutes exposed vs. non exposed Definitions must be exclusive (not overlapping) Must reduce potential sources of bias/confounding

plot an epidemic curve (Epicurve)

Graph the cases x-axis = time of onset y-axis = number of cases Interpret the curve Describe pattern Determine incubation period and perhaps mode of transmission Examine Outliers Point Source Transmission (common source) Most common type of food-borne outbreak Sharp increase, then tapers off Large population with a short exposure period Easier to identify incubation period Continuing Source Transmission Several peaks of cases Difficult to characterize incubation period Host-to-host transmission: Spread person to person Common pattern for viruses Initial rise, then tapers off gradually due to, e.g. Attenuation of virus Decreasing # of susceptibles Vector-borne Disease Cases start slowly Incubation period estimated by time between first case and the peak of cases Cases taper off slowly

determine the source of the epidemic

Identify the most likely cause and investigate the source to prevent future outbreaks If there is no obvious commonality, plot the geographic distribution of cases by residence/work/school /location and seek common exposures

characteristics of persons-marital status

In general, married people have lower morbidity and mortality rates than non-married people Lower mortality from many chronic diseases, many infectious diseases, and fewer motor vehicle accidents Marital status is also related to suicide rates Highest rates among widowed / divorced persons Lowest rates among married persons Marriage may be a protective and/or a selective factor for good health Being married can be helpful (e.g., physical, financial and/or emotional benefits) It is also possible that healthier people get married more than people who are unhealthy

data interpretation issues

In order to have confidence in our data, we must first be sure our estimates are scientifically valid In terms of study designs, validity specifically refers to the degree to which inferences drawn are warranted given the study methods and study population There are two types of validity: Internal and external validity

calculate Attack Rates & RR

Incidence proportion during a time period Typically expressed as a percentage (not a true rate!) Attack Rate = slide 15 *If everyone is susceptible, then this is the # ILL (with clinical illness) + # WELL Calculated for different exposures or settings Calculate the Relative Risk by comparing attack rates = slide 15 module 11 slide 16,17 Secondary Attack Rate Used to estimate to the spread of disease in a family, household, dorm or other group environment. Measures the infectivity of the agent and the effects of prevention (e.g. vaccines) formula slide 18,19

exposure routes- inhalation

Inhalation: taken into the body through respiratory system Breath in gases, vapors, dusts and aerosols Lungs = branching airways (bronchi) with clusters of air sacs (alveoli) Irritants come in contact with bronchi Particles (depending on size) may get deposited in bronchi/alveoli

ethical conduct- the IRB

Institutional Review Board (IRB) Goals is the protection of human subjects Comprised of a variety of people, including at least one: Scientific individual Non-scientific individual Community member not otherwise affiliated with the institution Reviews all types of epidemiologic studies; not just experimental designs Approve research protocols Monitor compliance of study team Ensure informed consent is acceptable (if applicable given the design - always needed for experimental research) Informed Consent describes: Overall experience that will be encountered Risks and Benefits of participation Alternatives to participation Extent to which the subject's information will remain confidential Compensation and/or expenses Informed consent is a process Obtain informed consent prior to enrollment A subject withdraw at any time without negative consequences Provide the subject with a list of people s/he can contact with further questions, including research-related injuries Collection and use of personal information is also governed by HIPPA (see earlier lecture)

validity of study designs

Internal Validity - Validity of inferences as they pertain to actual subjects in the study Depends on the absence of systematic error This is your primary concern as an epidemiologist! External Validity - Validity of inferences as they pertain to people outside the study population A matter of generalization, abstract inference Must have internal validity first Not always possible, depending on your study Large Randomized Controlled Trials (RCTs) usually have a highest degree of internal validity because of steps taken to minimize the risk of error Randomization, restriction, blocking and blinding Large sample size But, controlled trials often have poor external validity The study population may be a very specific part of the external population E.g., RCT testing osteoporosis treatment on females aged 50-60 years What we learn from the results may not apply to older females May not apply to individuals with comorbidities ranking- slide 35 module 10

sample of an outbreak investigation

Investigating an Epidemic: Oswego, NY On April 19, 1940 an outbreak of acute gastrointestinal illness was reported to the District Health Officer in Syracuse, N.Y. Dr. Rubin, was assigned to conduct the investigation. Dr. Rubin learned that all persons known to be ill had attended a church supper the previous evening, April 18. Family members who had not attended the church supper had not become ill. Accordingly, the investigation was focused on the circumstances related to the supper. Select the best case definition and find the error in the others: All participants in the Oswego church supper held in the basement of the church in Lycoming, Oswego County, New York, on April 18, 1940, between 6:00 PM and 11:00 PM; whether they attended church or not; whether they participated in food preparation, transport, or distribution or not; whether they ate or not. Persons who developed acute gastrointestinal symptoms within 72 hours of eating supper on April 18, 1940, and who were among attendees of the Lycoming, Oswego Church supper. Church members who developed acute gastrointestinal symptoms within 72 hours of the church supper held in Lycoming, Oswego on April 18, 1940. Calculate the overall Attack Rate for all attendees (assume all persons attending were susceptible) Solution: see Notes View Calculate item-specific Attack Rates using the supper menu (shown on next two slides) Using the item specific Attack Rates, calculate the relative risk of exposure between exposed and unexposed for each group (not shown - give it a try) The foods that have the greatest difference in attack rates may be the foods that were responsible for the illness. Greater the RR, the more likely to be the source slide 35 module 11

characteristics of persons-gender

Male vs. female Males have higher mortality rates than females Females have higher morbidity rates than males Males have higher rates of chronic handicap than females Females are more likely to attempt suicide, but men are more likely to successfully commit suicide Differences may be related to biological, lifestyle, behavioral and environmental factors Example: From 1930 to 1990 lung cancer mortality increased faster for women than for men

characteristics of place- within-country comparisons

Many countries have substantial variations which are related to the frequency of disease Environment (climate, geology, latitude, pollution) Population (density, ethnic/racial origins) Typical within-country comparisons in United States Pacific vs. Mountain vs. Central vs. Atlantic Northeast vs. Southeast vs. Southwest, etc. State vs. state County vs. county (see following slide for obesity differences) Current examples of disease variation in U.S. : Sunbelt states: highest rates of malignant melanoma and strokes AIDS: highest rates (1996-97) in states with high urban populations Multiple sclerosis more common in northern states (may have a genetic/environmental component) Higher death rates from leukemia in Midwest (may also have genetic/environmental component)

study designs

Many study designs available Differ in time required, cost, purpose, strengths and weaknesses Selection of appropriate design depends on goals of research When little is known, a quick economical study generally precedes lengthy and costly one As knowledge increases, undertake more rigorous and expensive study designs Study designs differ in several respects: Number of observations made One observation vs. observations at multiple points in time When and how exposure is measured Retrospective vs. Prospective designs Subject recall vs. direct monitoring of exposure Amount of time elapsed between exposure and disease Study designs differ in several respects: Data collection methods Previously collected vs. collect yourself Unit of observation Individual (information on individual people) vs. group (information on a population only, e.g., a state, a city, a town) Availability of subjects Ethical considerations including protection of vulnerable populations

characteristics of persons: race and ethnicity

Overlapping and often synonymous identifiers Used to describe ancestry and/or cultural differences Social classifications - describe self and others Also overlap with place of birth (nativity) and religion, SES Differences by race and ethnicity can be difficult to interpret Levels of diversity are often lost in race/ethnicity statistics Often seem to be diet and or stress related Some patterns currently observed in US: African Americans have the highest mortality of any racial group, especially for hypertension and stroke Diabetes incidence is higher for Native Americans than for Whites. Native Americans also have the highest infant mortality rates in US Asians have lower mortality (live longer) rates than Whites Unique morbidity and mortality profiles in Hispanic/Latino subpopulations Lower rates of coronary heart disease in Mexican Americans Higher rates of diabetes and obesity in Mexican Americans in American southwest.

