Research Flashcards
type I error
false +
you see a difference but there really isn’t one
overcalling it
type II error
didn’t get a difference but there should have been one
p value affects
*establish likelihood of type I error
*smaller p value=more likely a true accurate difference
under calling it
CI
narrow = less variable data
correlation coefficient
0-1
1=better correlation
sensitivity
measure of true +
false negative = 1-SN
use a negative test to rule out
associated with -LR
specificity
measure of true negative
use a + SP to rule in
false + = 1-SP
PPV- positive predictive value
patients who test positive who actually have the disease
NPV- negative predictive value
patients who test negative who actually don’t have disease
incidence
rate of new cases in specific time
RATE OF CHANGE to express risk of disease
prevalence
of cases at a specific time
CANT predict probability because it’s not a rate of change
likelihood ratios
closer to 1 = less useful
+LR = Sn / (1-Sp)
-LR = (1-Sn) / Sp
odds ratio
odds of event in control group / odds in experimental
P vs alpha value- stat significance
P value must be lower than alpha to be statistically significant
level I evidence
high quality from RCT, diagnostic studies, prospective studies
level II evidence
lesser quality diagnostic studies, prospective studies, RCT
such as improper randomization, <80% f/u, no blinding
level III evidence
case control study or retrospective
level IV evidence
case series
level V evidence
expert opinion
specificity
ability to be negative when a variable is absent
measures the proportion of negatives which are correctly identified.
sensitivity
ability to be + when variable is present
very sn = good screening tool
proportion of those with a positive test given that they have the condition being tested for
LR values
+ LR
>10 = large probability of condition
5-10 = moderate
<5 = small
1 = no change
- LR
<0.1 = large probability
0.1-0.2 = moderate
>0.2 = small
1 = no change
likelihood ratio definition
The likelihood that a test result would be expected in a patient with the target disorder compared with the likelihood of the results with a patient without the disorder
Tells you how much a test result changes the pre-test probability of being correct
+ LR = how much to increase suspicion of condition based on + test
-LR = how much to dec suspicion of condition based on in test
how to control type II error
statistical power
increase number of subjects to decrease error
how to control type I error
alpha - significance value
Kappa coefficient
0-1
0= not reliable
1= perfect
<0.4 = poor
0.6=fair
0.75=good
>0.75=excellent
1=perfect
inter vs intra rater reliability
inter: all PTs will get same (think INTERnet connects many people)
intra: same PT will get same results each time
effect sizes- small/mod/large
- 0.2 - small
- 0.5 - moderate
- 0.8 - large effect size
- sample size is related to the effect size; lower effect sizes require higher sample sizes to detect meaningful differences
Levels of Evidence 1-5
- 1: High-quality, randomized clinical trial (RCT), prospective or diagnostic study Systematic reviews with homogeneity of RCT
- Lesser-quality RCT, retrospective study, cohort, or untreated control RCT Systematic reviews of cohort studies
- Case-controlled studies or systematic reviews of case-controlled studies
- Case series
- Expert opinion
Internal validity
The ability of a study to correctly measure and identify differences
- Exists when changes in the dependent variable are due to changes in the independent variable
- Indicates good control of the research design
Selection Bias
- Refers to the use of improper subjects for the study, who are usually not representative of the population being studied
- Threat to external validity
Odds ratio
Gives the probability or odds of an event happening or not
- Used in case-control and epidemiological studies, and are determined by dividing the incidence in one group by another comparison group
Relative Risk
Determined by dividing the proportion of the outcome or incidence of the treatment group by the incidence in the control group
- A value of 0 means there was no effect
- Relative risk of <1 indicates a reduced risk or effectiveness of the intervention
- Relative risk of >1 indicates no effect or increased risk.
SN
- Ability to correctly identify those with the condition
- Values are noted as the percentage of patients who have the problem and test positive
- If a clinical test for an ACL tear is done in 100 patients WITH a tear and the test is positive in 60 patients, it has a 60 percent sensitivity
SP
- Ability to correctly identify those without the condition
- Values are noted as percentage of patients who don’t have the problem and test negative
- If a clinical test for an ACL tear is done in 100 patients WITHOUT a tear and the test is was negative in 95, the
test has 95 percent specificity
Pearsons R
-1 to 0 to 1
-1 = perfect inverse relationship: increase in one variable dependent on decrease in other
Grades of evidence
- A: Consistent, Level I studies
- B: Consistent Level II or III, or extrapolation of from Level I studies
- C: Level IV studies or extrapolations from Level II or III studies
- D: Level V evidence or troubling, inconsistent, or inconclusive studies of any level
hawthorne effect
alter behavior when you know you are being studied
nocebo effect
negative expectations of treatment cause it to have even more negative effect
john henry effect
control group perceives they are at disadvantage, so they work harder or may seek other treatment
best way to fix is to blind PARTICIPANTS
pygmalion or rosenthal effect
belief of authority figure may change outcome
ie teachers thought certain kids had higher IQ, those kids showed greater improvements during the year than everyone else
how to prevent- blind PTs administering treatment AND those evaluating subjects