Measure of Disease/Risk Flashcards
How is ratio measured? Example?
a:b
number of individuals diseased (cases) : number of not disease (non-cases/control)
4:1 - female:male in vet school
How is proportion measured? Example?
a / (a+b)
A fraction in which numerator is also included in denominator (# of case/ # of population)
How is probability measured? Example?
Proportion x 100 = %
Probability that you get heads when flip a coin is # of case / # of population = 1/2 x 100 = 50%
How is rate measured? Example?
Rate = a/x = disease / defined population at risk evaluated over a specific time period (a is included in x).
How to calculate crude mortality?
Death from all causes / population at risk for death
How to calculate cause-specific mortality?
Death from disease “X” / population at risk for death
How to calculate case fatality for disease “X”?
Death from disease “X” / Cases from disease “X”
Proportionate mortality for disease “X”
Death from disease “X” / death from all causes
Difference between prevalent cases and incident cases
Prevalent cases are all existing cases at a given point in time. Incident cases are new cases occurred during a given time interval.
How to calculate prevalent cases?
Number of cases observed at time t / total number of individuals at risk at time t
What measures the risk of being a case rather than becoming a case?
Prevalence
How does risk ratio measure?
Risk of disease in exposed group (either with or without outcome) / risk of disease in unexposed group (either with or without outcome)
Need to calculate 2 separate RISKs, then do a ratio
How does risk difference measure?
AKA attribute risk due to exposure factor.
Risk in exposed group - risk in unexposed group
Need to calculate 2 separate RISKs, then do subtraction
True but unknown values in populations called _______. We don’t know them unless we study 100% of a population.
parameters
We make interferences about the population _______ by measuring statistics on a _______ of the population.
parameter - sample
What is null hypothesis and how is it related to statistics?
Null hypothesis = assumes no difference, no association, equality of groups.
- Null hypothesis assume to be true, then we see how compatible our data are with null hypothesis.
- My logic: null hypothesis is true until proven wrong. Null hypothesis is another condition probability.
What is the question to ask when null hypothesis is assumed to be true at the beginning of the study?
What is the probability of me getting the data?
(You begin with a hypothesis and you ask about the probability of getting a data)
What is a p-value?
Probability of doing a study and obtaining the data actually collected (or data more extreme) when the null hypothesis is assumed to be true.
When it comes to probability of a scientific study, what probability are we talking about?
Probability of obtaining the data.
What does high and low p-value mean in terms of null hypothesis?
- High p-value tells us the data is compatible with null hypothesis.
- Low p-value tells us that we are UNLIKELY to observe the data when the null hypothesis is true. So in this case, null hypothesis can be rejected.
What is type 1 error?
Incorrect rejection of the null hypothesis.
What does “statistically significant” finding mean?
Finding that is real, reliable, and not due to chance.
Any difference can be found to be statistically _______ if the sample size is large enough.
significant.
(The bigger size the sample size, you’re more likely to get statistically significant result).
What is type 2 error?
Investigator fails to reject a null hypothesis that is actually false in the population (false negative).
What does confidence intervals tell you?
How confident that the confidence interval contains the true population parameter. “We’re 95% confident that the confidence interval contains the true population parameter.”
What is your sample profile for case-control study?
Select how many of cases and controls you want first (disease/undiseased), then measure exposure.
What can you NOT calculate in case-control studies?
You cannot calculate the incidence (aka risk) of disease in neither the exposed group nor the unexposed group in a case control study (because the number of cases and controls is determined by the investigator).
How to calculate odd of exposure among cases in case-control study?
Proportion of cases that are exposed divided by proportion of cases that are NOT exposed.
(A/(A+B)) / (B/(A+B)) = A/B
How to calculate odd of exposure among controls in case-control study?
Proportion of controls that are exposed divided by the proportion of controls that are NOT exposed
(C/(C+D)) / (D/(C+D)) = C/D
Cohort vs. case-control study
- If you assess the exposure first = cohort study
- If you assess the outcome first = case-control study
Define sensitivity
Ability of a test to correctly classify an individual as diseased when in fact that individual is disease.
A / (A+C)
Condition on HAVING the disease.
P (test pos | dz pos)
Define specificity
Ability of a test to correctly classify an individual as not diseased when in fact that individual is disease-free.
Condition on individual NOT having a disease. P (test neg | dz neg)
D / (B+D)
Define true prevalence
How common is a disease in a population at some point in time. A sense of inventory at that time.
