Section 1
Preview this deck
difference between confounding and bias
Front
| 5 | 0 | ||
| 4 | 0 | ||
| 3 | 0 | ||
| 2 | 0 | ||
| 1 | 0 |
Active users
0
All-time users
0
Favorites
0
Last updated
7 years ago
Date created
Mar 1, 2020
Cards (182)
Section 1
(50 cards)
difference between confounding and bias
confounding: not preventable bias: preventable
cumulative incidence is a measure of..
risk
experiment
investigator-initiated intervention that modifies the exposure through prevention, treatment, or removal should result in less disease ex: smoking cessation programs result in lower lung cancer rates provides strong evidence for causation but most epidemiological studies are observational
risk difference
ranges from -1 to +1 (risk in Exposed)-(risk in Non-Exposed) (cumulative incidence in Exposed)-(cumulative incidence in Non-Exposed) Risk difference (good for public health action)
strength of association
larger the association, more likely the exposure us causing the disease ex: if relative risk of lung cancer in smokers is higher than in non smokers there is greater chance it is causal strong associations are more likely to be causal because they are unlikely to be due entirely to bias and confounding
consistency
the association is observed in different persons, places, times, and circumstances, replicating the association in different samples with different designs gives evidence of causation ex: smoking has been associated with lung cancer in at least 29 retrospective and 7 prospective studies
sufficient and necessary cause framework
there may be a number of sufficient causes for a given disease or outcome A component cause that must be present in every sufficient cause of a given outcome is referred to as a necessary cause ex: there are many things needed for the contractions of TB (sufficient), but exposure to TB is the only one that is necessary necessary but not sufficient: + sufficient but not necessary: or
information bias
bias due to the procedures used to collect or analyze information that results in misclassification of exposure and/or outcome status
primary, secondary, tertiary prevention
1: vaccines 2: screening 3: treatment
outbreak definition
occurrence of disease that is greater than would be expected at a particular time and a particular place
ways to express prognosis
case fatality rate, deaths/person-years, 5 year survival rate
basic reproductive number
The basic reproductive number (R0) is the average number of secondary cases a primary case generates Infection will spread in the population if R0> 1 Infection will die out in the population if R0< 1 Infection will reach an endemic phase if R0= 1
controlling confounding through stratified analysis
Stratification allows the association between exposure and outcome to be examined within different strata of the confounding variable
statement of risk with explicit comparison
risk difference relative risk odds ratio
probability vs odds
probability: # cases/ entire population (proportion) odds: # cases / # non-cases(ratio)
Bias: what kind of error
bias is a systematic error it occurs the same amount regardless of the study size (random errors decrease as study size increases)
case fatality rate
number people die/number people with disease a cumulative incidence technically a proportion
5-year survival rate
percent of patients still alive after 5 years
index case, primary case, secondary case
index: first case identified primary case: first case in population, often identified after secondary: cases infected by index case
principles of epidemiology
distribution, health related status, population
measures of disease occurrence
cumulative incidence (new cases/pop at risk) incidence rate (takes into account time) point/period prevalence
rate
A proportion with the specification of time (# new cases of cancer) / population size * years of follow-up
mortality versus morbidity
mortality: death morbidity: illness
example of information bias
recall bias
cumulative incidence risk ratio vs odds ratio
CI RR: cumulative indicence in exposed (a/a+b) / cumulative incidence in unexposed (c/c+d) OR: odds of disease among exposed (a/b) / odds of disease among unexposed (c/d)
Proportion:
A ratio where the numerator is included in the denominator preterm births / all births
definition of confounder
associated with exposure associated with outcome not an intermediate in causal pathway
temporality
the causal factor must precede the disease in time only criteria that everyone agrees with prospective studies do a good job establishing the correct temporal relationship between exposure and disease ex: a prospective cohort study of smokers and non-smokers stars with the two groups when they are healthy and follows then to determine the occurrence of subsequent lung cancer
analogy
has a similar relationship been observed with another exposure and/or disease? ex: effects of thalidomide and rubella on the fetus provide analogy for effects of similar substances on the fetus.
