Exam 2 BIO Flashcards
Case control
disease is rare and exposure is common
Cohort
disease is common and exposure is rare
case-control
disease has a long latent period and the exposure information is very expensive to obtain
odds ratio
AD/BC
interpreting odds ratio
compared to normal, patients with ___ had 3.9 times the odds of having _____
measures of association
comparison of disease frequency
absolute measure
calculate difference between 2 measures of disease freqeuncy
relative measure
calculate the ratio of 2 measures of disease frequency
if there is NO association between the exposure and disease then,
RD=0
if the exposure is associated with increased risk of disease then,
RD>0
if the exposure is associated with DECREASED risk of disease then,
RD<0
Cumulative Incidence (CI) -Risk difference
Exposed (Disease) -Unexposed (No Disease)
If RR is less than 1 (like 0.55), then
(1-RR), so 1-0.55=0.45—–45% decrease in risk (in whatever like getting a vaccine)
Risk Ratio
Ratio of cumulative incidence between (exposed/unexposed)
Excess Relative Risk
(RR-1) X 100%—-meaning ___ % have an increased risk
RD is Absolute Measure
Represents measure of public health impact of exposure on disease occurrence
RR is Relative Measure
Represents measure of strength or magnitude of the association between an exposure and a disease—-used in etiologic research
There are 2 types of epidemiology
Descriptive and Analytic/Scientific
Descriptive
-Identifies and counts cases of disease/population according to person, place, time
Case reports and series
Cross-Sectional study
Ecologic Study
Analytic/Scientific
-Compares group and systemically determine if there is association
Clinical trial
Experimental Study
Case-Control Study
Cohort Study
Prospective Cohort Study
Weakness: -More expensive, time consuming, and not efficient for diseases with long latent periods
Strength: Better exposure and confounder data and less vulnerable to bias
Retrospective Cohort Study
Weakness: Exposure and confounder data may be inadequate and more vulnerable to bias
Strength: Cheaper, faster and efficient with diseases with long latent period
Start of the study: Compare incidence of disease
Cohort Study Limitations
may need large number of subjects to be followed for long periods of time-difficult, expensive and time consuming
Undermines validity (loss to follow)
Not good for rare diseases/long latency
Not good when exposure data are expensive to obtain
Case-Control Study
Investigator identifies cases and selects controls who represent a sample of the population —>compares exposures
Cohort Study
Investigator identifies exposed and unexposed groups—->compares incidence of outcome
When to conduct a case-control study?
-When exposure data are expensive or difficult to obtain
-when disease has long latent period/decades for results
-disease is rare
-population is difficult to follow/high loss
-little is known about the disease/need to evaluate many exposures
Case control studies need source population:
FIxed or Dynamic but are good in dynamic populations
Controls
to estimate the exposure distribution in source population that produced the cases
must come from same source population
must be selected independently of exposure
Would Criterion
Would it be enrolled as a case? if yes then selection bias is unlikely
To be yes: controls need to have SAME population as the cases that were selected
Nested Controls
Controls selected from an existing cohort population (sub-set)
Population Based Controls
Controls selected from general population-suitable when cases are from well defined geographic area
Hospital or Clinic Based Controls
Controls selected from among patients at a hospital or clinic
(should be unrelated to exposure , should not cause the exposure)
3 ways to sample controls in case-control study
- survival sampling
- Case-Cohort Sampling
- Risk-Set Sampling
Survivor Sampling
Select Controls at END of follow-up
Case-Cohort Sampling
Select controls from starting Cohort at beginning of follow-up
Risk-set sampling
Longitudinally select controls as cases arise during follow-up
Risk-set sampling
Longitudinally select controls as cases arise during follow-up
Case-Control Studies Strength
Fewer ethical concerns, more efficient than cohort study (less time, fewer subjects needed, no need to wait for long latency diseases to develop), easy to explore effect of many exposures on an outcome (ID outbreak, or disease with little info)
Case-Control Studies Limitation
limited to a single outcome, inefficient for rare exposures, more opportunity for systematic bias, cannot calculate absolute measure of association
Nonparametric tests
when data is not normally distributed
When to use nonparametric tests
-when sample size is too small
-when data has outliers that cant be removed
