Exam2 Flashcards
Advantages to randomization
1) Assignment of a patient to a study group is determined by chance
2) Unpredictability
3) Rules out self-selection of subjects
4) Provides control over confounding, even by factors that are hard to measure or unknown to the investigator
5) Distributes potential confounders similarly across comparison groups (exposed/non-exposed)
6) Randomization does not guarantee the comparability of groups – therefore stratified matching
Effects of non-compliance
Drop-outs do not take the treatment, either knowingly or unknowingly when they shouldn’t while drop-ins either knowingly or unknowingly the treatment when they shouldnt. The effect is that there is less difference seen in treatment effect, the groups will be less different than they should have been.
Intention to treat analysis
Analytical method for randomized trials. primary type of analysis done, all individuals randomly allocated to treatment are analyzed regardless of whether they received treatment or not.
Efficacy analysis
Analytical method for randomized trials. Used for reduction of risk. Determines the treatment effects under the ideal conditions in those who take the full treatment as directed (with compliance) but excludes those without compliance. excluding non-compliers usually over-estimates the effectiveness of a therapy
Matching in case control
individual matching
frequency matching - case and control groups
DISADVANTAGES
cannot study the effect of the matching factor
may reduce statistical power
matching factor must be included in the analysis
may be difficult and time consuming to match
Advantages of intention to treat
- It preserves the benefits of randomization
- Maintains the statistical power of the original study
- Helps ensure that the study results are unbiased
- Gives info on the effectiveness of a treatment under everyday practice condition
- Excluding non-compliers would over-estimate the success of intervention (conservative approach)
Number needed to treat
number needed to treat = number of patients who need to be treated in order to prevent one additional bad outcome
NNT = 1/(rate in untreated – rate in treated)
Rate = mortality, incidence (?)
Efficacy
used to analyze the extent of the reduction in outcomes by treatment - vaccines
(Rate in controls – rate in treated)/(rate in controls)
AKA reduction in risk
type I errors
The treatments do not differ but we conclude that they do (alpha)
type II errors
The treatments differ but we conclude that they do not (beta)
alpha
probability of making a type I error, concluding that treatments differ when they don’t
the level of statistical significance
beta
the probability of making a type II error, concluding that the treatments do not differ when they do differ
Strengths of randomized trials
1 - Control over the assignment of the treatment
2 - Randomization ensures the treatment and control groups are balanced, reducing bias and confounding
3 - Blinding minimizes bias
4 - Prospective design allows for temporality and causal relations
5 - Provides firm basis for statistical hypothesis testing - gold standard
Weaknesses of randomized trials
1 - generalizability - testing is done on a small number of motivated volunteers
2 - close monitoring may not be the case in a community setting
3 - expensive and labor intensive
When does the odds ratio approximate the risk ratio?
when the incidence is low, when the disease is rare
cohort effect
the influence of membership in a particular cohort - shared temporal experience or common life experience
Advantages of cohort studies
temporality
Direct determination of risk
Can design the study to follow the exposures you specifically want
Size of the cohort under control by study investigators
Can study rare exposures
Disadvantages of cohort studies
Takes a long time Expensive Need a lot of justification and supporting scientific data Not great for studying rare outcomes Subjects lost to follow up
Biases for cohort studies
Information bias on exposure and related factors
Bias in assessing outcomes
Bias in analysis and reporting
Bias from non-response and loss to follow up
Advantages of retrospective cohort
Faster – you don’t have to wait for cases to accrue
Less expensive than cohort study
You have a lot of exposure data to choose from
Advantages for case control studies
Rare diseases
Disease with long induction/latency periods (cancers)
Lower cost than cohort studies
Faster than cohort, no follow-up time needed
Less expensive than cohort studies
Can be exploratory (little known about disease)
Can assess multiple exposures – good for diseases about which little is known
Tend to use smaller sample sizes than cohort studies
Good for dynamic populations
Good for when exposure data are expensive or difficult to obtain
Disadvantages of retrospective cohort study
relies on available exposure info only, may not be in info that you want/need
Disadvantages of case control studies
No incidence or temporality
Information on previous exposures may not be available or accurate
Difficult to obtain appropriate controls
Representativeness of cases and controls is often unknown – selection bias
case-control biases
Information bias – interview cases more than controls
Recall bias: controls don’t remember as well
Misclassification bias (differential vs. non-differential)
Response bias: some groups of people may respond better than others
Selection bias -
Control selection – controls should be cases if they had been exposed
Controls should be from the same place as cases, or else there will be bias
Why try to choose incident cases for case control?