Hill's criteria
guidelines for judging whether an observed association is causal
count
number of occurrences
herd immunity
Indirect protection from infectious disease that occurs when a large percentage of the population is immune (e.g., through vaccination) Once a certain threshold has been reached, herd immunity will result in the elimination of the disease from the population
statements of risk without comparison
absolute risk
types of sources: point source, continuous source, propagated
point: one outbreak continuous: ongoing exposure propagated: person to person
epidemic versus endemic
epidemic: disease in excess of normal endemic: usual presence of disease
3 types of internal validity and 1 type of external validity
internal: confounding, selection bias, information bias study population poorly reflects total population, but is truly randomized external: generalizability study population accurately reflects total population, but is not randomized
risk ratio
range from 0 to positive infinity risk in exposed/risk in unexposed Risk ratio (good for questions of "causation")
class definition of bias
any systematic error in the design of a study that results in a mistaken estimate of an exposure's effect on the risk of disease
clinical-onset serial interval
Defined as the average time between symptom onset in an index case and a secondary case(s) Derived from analysis of symptom onset times between index and secondary case
3 critical factors measuring frequency of disease
number of people affected by disease size of population that gave rise to cases length of follow up time for the population
epidemic curve
We use an epidemic curve to measure and track an outbreak histogram of the number of cases against the time of onset of disease
plausibility/coherence
biological or social model exists to explain the association. association does not conflict with current knowledge of natural history and biology of disease ex: cigarettes contain many carcinogenic substances often times the cause-effect relationship is identified before the biological mechanisms are identified
specificity
a single exposure should cause a single disease (has many exceptions) ex: smoking is associated with lung cancer as well as many other diseases. In addition, lung cancer results from smoking as well as other exposures when present, specificity provides evidence of causality, but its absence does not preclude causation
survival analysis: life tables and kaplan-meier curves
calculates survival at the time where the death/event occurs rather than in pre determined intervals (like in a life table) KM>life table
deaths/person years
just another way to express prognosis
attack rate
cumulative incidence of infection in a group of exposed susceptible hosts Measured from beginning to end of outbreak Attack rate is equal to the number of persons infected, divided by the total number of exposed, susceptible persons during the outbreak
incubation period
time from exposure until the onset of disease
biological gradient
a "dose-response" relationship between exposure and disease. persons who have increasingly higher exposure levels have higher risks of disease ex: lung cancer deaths rates rise with the number of cigarette smoked some exposure may not have dose-response, but instead have threshold
ratio
Relationship between two numbers women with depression / men with depression no units
selection bias
if the way in which study participants were selected results in "a mistaken estimate of an exposure's effect on the risk of disease" all study designs are subject to selection bias
Section 2
(50 cards)
why is a comparative trial needed to confirm efficacy
need to compare to standard of care to prove the value of the intervention otherwise, you will not know what would have occurred to the participants in the absence of the intervention the comparison groups is used to estimate the counter-factual
characteristics of a clinical trial
prospective: follows subjects moving forward must be an intervention: changes/ alters/ controls something must have control group: way to estimate the counter-factual
differential attrition
loss of participants from various comparison groups in unequal amounts can lead to a heavily biased result type of selection bias
random selection
compared to randomization (which aims to create two equal and random groups), random selection is used to draw a representative sample from a target population (often times the first step before randomization)
priorities: internal versus external validity
in observational studies epidemiologists prioritize internal validity (lack of bias) over external validity (generalizability)
prospective v retrospective cohort
both study designs are identical prospective: looking from current year and following into future retrospective: looking from current year into the past (different from case-control because still exposure is primary thing looking for, and then looking at outcomes)
more on phase 3
should only be conducted once phase 1 and 2 work is complete design based on phase 1 and 2 results compares intervention with control group control group may be standard or care or no intervention choice is based on ethics, and an analysis of the literature usually larger sample size designed to detect a clinical meaningful difference depending on length of trial, may detect long-term risks of intervention
randomization
participants are assigned to one group or another on a random basis randomization produces study groups that are comparable with respect to unknown risk factors (this avoids confounding)
benefits of observational trials
more practical/ easier to conduct results more generalized than an experiment conducted on one population sometimes there are natural experiments (John Snow and Pump) often you can adjust for potential biases either in the study design phase or after
phase 4 clinical trials definition per NIH
studies done after the intervention has been marketed these studies are designed to monitor the effectiveness of the approved intervention in the general population and to collect information about any adverse effects associated with widespread use may be observational or RCTs
attrition (loss to follow up)
attrition in cohort studies and clinical trials is source of potential selection bias complicates the interpretation of the study findings
equipoise
the ethics of clinical research requires equipoise: a state of genuine uncertainty on the part of the clinical investigator regarding the comparative therapeutic merits of each arm in a trial you want to give a placebo only when the new treatment does not have any sort equal (this is because you want to compare any new treatment to the most recent most similar alternative in order to show it is better and to be ethical)