case-control studies

Overview: Case-control studies are an observational, analytic design that seeks to identify causes of disease by studying how risk factors affect diseased and non-diseased people differently. At baseline Disease status is known Exposure status is unknown How it works: Subjects selected based on disease status Cases =Disease of interest Controls = No disease of interest Collect data on past exposures Compare frequency of exposure among cases and controls with calculation of odds ratio (OR) for exposure(s) of interest

host

Person or other living animal that affords subsistence or lodgment to an infectious agent under natural conditions May be population or group Includes numerous physiologic and personal factors Immune status Immunity can be Natural = results through having infectious agent Artificial = by vaccine that stimulates antibody production Immunity may be acquired Actively = produce own antibodies Passively = preformed antibodies Immunity may be of Short duration = a season, a year, etc. Long duration = last a life time or many years

reducing confounding

Prevent by Design Randomization of a large numbers of subjects - distributes known and unknown confounders equally between the intervention and control groups (experimental designs only) Restriction- Inclusion criteria only allows people without the known confounder into the study Matching - For each case with a confounder, include a control who also has the known confounder Control through Analysis Stratification - Analyze results by breaking the study into subgroups based on presence or absence of the potential confounder Standardization - Obtain summary measure through use of adjustment Multivariate analysis (e.g., regression, ANOVA) - Simultaneously take several variables into account during the analysis (details are beyond the scope of this class!)

SMR- key limitations

Problems using SMRs to study change in a population over time: The longer you follow a study population, the less information the SMR yields Everyone in the population will eventually die, and the SMR will tend to equal 100% over time SMRs can hide differences in subgroups Example: An SMR of 100% indicates that the morbidity/mortality in the overall study population doesn't vary from the general population, but it doesn't tell you what is going on within the subgroups of the study population.

case control studies- advantages and disadvantages

Relatively quick and inexpensive It's a good method for studying: Rare diseases Diseases with long latency periods Diseases for which people seek medical care Examining multiple exposures Temporal relationship between disease and exposure may not be clear or known It cannot provide direct risk estimate of exposure and disease relationship - more on this in our next module Inefficient for studying rare exposures Recall and interviewer bias may occur

casual pies/ understanding prevention

Removing one risk factor will not result in the complete elimination of the disease unless this factor is present in every causal pathway. Which causal pathway would still remain even if we were able to block or eliminate risk factor C (e.g., smoking)? Cause 2 remains because it does not contain component factor C. In other words, smokers could still develop stomach cancer from another causal pathway (Cause 2) even if they quit smoking. Prevention/policy implications: Prevent any one risk factor and you prevent the disease occurring from that sufficient cause Any remaining risk is due to factors forming other sufficient causes. Epidemiologists like to account for other pathways to disease when estimating the benefits of prevention efforts for a specific risk factor. To do so, we calculate the Etiologic Fraction

rd key formula

Risk Difference (or Rate Difference) (RD) Formula: Difference between the incidence of disease in exposed (Ie) and the incidence in the nonexposed (Ine) Difference between incidence proportions = Risk Difference Difference between incidence rates = Rate Difference RD = Ie - Ine Interpretation: The RD estimates the excess disease associated with exposure compared to the nonexposed Incidence proportion data interpreted as 'excess risk' Incidence rate data interpreted as 'excess rate' example slide 5 module 9

interpretation of a SMR

SMR > 100% Means the study population has a higher rate of death/illness than the comparable standard population. In other words, the study population is in general less healthy than the standard population SMR = 100% (Null effect) Means that there is no difference between the number of outcomes in the study population and what you would expect to see if the study population was like the standard population SMR < 100% Means the number of outcomes in the study population is less than what you would expect to see in the standard population. In other words, the study population is in general more healthy than the standard population sample interpretations: Example: SMR = 200% The study population has a mortality rate 2x or 100% higher than that which you would expect in the general population Example: SMR = 150% The study population is dying at a rate 1.5x or 50% higher than the general population, slide 44

characteristics of time- secular trends

Secular trends = Changes in frequency and patterns of disease over long periods of time (10+ years) Trends may reflect changes in exposure Increase in lung cancer among women Follows increase in women smoking, especially in post-war U.S. (so, 1950s and after) Trends may also reflect long-term impact of public health programs, diet improvements, better treatment and detection programs Decrease in mortality from heart disease, 1950-2006 Progress made in medicine and health interventions - Framingham study contributed to our knowledge base

occupational cohort study

Select a Study Population: Companies Unions for specific occupations Sources of data: Company employment records Retiree databases Worker compensation records Worker associations (e.g. unions) Interviews Personal identifiers and other demographic data Record linkage with SSI Death certificates Characterize exposure Length of employment Work history Check for synergy or synergism Information about potential confounders, including Medical history Alcohol use Smoking habits

characteristics of persons-SES

Social class variations in health observed formally and informally since beginnings of organized society Relationships fairly consistent: Low-income individuals typically suffer higher rates of morbidity, higher infant mortality, and have lower life expectancy than the wealthy See different patterns by type of disease, too See different types of illnesses by social status Example: mental illness Upper class neurotic (e.g. obsessive-compulsive disorder) Lower class psychotic (e.g. schizophrenia) See differences in behavioral risk by social status, too Typical SES measures include occupational or social position, educational attainment and income Gout = associated with excessive consumption of rich foods and drinks Plague = associated with unsanitary living conditions

environmental media- soil

Soil Contamination Contamination from pesticides, petroleum hydrocarbons, solvents, lead, chromium VI due to industrialization and chemical usage Contamination can arise from many sources Of particular concern in residential areas including schools, parks, backyards

testing for confounding

Some Epidemiologists like to test for confounding Reasonable rule of thumb is when crude and adjusted measures differ by more than 10% : Calculate the crude measure of association (e.g. OR, RR) Stratify the analysis based on potential confounders Calculate an adjusted measure of association Chance in the estimate of effect by more than 10% suggests the influence of a confounder slide: 55 module 10

exposure pathway model

Source = Type , strength and amount of the chemical, physical or biologic agent Concentration= transport in the environment such as air, water, soil, food and any transformation process Exposure = route into the body (e.g., inhalation, ingestion, dermal) Dose = how the amount gets modified once in the body before reaching the target organ Host factors = includes age, weight, gender, genetics, etc. slide 10 module12

experimental studies-design

Subject selection Study population is defined a priori In consideration of research hypothesis Restrictions are used Inclusion criteria Exclusion criteria No subjective decision making! Randomly assign people to two groups (minimum): Intervention group: exposed to intervention Control group: not exposed to intervention Placebo or other intervention More on Randomization Occurs after informed consent is provided (more on this later in this lecture) Usually done on large number of subjects This tends to distributes confounders (age, sex, etc.) evenly among treatment and control groups To be sure confounders are evenly distributed, use Blocking Blocking example: Enroll 400 individuals ages 30-49, of both sexes, who are white or African American Block on age groups 50% 30-39 years of age 50% 40-49 years of age Block on gender 50% Male 50% Females Block on race 50% White 50% African American Outcomes Clinical end points (e.g., disease, recovery, death) Clearly defined before the study starts Measured the same way in both intervention and control groups To assess results, compare rates Rate of disease Rate of recovery Death rates Or, other outcome of interest Data Collection Intervention trials have extensive data collection at both baseline (when study starts) and during the trial Many questions need to be resolved before a trial starts What type of data should be collected? Who collects the data? When/how often are data collected? Who monitors data collection? Potential Problems Strict inclusion/exclusion criteria The study population may have limited generalizability Observer Bias If subjects' group assignment is known, they may be treated differently Non-compliance and drop outs People may not stick to group assigned to People may withdraw from the trial Blinding (Masking) Prevents subjects and study personnel from knowing who is in treatment and control groups Single-blind: only study subject is not aware of group assignment Does not protect against observer bias Double-blind: neither study subject nor experimenter/observer are aware of group assignment Best protection against observer bias Studies may also be Triple-blind, where the person assigning the drug doesn't know. For various practical reasons, this may not always be possible. Strategies to maintain compliance with the intervention Home visits Payment at time of visit Telephone reminders Calendar pill packs Daily logs Pre-study compliance checks Pill-counting Interview living companions Laboratory studies Document reasons for non-compliance