(A+C) / (A+B+C+D)
Dz pos / total population
Define apparent prevalence
How common is a positive test result in a population at some point in time.
Test pos / total population
(A+B) / (A+B+C+D)
Define positive predictive value
Average conditional probability that a patient whose test is positive actually has the disease.
P (dz pos | test pos)
A / (A+B)
Conditioned on Positive test result
Define negative predictive value
Average conditional probability that a patient whose test is negative is actually disease-free.
P (dz neg | test neg)
D / (C+D)
Conditioned on Negative test result
What is unconditional probability?
The probability of an event/outcome occurring that does NOT depend on any other results, findings, or states.
P(A)
Why average risk?
Because risk is the unconditional probability of developing disease in SOME specified period of time.
- Average risk = sum (new case/population at risk)
Review your 2x2 table
Dz in row, Test in column
A (true pos) - B (false pos)
C (false neg) - D (true neg)
MATH question: 99% sensitivity, 99% specificity, 25% prevalence. What is the P (dz pos | test pos)?
Assume total is 100K. Calculate prevalence yields 25K for Dz (+) and 75K for Dz (-). Then use 99% sensitivity to calculate A/(A+C) and 99% specificity to calculate D/(B+D).
Answer: A / (A+B) = 97.1%
MATH question: 99% sensitivity, 99% specificity, 0.1% prevalence (rare disease). What is the P (dz pos | test pos)?
Assume total is 100K. Calculate prevalence yields 100 for Dz (+) and 99,900 for Dz (-). Use 99% sensitivity to calculate A / (A+C) and 99% specificity to calculate D / (B+D).
Answer: A / (A+B) = 0.090 = 9%
What does high and low prevalence mean?
- High prevalence: positive test is more likely true positive case.
- Low prevalence: positive test is more likely to be a false positive non-case (dz negative).
Accuracy
= the total number of true positives and true negatives DIVIDED by the total number of tests
Explain valid vs. precise?
- Precise: How close the measurements are to each other.
- Valid: The extent to which the study measures what it is intended to measure. Are the values describing what was supposed to be measured?
Distinguish between experimental and non-experimental (also called observational) study design
- Experimental studies: investigator intervenes and assigns exposure (tx) to subjects.
- Non-experimental or observational studies: investigator observes and records exposure and outcome.
Understand the role of a control group in experimental study designs (example clinical trial) compared to in an observational study (such as case-control)
???
Understand the concept of randomization of a treatment or exposure to patients in a clinical trial
Randomization = assign participants to treatment and control groups, assuming each participant has equal chance of being assigned.
Randomization avoids confounding. Larger sample size achieves comparability between groups.
List the major types of observational studies
- Cohort
- Case-control
- Cross-sectional
Distinguish between two important observational study design, namely the cohort study Vs case-control study
- Cohort: start as a single cohort of disease-free subject. Assess their exposure. Followed over time. Disease (outcome) determined in exposed vs. unexposed subjects.
- Case-control: Assess their disease (outcome) first. Followed over time to determine exposure status. Controls provide frequency of exposed and unexposed in source population.
Define and identify confounders
Cofounding is bias that creeps into the study because of the study design - it’s not true in nature. It is attributing an effect to a factor when it is due to another correlated factor.
Understand ecological bias associated with ecological studies
Ecological fallacies (biases) occur when we try to draw conclusions about individuals based on data collected at the group level.
How to calculate odd ratio?
Odds that a case was exposed / odds that a control was exposed
AC / BD
How to calculate odd of exposure in cases?
number of cases with exposure / number of cases without exposure
E.g.: number of azotemia with methadone exposure / number of azotemia without methadone exposure
How to calculate odd of exposure in controls?
number of controls with exposure / number of controls without exposure
E.g.: number of healthy with methadone exposure / number of healthy without methadone exposure
95% confidence interval for the odd ratio include 1 in the odd ratio. What does it mean?
Because the lower bound of the 95% confidence interval includes 1.0, we cannot conclude with 95% confidence that the true (population parameter) odds ratio does not equal 1.0. This also means that a statistical test of the null hypothesis that the odds ratio = 1 would not be significant at the 5% level of significance.
95% confidence interval for the odd ratio EXclude 1 in the odd ratio. What does it mean?
Because the lower bound of the 95% confidence interval excludes 1.0, we can conclude with 95% confidence that the true (population parameter) odds ratio does not equal 1.0. This also means that a statistical test of the null hypothesis that the odds ratio = 1 would be significant at the 5% level of significance.
