Epidemiology Final Exam Study Guide

descriptive epidemiology - epidemiologic studies concerned with characterizing the amount and distribution of health and disease within a population WHO is getting sick? (ex sex, age, race/ethnicity) WHERE are they getting sick? (ex country, rural/urban, near factory) WHEN are they getting sick? (over a day, year, decade) descriptive study designs - cross-sectional, case reports, case series, ecologic analytic epidemiology - examines causal hypotheses regarding the association between exposures and health conditions WHAT is making them sick? (ex virus, pollution, radiation, stress) observational analytic study designs - ecologic, case-control, cohort experimental analytic study designs - clinical trials, community intervention case reports/case series - individual-level observations; describe a particular clinical phenomenon in a single patient (case report) or a group of patients with similar problems (case series); goal to provide a comprehensive and detailed description of case(s) under observation; observational study ecologic studies - looks at group level data or population level data; often country level ecological fallacy - assuming that individuals within those populations follow the same trend that is seen in population-level data cross-sectional studies - exposure and disease outcome are determined simultaneously in each subject/patient "Prevalence study" or "snapshot in time" serial cross-sectional studies - trends in disease prevalence over time; surveying different individuals in same general group case-control studies - start with cases (those with disease), and controls (those without disease), look backward for past exposures (through interviews, medical records, etc) Start with disease status, look backward for exposure Cannot calculate prevalence or risk Good for rare diseases uses Odds Ratio biases in case-control studies - Selection bias: sources of cases (from one place may not be able to generalize), using incidence (wait for diagnosis and may die before diagnosis) or prevalence cases (risk factors related to survival instead of development of disease), selection of controls (best friend control, patients in hospital too similar) Information bias: can't remember information, lead to misclassification of exposure (put in wrong exposure group) = will bias results towards null finding Recall bias: one group systematically has better recollection of exposures than other group (usually case group) group matching in case-control studies - proportion of controls with a certain characteristic is identical to the proportion of cases with the same characteristic individual matching in case-control studies - for every individual case, a control is selected that matches the case in terms of all desirable variables (sex, age, etc.) = matched pairs Odds Ratio - Measure of association between frequency of exposure and frequency of outcome (exposed cases/not exposed cases)/(exposed controls/not exposed controls) OR 1 - positive association between exposure and disease OR = 2 - odds of disease are about 2x higher among exposed than among non exposed OR 1 - exposure may be protective factor OR = 1 - no association between exposure and outcome cohort studies - start with exposure status (exposed vs not exposed) and look for disease outcome (disease develops vs disease does not develop) Start with exposure, look forward to disease outcome Determines incidence/risk Good for rare exposures; uses relative risk ratio prospective cohort study - Start with one group of people with a current exposure and one similar group of people without the exposure Follow them forward in time to see who develops the disease retrospective cohort study - Start with one group of people with a past exposure and one similar group of people without the past exposure Follow them forward in time (often to present) to see who developed the disease We are still looking forward in time after the exposure longitudinal study - Start with one defined population (no exposure yet) Follow them to see who gets exposure Continue following to see who gets the disease Ex: framingham study Where we obtained most of our knowledge about coronary heart disease Relative Risk (RR) ratio - ratio of incidence rate of a disease or health outcome in an exposed group to the incidence rate of the disease or condition in the non exposed group ((exposed, gets disease) / (exposed, gets disease) + (exposed, doesn't get disease)) / (not exposed and gets disease/ (not exposed and gets disease + not exposed doesn't get disease) Incidence of disease in exposed/incidence of disease in non-exposed RR = 1 - risk of disease among exposed is same as risk of disease among non exposed RR 2 - risk is more than 2x as high among exposed compared to nonexposed (positive association between exposure and outcome) RR 1 - exposure has protective effect and exposed group is less at risk than nonexposed group attributable risk - difference between incidence rate of a disease in exposed group and incidence rate in non-exposed group population risk - difference between incidence in total population and incidence in non-exposed segment potential biases in cohort studies - selection bias (One group more likely to refuse to join, one group more likely to be lost to follow-up) information bias (Quality differs between groups, Researcher who assigns people knows exposure history, Researcher has strong preconceptions) experimental studies - start with study population, then randomly assign to treatment or control, look to see who improves, who does not. has an intervention and often has randomization experimental study subject selection - selection criteria must be completed before starting the subject, must be in writing, must be precisely described (anyone picking up criteria should be able to know exactly who may be enrolled) experimental study treatment groups - First group: receives intervention (can be called treatment, experimental or intervention group) Second group: does not receive intervention (can be called control or comparison group) Either receives no treatment, current treatment or standard of care, or different treatment experimental study possible control groups - historical controls, simultaneous nonrandomized controls, simultaneous randomized controls historical controls - Use records of patients with same disease who were treated before availability of new therapy useful for fatal diseases when a new drug becomes available Problems: lack of meticulous study protocols for data collection, medical records may be incomplete/have errors, differences could be from data collection differences, secular changes simultaneous non-randomized controls - Use simultaneous controls that are not selected in a random manner Problems: groups may be very different (sicker people as treatment group), Predictable assignment system simultaneous randomized controls - Use simultaneous controls that are selected in a random manner; Best approach Done through computer, table of random numbers Method used must be spelled out in writing before randomization begins Potential problems: physician conflict randomization - process whereby chance determines the subjects likelihood