information bias

Systematic differences in data collected on outcome/or exposure between different groups in the study Examples of Information Bias Recall bias When assessment relies on subjects' memories, cases may search for an explanation for their disease and may be more likely to report or misreport information than controls. Example: Mothers of children with childhood leukemia may recall more detail (accurately or not) about what they ate and drank while pregnant than mothers of healthy children. Result: frequency of dietary exposures much higher for cases than controls Interviewer bias When an interviewer has knowledge of a subjects' disease status, he/she may ask cases certain questions differentially or in a different manner than of the comparison group. Examples: An interviewer gives leukemia cases more assistance in recalling possible environmental exposures than healthy controls. Result: frequency of environmental exposure higher than in controls Observer bias When an observer has knowledge of a participants exposure status, they may follow a participant who is exposed more closely and report more detail than a participant who is known to be unexposed RCT example: Observer records more detail about side effects in women assigned to new IV-based treatment than women assigned to older oral tablet. Ways to reduce information bias Establish strict data collection guidelines Train study personnel in data collection Reduce recall bias by: Use memory aids and validate exposures Mask/blind participants as to study goals and their status Reduce interviewer/observer bias by: Randomly allocate interviewer/observer assignments Mask/blind interviewers/observers to subject status

selection bias

Systematic differences in the procedures used to select study subjects or from factors that influence participation. Examples of Selection Bias: Participation bias People who participate are different than those who refuse to participate. Often a matter of self-selection. Examples: Study volunteers may be different from those who are enlisted. People who get screened may be different from those who don't get screened. Exclusion bias Occurs when eligibility criteria for groups differs Not due to self-selection by participants, but rather an error in the study design Example: Criteria for cases is different than criteria for controls Some characteristics or conditions may be acceptable in cases but not in controls. Loss to follow-up People who are lost-to-follow-up or who withdraw are different from those who continue Examples: In a cohort study to look at smoking exposure and lung cancer, people who develop cancer may be more likely to leave the study. In a clinical trial, people who feel the study drug isn't helping may stop participating in the study. Ways to reduce selection bias Clearly define subjects at the start of the study Strive for high participation rates and minimize loss to follow-up Effective case ascertainment* Careful thought to selection of controls*

cohort studies-advantages/ disadvantages

The main advantages of cohort designs are Address temporality issues Minimizes recall bias Good for studying rare exposures Can study more than one disease (outcome) at the same time Quantify incidence of disease The main disadvantages of cohort designs are: Time consuming Loss to follow-up a problem- people may drop out of the study over time. Not good for studying rare diseases Can only study one exposure at a time Expensive!

experimental designs- advantages /disadvantages

The main advantages of experimental designs, and RCTs in particular are: Randomization makes it more likely that the study groups will be similar with respect to risk factors Investigators have control over the way the study is done, including the amount of "exposure" subjects receive Best control of bias and confounding The main disadvantages or limitations are: Artificial setting for administration and evaluation of treatment May be difficult to keep people in the study, especially if the treatment has negative side effects With-holding a potentially beneficial treatment presents ethical issues Very expensive!

environmental epidemiology

The study of how physical, biologic, and chemical factors in the environment affect human health Key factors: Chemical agents Heavy metals Non-Ionizing radiation Ionizing radiation Pollution Environmental and workplace hazards are associated with a wide range of heath effects, e.g.: Cancers Lung disease Reproductive problems and birth defects Dermatological problems Injuries and trauma Neurotoxicity

occupational epidemiology

The study of how workplace factors affect human health Key factors: Chemical and biological exposures Stress and other psychosocial aspects of work Physical plant/layout of work stations Safety procedures Building ventilation

food-borne outbreaks

The vast majority of outbreaks are food-borne Most undiagnosed and unreported CDC estimates: 76 million people per year in U.S. ill Results in 5,000 deaths per year Food-borne outbreak = An incident in which (a) two or more persons experience a similar illness after ingestion of a common food, and (b) epidemiologic analysis implicates the food as the source of the illness Types of Food-borne Contamination Physical Glass, metal fragments, tacks, dirt, bone, etc. Chemical Pesticides, cleaning compounds, poisonous metals, additives and preservatives Biological Bacteria, viruses, fungi, yeast, molds, parasites, poisonous fish and plants, insect and rodents Major Reasons for Food-borne Illness Foods are cooled or (re)heated improperly Improper holding temperatures Food handlers are infected Poor hygiene Cross-contamination Food-borne Illness Incubation Periods Intoxications - shorter incubation periods, upper gastrointestinal symptoms E.g.: Staphylococcus aureus ½ to 8 hours, usually 2 to 4 hours Causes severe nausea, vomiting, cramps, often diarrhea No significant fever; body temperature often drops Infections - longer incubation periods, typically lower gastrointestinal symptoms E.g.: Salmonella (several serotypes) 6 to 72 hours, usually 12 to 36 hours Causes gastroenteritis with cramping, diarrhea, abdominal tenderness, vomiting, and fever The diarrhea is usually watery, but may contain blood or mucus

types of intervention

Two main trial types: Prophylactic trials - test interventions designed to prevent disease Therapeutic trials - test interventions designed to reduce or cure disease In addition, interventions may be tested again for: New uses for old drugs New doses New populations Equivalency for generic formulas

the odds ratio

Two types of Odds Ratios: Exposure Odds Ratio = (a/c)/(b/d) Odds that case was exposed (a/c) Odds that control was exposed (b/d) Calculated in case- control studies Disease Odds Ratio = (a/b)/(c/d) Odds that exposed developed disease (a/b) Odds that non exposed developed disease (c/d) Sometimes calculated in other study designs Both reduce to the same key formula: (a*d)/(b*c) Commonly called the Odds Ratio (OR)

observational research

Used when experiment impractical or unethical Investigators observe the natural exposure-disease relationship Key factors NO manipulation of study factor NO randomization of study subjects Relies on careful measurement of exposure-disease patterns to draw inferences about etiology Examples: Case Report, Ecologic Study, Cross-sectional study, Case Control Study, Nested Case Control, Cohort Study

syngery or synergism

When the combined effects of the exposures are greater than the sum of their parts. Example: The relative risk of lung cancer has already been calculated for our hypothetical cohort and we want to compare the risk of those with asbestos and/or smoking exposure: Risk among those with asbestos exposure alone: (2/1) -1*100= 10% Risk from smoking alone: (3/1) -1*100 =20% Risk from smoking combined with asbestos exposure: (7/1) -1*100 =60%

healthy worker effect (HWE)