cohort study: who participates, what to measure with participants, comparison of interest
participants free of disease measure exposure at study entry and follow them for new onset disease compare rate/risk of disease among the exposure vs unexposed= risk ratio
single blinding v double blinding
ideal trial design is double-blind because either subject nor person running trial knows the intervention arm
how to interpret the odds ratio from a case control study
we calculate and describe the odds fo exposure among those diseases compared to the odds of exposure among the controls but we think about this as an estimate of relative risk- among those exposed, how the risk of disease compares to the risk for those not exposed
more on phase 2
compared to most phase 3 studies, fairly small uses dose or technique discovered in phase 1 trial, sometimes evaluated several doses determines whether or not the medication works at MTD (maximum tolerated dose) determines feasibility and intervention efficacy (i.e., does the intervention appear to improve condition) collects information on rates of adverse events
clinical trials: definitions per NIH
biomedical or behavioral research study of human subjects designed to answer specific questions about biomedical/behavioral interventions
selection of controls: purpose, what is required of
purpose: provide estimate of distribution of exposure in source population only works if controls: come from same source as cases are sampled independently of exposure have to be at risk of developing outcome
why conduct a cohort study?
generate measures of disease incidence seek to answer: what is the risk of disease comparing disease incidence in exposed and unexposed generates incidence ratios and other effect estimates from which we infer causation public health implication: prevent new cases of disease in the future
effectiveness
whether a drug/treatment works in real life i.e., if the drugs work when taken the way people normally take them and when taken by norma people, if drug is effective then most people who have the disease would improve if they used the treatment
cohort study
group of healthy people identified and followed over time period ascertain occurrence of health-related events in order to investigate if the incidence of an event is causally related to an exposure of interest exposure -> outcome (over time) KEY: exposure precedes outcome
phase 1 design
study population - intervention- maximum tolerated dose
case control studies: estimating an incidence rate ratio: what we need to know
need to know: ratio of exposure to unexposed cases need to know: an estimate of the ratio of exposed to unexposed person-time i.e., how many cases, how many total, how much time (in order to calculate incidence rate and then make ratio)
misclassification in case-control studies
misclassification can happen happen too, but whether classification of exposure happens comparably or differently among diseased or non-diseases (opposite from cohort)
phase 1 toxicity studies
tests tolerance to new agent, assess what a toxic dose is estimates maximum tolerable dose
case control versus cohort
in case control start with cases and controls then determine exposure history
example of phase 4 trial
woman's health initiative (estrogen replacement therapy) looking for adverse outcomes of an intervention that was already on the market participants randomly assigned to placebo or treatment and monitor rate of outcomes of interest (symptom relief)
blinding
not disclosing to patients and outcome assessors the treatment allocations after random allocation avoids bias
phase 2 clinical trials definition per NIH
study in a larger group of people (several hundred) to determine and further evaluate safety
cumulative incidence ratio or risk ratio incidence rate ratio
(disease developed in exposed/ total exposed) / (disease developed in unexposed/total unexposed) ratio of incidence rates
source population
hypothetical population in which cohort study that produced the cases might have been conducted
weaknesses of cohort studies
loss to follow up expensive
allocation concealment
a technique used to prevent selection bias by concealing the allocation sequence (order in which participants are allocated to treatment) from those assigning participants to intervention groups until the moment of assignment prevents researchers from influencing which participants are assigned to a given intervention group
components of clinical trials
manipulate the exposure/ diagnostic tool/ treatment randomly design who gets exposure
negative confounding
crude measure of association is smaller than stratified association ex: confounding results in an under estimation of association between exposure and disease
non-compliance and ITT analysis
intent to treat analysis: include individual participant in the arm to which they were assigned by randomization, irrespective of actual treatment received because those who "select" to comply with the treatment may differ from those who do not comply - introducing potential selection bias (and eroding the benefits of randomization) "once randomized, always analyzed" - to reduce selection bias
biases in cohort studies
information bias: misclassification of the outcome quality of information obtained should be comparable in both groups particularly common with subjective outcomes ensure that person who conducts the outcome ascertainment is makes with regards to exposure status
phase 1 clinical trial definition per NIH
testing small group of people (20-80) to determine efficacy and evaluate safety
odds ratio vs IRR (case control vs cohort)
the odds ratio from a well conducted case control study provides an estimate of the incidence rate ratio from a comparable cohort study with the same source population for this to be a good estimate, the controls must be samples properly
phase 3 clinical trials definition per NIH
study to determine efficacy in large groups (several hundred to several thousand) by comparing the intervention to other stranded or experimental intervention, to monitor adverse effects, and to collect information to allow safe use
rothmans general model of causation
for each exposure-outcome relationship, there is a causal pie comprised on different component causes, if the pie is complete then the outcome occurs
if we already know outcome, what are we trying to determine?
