Reading...
Play button
Play button
Use LEFT and RIGHT arrow keys to navigate between flashcards;
Use UP and DOWN arrow keys to flip the card;
H to show hint;
A reads text to speech;
32 Cards in this Set
- Front
- Back
|
Do EBP principles for therapy apply to any intervention? If so, which ones?
|
Yes.
1) Therapeutic 2) Preventive 3) Performance enhancement |
|
|
For any given clinical task, a good question will contain at least 4 basic elements. What are they?
|
PICO:
1) The patient or problem 2) The intervention 3) The comparison intervention 4) Outcome(s) |
|
|
What is the first question we should ask ourselves when analyzing a study?
|
ARE THE RESULTS VALID??
|
|
|
What are some threats to validity?
|
1) Patients not randomized
2) Randomization was concealed 3) Baseline homogeneity (patients in treatment and control groups are similar with respect to known prognostic factors) 4) No blinding (patients, treaters, assessors) 5) Incomplete F/U 6) No "intention to treat" (every subject was treated in the group they were originally intended to be treated in) |
|
|
Why do you want randomization to be concealed?
|
Because you don't want people who are asking for volunteers to know which patient would end up in which group (leads to artificial inflation of effect sizes)
|
|
|
Is it better to blind the PT, patient, or measurer?
|
Measurer because it's very difficult to blind PT and patient about what treatment they're receiving but the measurer can always be blinded
|
|
|
The appropriate length of a study F/U depends on what?
|
The nature of the target disorder
|
|
|
What is imputation?
|
Making up fake data when you don't have any actually observed data (e.g. using the last available data in order to continue with intention-to-treat analysis)
-With nominal questions, assume the answer you DON'T want as a researcher because that increases validity of study |
|
|
In studies with 100% F/U (no dropouts), what is the per-protocol analysis?
|
Intention to treat (all patients analyzed in groups to which initially assigned)
ITT principle is satisfied by analyzing data for subjects completing the study (which happens to be every enrolled subject when there are no dropouts) |
|
|
What is the "TLC" principle?
|
When extra care or empathy is given to one treatment group and not the other
Related to blinding (creates bias) |
|
|
In the classical two-way mixed model ANOVA what are the 3 effects?
|
2 main effect p-values and 1 interaction effect
|
|
|
How do you interpret a significant interaction effect?
|
If there are changes across levels of one factor, those changes depend on the other factors
|
|
|
Which interaction effect is of the greatest interest to us?
|
The group x time interaction effect
|
|
|
What are some differences between raw and standardized effect sizes?
|
Raw has normal units while standardized doesn't have units
E.g. WOMAC scores in placebo group and therapy group - to get raw (within-group) effect size, subtract one group's scores (e.g. Placebo WOMAC scores) at week 4 from those at baseline To get the standardized effect size divide raw effect size by associated SD (which is why they're unit-less) |
|
|
What is the minimal clinically important difference (MCID)?
|
It is the smallest change in score associated with a patient's perception of a change in health status
It's important when looking at effect size and clinical relevance of research findings |
|
|
What attributes does MCID depend on?
|
1) Outcome scale
2) Clinical context 3) Clinical judgment |
|
|
What does a "positive trial" mean?
|
It means that the null isn't included because you've found that the results are statistically significant
Can then decide whether it's clinically meaningful (based on MCID) |
|
|
What does it mean if the MCID is below the average (CI)? How about above?
|
If it's below, you can be SURE the test is effective
If it's above, you're SURE it's NOT effective |
|
|
What can you conclude if the lower boundary of the CI is GREATER THAN the MCID?
|
The trial is a DEFINITIVE positive trial
|
|
|
What can you conclude if the CI includes values less than the MCID?
|
The possibility has not been ruled out that the treatment benefit might be trivial (not a definitive positive trial)
|
|
|
What does a "negative trial" mean?
|
Has a p-value > 0.5 and includes the null hypothesis
|
|
|
When you see that a trial is potentially negative, what part of the CI do you want to focus on?
|
The upper end of the CI because if the upper boundary excludes any important benefit to treatment, you can conclude that the trial is DEFINITIVELY NEGATIVE
|
|
|
What does it mean if the CI includes an important benefit when analyzing a potentially negative trial?
|
The possibility hasn't been ruled out that the treatment still might be worthwhile
|
|
|
For continuous variables (e.g. mean scores) what statistical test do you want to use to determine if the treatment group had a better outcome than the control?
|
Group x Time mixed-model ANOVA (p-value for interaction effect)
Also want the point estimate for post-treatment between group difference to be between means (with 95%CI) |
|
|
For dichotomous variables (e.g. death, fracture, failure to lose weight) how can you tell if the treatment group had a better outcome than the control?
|
Use risk reduction and number needed to treat (with 95%CI)
|
|
|
Dichotomous outcomes are also sometimes called _____
|
Events
|
|
|
Are events operationally defined as negative or positive outcomes?
|
Negative. An event is one patient with a bad outcome
|
|
|
When do you compute amount of reduced risk (ARR)?
|
When you get a bad outcome conferred by the treatment
|
|
|
Why do you compute number needed to treat (NNT)?
|
To see how many patients must be treated in order to prevent one bad outcome
|
|
|
What is controlled event rate (CER)?
|
The % of patients in the control group with a bad outcome
|
|
|
What is experimental event rate (EER)?
|
The % of patients in the experimental group with a bad outcome
|
|
|
How do you determine how much RRR or NNT is enough to care about?
|
It's a judgment call. Look at widths of CIs
E.g. NNT of 7 sounds good but not if CI is 1-200 E.g. RRR of 50% sounds good but not if CI is 3-97% |