Workers usually exhibit a lower overall death rate than the general population. Because the severely ill or chronically disabled leave the workforce. Only the most healthy are left. Death rates in the general population include the sick and disabled. So, we expect a SMR of less than 100% in working populations. This creates a special problem when try to evaluate workers using an SMR This problem is called "the Healthy Worker Effect" Epidemiologists use additional methods to evaluate these populations slide 46,47

synergism

- combined effects of exposure greater than sum of the parts, e.g.: Risk of cancer with asbestos and/or smoking exposure Risk from asbestos exposure alone =10% Risk from smoking alone =20% Risk from smoking combined with asbestos exposure =60%

chain of infection- portal of entry

- the route the agent uses to get into the new host Inhalation Ingestion Dermal Injection (blood borne) Mucous membranes Clinical Classification of Infectious Diseases slide 31 module 11

chain of infection

-how infectious diseases spread and the points where they may be contained Portal of Exit - path by which an agent leaves its human or animal source host Respiratory tract Urine Feces Conjunctiva Skin lesions Percutaneous Placental example- slide 20,21 22module 11

Steps in Scientific Research

1.Review literature, define the question, develop hypothesis 2.Design the study, collect data through observation or experimentation 3.Analyze data 4.Interpret data and draw conclusions (inference) Serve as a starting point for new hypothesis Or, as the basis for a comparative study 5. Publish results

epi triad- agent

A factor whose presence, excessive presence, or relative absence (if a deficiency disease) is essential for the occurrence of disease. In terms of infectious disease, the microorganism or microbial factor In terms of causal pies, it's a necessary cause Common to classify them by type- slide 5 module 11 Agents differ in: Infectivity vs. Pathogenicity vs. Virulence Also: Toxigenicity = produce a toxin or poison Resistance = survive adverse environmental conditions Antigenicity = induce antibody production in host Immunogenicity = produce specific immunity INFECTIVITY: Ability to enter and multiply in susceptible host and thus cause disease Measured by : Number of infected * 100 = % Number of susceptibles Example: 300/1000 = 0.3 * 100 = 30% PATHOGENICITY: Ability to cause clinically apparent disease in the infected Measured by: number with clinical disease *100 = % number of infected Example: 150/300 = 0.5 * 100 = 50% VIRULENCE : Ability to produce severe clinical disease or death in those with clinical disease Measured by: Number of severe cases * 100 = % Number with clinical disease Example: 15/150 = 0.1 * 100 = 10% Can also be measured by Case Fatality Rate CASE FATALITY: % of deaths caused by a disease among those who have clinical disease Measured by: CF(%) = # deaths due to disease * 100 = % # with clinical disease Example: CF= 5/150 * 100 = 3.3%

incident cases

Advantages: New subjects have better recall of exposure history Reduced likelihood that exposure has changed as a consequence of disease Disadvantages: Fewer cases likely to be available at start of study Matters most if you are collecting own data as opposed to using data from disease registry

characteristics of persons -age

Age is perhaps the most important factor to consider Greatest variation in specific disease rates Age related difference can be due to many factors, e.g.: Changes in immune system over lifespan Genetic conditions triggered by aging process Latency period between exposure and development of disease Differences in lifestyle or behavior Examples of patterns observed: Leading causes of death vary by age, e.g. Infants: pre-maturity Ages 1-14: Accidents (other than auto) Ages 15-24: Motor vehicle accidents Ages 25-65: Heart disease Ages 65 and older: Heart disease Certain diseases are more prominent in children Chicken pox, otitis media Certain diseases peak at more than one age Hodgkin's: mid-20s, early 70s Bimodal peaks may suggest two different causal mechanisms Population Pyramids Show age and sex distribution of a population Distribution reflects many factors including birth rate, leading causes of mortality, and state of medical care Useful for getting picture of a population at a particular point in time Shape of pyramid will change over time Populations are dynamic Economic and social conditions can change Population pyramids are formed by taking the age distribution of males and the age distribution of females, and joining them. Interpreting population pyramids: Triangular population distributions (e.g., Kenya) Decreased life expectancy Higher birth rate High childhood mortality Associated with less (economically) developed countries Rectangular population distributions (e.g., Italy) Increased life expectancy Lower birth rate Lower childhood mortality Associated with more (economically) developed countries

experimental studies

Also called Intervention studies Controlled clinical trials, randomized clinical trials (RCTs) Community interventions Goal: Test the efficacy of a treatment or other intervention in reducing or preventing disease through a planned experiment Multi-centered, randomized, controlled clinical trials are the most powerful type of epidemiological study design How it works: 1. Subjects enrolled on disease status All are healthy Or, all have a specific disease 2. Subjects assigned to intervention (treatment) or control Randomization is best method Researchers determine the dose and time frame 3. Monitor subjects to see if the intervention helps to prevent or reduce the severity of disease

nested case control studies

a case-control study conduced within the context of a cohort study In other words... select cases and controls from a cohort that is already being studied Advantages: Reduces cost Usually allows for the study of additional outcomes and additional exposures (data permitting) example: Step 1: Cohort Study is undertaken Hypothesis #1: HIV+ individuals with Hepatitis C have a higher risk of getting liver cancer than HIV+individuals without Hepatitis C Begin with a cohort of 10,000 HIV+ individuals who don't have liver cancer Test for Hepatitis C Classify people with Hepatitis C as exposed and people without it as unexposed Follow the group for 10 years to see who gets liver cancer Calculate RR to assess association of HIV+ Hep C with Liver cancer Step 2: Case-Control is conducted within the Cohort using additional data Hypothesis #2: HIV+ individuals with liver cancer are more likely to have taken AZT for 5+ years than those without liver cancer From the same cohort, select people who have liver cancer and match them to controls Go back and look at the medical records to see who used AZT for 5+ and who didn't Calculate the OR to assess association between AZT use and liver cancer

systematic errors

are non-random errors (also called biases) in the design, conduct or analysis of a study due to: Selection of study subjects Gathering information on exposure or disease Confounding effects of other variables Consequence: observed study results will tend to differ from true results Major problem in observational studies Less of a problem in experimental studies

population based controls

ensures the distribution of exposure among controls is representative of the target population selected by: 1. door to door recruitment Start in the case's neighborhood and systematically knock on doors to nearby houses in search of controls Disadvantages/problems: Expensive Time consuming People don't answer their doors Often a poor yield for the time involved Exposure history may be too similar for cases and controls 2. population/community lists Randomly select controls from a list of names/addresses from the same area E.g., Driver's license lists, Voter Registration, white pages, Medicare service lists, Lexis-Nexis Advantage: Some lists are inexpensive, but may be incomplete; better lists incorporating multiple sources cost a lot of money Disadvantage/problem: Lists are not all-inclusive of the population 3. random digit dialing Computer randomly generates last digits of a phone number based on the first three digits of case phone numbers Advantage: Random sample of the population Disadvantages/problems: People with caller ID might not answer Not everyone has a phone or a land-line anymore Expensive

infectious diseases

illness due to a specific microbial agent or its toxic products Arises through transmission of that agent or its products from infected person, animal, or reservoir to a susceptible host Some are contagious, e.g. transmitted person to person Many, but not all are reportable diseases Measles - reportable Head colds - not reportable See earlier lecture for information on notifiable diseases Must be a public health concern Required by law to be reported

famous cohort studies

iowa womens study Cohort = 41,837 women in study Began with random sample of women aged 55 to 69 (94% women in Iowa in this age range) in January 1985 Health questionnaire and record linkage Purpose of Research Look at mortality and cancer occurrence in older women in Iowa Primary hypothesis tests if distribution of body fat associated with increased risk of cancer Questionnaire included medical and reproductive history, personal & family history of cancer, usual dietary intake, smoking, exercise habits, medical use, weight, body measurements Records later linked to data in Iowa Cancer Registry (SEER participant) to check for incidence of cancer Nurses' Health Study Cohort =122,000 nurses residing in 11 states, 1976- Targeted due to motivation and knowledge Questionnaires and biological samples Purpose of research What are long term effects of oral contraceptives Smoking, hormone use, menopausal issues Study later expanded to look at diet and quality of life Women's Health initiative Cohort = 93,000 postmenopausal women between ages of 50-79, each followed for ~ 9 years, 1991- Health forms and clinical visits Purpose of research How much do known risk factors predict coronary heart disease, breast and colorectal cancer, osteoporotic fractures in women Identify new risk factors