not interested in rate of exposure among controls, we want to know if cases have higher levels of exposure than other persons in source population
non differential and differential misclassification
non: mistakes happen comparably among exposed and non-exposed typically biases results towards the null value (no association, relative risk= 1) diff: mistakes in classifying diseases happen differently among exposed and non-exposed can bias study results in any direction (towards or away from null value)
more on phase 1
generally first test on humans healthy participants or people with no hope of recovery sometimes gives early evidence of effectiveness largely to determine safety detect side-effects with increasing doses
efficacy
whether a drug/treatment works under the best possible conditions i.e., just testing to see if the drugs works when taken properly and taken by an ideal population
3 features of RCTs that protect against threats to internal validity
1. true randomization (no one know the randomization pattern) 2. blinding so that no one is looking too closely at anything 3. use the placebo 4. consider equipoise
phase 2 design
study population - intervention - feasibility efficacy
positive confounding
males more likely to be diseased, males more likely to be exposed crude association between E and D is confounded due to gender and stratified results indicate no association between E and D
crossover trial design
crossover designs: each person acts as their own control, useful for chronic conditions or for shorts term, frequently occurring things (like common cold)
case control study: who participates, what do you measure with participants, what is comparison of interest
cases (those with disease) and controls (those without disease) measure exposure before disease occurred compare rate of exposure among cases versus controls typically compare odds= odds ratio
Section 3
(50 cards)
what type of bias does blinding eliminate
information bias
positive predictive value (PPV)
if i test positive, this is the chance i actually have the disease
when we screen we hope to;
Detect disease at an earlier stage provide treatment at an earlier stage
major advantages of ecological studies
relatively quick and inexpensive: secondary data sources often readily available may be only appropriate design for associations at group level
odds ratio in cohort study vs odds ratio in case control study
cohort: odds ratio: odds of disease among exposed / odds of disease among those unexposed case control: odds ratio: odds of exposure among those diseases / odds of exposure among those without disease
pre-clinical phase
the time from biological onset of the disease to the development of signs and symptoms
when to conduct case control
when exposure data are expensive when disease has long induction and latent period when disease is rare when little is known about disease
disadvantages of ecological studies: ecological fallacy
ecological fallacy: using an association at the group level to make inferences at the individual level- individuals are characterized only by the average value of the group can work in reverse= using associations at individual level to make inferences at group level (atomistic fallacy) both called cross-level bias
other types of case control: case cohort study
nested within a cohort
clinical phase
the time from when signs and symptoms develop to an ultimate outcomes such as possible cure, control of the disease, or death
key features of cross sectional study
main purpose is to generate or test preliminary hypotheses frequently used when little is known about causes of particular outcome most use population survey methods: questionnaires, interview, telephone interviews, physical exams
validity of screening tests
screening tests should be: sensitive, specific
matching
can be used tonsure that important covariates are similarly distributed in cases and controls stratified analyses or modeling can be used to adjust for confounding
limitations of case control
possibility of selection bias possibility of recall bias temporal association inefficient for rare exposures survival bias
differential recall
cases remember better/worse than controls bias leads to incorrect associations
in some cases, we might use simultaneous testing:
individuals undergo two screening tests (with different sensitivities and specificities for detecting the disease) at the same time
what is screening
screening involves the identification of unrecognized disease using tests, examination, or other procedures
when should we screen?