incident cases have less chance of change of exposure
difficult to assess temporality with prevalent cases
survivorship bias for prevalent cases - tend to be longterm survivors
how to choose controls - case-control
controls should be comparable to cases except with no disease and exposure experience
potential for exposure should be the same
controls should come from the same population as cases
controls should represent those who would have been cases if they were exposed
hospital based controls - pros and cons
easy to ID, more willing to participate, if from the same source population, minimal bias
However, not population based, may be from different source populations, exposure of interest may be associated with other diseases that are used as controls
Cross-sectional advantages
Estimate the magnitude and distribution of a health problem
Good for hypothesis generation
Useful for planning interventions
Repeated cross-sectional studies can show changes in trends in disease and risk factors
Low cost and generalizable
Cross-sectional disadvantages
No incidence data
No temporality
Healthy worker survivor effect – long term prevalence – long term survivors favored
Not good for disease with low prevalence (rare or short duration)
ecologic fallacy
association observed at an aggregate or group level does not necessarily represent the exposure-disease relationship at an individual level
Ecologic studies Advantages
estimates disease rates at population level or global measures
good approach for generating hypotheses when a disease is of unknown etiology
quick, simple, inexpensive
Ecologic studies disadvantages
ecologic fallacy
imprecise measurements of exposure and disease
non-concurrent prospective study
retrospective cohort
prevalence study
cross sectional study
Relative risk and odds ratio vs. attributable risk
relative risk = strength of association and potential for causation - important for deriving causal inference - etiology studies
Attributable risk = the potential for prevention if the exposure were eliminated, how much of the disease that occurs can be attributed to exposure? good for elimination strategies
attributable risk = risk difference
incident rate of disease in exposed - incident rate of disease in unexposed
etiologic fraction
used in conjunction with attributable risk
use only when there is certainty of causation
excess fraction
used in conjunction with attributable risk
use when there is no absolute certainty of causation
Attributable risk interpretation
1) the excess risk of disease associated with exposure is xx or xx*100%.
2) xx of the yy incident cases of disease with exposure are attributable to their exposure
where xx = attributable risk and yy = incident cases of disease in exposed
3) If there were a prevention program for exposure, we could hope to eliminate xx of yy incident cases of disease that experience exposure
Percent attributable risk
(incident rate of disease in exposed - incident rate of disease in unexposed)/incident rate of disease in exposed
What proportion of the risk in exposed people is due to exposure
Percent attributable risk interpretation
xx% of the cases of disease with exposure can be attributed to the fact that they were
If prevention programs were 100% effective, there would be xx% fewer cases of disease
Population attributable risk = population risk difference
Incidence in total population must be known or proportion of exposed in total population
incidence in total population = incidence in exposed prevalence of exposure + incidence in unexposedprevalence of unexposed
PAR = incidence in population - incidence in unexposed
Population attributable risk interpretation
if there was an effective prevention program, we could hope to prevent xx of the yy incident cases of disease in the total population
Population attributable risk percent
(incidence in total population - incidence in unexposed)/incidence in total population
Population attributable risk percent interpretation
xx% of cases of disease in the total population may be attributable to exposure and could be eliminated by eliminating exposure in the population
attributable risk using relative risk
[(RR-1)/RR]*100%
attributable risk using odds ratios
[(OR-1)/OR]*100%
You can only assume that the exposure caused the disease
Population attributable risk using odds ratios
Pe(OR-1)/Pe*(OR-1)+1
where Pe = exposure prevalence in the target population
Can only assume that the exposure caused the disease
Infectious disease model
host > environment > agent > (vector)
chronic disease model
multifactorial model - can be very complicated
risk factor epidemiology > modifiable vs. non-modifiable risk factors
genetic disease model
multiple genes -> multiple environmental factors -> disease
clusters of disease in one family do not necessarily imply genetic disease
descriptive epi > family based studies > population based studies (best)
genes can be determinants of disease or determinants of environmental susceptibility leading to disease
social epidemiology model
look at many different factors to see if there is a development of a certain outcome
look at the material and social conditions of life
correlation
changes in two factors are related
association
there is a relationship between an independent factor and an outcome
effect
a causal action of a factor on an outcome
causal inference steps
- develop evidence - observations of people, human experiments, other research
- synthesize - systematic reviews, meta-analysis, other evidence
- evaluate - expert judgment, causal criteria
necessary
must be present to cause disease
sufficient
can independently cause disease
Causal criteria
- strength of the association
- temporality
- dose response
- consistency
- specificity
- plausibility
- replication
- cessation of exposure
- consideration of alternate explanations
Causal criteria: specificity
x
Causal criteria: dose response
x
Causal criteria: temporality
x
Causal criteria: replication
Other people following your methods can replicate your findings - you can replicate the findings of others’ work by following their methodology
Causal criteria: biological plausibility
there is a biological basis in the theory
Causal criteria: cessation of exposure
x
Causal criteria: consideration of alternate explanations
x
Causal criteria: strength of association
x
Causal criteria: consistency
x
why randomize?
1) rules out self-selection of subjects
2) provides control over confounding, even over factors that are unknown
3) distributes potential confounders similarly across comparison groups
non-compliance types
drop-out: people in treatment group do not take the treatment intended, either on purpose or by accident
drop in: take the treatment, either on purpose or by accident if they are supposed to be controls
Analytical approaches to randomized studies
Intention to treat: analysis comparing outcomes between comparison groups that were formed by randomization, all individuals are analyzed regardless of whether they completed the treatment or not, most popular
Efficacy analysis: reduces risk, determines the treatment effects under the ideal conditions in those who take the full treatment as directed but excludes those without compliance
Subgroup analysis: used to determine the treatment effects among different subgroups - treatment/intervention may work better in one group than another, predetermined subgroups
Risk ratio interpretation - cohort study
1) the risk of the outcome is xx times greater in exposed compared to unexposed
2) there is a xx-1*100% greater or lesser risk of outcome in exposed compared to non-exposed
Odds ratio interpretation - case control
the odds of exposure is xx greater in cases compared to controls
power
1-beta
the probability of correctly concluding that the treatments differ