ecological studies

measure correlations and trends using group data Unit of observation = group (e.g., a state, a town, a city) Typically use aggregate population data from readily available sources Exposure data often assigned geographically Typically, examines if exposure correlates with incidence or prevalence or mortality Examples: Are daily variations in mortality in Boston related to daily variations in particle air pollution? Are deaths from colorectal cancer associated with UV radiation? Ecological studies only have data on totals Disease totals (a+c)/(N) Exposure totals (a+b)/(N) But, no data on components such as # exposed cases Advantages Conducted at the group level, typically using readily available data Compare geographic areas Look at trends within one area Combination of the two May help identify high-risk populations for prevention efforts May generate hypotheses for analytic studies Disadvantages/Limitations Cannot illustrate dose-response relationship Data represent average exposure, not individual exposure Cannot calculate a true measure of association Have no real comparison group, esp. unexposed cases Cannot adequately control for potential confounders/bias Cannot infer causal association Temporality not possible to satisfy since data is not at the individual level Disadvantages/Limitations, cont'd Ecological fallacy "Term used when spatially aggregated data are analyzed and the results are assumed to apply to relationships at the individual level." (Everitt: Medical Statistics from A to Z) Example: Higher mortality from colorectal cancer is observed in geographic areas with lower levels of vitamin D effective UV radiation Do not know individual exposure levels which are highly variable due to skin photosensitivity and behavior, such as amount of time spent outdoors, sunscreen use, age, weight, etc. Also cannot account for confounding factors such as family history/genetics, diet, etc.

cross-sectional studies

measure prevalence and show relationships between risk factors and disease Also called Prevalence Studies Unit of observation = individual Simultaneously measure disease and exposure Who is sick? Who is not sick? Who is exposed? Who is not exposed? Single period of observation time Point or Period in time "Snapshot" of the population Examples: Ask students entering the student center if they study at least two hours a day and if they wear contacts Ask all teens in the high school if their parents are divorced and if they smoke Ask everyone in the supermarket if they eat red meat and if they have high blood pressure Advantages: Good method for identifying prevalence of common diseases Repeated cross-sectional surveys allow you to look at prevalence trends over time Describe magnitude of health problem Useful for prevention and planning purposes Generate hypothesis about various risk factors/exposures and disease outcome Follow-up with analytic study to test hypothesis for purposes of inferring causal association Cross-sectional studies have individual data 'N' may represent the total population or just a sample Know distribution of a, b, c, d Can calculate measures of association between disease and exposure (typically, an 'odds ratio') But due to limits of study design, interpret as 'correlates' not causal factors Disadvantages/Limitations: No temporality You can never infer causality from cross-sectional designs! Cannot calculate disease incidence, only prevalence Not directly interpretable as average risk of getting a disease Not good for studying rare or deadly diseases Under-represent people with short duration of disease Results are influenced by survival factors

randomized clinical trials

occur in phases Phase I Establish effects of new drug in humans using healthy volunteers Phase II Determine therapeutic efficacy in diseased group Phase III Test against other therapies or placebo Phase IV Test for other uses of the therapy Sample size in RCTs: Phase I and II studies are generally small, with a handful of people Phase III and Phase IV studies generally have a large sample size Sufficient sample size necessary to: Improve ability (statistical power) to detect a difference Reduce random error

factorial design

simultaneously test two or more interventions Use the same study population to test Drug A & Drug B at the same time Assume: The outcomes for each drug are different Modes of action are independent Advantage: If you need to terminate the study of Drug A, you can continue the study to determine the effects of Drug B instead of beginning a new study Example: Physician's Health Study Test aspirin as a way of preventing cardiovascular disease Test beta-carotene as a way of preventing cancer Aspirin arm terminated early due to a significant drop in the risk of first heart attack Beta-carotene arm continued to study completion example: slide 26 module 10

modes of transmission

slide 23 module 11 Direct Contact - spread through physical contact or droplet spread Physical contact - person-to-person contact such as kissing, skin to skin contact and sexual intercourse. Examples: HIV/AIDS; Chlamydia ; Genital warts; Gonorrhea ; Hepatitis B ; Syphilis; Herpes Droplet spread - spray by short range aerosol droplets onto another person through coughing sneezing, talking or singing Examples: Bacterial Meningitis ; Chickenpox Common cold ; Influenza ; Mumps; Strep throat ; Tuberculosis ; Measles ;Rubella; Whooping cough Indirect Contact: spread of infectious agent through intermediary source Vehicle borne - disease transmission through food, water, blood, tissues, organs, e.g. salmonellosis, cholera, Hepatitis A, Polio, Rotavirus (many through Fecal-Oral route) Fomites- disease transmission through inanimate objects such as bedding, toys, doorknobs, combs, clothing, drinking glasses, cooking utensils, pencils, straws, or surgical instruments, e.g. viruses or bacteria Vector borne - disease transmission by a living carrier, e.g. plague Vector Borne Transmission Agent is carried by a vector Vector = an insect or any other living carrier that is involved with transmission of disease Most vectors are arthropods (insects) such as mosquitoes, flies, ticks or fleas, but they may be animals Vector-borne transmission may be mechanical and/or biological examples slide 27 module 11 Mechanical Vector borne Transmission Transport of the vector through mouthparts, antennae, limbs There is no multiplication of the agent within the vector Example: The fly lands on Shigella-contaminated items, carries the agent on its legs, antennae, etc. to a new location Biological Vector borne Transmission Transmission of the infectious agent to susceptible host by bite of blood-feeding (arthropod) vector E.g.: malaria, schistosomiasis Agent may undergoes part of its life cycle within the vector, multiplying or undergoing physiologic change

determining exposure status

t's important to clearly define exposure: ...in other words, how much exposure makes one "exposed"? Example: If exposure is smoking, how is it defined? Non-smoker vs. Low vs. Medium vs. High Smoke > 100 vs. < 100 cigarettes in a lifetime Smoke cigarettes only, cigarettes or cigars, etc. Smoke at least 1 pack/day vs. less than 1 pack/day Ways to clearly define exposure: Divide cohort into groups based on level of exposure Exposure may be broken into levels Extreme levels typically of interest Specify way exposure is to be measured Self-reports (e.g. surveys) Diagnostic tests (e.g. blood pressure cuff) Environmental tests (e.g. water sampling)

exposure routes-ingestion

taken into the body through the digestive system Typically by swallowing dusts or liquids Absorbed at various points Deposited in organs such as liver Transported to others, such as kidney and bladder

exposure routes- ocular

taken into the body through the eye Eye is a membrane Substances can burn or irritate it Some can be absorbed through eye and enter the bloodstream

spectrum of disease

the progress of a disease with no intervention exposure--> pathological changes-->symptoms--> clinical illness--> recovery, disability or death Inapparent infection - has subclinical infection, does not yet have obvious symptoms but could transmit disease to other susceptible hosts Incubation period Time between infection and the appearance of first signs or symptoms of disease (onset of clinical illness) Latency period Time between infection and when the individual becomes infectious to others Infectious period Time during which the infectious agent may be shed (host need not be symptomatic) Example: Measles "It can take seven to eighteen days between the moment the Measles virus enters a person's body and the appearance of symptoms. A person is contagious from three to five days before symptoms appear to about four days after the rash shows up." What's the incubation period? What's the latency period? What's the infectious period?