-when you want to test for median rather than the mean (skewed distribution)
-Outcome is continuous and not normally distributed
Non-parametric tests include:
Spearman Rank Correlation, 1-sample sign test, mann-whitney test, wilcoxon signed rank test, kruskal-wallis test, friedmans ANOVA
Use nonparametric if sample size:
Less than 20
Mann-Whitney U test
Test 2 independent samples with the dependent ordinal/continuos not assumed to follow a normal distribution
If H0 (null) two populations are equal; if H1 , two populations are not equal
Sign Test (pos/neg)
-Continuos outcome measured in matched or paired samples
-SUBTRACT After - Before
-Can be used for ordered (ranked) categorical data
H0=Median difference is zero
H1=Median difference < or > zero
Reject H0 if the smaller of the number of pos/neg signs is less than or equal to the critical value
Wilcoxon Signed Rank Test
-Examines signs, score differences, and magnitude of observed difference
-SUBTRACT Before - After
-Test statistic is W and we use the smaller of W+ and W-
Reject H0 if W is < or = to critical value
Kruskal Wallis Test
sample size does not need to be equal
K>2
H0=K population median are equal
H1=K population median are not equal
Test statistic is H
Reject H0 if H is >/= critical value
logistic regression
used when the independent variables include both numerical and nominal measures and the outcome variable is BINARY (Dichotomous)
-YES/NO
-Sruvived/Deceased
Logistic Regression Variables:
Two Indepedents: X1=dichotomous and X2=Continuous
Outcome Variable: Y=Dichotomous
Odds Ratio
If OR is >1: greater odds of association of exposure and outcome
If OR=1: there is no association between exposure and outcome
If OR<1: there is a lower odds of association
If CI includes 1 then results are not statistically significant
If OR=1.2
1-1.2 times 100=20% chance increase in odds of an outcome happening
If CI includes 1 then
cant be significant because can be either lower or higher cant pick a side in other words
Analyzing Hazard Ratio: HR=0.87
(1-HR)X 100= 13% reduction
Survival Analysis
situations where the outcome is the length of time that elapses until the event of interest occurs
For example:
1. Length of time until death following diagnosis of a disease
2. Length of time to cancer remission
3. Length of time until death following initiation of therapy
Survival Analysis what does it measure?
Outcome is time to event
measures whether person has event or not (Yes/No) and time to event
estimate survival time
determine factors (pt characteristics) associated with longer survival
survival analysis used to:
determine if the risk of experiencing a particular event over a specific period of time differs between groups
Log rank test
determines if the difference between two survival curves is statistically significant
-the larger the difference, the more likely it is statistically significant
Survival Curves:
Estimated using Kaplan-Meier Method and compared statistically using log-rank test
hazard ratio
measures an effect of the intervention on an outcome of interest over time
If negative outcome (death) and hazard ratio <1:
this is desirable and is a percent reduction in risk
if HR was 0.25, then 75% were less likely to die
Cox Proportional Hazards Regression Model
outcome is still time to event but it is an extension to survival analysis by including control variables in the regression model
Repeated Measure ANOVA
technique used to test equality of means
-Need to measure 3 items
When to use repeated Measure ANOVA
measuring performance on the same variable over time
(changes in performance in training or after a treatment)
-same subject is measured multiple times under different conditions
-same subject provides measures/ratings on different characteristics
superior trial
aims to show that the new drug is better than standard treatment
-like traditional hypothesis testing
-if 0 is not contained within the CI we can reject H0–>concluding new drug is superior
Non-inferiority Trial
Aims to show that the new drug is no worse (not inferior) than standard treatment (existing med) by a prespecified amount.
Prespecified amount=non-inferiority margin and is sometimes expressed as M
-There is no placebo group: control group takes existing medication that already demonstrated effectiveness
-compare CI to M2 and to zero
two noninferiority margins (M1 and M2):
M1 reflects the difference between control drug and placebo. Reflects the full efficacy of control drug
M2 a smaller amount that reflects how much efficacy the investigators are willing to give up in return for other advantages the new drug may offer.(like improved safety)
Equivalence Trials
aims to show that the new treatment is no better and no worse
-rejecting the H0 is equivalent to concluding that the test drug and control drug DO NOT DIFFER from one another