final exam Flashcards
The average value of a data set (the sum of all values in the dataset divided by the number of values)
Mean
the mean, while excellent at identifying the middle of a dataset can be skewed by _______. This can render a false impression of our dataset
outlier data points
the middle value in the data set.
Median
- In odd numbered datasets, this will always be a value in the data set.
- In even numbered datasets, the median is the average of the two data points closest
to the middle
the most commonly appearing value in your dataset (i.e. it appears more often than other data points)
Mode
Bimodal is the term used to describe what in statistics?
When there are two modal values in a data set
the broadness of the distribution- it is defined by the upper and lower values of a dataset
Range
A narrower range of data will produce a _____ curve, while a broader range of data will produce a _____ curve
Peaked, flattened
The tendency of a curve to “lean” one direction or another because of the distribution of the data.
Skew
A _____ (+/-) skew has a mean that leans to the left and has a long tail to the right.
positive
A _____ (+/-) skew has a mean that leans to the right and has a long tail to the left.
negative
The “peakness” of a curve
Kurtosis
The more “peaked” the curve describing the date, the tighter the range and closeness of the values. This is called ______
Positive Kurtosis
The flatter the curve, the broader the range of data points. This is called _____
Negative kurtosis
The measure of dispersion of data in a dataset (how spread out it is from the mean
Standard deviation
If the data points close to the mean the standard deviation will be small. If they are spread out away from the mean, the standard deviation will be larger
What is the benefit of using standard deviation when making inferences from data?
It is less sensitive to the effects of outliers and other anomalies of distribution and combines the qualities of mean, median and mode.
Standard deviation serves as a marker of what range of values we can expect to fall within ____% of the curve’s volume.
95
_____ is used as the best estimator of a dataset’s fitness to be compared to other datasets
Standard deviation
Most statistics in RCT’s rely on data being normally distributed. The standard deviation is a part of how we measure treatment effects against each other
Can no-normal distributions of data be characterized using standard deviations?
Yes, but but a statistical “correction” must be done to create a “normalization” of the data
Research results are measure in outcomes. What are outcomes?
Outcomes are numerical measures of results that can be calculated, compared and assessed
for their “truthfulness”
From outcomes, we want to know THREE THINGS:
- statistical significance
- precision
- clinical significance
What does it mean when outcome results have statistical significance?
range of results and the point estimate of their average is “true” and not due to chance
What does it mean when outcome results have precision?
the range of results is “tight” around the point estimate of their average, and not spread across a wide range
What does it mean when outcome results are clinically significant?
the results matter clinically
_______ are used to establish whether a point estimate of an outcome is likely to be due to chance or not.
P-values
When considering P- values, we generally want to see a result that has less than a ____% chance of being due to random chance or error. This would be reported as a P- value of _____.
5%, p = 0.05
P-values that exceed 5% likelihood of being due to error are not generally accepted as being “statistically significant”
P = 0.005 is a way of stating that the outcome has a _____% probability it was due to
chance (a _____% chance the results were NOT due to random error or chance)
1.2% : 98.8%
In health care , we generally use p= _____ as our standard maximum p-value (95% certainty that the outcome is not due to random chance or error)
. 05
The number of subjects in a study (N) can affect P values. How will a high N affect it?
the less likely that an
outlier will affect the results and the lower the probability that the
outcome is due to chance.
i.e. lower P-value
Does a low P- value predict success?
No, it simply says the outcome is very likely the truth
A ________ is done before a study to identify the number of subjects
needed to be certain the results are not due to anomalies in the statistics caused by too few observations (N being too low)
“power calculation”
To conduct a power study, what 3 things need to be identified in order to determine the pool size needed to detect the change in patient’s condition?
- smallest change in primary outcome that is worthwhile detecting
- how variable the results might be based on previous research
- least degree of acceptable accuracy
Individual practitioners mat be inclined towards interventions that in personal experience or patient preference have been beneficial, even if the p-values of available data are above 0.05. What other elements must be considered in the decision making for implementing those treatments that do not meet the statistical threshold?
- risks
- benefits
- costs
- alternative interventions
The more measures one does, the greater the likelihood that at
least one of them is positive for something. This is called the “blind squirrel phenomenon”. What causes the blind squirrel outcome?
- variation in measurement that produces false positive or negative
- variation in person, equipment, facility, etc
- interpersonal variation (ex: difference in pain tolerance)
- intrapersonal variation (ex:personal experience varying over time)
- random error/anomaly
The range of values of an outcome that will fall within the 95th percentile of significance
Confidence intervals
Confidence intervals tell us what two things?
- if the results are precise
- if the results are statistically significant
What does a wide range of values in a confidence interval tell us?
the result is statistically
significant, it is not precise
When reporting confidence intervals in the format:
.30 (95% CI 0.28 - 0.33)
What does 0.30 indicate?
Outcome
When reporting confidence intervals in the format:
.30 (95% CI 0.28 - 0.33)
What does 95% indicate?