crossover trial design

the same person serves in both the intervention and control groups Patients serve as their own "control" This design has often been used to test inhalers in people with asthma Randomization is to order in which the subject receives treatment/placebo (or other treatment) Subjects begin the study on Treatment A and later switch to Treatment B/Placebo Or, start with B and later switch to A Residual drug carryover can be a problem There must be Washout period between treatments or this design cannot be used example: slide 20 module 10 1. Enrollment Inclusion criteria: Dogs under 3 years of age who suffered from carsickness but were otherwise healthy Incentive was $200 in free vet care Dog met basic inclusion criteria and was allowed to participate Vet recorded his exact age, weight, breed, sex, and other factors prior to start of trial 2. The actual experiment: A. Dog owner (me) given two pre-sealed, numbered envelopes Each envelope contained a pill Treatment A, Treatment B Random order - possible that Treatment B would be taken before Treatment A One was test drug, one was placebo Vet and dog owner (and dog!) blind to order Packets only labeled "1" and "2" Prepared by lab worker not connected with subject selection/evaluation B. Owner asked to take two rides with dog one week apart, that end up at animal hospital where Vet works Same car, same driver, no passengers, same route, same day of week, same time of day, same length of time (60 minutes hour) Dog given no food for 12 hours prior; no water for 2 hours prior. Given one pill exactly 1.5 hours prior to each trip C. Dog checked for car-sickness at end of 60-minute trip by animal hospital staff and by owner report 3. Results compared with other dogs participating in study

chemical agents

vapors, fumes, liquids Pesticides (workplace exposure, residues in food or water) Organochlorides (DDT) and breast cancer Organophosphates and sterility Also linked with depression, cognitive deficiencies, birth defects Vinyl Chloride (plastics industry) Angiosarcoma of liver Benzene (gas station workers) Leukemia

steps in significance testing

1. Set the null and alternative hypotheses Ho: null hypothesis: There is no effect or no difference Ha: research (alternative) hypothesis: There is an effect or difference 2. Collect the data you wish to test 3.Set your significance level, e.g. 5% (look up the critical value(s) in the appropriate table or use statistical software) 4. Calculate the p-value for your data using appropriate test and compare it 5. Based on results, reject or fail to reject null hypothesis

cases-selection

1. define the cases Definition of disease List diagnostic criteria Severity of disease Subjective vs. objective criteria Symptoms vs. lab tests/values Patient vs. doctor 2. identify the cases ideally, all cases have an equal probability of entering the study and no false cases enter dont need every case for study to be valid cases can come from: clinics, hospitals or private medical practices, screening events, disease registries, vital statistics data bases

confounding

A distortion in the estimate of the effect (e.g., RR, OR) due to the mixing of effects by other extraneous factor(s) Depending on the relationship, it results in either an overestimation or underestimation of the true association between exposure and disease To be a confounder, a factor: 1. Must be a risk factor for disease 2. Must be associated with exposure 3. Can not be an intermediate step in the causal path between exposure and disease In other words, a confounder has this relationship with exposure and disease: Example: Study testing if hypertension causes End Stage Renal Disease (ESRD) (van Stralen et al. 2010) Could Obesity be a Confounder? Is it associated with exposure? Obese people are more likely to have hypertension Is it associated with disease? Obesity is a known risk factor for ESRD Is it on the causal pathway? No, hypertension does not cause obesity

limits of significance testing

A significant statistical association does not show cause It just means your results probably didn't happen by chance alone A non-significant association can be due to reasons other than chance Having too few people in your study can result in a lack of statistical power Unable to reject the null even if the null is false Decision making is ingrained in statistical testing, yet policy decisions or causality never inferred from just one epi study Every time we reject the null hypothesis we risk being wrong This is called a Type 1 error Every time we fail to reject the null hypothesis we risk being wrong This is called a Type 2 error

prevalent cases

Advantages: More cases likely to be available at the start of the study Disadvantages: People who die quickly with disease are under-represented in the study population Older cases may have "worse" memory than new cases Exposure may have

environmental media- air

Air Pollution Contamination by hazardous pollutants such as Criteria Pollutants, Hazardous Air Pollutants and other sources, including environmental tobacco smoke Some are associated with increased morbidity and mortality from: Respiratory disease Lung cancer Cardiovascular incidents Differences between Indoor vs. outdoor sources, and seasonal trends

biological agents

Allergens, including Mold Provoke allergic reaction in susceptibles Ranges from dermatitis to acute systemic shock Allergens are a huge burden to the healthcare system Common allergens include fur, dust mites, pollen, etc. Mold is an increasing health problem Can cause serious respiratory illness in homeowners

outbreak investigations

An outbreak occurs when there are more cases than you'd expect in a given area or among a given population in a specific period of time Not defined by an absolute number Relative to background levels

heavy metals and metallic agents

Arsenic (soil, water, wood preservation & manufacturing, mining and smelting metals) Varies in toxicity depending on chemical form Linked to skin, bladder, kidney and liver cancers Studies suggest association between arsenic in drinking water and bladder cancer E.g., Dry milk and soy sauce in 1950s Japan - 1000s of cases Mercury (Fish, dental fillings, historical medical treatment) Associated with neurological conditions including numbness in extremities, deafness, poor vision, drowsiness, death Example: Minamata Bay, Japan : about 3000 cases from ingestion of fish containing methylmercury, which bio-accumulates. Minamata disease- first identified in 1956 (see illustration at left). Mercury in industrial wastewater from Chisso chemical company, released 1930s to 1960s Lead (Once widely used in paint and gasoline) Central nervous system effects Adverse effect on intelligence, behavior, development E.g., As children's blood lead levels increase, children's IQ decrease Low Frequency, non-ionizing Radiation Electric and Magnetic Fields (EMF) Induces weak electric currents in body Does not appear to be genotoxic but may influence cellular growth/proliferation May be associated with Acute non-lymphocytic leukemia in children

minerals and dusts

Asbestos (shipyards, insulation, and older brake linings) Mesothelioma and asbestosis Silica dust (sandblasters, miners) Silicosis Coal dust (miners) Black Lung (pneumoconiosis)

characteristics of place- international comparisons

Both infectious and chronic diseases show differences by country Differences in standards of living play major role Differences in climate/environment Differences in cultural factors including attitudes towards smoking and national dietary practices At present: More developed countries have the high rates of deaths from diabetes, hypertension, heart disease, stroke Less developed countries have the highest rates of deaths from infectious diseases Some examples: Highest rates of lung cancer are in Eastern European countries; Overall rates in Britain and US are declining (associated with number of smokers) Japan has highest rate of stomach cancer in developed world (associated with consumption of preserved, cured or salted foods) Belarus, Lithuania and Russia are among countries with highest suicide rates (associated with depression in males, economy, culture) Countries in tropical Africa account for more than 80% of all clinical cases of parasitic infections Many of these countries have extremely high infant mortality

etiologic fraction

Calculated as the proportion of disease in the exposed that could be prevented or reduced by blocking a particular exposure, after accounting for other pathways to the disease. These other pathways are estimated from the incidence in the nonexposed (e.g., non-smokers), which represents the amount due to other causal mechanisms. Example: How much stomach cancer could be prevented in smokers, accounting for other pathways they could contract the disease? (e.g., Helicobacter pylori and diet have all been independently linked to stomach cancer) Etiologic Fraction Key formula: Difference between the incidence in exposed and the incidence in nonexposed, divided by incidence in exposed (Iexposed-Inonexposed) / Iexposed Can also be calculated other ways that are mathematically equivalent: RD/Ie Multiply answer by 100 to get % RR -1 / RR Multiply answer by 100 to get % example slide 14 module 9 Interpretation: In this study, approximately 35% of the incidence of stomach cancer in subjects who smoke could be prevented or reduced if the subjects did not smoke That means 65% of the incidence of stomach cancer among smokers can not be attributed to smoking exposure but must be due instead to other causal mechanisms without this particular component factor. In other words, we can not prevent or reduce 65% of cases among smokers by eliminating smoking exposure alone.