of assignment to either the intervention or control group; goals: non-predictability of next assignment, removal of biases in assigning participants to group, increases likelihood that groups will be comparable in relevant characteristics stratified random sampling - an option to assume comparability on certain characteristics; Stratify study population by each variable that is considered important, and then randomize participants to treatment groups within each stratum experimental study data collection - treatment (treatment group assigned, treatment actually received), outcomes (improvements, side effects, public health outcome) single blind masking - participant does not know which group they are in; Knowledge of receiving treatment may make subject believe they are improving or experiencing side effects double blind masking - neither participant nor researcher knows which group participants are in; Observer bias: researchers expectations affect how they perceive behaviors and take measurements and record results planned crossover design - participants are randomized to intervention/control groups, intervention is administered to intervention group, results are taken, participants switch groups, intervention is administered to new intervention group, and results are taken participants serve as own controls problems: carryover (residual effect of first treatment; can be fixed by washout period), order of treatments; impossible for surgical interventions or interventions that cure disease unplanned crossover design - participants start out in one group but switch to the other May change their minds if blinding is not possible Analyze their outcomes by either intention to treat or as treated (usually intention to) factorial design - using one study population to test two different interventions Allows researchers to study effects of two or more interventions, as well as their interactions, in one study noncompliance - when participants don't comply with the assigned treatment; net effect of noncompliance will drive the difference in results towards null overt noncompliance - participant makes it known they are not complying, drops out of study covert noncompliance - participants stop taking assigned agent without making it known, drop ins- controls inadvertently take intervention four possible conclusions when testing whether or not treatments differ - if treatments do not differ in reality: may correctly conclude they do not differ, or in error may conclude that they do differ if treatments do differ in reality: may correctly conclude they do differ, or in error may conclude that they do not differ type I error - alpha; when in reality treatments are not different, but we conclude that they are type II error - beta; when in reality treatments are different, but we conclude they do are not alpha - probability that we will make a type I error; also known as the P value can choose any number but usually 0.05 alpha 0.05 - probability that difference occurred by chance and not because of actual difference is 0.05 beta - probability that we will make a type II error can choose any number, but usually 20% power of a study - 1 - beta how good our study is at correctly identifying a difference between the therapies if in reality they are different sample size - need enough people so sample is representative of population; the larger the sample the more likely it is representative one-sided test - if new treatment is better two-sided test - if new treatment is better or worse power (1-beta) - usually 80% to calculate sample size - 1. determine if one or two sided test 2. calculate difference in response rates (new treatment expected cure rate- current treatment cure rate) 3. determine which cure rate is lower 4. find what number meets in middle of table; multiply by 2 for one control and one intervention group recruitment and retention issues - not enough people = not statistically valid results not enough volunteers, not enough people willing to be randomized, not enough people stay for entire trial (drop out, move, or loss to follow-up) research ethics - participation requires informed consent, participation must be voluntary, participants must be given adequate information (benefits and risks), participants must understand information (vulnerable groups) clinical trial - a research activity that involves the administration of a test regimen to humans to evaluate its efficacy and safety prophylactic trial - designed to test preventive measures therapeutic trial - evaluates new treatment methods randomized controlled trial (RTC) - subjects randomly allocated into test and control groups, may be switched between treatment groups (Crossover) quasi-experiments - investigator manipulates the study factor but does not assign individual subjects randomly to the exposed and non-exposed groups; some assign study units (communities, schools etc) randomly to the study conditions; program evaluation hawthorne effect - Participants' behavior changes as a result of their knowledge of being in a study recall bias - Cases may remember an exposure more clearly than controls type of information bias healthy worker effect - The observation that employed populations tend to have a lower mortality experience than the general population confounding variables - factors that cause differences between the experimental group and the control group other than the independent variable (like age) efficacy - reduction in risk under ideal conditions effectiveness - reduction in risk under real-life situations number needed to treat (NNT) - number of patients who would need to be treated to prevent one adverse outcome (such as death) number needed to harm (NNH) - number of people who would need to be treated to harm one additional person generalizability - external validity; ability to apply results found in our study population (our sample) to the overall population ability to generalize will depend on how representative the sample is to the overall population internal validity - ability to conclude that the study design, conduct, and analysis answer the research questions without bias or mistakes comparative effectiveness research (CER) - two or more existing interventions are compared beneficial to: determine which treatment would work best in a given population or particular patient or if a cheaper alternative is just as effective four phases in new drug testing in US - phase I trials, phase II trials, phase III trials, phase IV trials phase I trials - clinical pharmacology studies small studies (20-80 people) safety issues dosages side effects phase II trials - clinical investigations 100-300 people efficacy further safety phase III trials - large scale (+ people) randomized controlled trials efficacy and relative safety often multi-centered phase IV trials - post-marketing surveillance looking for outcomes that may not have been identified previously (take a long time to materialize, or infrequent outcomes) publication bias - an erroneous conclusion that a drug (or other intervention) has only shown beneficial results (and possibly no side effects) because only the studies showing beneficial results are published and available to researchers/the public