Content Flashcards
descriptive
surveillance, ecologic, cross-sectional
no advance hypothesis
accept association may be causal or not
use pre-existing data
analytic
hypothesis testing
hypothesis poses casual link
collect new data or make use of existing data
observational studies - case-control, cohort
experimental studies - RCT, interventions
John Graunt
1662
demographer
counted number of people who have died by looking through by age and sex –> looked at patterns
analyzed weekly mortality reports
contributions: use of existing data, descriptive epi, describe mortality rates by person, place, time
James Lind
1747
surgeon
First clinical trial - sailors with scurvy in different treatment groups
contributions: develped method of experimental study design, tested hypothesis, group assignment, intervention
William Farr
Collected vital statistics and reported death by time, location, age, sex, occupation and cause
supported miasma
association between altitude and cholera mortality
contributions: collection of accurate health statistics data, interpret data to identify etiology of disease, developed method for standard mortality rate
Miasma theory
mid-1800s
cholera caused by polluted gases from decaying organic matter inhaled
prevented by better sanitation
John Snow
1854
Medical practitioner, anesthesiologist
Challenged miasma theory of cholera transmission
Grand experiment - water companies moved water source above tidal flow of thames river
Father of modern epi
formation of causal hypothesis based on prior knowledge and observation
design study to test causal hypothesis
collection of primary data
address problems of confounding, adherence, missing data
Austin Bradford Hill
1940
Randomized Control Trial of streptomycin for treatment of TB
patients with TB assigned to strep + bed rest vs bed rest alone
Contributions: randomization, comparison to standard of care, blinding, equipoise
Richard Doll
1950s
Prospective cohort study of heavy smokers –> lung cancer
case control study of people with and without long cancer and previous smoking history
contributions to epi: study design development, obtain outcome data from existing data, matching, causality - Hill’s criteria
Hill’s criteria for causality
temporal relationship, strength of association, dose-response relationship, biological plausibility, consideration of alternate explanations or confounding factors, replication of findings, cessation of exposure, consistency with other knowledge, specificity of association
Framingham Heart Study and Nurses’ Health Study
1958 and 1980
Framingham: establish risk factors for CVD
Nurses’: prospective cohort study for long-term effects of oral contraceptives
Contributions: landmark prospective cohort studies, detailed prospective data collection for prospective cohort study, analytic method
Epi objectives
determine extent of disease in pop
identify patterns and trends in disease occurrence
identify causes and risk factors for disease
evaluate effectiveness of preventive and therapeutic interventions and public health programs
applications of epi
study history of disease and describe health events
identify risk factors and causes of diseases
monitor and evaluation
identify control and preventive measures
odds
ratio of probability of occurrence of event to probability of nonoccurrence
part/non-part or p/(1-p)
p=probability
odds vs proportions
small proportions (part/whole) aka rare events = approximates odds large proportions = value does not approximate value for odds (non-rare events)
prevalence
of existing cases at point in time / # in total pop
factors that increase prevalence
increase incidence, longer duration, prolongation of life without cure, in-migration of cases, in-migration of susceptible people, out-migration of healthy people, better diagnosis of reporting
factors that decrease prevalence
decreased incidence, shorter duration, high case fatality, in-migration of health people, out-migration of cases, improved cure rates
incidence
of new cases of disease / # of population initially at risk
cumulative incidence
# of new cases over specified time / # at risk in population value increases with increased period of follow-up
attack rate
subjects who develop infectious outcome / # at risk during outbreak
case fatality rate
subjects who die / those who develop a disease
survival rate
1 - cumulative incidence for death # of participants who don't die during follow-up / number of subjects at risk for dying
concerns of cumulative incidence
assumes every is followed for same period of time
lost to follow up (censoring aka lack of knowledge for outcome of the individual)
competing risks (removes participant from being at risk from outcome of interest)
incidence rate
solution to difficulty with cumulative incidence # of new cases of disease / total person-time of observation
odds ratio
odds of developing the disease among exposed compared to odds of developing the disease among unexposed case-control small risks: OR = RR larger risks (common): OR =/= RR
risk ratio
relative effect associated with exposure
null value = 1
>1: risk or odds of disease is greater in exposed
risk difference
absolute effect associated with exposure
null value = 0
>0: risk of disease greater in exposed
Cochrane
gold standard for causal inference
trasnparent method and guidelines from WHO
standardized approaches
look at all the evidence to date and make decisions according to this
ecological study
look at associations and interpret data
strengths - hypothesis generating, less time and resource intensive
limitations - country-level data, ecological fallacy, confounding
cross-sectional study
snapshot of population
strengths - one stop shopping, less expensive, can detect effect of exposure that do not vary over time
limitations - temporal relationship not clear, prevalence in function of incidence and duration, prevalent cases may be over-represented by cases with long disease duration
case-control study
investigator enrolls cases and controls and records exposure status