outcome measures

Case control studies have individual data Start by selecting Cases (a+c) and Controls (b+d) Determine distribution of a,b,c,d

hospital controls

Cases and controls may be selected from hospital populations Controls may come from any department in the hospital and may be in-patients or out-patients Advantages: Patients may be more likely to participate than people from the general population Disadvantages/serious problems: What diagnosis should the controls have? People in the hospital are sicker than the general population

characteristics of persons- religion

Certain religions stipulate proscribed lifestyles or practices that have health impacts Examples: Seventh-Day Adventists have low rates of CHD Lacto-vegetarian diet, abstain from alcohol and tobacco products Mormons have lower incidence and mortality rates of cancer Restrict coffee, tea, and meat consumption Promote exercise, social support, and stress-reducing ideology

descriptive epidemiology

Characterize amount and distribution of disease in a population* Evaluate health trends and make comparisons Provide a foundation for planning, provision and evaluation of health services Generate hypotheses to be studied by analytic methods *This kind of data is typically gathered using descriptive study designs, such as prevalence (cross-sectional) as well as ecologic study designs.

cohort studies-data and outcome measures

Cohort studies have individual data Start by selecting population for cohort N Identify exposed (a+b) and unexposed (c+d) May add up to total N or represent a sample Follow for development of disease From cohort data we can calculate Incidence in exposed Incidence in unexposed Depending on the type of incidence data available, we can calculate a Risk Ratio or a Rate Ratio Incidence Proportion (risk) data gives us Risk Ratios Incidence denominator is population at risk Incidence Rate data gives us Rate Ratios Incidence denominator is person-time Both Risk Ratios and Rate Ratios share the same general formula* rr=formula slide 26 module 9 *Just keep in mind that with Incidence Rate data, the denominators (a+b) and (c+d) represents the person-time of the study population

recommend control measures

Control the present outbreak Prevent future similar outbreaks May include: Sick: isolation/quarantine, treatment Well: immunization, sanitary measures

characteristics of time- cyclic variation

Cyclic variation = increases and decreases in frequency of disease over period of years or within a year (seasonal) Example: pneumonia and influenza Show both annual peaks and epidemics every few years Increase and decrease could be due to changes in host, seasonal climatic changes, virulence of infectious agent Cyclic variations seen in infectious as well as chronic diseases Seasonal variation Disease peaks in one season, hits a low point in another Same pattern observed year after year Peak and dip usually observed 6 months apart Examples of seasonal trends Flu peaks in February Meningococcal disease peaks in winter, declines in summer Heart disease peaks in winter Birth rates increase in summer Cohort effects and clustering may suggest a relationship between a common exposure to an etiologic agent and development of morbidity and mortality Cohort effects = long term secular trends Example: Thyroid cancer in teenagers in Belarus, who were born shortly after Chernobyl accident Leukemia after atomic bomb dropped on Hiroshima and Nagasaki Clustering = unusual aggregation of health events in time and place Examples: John Snow and cholera in Broad Street pump area Liver cancer among workers exposed to vinyl chloride

exposure routes- dermal

Dermal: taken into the body through the skin Contact typically due to spillage or immersion into liquids or dusts Skin is protective barrier, but some substances can penetrate the skin Substance may burn, irritate, or change the skin Skin varies in terms of absorption rates Depends on thickness and other qualities Least absorbent Forearm Palm of hand Ball of foot Most absorbent Genital area Ear canal Forehead Scalp

place- urban vs rural comparisons

Differences in urban vs. rural areas Could be related to diet, occupation, types and levels of physical activities, environmental exposures, lifestyle Not always predictable, but in general: Urban areas Higher rates for diseases transmitted by person-to-person contact Higher rates of tuberculosis, cirrhosis of the liver, and heart disease than rural areas Higher rates of violent deaths Rural areas Higher rates of cancers of the lip (male/females) and eye (males) Higher rates of mortality from heart disease and COPD Higher rates of total tooth loss

measuring disease

Disease endpoint(s) must be the same for exposed and unexposed Endpoints must be well-defined and unambiguous Example: lung cancer = local disease? metastasized disease? asthma= mild? severe? If multiple endpoints are used, each must be clearly defined Use objective measures instead of subjective whenever possible Observer bias can also play a role in cohort studies, too!

levels of disease

Endemic - background cases; expected number of cases Epidemic - excess number of cases in a localized area Pandemic - excess number of cases worldwide Often occurs because... Entire populations are susceptible There are no effective treatment or control measures

random error

Error that we cannot predict 3 major contributors Lack of Precision - not enough sample size to get clear result Sampling error - due to variability in sampling of study subjects Variability in measurement - lack of consistency in measurement Way to reduce Random error: Increasing sample size Increases both precision and reduce sampling error Increase number of measurements Better captures variation Improve the reliability and objectivity of data collection Reliability = reproducibility Use objective not subjective (e.g., relying on human memory or observation) measures

verify the outbreak

Establish a case definition Who (person) What (symptoms) Where (place) When (time) Identify, confirm, and quantify cases Determine if there are more cases than expected

analytic designs

Examine associations between exposure and disease Look at specific causal relationships Account for temporality (in varying degrees) Test associations between risk factors and disease for the purposes of causal inference

prospective cohort study

Exposure determined at baseline Follow cohort forward in time from the start of the study and document who gets sick Example: Follow healthy RU smokers and non-smokers for the next 20 years to see who gets asthma

historical prospective

Exposure occurred in the past Follow-up for disease occurrence from past to the present and then into the future Example: Look at students who entered RU 10 years ago (Fall 1996) and see who smoked and who didn't at that time. Look at each group to see who got asthma between then and now. Keep following them for the next 10 years to see who else in the cohort gets asthma.

frequency matching

For every case in a specific category, a control is selected in the same category Examples: For every case in the 20-25 year age group, select a control in the 20-25 year age group For every case that is a "heavy smoker," select a control that is a "heavy smoker"

dead cases and controls

If a case died from a specific cause, then controls may have died from another cause Advantages: You can use death certificate data to get a lot of information Disadvantages/serious problems: May not have same background exposure as live controls How do you assess exposure in dead people?

clinical vs statistical significance

In epi, we also consider the clinical significance in conjunction with the statistical significance Clinical significance - describes the ability of a factor to impact health Questions to consider: What is the benefit or harm to people? Does the finding have a practical application? Is there a clinically relevant benefit or harm? Are we sure the benefit is not outweighed by something negative?