Level of statistical significance
When reporting confidence intervals in the format:
.30 (95% CI 0.28 - 0.33)
What does CI indicate?
Confidence interval
When reporting confidence intervals in the format:
.30 (95% CI 0.28 - 0.33)
What does 0.28-0.33 indicate?
The range of values that fall within the 95th confidence interval
The wider the range, the lower the precision
If a result is statistically significant, does that mean that it is also precise?
No, Just because it is statistically significant does not mean that it is
precise and just because a result is precise, doesn’t mean it is accurate
What does it mean when the confidence interval presses the “no effect” line?
It is not statistically significant
What is Minimal Clinically Important Difference
This is the minimum amount of change necessary before we consider the difference to be clinically meaningful to the patient
How is minimal clinically important difference determined?
This is sometimes an individual decision; sometimes it is driven by prior research; sometimes it is stated as a consensus of experts. Regardless, it needs to be stated in a RCT. Better papers will explicitly state the threshold they used as the MCID
When an MCID has not been previously established, researchers will sometimes
use a ______ improvement as a rule of thumb
30% (or even 50%)
Which is more robust data, especially in the chronic pain literature; the difference in treatment effectiveness between two interventions or improvement over baseline (i.e., intra-group)?
Improvement over baseline
If two therapies perform nearly equally, what other factors should be weighed when choosing one therapeutic intervention?
- cost
- risk
- acceptability
- invasiveness
- availability
What is a risk ratio
a comparison of risk/probability of an outcome, desirable or not, in one group compared to the risk of the same outcome in a control group
What does a risk ratio of 1 mean?
No effect/no excess risk of a different outcome from one group to the other
What does a risk ratio above 1 indicate?
there is a greater risk of the outcome (usually a beneficial treatment effect) as compared to the control or other intervention group- good outcome
What does a risk ration below 1 indicate?
there was a lower likelihood of the desired outcome in the
experimental group than the control… bad outcome
Using a risk ratio, how would you state that the probability of the outcome doubled with the treatment being studied?
Risk ratio = 2.0
Using a risk ratio, how would you stat the that the treatment increases the probability of the outcome by 60%?
RR =1.6
Using a risk ratio, how would you state that the probability of the treatment producing a positive outcome was half of the comparison group?
RR = 0.5
What does this risk ratio mean?
fatigue (RR = 1.85, 95% CI [1.12 to 3.05]
The RR of patient developing fatigue with the treatment was 85% higher than the control
NOTE: this estimate is moderately precise due to CI
What does this risk ratio mean?
dizziness (RR = 1.99, 95% CI [1.17 to 3.37])
The relative risk of a patient developing dizziness was almost doubled with treatment compared to the control
NOTE: CI is not as precise due to range
What does this risk ratio mean?
visual disturbances (RR = 5.72, 95% CI [1.94 to 16.91]
The relative risk of a patient developing visual disturbances was 5.72 times greater in treatment group compared to control.
NOTE: CI was not precise AT ALL due to HUGE range
When comparing an exercise program and control (no exercise), and looking at recurrence of LBP, what would a RR of 0.43 indicate?
the risk for recurrence was reduced to
43% of what it would be without doing the exercises, so the relative Risk Reduction for this statistic would be 1 - 0.43= 0.57
I.e. the exercise program reduced the risk of recurrence of low back pain by 57% compared to no exercise/control
How do you calculate the relative risk reduction?
1 - RR = RRR
If RR = 0.6, then RRR = 0.4 which means that the risk of outcome was reduced by 40%
T/F: RR and RR reflect the probability of an event/improvement occurring
True
T/F: RR and RRR reflect the degree of the event/improvement
False, only reflects probability of outcome
T/F: RR and RRR can be generalizable to comparators other than the thing it was compared to.
False
T/F: RR and RRR offer perspective about the absolute improvement of one’s chances of achieving an outcome
False
What is ARR?
Absolute risk reduction - the % change of one population vs. % change of another population. it is the NET change in percentage change (risk) between the two groups.
What is the benefit of using ARR over RRR?
RRR can seem large, especially in low risk populations, and so it can create the appearance of value in treatment that may not be warranted. AAR reflects the net change in risk especially in low risk populations
When evaluating low risk populations, which is a more reliable indicator of treatment value, ARR or RRR?
AAR because if reflects net change in risk
What is NNT?
The number needed to treat in order for one additional patient to achieve a beneficial outcome compared to no treatment or alternative therapy
What is the perfect NNT?
1, because that means that every additional patient treated would achieve a beneficial outcome
We want treatments to have NNTs less than or equal to _____
5
Unless it is for a prevention study, in which case we accept much higher NNTs
Does NNT tell you how many people will respond, how large the effect is, or both?
Only how many people will respond
Does the NNT value indicate what counted as treatment success?
No, the value alone does not clarify between complete resolution or reduction in symptoms. More detail and context (especially what constituted success) should be reported along with the NNT value