strengths - efficient design for rare outcomes, quick to complete, inexpensive, easy to study multiple exposures, disease has long induction or latent periods, exposures and risk factors are not known well, exposure is difficult to obtain
limitations - no estimate of incidence is possible, control choice may be different, recall bias is possible, can’t infer something about overall population
ODDS RATIO
cohort study
observational; enroll participants free of disease, but at risk; follow up over time
strengths - exposure measured before outcome, estimate incidence in exposure group, easy to study multiple outcomes, allow study of rare exposures (risk ratio, hazard ratio)
limitations - inefficient for rare outcomes, costly and time consuming, confounding, size of exposure and unexposed may not reflect source population
Calculate - incidence rate, risk ratio
nested case control study
use data from cohort, but info you are looking at hasn’t been studied
Faster and easier, because the data is already there
randomized clinical trial
gold standard for causality, random allocation, equipoise, blinding
strengths - gold standard for causality, avoid bias by confounding, clear that exposure precedes outcome, estimate incidence, can study several outcomes
limitations - not always possible or ethical (equipoise), inefficient for rare or delayed outcomes
case cohort study
controls selected from cohort at beginning of time period
density case-control study
controls sampled each time case occurs (risk set sampling)
cumulative case-control study
controls sampled from disease-free for entire period (chosen at end of study)
prospective cohort study
follow-up period occurs chronologically after start of study period
retrospective cohort study
follow-up period occurred chronologically before start of study period
ambidirectional cohort study
follow-up period began chronologically before the start of the study period, but continues into the future
open cohort
dynamic - (membership defined by changeable characteristic (location, relationship status)
exposure status can change over time
follow up - new participants added or eliminated during follow up
incidence rate
fixed cohort
defined by irrevocable element (exposure to man-made or natural disaster, inhabitants of a location at a specific point in time)
exposure - doesn’t change
may have loss to follow up, but no enrollees
incidence rate
closed cohort
irrevocable event
no losses to follow up
cumulative incidence or incidence rate
induction period
time between exposure and development of disease
latent period
time period between disease development and detection
empirical latent period
induction period + latent period
equipoise
uncertainty about risks and benefits of treatment
PICO
population, intervention, comparison, outcome
community RCT
randomize groups or clusters to groups
less statistically significant
people within groups are more similar to each other than people in other groups
RCT Phases
Phase 1: Formulation - safety and toxicity, dosage, side effects
Phase 2: Activity - Evidence for potential treatment, effect of intervention
Phase 3: efficacy - RCT, effects of intervention on outcome
Phase 4: effectiveness - investigate wide-scale use and effectiveness
effectiveness
strategy practical to use in real world
realistic
broad sample from potential target population
efficacy
theory predicts it will work
testable in ideal and controlled conditions
greatest effect of intervention; high adherence
sensitive to predicted effects; may be surrogate
parallel design of RCT
two or more arms of treatment
different participants in each
crossover design of RCT
two or more arms
participants receive all treatments
outcome must be acute and reversible
key assumption: participants at beginning second phase returned to initial status (washout period)
simple
single new treatment vs comparison treatment
treatment vs standard of care
factorial
multiple treatments
2x2 factorial design (each participant undergoes separate randomization assignment for 2 treatments)
potential to examine effect of multiple treatments within signle study design
simple randomization
no guarantee of equal size groups
coin flip, random number generator
still possible for imbalance, if small trial
blocked randomization
within block, 1/2 get intervention, 1/2 get placebo
block sizes vary under randomization process
stratified randomization
ensure balance by key characteristics in all groups
assign patients to strata, according to baseline cahracteristics
within strata, assign patients using blocked randomiation
Random error
error caused by some factor that changes from one measurement to another
decreases with larger study size
selection bias
error due to systematic differences between the characteristics of the people selected for a study and those who are not (follow up and selection)
source:design, sampling, Berkson’s bias, self-selection, healthy worker effect, non-response
Problems: selecting inappropriate control group, exposure influences detection of cases, outcome influences choices of exposed and non-exposed participants, loss to follow up related to exposure and outcome
information bias
Error due to differences in manner in which data for exposure or outcome are obtained from various groups, which may lead to misclassification of study participants.
sources: recall, prevarication, reporting, loss to follow up, missing data, digit preference, boserver, instrumental, detection, lead time, length
random/non-differential information bias
variability in data that is not readily explained
measurement of disease is not different for exposed and unexposed; diseased and non-diseased
validity and reliability
tends to result in measure of association that is biased towards null value
systematic/differential information bias
effect estimate may be incorrect
case-control study - cases recall or report exposure differently from controls
avoid this via blinding, change control group, etc.