types of cases

Index Case - first person that is sick Primary Case(s) - people who get sick from being directly exposed to the agent Measured by the attack rate Secondary Case(s) - people who get sick from being exposed to a primary case Measured by the secondary attack rate

occupational or environmental cohort study

Key problem: selection of a comparison population for calculation of the relative risk Difficult to compare within the cohort as everyone may be exposed Sometimes external controls are used, e.g., workers from another company without occupational exposure Most often compared to rate in general population Calculation of a Standardized Mortality Ratio or SMR Lets you compare a study population to an overall standard How much do they differ from the population standard? Are they more or less healthy? The method used to calculate the SMR is called "indirect standardization" and should be compared to "direct standardization" discussed in Lecture 4: Working with Rates

place-local comparisons

Local outbreaks or elevation in rates may be due to unique environmental or social conditions Examples: Goiter common in landlocked areas of U.S. where seafood was not consumed High radon exposure in Ohio communities Fluorosis common in areas with high fluoride content in the water

calculate SMR

Locate the total number of deaths observed in your study population. Calculate an "expected" number of events (e.g., deaths) for your study population Use the population structure of your data And apply the rate from the standard population easily obtained from national mortality data, e.g. from "U.S. Vital Statistics" Compare the observed total events (e.g., deaths) with the expected number to get an "Standardize Mortality Ratio" SMR = Observed deaths/Expected deaths * 100 Answer is a ratio Expressed as a percentage % example: slide 40,41

food-borne illnesses- bacterial requirements

Most bacteria have an optimal temperature range that allows for growth, ranging from 40°F - 140°F To avoid bacterial growth: Keep refrigerated foods below 40°F Maintain hot cooked food at 140°F or higher Reheat cooked food to at least 165°F High protein/moist food provide a potentially hazardous environment for bacterial growth: Milk or milk products Eggs Meat, poultry, fish, shellfish, crustaceans Raw seed sprouts Heat treated vegetables and vegetable products Some fruits slide 27

interpreting odds ratio

OR > 1.0 Increased risk of exposure in cases Disease is positively associated with an exposure, which is typically interpreted as a risk factor for disease OR = 1.0 No difference in exposure between cases and controls Disease is not associated with exposure OR < 1.0 Decreased risk of exposure Disease is negatively associated with an exposure, which is typically interpreted as protective for disease example: OR = 1.69 Interpretation: People with the disease are 1.69 times (69%)* as likely to have been exposed relative to those without the disease. **Always subtract the null value of 1 before multiplying by 100 to get the percentage. 1.69-1 = 0.69 * 100= 69%

cohort studies

Overview Cohort studies are an observational, analytic design that seek to identify cause(s) of disease by studying how exposed people differ from unexposed people with respect to incidence of disease At Baseline Exposure status is known for all subjects All subjects are free of the disease of interest How it works: 1. Subjects selected based on exposure status 2. Follow subjects over a set period of time and document development of disease 3. Compare incidence of disease in exposed to incidence of disease in unexposed Cohort = group of people followed over time Good method for studying Rare exposures Diseases with short latency periods 3 subtypes Prospective cohort Retrospective cohort Historical cohort

chimney sweeps and scrotal cancer

Percivall Pott (1714-1788) 1775: described the occurrence of cancer of the scrotum "sootwart" in a number of his male patients, whose common history included employment as chimney sweeps when they were young. Concluded that their prolonged exposure to soot was the cause (latency~20 years). Impacted diagnosis and treatment, inspired some minor reforms in child labor practices in England

interpretation of risk difference

RD>0 Exposure increases risk of disease Exposure is a risk factor for disease RD=0 (this is the NULL value) No association between exposure and disease Exposure is not a risk factor for disease RD<0 Exposure decreases risk of disease Exposure is protective for disease RD = 0.06 *100 = 6% Interpretation: "Subjects with smoking exposure have an incidence of stomach cancer that is 6% higher than subjects without smoking exposure

interpretation of RR

RR > 1.0 Increased risk of disease in exposed Exposure is a risk factor for disease RR = 1.0 (This is the NULL value) Risk of disease is the same for both exposed and unexposed Exposure is not a risk factor for disease RR < 1.0 Decreased risk of disease in exposed Exposure is protective for disease example: RR = 1.57 Interpretation: Subjects with exposure are 1.57 times as likely to develop disease than subjects without exposure. Or, subjects have 57% greater risk of developing disease than subjects without exposure. The example on the previous slides uses incidence proportion or cumulative incidence data If we had incidence rate data instead, the RR would be calculated in units of person-time Person-time data has the same general interpretation provided that the rate~=risk Rate ~ = Risk when time period is short (e.g. yearly) and incidence is fairly constant, and there is no loss to follow-up (e.g., people leaving the cohort)

risk ratio vs odds ratio

Recall that with case control studies we calculate the Odds Ratio It is not a ratios of risk For example, disease OR = (a/b)/(c/d) Usually do not have total population or person-time data in case control studies A ratio of two ratios derived from two odds Contrast this with a risk or rate ratio, calculated for cohort studies using incidence proportion or incidence rate data: RR = (a/a+b)/(c/c+d) A ratio of two proportions BUT, the Odds Ratio can approximate the relative risk (i.e., risk ratio) associated with a given exposure when: Controls are representative of target population Cases are representative of all cases Frequency of disease in the population is low When 'a' and 'c' are not frequent (rare - literally, a very small number compared with their denominator), then RR (a/a+b)/(c/c+d) ~= (a/b)/(c/d) = OR (ad)/(bc) This is called the Rare Disease Assumption slide 34 module 9

relative risk

Risk Ratios and Rate Ratios are more generally called the "Relative Risk" These measures are generally interpreted as the risk of developing disease in the exposed group as compared to those who were unexposed Used to infer a causal relationship between the exposure and the disease

The Scientific Method

Techniques for investigating phenomena, acquiring new knowledge, or correcting and integrating previous knowledge. Consists of the collection of data through observation and/or experimentation, and the formulation of hypotheses which are testable and open to falsification. Utilizes methods to minimize or eliminate bias in data collection and analysis. Requires one to document, archive and share all data and methodology to allow for reproducibility.

quantifying the benefits of prevention

The Risk Ratio (or Rate Ratio) is used to calculate the strength of association between the exposure and the outcome in etiologic research Data from the same 2x2 table can also be used to estimate the benefits of prevention using these measures: Risk or Rate Difference Etiologic Fraction

background: disease trends and the environment

The age-adjusted incidence of all forms of cancer increased 43.5% from 1950 to 1990 (the peak, for many cancers) These changing trends in pattern of disease could be related to any of these: New screening techniques New diagnostic tests "Lifestyle changes", including diet, smoking and exercise Changes in environmental and/or occupational factors that play a role in the etiology of disease

herd immunity

The decreased possibility of a group or community developing an epidemic because there is a specific level of immunity or resistance to that disease in the population Basic principle: the entire population does not have to be immunized to prevent the occurrence of an epidemic. Immunized persons (black circles) act as a barrier to spread to persons who are susceptible to the infection (white circles). In the illustration below, note how many white circles are touching each other. This indicates the disease can be spread person to person between susceptibles.

physical and mechanical energy

Trauma In home/workplace Noise Hearing loss/deafness Vibration Whole body or hand/arm Temperature Extreme cold or heat

retrospective cohort study

Use past exposure status to do the study "Follow-up" for disease development is done from time of exposure (past) until present (time of the study) Example: Look at students who entered RU 10 years ago (Fall 1996) and see who smoked and who didn't at that time. Look at each group to see who got asthma between then and now.

environmental media-water

Water Pollution Contamination by microorganisms, particulate matter, inorganic solvents, organic solutes, radionuclides, etc. Exposure by ingestion, dermal contact, inhalation of vapors Water-borne enteric (gastro-intestinal) illnesses are of particular concern, so various water treatments are used

case reports

describe individual occurrences of disease Describe new case(s) of disease Who - patient demographics What - symptoms When - time course of disease Where - location of affected person Only present data on individuals with disease; they do not include a comparison population or evaluate the exposure-disease association Are used to highlight unusual/new disease presentations Are often followed by a cross sectional or case control study, which is used to conduct the next stage of an investigation

threshold case

lowest dose at which particular response may occur, not all substances have a threshold dose

dose response curve

models relationships between changes in dose and changes in health outcome

latency

time period between initial exposure and measurable response (e.g., symptoms, disease) Long latency periods are problematic for measuring exposure