Epi/Biostats Flashcards
Negative predictive value
Chances that a person does not have disease when test is negative.
TN/(TN+FN)
Inversely proportional to pretest probability (and prevalence)
Incidence
Looks at new cases
Number of new cases per year / total population at risk (not including those who already have it)
How are prevalence and incidence related?
Prevalence = (incidence) x (time)
- This is given a stable population with little migration
How can a population have a stable incidence but a rising prevalence?
Improved quality of care means higher survival, so the same number of people get it but fewer die each year.
Number needed to harm
Number of pts exposed for 1 pt to be harmed.
NNH = 1/attributable risk
Attributable risk
The difference in risk between exposed and unexposed groups. Proportion of disease occurrences that are attributable to exposure.
AR = (event w risk factor / total w risk factor) - (event w no risk factor / total w no risk factor)
Absolute risk reduction
Control event rate minus treatment event rate. Difference in risk attributable to intervention, compared to control.
Just subtract the percentages!
Which of the following are affected by disease prevalence: sensitivity, specificity, positive predictive value, negative predictive value.
Only PPV and NPV
What kind of bias is loss to follow up, aka attrition bias?
This is a form of selection bias. Common when studying diseases with early mortality
Selection bias
Bias involved in recruiting and assignment to study group. Includes Berkson bias, attrition bias, and healthy worker/volunteer bias.
Berkson bias
A type of selection bias where a study only looks at inpatients
Healthy worker/volunteer bias
Selection bias where study populations are healthier than general population
Recall bias
Inaccurate recall of past exposure status based on having disease. Common in retrospective studies.
Can be reduced by getting info very soon after exposure.
Measurement bias
Information is gathered in a way that distorts it. Hawthorne effect is when people behave differently when being studied.
Reduced by using placebos and blinding
Procedure bias
Subjects in different groups are not treated the same
Observer bias
Researcher’s beliefs change outcome. Self-fulfilling prophecy. Prevented by blinding investigators
Confounding bias
A factor is related to exposure and outcome, but not on the causal pathway.
Pulmonary disease common in coal workers confounded by smoking, which is also more common in coal workers.
Reduce with multiple studies, crossover studies, and matching treatment and control groups
Lead-time bias
A screening test diagnoses a disease sooner than it would appear clinically. Falsely increases survival time, seen with improved screening tests.
Reduce by adjusting survival according to severity at diagnosis.
Cohort study
Compare a group with a given exposure to one without the exposure. Can be prospective or retrospective. Observational.
Given this exposure, who will develop disease?
Cohort study is good for measuring
Relative risk
Case-control study
Observational and retrospective. Compares a group with disease and group without disease.
Given disease status, who had exposure?
Case control study is good for measuring:
odds ratio
Cross sectional study
aka prevalence study. A snapshot in time, measures exposure and outcome simultaneously. Good for measuring prevalence. Inexpensive and easy to do.
Note: can show association, but NOT causality
Attributable risk percentage (ARP)
The excess risk in the exposed population that can be attributed to the risk factor.
ARP = (RR-1)/RR
Relative Risk
The risk of developing disease in the exposed group divided by risk of developing disease in non-exposed group. Used in cohort studies.
RR = [a/(a+b)] / [c/(c+d)]
Odds Ratio
Odds that the group with the disease had exposure, divided by odds that group without disease had exposure. Used in case-control studies.
OR = (a/c) / (b/d)
Crossover study
Subjects randomly assigned to a sequence of treatments given consecutively, with washout in between.
Each subject serves as his/her own control.
Relative risk reduction (RRR)
Percent reduction in absolute risk between treatment and control.
RRR = [control absolute risk - tx absolute risk] / control absolute risk
RRR = 1-RR
Incidence and deaths of cancers in women
Breast: 1st and 2nd
Lung: 2nd and 1st
Colon: 3rd and 3rd
Some common confounders
Gender, smoking, socioeconomic status
Confidence interval
Range of values in which a specified probability of means of repeated samples is expected to fall.
CI = mean +/- Z(SEM)
For 95%, Z = 1.96
For 99%, Z = 2.58
Remember that SEM = Std dev / sqrt(n)
Confidence interval for mean difference between 2 variables includes 0:
Null hypothesis is not rejected - no significant difference
Confidence interval for mean difference between 2 variables includes 1:
Significant difference - null hypothesis is rejected
Confidence intervals between two groups overlap/don’t overlap
If they overlap, there is no significant difference
If they don’t overlap there is a significant difference
95% CI, p=?
p=0.05
Standard error of the mean and standard deviation
Std dev: how much variability exists from mean in a set.
SEM: estimation of how much variability exists between sample mean and true population mean. SEM = std dev / sqrt(n)
Mean, mode, median in a positive skewed distribution
Asymmetry with longer tail on right
mean > median > mode
Mean, mode, median in a negative skewed distribution
Asymmetry with longer tail on left
Mean < median < mode
What does RR = 1 mean? Greater than 1? Less than 1?
RR=1 Null value - no association
>1 exposure is associated with increased disease occurence
<1 exposure associated with decreased disease occurence
If 95% CI contains 1.0:
If it does not contain 1.0:
Contains 1 = p>0.05
If it does not contain 1, p < 0.05
Most effective preventative intervention almost all the time
STOP SMOKING
In a gaussian curve, percentage that falls within 1, 2, and 3 standard deviations from mean.
1: 68% (16 on each side)
2: 95% (2.5 on each side)
3: 99.7% (0.15 on each side)
Null Hypothesis
No association with disease and risk factor
Hypothesis testing: correct result
Stating there is a difference when there really is one
Stating that there is not a difference when one doesn’t exist
Type 1 error (a)
False positive error. Study says there is a difference when there is none. alpha is the probability of making this error.
Type 2 (b) error
False negative error. Saying there is no difference when there really is one. Related to power (power = 1-b)
Power
(1-b). Probability of rejecting the null hypothesis when it is false.
Things that increase power, decrease b
Increased sample size, expected effect size, and precision
t-test
Check differences between two groups
ANOVA
Analysis of variance. Check difference between means of 3 or more groups.
Chi-square
Check difference between 2 or more percentages of categoracal outcomes, not means.
eg compare percentage of people of different ethnic groups with hypertension
Primary, secondary, and tertiary disease prevention
Primary: prevent occurrence (vaccination)
Secondary: screen early for disease
Tertiary: treatment
Correlation coefficient (r)
Measure of strength and direction of linear relationship between two variables. Between -1 and 1, closer to 1 = stronger correlation
Ofter reported as r^2, coefficient of determination