measures of association can be biased in any direction
confounding
associated with exposure, associated with disease, not part of causal pathway of exposure
calculate adjusted measure of association to only reflect effect of exposure (without confounder)
(RRcrude - RRadjusted) / RRadjusted > 10%
3C’s of confounding
comparability - distribution of Z differes between exposed and unexposed
collapsibility - 2x2xk table can be collapsed into 2x2 table without changing effect estimate (stratification)
counterfactual - factual and counterfactual outcomes of exposed participants are not exchangeable
positive confounder
variable that is positively related to disease and positively related to exposure or both inversely related
RRadjusted < RRcrude
negative confounder
variable that is either positively related to disease and inversely related to exposure or vice versa
RRadjusted > RRcrude
How to reduce for confounding
study design - randomization, restriction (limit study to one category of potential confounder), matching
analysis: stratification, change in effect estimate (adjust after stratification), multivariate analysis
effect measure modification
effect of exposure on outcome.disease is modified depending on value of third variable called “effect modifier”.
want to better understand and measure – effect is different for different people.
exposure having a different effect on outcome in different groups of patients.
detected by stratum-specific estimates of measure of effect
generation time
time interval between one person getting infected and another person getting infected from the firt
reproduction number
average number of infected persons resulting from contact with single infected person occur from contact with primary case
secondary attack rate
risk of infection among susceptible individuals exposed to an infected source
transmission probability
probability of transmission from infected person to susceptible person during contact
virulence
degree to which a pathogen can cause disease and death
direct transmission
airborne, direct contact, fecal-oral, STD
indirect transmission
zoonoses, vector-borne diseases, environmental pathogens, intermediate host (tapeworm)
serial interval
time between successive cases of disease
basic reproductive number
average number of infections that would be caused by one infected person when everyone else is susceptible
effective reproductive number
average number of infections resulting from one infected person given that not everyone is susceptible
Reproductive rate
beta x k x D
beta = likelihood of transmission per individual contact
k = number of contacts a person has
D = duration of infectivity
– decreased by immunization rate (leads to herd immunity)
What’s different about ID epi?
A person with infection is both the outcome and source of transmission.
A person with infection may resolve an infection and become immune
A person with an infection may have no symptoms yet still infect others without being recognized as a source
Since ID can be an epidemic, rugency may exist and preventive action must be taken.
There is strong biological basis for investigation and action (Koch’s and Hill’s)
People can become immune from having the disease or through immunizations
Why is access to HIV diagnosis and treatment important?
life extension
reduce infection
HIV has short latency period and long incubation period
Multicenter AIDS Cohort Study
longest US-based study of HIV-infected individuals
over 1000 publications
discoveries: how to best diagnose HIV infection, direct association between viral load and HIV disease progression, connection between low CD4 T-cell counts and progression to clinical AIDS, central role of immune activation in HIV disease, how to best treat and care for HIV
trends in HIV infection
antiretroviral therapy retention decreases as months go on
new infections common in women in sub-saharan Africa
infections slightly more in men (overall)
prevalence is higher among females
East and Southern Africa - more common among young women (prime time for pregnancy and repro)
Risk of MTCT of HIV
5-10% in pregnancy/in utero (medication)
10-20% labor and delivery / intrapartum (C-section)
10-20% breastfeeding / postpartum (supportive care during breast feeding)
overall: 30-45%
Risk of MTCT of HIV
5-10% in pregnancy/in utero (medication)
10-20% labor and delivery / intrapartum (C-section)
10-20% breastfeeding / postpartum (supportive care during breast feeding)
overall: 30-45%
how to design a successful screening program
- suitable disease (high prevalence, high burden, detectabl, treatment is available, early treatment is beneficial)
- Suitable test (simple, rapid, inexpensive, safe, acceptable)
- Suitable screening program (reliable, valid)
sensitivity
% of individuals with the disease correctly classified by screening test as having disease
denominator would be dependent on another (diagnostic)test
specificity
% of individuals without the disease correctly classified by screening test as not having the disease
how can sensitivity and specificity of a test be improved?
retrain screeners, recalibrate screening instruments, use a different test, use more than one test, develop better screening tests
positive predictive value
% of individuals classified by screening test as having the disease who actually have the disease
negative predictive value
% of individuals classified by screening test as disease-free who do not have the disease
lead time bias
amonunt of lead time gained by screening and degree that these lead times improve effectiveness of treatment
time between disease detection with screening and its usual clinical presentation and diagnosis
does the length of time before getting clinical symptoms matter?
length time bias
evaluation of screening must also take into account length-biased sampling
screen-detected cases may not be representative of all cases (may have longer preclinical phases, biologically slower progression and somewhat better prognosis)
screening will preferentially identify those slowly developing disease
volunteer bias
people who choose to participate in screening program may be healthier than those who do not participate
evaluations must be based on measures that are not affected by early diagnosis except to the extent that early treatment is beneficial
healthy worker effect
special case of selection bias in which mortality related to occupational exposures is underestimated
Berkson’s bias
special case of selection bias involving hospital cases and controls, where the higher exposure rate among those individuals admitted to a hospital leads to a distorted odds ratio effective size