Can disease be detected early? what is the sensitivity and specificity of the test? what is the predictive value of the test? How serious is the problem of false positive results? what is the financial and emotional cost of early detection? are patients harmed by screening tests? do individuals benefit from early detection?
what can ecological studies do
provide the first evidence of a relationship between exposure and disease because they are easy to conduct because they rely on secondary data sources
specificity equation:
true negative/(true negative+false positive)
disadvantages of cross sectional study
difficult to establish temporal sequence: which comes first? unsuitable for rare exposures or outcomes potential for incidence-prevelence bias
advantages of cross sectional studies
require less time and resources than other epidemiological or experimental studies easier to undertake: no follow up time, usually less stringent inclusion and exclusion criteria
what does a placebo approximate
the counterfactual
types of ecological variables- aggregate ecological variable
aggregate ecological variable- individual observation summarized at the group level mean age, mean blood pressure, median income, proportion that are current smokers
potential for information bias
limitation in recall: relying on patient memory memory is better for some exposures (current OC use) and not others (episodes of anger)
cross sectional study: defining characteristic
exposure and disease assessment occurs at the same point in time
sequential (two-stage) testing:
step 1: a less expensive, less invasive or less uncomfortable screening test is used first for all individuals to provide a preliminary result step 2: individuals with positive results from step 1 will return to receive a second test that is more expensive, invasive and has a greater sensitivity or specificity
results from two stage sequential testing
sequential testing leads to a net increase in specificity
PPV equation
true positive/(true positive+false positive)
what type of bias does allocation concealment eliminate
selection bias
where to find controls
random digit dialing: variable because people don't answer phones neighborhood control: might be associated with exposure, neighbors may not be in source population hospital based: referral patterns may differ by condition, tend to differ from people in community, control conditions must be unrelated to exposure friend: not likely to be samples independent of exposure dead: not at risk for disease under study
key features: ecological study
unit of observation: group unit of analysis: group overall summary measure of exposure overall summary measure of outcome
null value for risk difference
0
selection bias in case control studies
when controls don't provide good estimate of distribution of exposure in source population
example of large cross-sectional study
NHANES- national health and nutrition examination survey
NPV equation
true negative/(true negative+false negative)
trade-offs between sensitivity and specificity
increasing one with inevitably decrease the other, so you want both as high as possible (above 90%) when there is continuous yes and no scales, there has to be a point at which we cut off what is a yes and what is a no, wherever we put the cut off point is where we are telling someone they have it or do and we cannot have 100% of both
how to evaluate screening trials for efficacy
doing an RCT, we can look at bias and blinding in the same way as we do other trials, in addition to the specific biases we examine within screening
misclassifications are what types of bias, except for differential attrition is what type of bias
information bias selection bias
other types of case control: case-crossover
case and control information comes from same person but at different points in time useful for studying acute effects (anger, drugs, exercise)
negative predictive value (NPV)
if i test negative, this is the chance I actually don't have the disease
sensitive: able to correctly identify those with a disease
specific: able to correctly identify those without a disease
sensitivity equation
true positive/ (true positive+false negative)
where to find cases
registries administrative records physician practices hospital admissions
as prevalence of a disease increases...
PPV is likely to go up as there are more positive tests NPV is likely to go down as it gets harder to find a negative negative tests (holding specificity and sensitivity the same)
non differential errors in recall
comparable between case and controls bias toward null (no association)
analysis of cross sectional studies
set up 2x2 table measuring prevalence can use either prevalence ratio or odds ratio
residual confounding
we don't measure all confounding, leftover distortion in the study
null value for risk ratio, odds ratio
1
incident cases and prevalent cases and which one is preferred
incident: newly diagnosed prevalent: existing cases since we are usually interested in identifying risk factors for disease incidence, we generally prefer to study incident cases
Section 4
(32 cards)