The flashcards below were created by user
kjp19
on FreezingBlue Flashcards.
-
: reports on the treatment of individual
patients and uses no control groups to compare outcomes (little statistically
validity)
Case series/case reports
-
·
patients
who already have a specific condition are compared with people who do not have
the condition. The researcher looks back to identify factors or exposures that
might be associated with the illness
Case control studies
-
·
identify
a group of patients who are already taking a particular treatment or have an
exposure, follow them forward over time, and then compare their outcomes with a
similar group that has not been affect by the treatment or exposure being
studied
Cohort studies
-
-
uses
information that has been collected in the past and kept in files or databases.
Patients are identified for exposure or non-exposure and the data is followed
forward to an effect or outcome of interest
Retrospective cohort (or historical cohort):
-
focus on a clinical topic and answer a specific
question. Summarizes results
Systematic reviews
-
mathematically combines the results and
reports them as if they were one large study
Meta-analysis
-
·
describes
the relationship between disease and other factors at one point in time in a
defined population. They lack any information on timing of exposure and outcome
relationships and include only prevalent cases and are often used for comparing
diagnostic tests
- Cross-sectional studies (aka: prospective, blind
- comparison to a gold standard):
-
·
describe,
explore and explain the health-related phenomena being studies
Qualitative research
-
Key issues in diagnostic studies
Diagnostic uncertainty
Blind comparison to gold standard
Each patient gets both tests
Patients should be included which have high, medium, low probability of disease
-
Key issues in prognostic studies
- Well-defined sample
-
- Similar prognosis
Objective and unbiased criteria
- Prognostic results can be presented in the following ways
- 1Absolute terms: 5 year survival rate
- 2Relative terms: risk from prognostic factor
- 2Survival curves: cumulative events over time
-
Key issues for harm studies
Similarity of comparison group
Outcomes and exposures measured same for both groups
Follow-up sufficient length (>80%)
For cohort studies: detection biases
- For case control studies: exposure is critical and blinding interviewer/using
- objective data can help eliminate bias
-
Key issues for systematic reviews
Focused question
Thorough literature research
Include validated studies
Selection of studies reproducible
-
Types of outcomes
- Categorical:
- complete, partial, no, etc (commonly represented as frequencies in tables)
- Continuous:
- lipid levels, A1c (but remember these can still be categorized!)
-
Comparing continuous variables
compares differences in means and assumes that you have a bell curve, independent
observations, large sample size (how likely is the differences between means
due to chance?)
t test
-
Comparing continuous variables
compares observations that are not independent (ex: measurements before/after treatment, matched studies, parent-child measurements)
Paired t test
-
Comparing continuous variables
assumed that you don’t have normal distribution
Nonparametric tests
-
Comparing continuous variables
Nonparametric tests
Unpaired data
Mann-Whitney U test and Wilcoxon Rank Sum test
-
Comparing continuous variables
Nonparametric tests
Paired data
Wilcoxon Signed Rank test
-
Comparing continuous variables
Nonparametric tests
ANOVA (analysis of variance):
- used to compare more than 2 means (ex: clinical trial with 3 arms or multiple
- measurements in 2 groups)
- Kruskal-Wallis test: nonparametric counterpart to ANOVA and is commonly used with small
- samples
-
large sample comparing continuous
unpaired data?
Paired data?
>2 groups being compared?
Unpaired data: T test
Paired data: paired t test
>2 groups being compared: ANOVA
-
small sample comparing continuous
normal distrubution
unpaired data?
paired data?
>2 groups being compared?
unpaired data: Mann-Whitney U test, Wilcoxon Rank Sum test
paired data: wilcoxon signed rank test
>2 groups being compared: Kurskal-wallis test
-
comparing categorical variables
comparison of observed vs expected frequencies?
used when expected frequencies in cells are small?
tests for linear association between row and column variables?
chi squared test: comparison of observed vs expected frequencies
fisher's exact test: used when expected frequencies in cells are small
mantel haenszel: tests for linear association between row and column vairiables
-
Experimental Group Risk (EGR)?
Control Group Risk (CGR)?
ExperimentalGroup Risk (EGR) = A/T1
and
Control Group Risk (CGR) = C/T2
-
Absolute risk reduction (ARR) or risk difference?
- : CGR-EGR
- Control Group Risk (CGR) - Experimental Group Risk (EGR)
-
Relative risk (cumulative incidence ratio, risk ratio): the risk of exposed/risk of unexposed *commonly used in RCT or prospective cohort studies
(example: the outcome occurs 3 time more in those exposed versus those
unexposed)
[EGR/CGR]
Experimental Group Risk (EGR)/Control Group Risk (CGR)
-
Relative risk reduction:
1- [EGR/CGR] x 100 (the increase in the ratio the more effective the therapy)
-
Relative benefit increase
:[EGR-CGR]/CGR
-
Hazard ratio
relative risk over a period of time
-
Odds ratio
- -
- odds exposed cases/odds exposed
- controls = [A/C]/ [B/D]*commonly
- used in case control or retrospective studies (example: cases were 3 times more
- likely to have been exposed than were control patients)
-
P-value < 0.05
means recect the null and the differences between the groups is not due to chance
-
Confidence intervals
differences in means: cannon include __
differences in proportions: cannot include __
differences in ratio measure (RR, OR, HR, PR) cannot include __
increase in confidence interval with decrease in sample size (less precise)
- differences in means: cannon include 0
- differences in proportions: cannot include 0
- differences in ratio measure (RR, OR, HR, PR) cannot include 1
-
sensitivity=?
specificity=?
- Sensitivity = true positive / all disease
- positives (probability that a person with the disease will have a positive
- result)
-
- Specificity = true negative / all disease
- negatives (probability that a person without the disease will have a
- negative result)
-
Likelihood ratio (LR)
(LR +) =
(LR-) =
- positive test in patients with disease (true +) / positive test in patients without disease (false +)
- negative test in patients with disease (false -) / negative test in patients
- without disease (true -)
-
LR + =
LR - =
Increases odds of having the disease after a positive test result
Decreases the odds of having the disease after a negative test result
-
confounding
Definition:
Addresses in 2 ways:
the association between the two groups may be due to differences in the control and experimental groups
randomization and multivariate analysis
-
Confounding
Regression analysis
good for controlling for multiple variables
Logistic regression: used for dichotomous outcomes and generates an odds ratio
Survival analysis/proportional hazards modeling/cox proportional hazard model: takes into account time to event
-
confounding
log rank test
cox proportional hazard model and log rank test
compares kaplan meier curves (small p value indicates that the curves are statistically different)
compare exposure groups but do not take into account other variables
-
|
|
