Epidemiology (IX) Miscellaneous Flashcards
Increasing the cutoff will do what to specificity & sensitivity?
Increasing the cutoff decreases sensitivity but increases specificity (fewer false positives, but more false negatives)
Decreasing it would do the opposite.
PPV increases if ___ or ___ increases
prevalence or specificity
Calculating net sensitivity in sequential testing?
of people who tested positive on test2/# of people who truly had the disease
Calculating net sensitivity in parallel testing?
# of people who tested positive on BOTH tests + ppl who tested positive on test1 + ppl who tested positive on test2 / # of people who truly have disease
Calculating net specificity in sequential testing?
# of people who tested negative on first test + # of people who tested negative on second test / # of people who actually don't have disease
Calculating net specificity in parallel testing?
# of people who tested negative on BOTH tests / # of people who actually don't have disease
In calculating the kappa statistic,
How do you find “observed agreement (%)”?
It’s overall percent agreement (a+d)/(a+b+c+d)
In calculating the kappa statistic,
How do you find “Agreement expected from chance alone (%)”?
Apply Observer A’s percentages for positive and negative to Observer B’s totals for positive and negatives.
Now, add these results and divide by the total # of cases.
Kappa can have no agreement better than chance alone, poor agreement, intermediate agreement, or excellent agreement. What is the range for intermediate agreement?
0.4 - 0.75
Epidemic Curve
# of disease cases on Y-axis Time on the X-axis
Investigation of an outbreak
define the epidemic examine the distribution of cases look for combinations of relevant variables develop hypotheses test hypotheses recommend control measures
Annual incidence rate
of new cases in a specified population during a given year
/
Midyear population estimate
Mortality rate/ Annual Crude Mortality
of deaths in a specified population during a specified time period
/
Person-time contributed by the population during the time period OR midyear population
*Either way, this is a RATE- never expressed as a percentage
Indirect adjustment tells us how risk compares between one population and a standard population. How do you calculate it?
SMR/SIR =
Observed # of deaths or disease per year
/
Expected # of deaths or diseases per year
where expected # is determined by multiplying a known population’s rates by the study population size
What’s the difference between mortality rate from disease X and case fatality from disease X?
Both have the numerator as the people who died from disease X, but
Mortality rate has the entire population in the denominator.
Case fatality has only those who were diagnosed with disease X in the denominator.
Increasing proportionate mortality doesn’t necessarily mean the disease-specific risk has increased because
People may simply be dying of other causes, which makes the denominator (total all-cause deaths in population during time period) larger.
The first step in forming a cohort study’s study population is
to exclude prevalent cases from the reference population!
Confounding
Exposure is associated with other factors that might influence the outcome, thus distorting the association between exposure and risk
Information bias
AKA measurement error.
Systematic mis-measurement of exposure or disease, especially if the quality or extent of information obtained is different between groups.
Selection bias
Including nonparticipants or people lost to follow-up
Total population vs defined population vs study population
Total population: overall population group
Defined/source/reference population: Population that the trial is meant to be applied to
Study population: people actually in the study
Primary goals of randomization in cohort studies
- Removing investigators’ biases
- Increase study groups’ comparability in target and non-target characteristics (can also be done by having a larger sample size), though not always guaranteed
How can you guarantee comparability in cohort study arms/groups?
Stratify before randomizing!
Ex) Distribute half the men and half the women to the control group and the other halves to the experimental group so that both groups have equal amts of men and women.
Data and safety monitoring boards must:
- Ensure that any risks to participating are minimized
- Ensure integrity of data
- Stopping trials for safety reasons; if objectives won’t be met; or if it becomes clear objectives will be met and it’s unethical to continue
Factorial design
There are two treatments, so you have to randomize twice into four possible treatment groups.
Problems with randomized controlled trials
Nonparticipation
Noncompliance
Crossover
Loss to follow up
What are the two forms of noncompliance and how do you deal with it?
Drop-ins: subjects switch from control to treatment group
Dropouts: subjects stop adhering to their treatment
Deal with it by “intention to treat analysis”, which analyzes all patients together, regardless of whether or not they received the prescribed regimen.
Avoid breaking randomization by excluding/reclassifying noncompliancers (on-treatment analysis)
Efficacy vs effectiveness vs efficiency
Efficacy: how well does it work in the study environment
Effectiveness: how well does it work in the real world
Efficiency: cost-benefit analysis of a treatment once implemented in the real world
Number needed to treat
The number of patients who need to be treated to prevent one adverse event over a specified time period
NNT = 1/(CI control - CI treatment)
Preclinical vs clinical phase
Preclinical is before signs and symptoms start, but clinical is after.
Case fatality is not appropriate for what types of disease?
Long term chronic disease where death from other causes is likely
What is the issue with mortality rates?
Like incidence, it assumes that risk is constant such that person-time is contributed equal, regardless of WHEN it was contributed.
Ex) The experience of 1 person observed for 10 years = 10 ppl observed for one year each. It may look like risk of death is greater in the 10 people.
Ecologic studies
Are descriptive; they study the association between an exposure and an outcome in which the unit of analysis is groups of people, often defined by geography.
May be at one point in time or over several time points.
Ecological fallacy
erroneously assuming that aggregate (population) data applies to individuals.
You can never infer ___ from ecological studies because there’s no individual level data. It’s hypothesis-generating.
Causation
Two measures of association in ecological studies
Risk/rate ratio
Correlation coefficient (r or r^2): measures the strength of a linear relationship
Cross-sectional studies
Are observational and sometimes descriptive; they study exposure, disease, and/or the association between them all at the same point in time using individual-level data.
The measure of cross-sectional studies is
Prevalence ratio =
Prevalence of disease among the exposed
/
Prevalence of disease among the nonexposed
(Similar to cohort studies, but uses prevalence instead of incidence, so it uses existing, not new, cases)
Problems with cross-sectional studies
1) Temporality- Can’t tell if exposure came before disease, so you can’t confirm causation
2) Prevalent cases estimate both risk AND duration/survival. We may wrongly attribute exposure to risk, when it really might be causing survival.
3) Incidence-prevalence bias: because severe-disease samples will die more quickly than moderate-disease samples, we’re less likely to capture them.