Unit 3 Flashcards
What are the 3 attributes of study variables?
Order / Magnitude
Consistency of scale / equal distances
Rational absolute zero
What are the 3 levels of data measurement?
Nominal
Ordinal
Interval/ Ratio
Nominal
Dichotomous/ Binary
Non-ranked (non-ordered)
Named categories
- categorical data - can be more than 2 categories - ex: 2 genders, 2 age groups
No order/ magnitude
No consistency of scale or equal distances
Nominal variables are simply labeled- variables without quantitative characteristics
Examples of Nominal variables
What is your gender?
- Male or Female
What is your hair color?
- brown vs black vs blonde vs grey vs other
Education level (if made binary)
Smoking vs non-smoking
Ordinal
Ordered and order-able
Rank-able categories
Non-equal distance between ranges
- technically can be equal and unequal
Unitless
Yes order/ magnitude
No consistency of scale or equal distances
- No units or scales - No even spacing between them
Data is collected in categories and can be ordered
Examples of Ordinal variables
Pain Scales
- patient decides what each value means
Strongly agree > somewhat agree > Neither > somewhat disagree > strongly disagree
SES
- unitless, broken into categories
Interval/ Ratio
Order/ magnitude
Equal Distances
- unitless
- equal spaces between scales
Interval
- Arbitrary 0 value - 0 doesn't mean absence - Can be 0 or negative values
Ratio
- Absolute 0 value
- 0 means absence of measurement value
- No negative values
- Ex: physiological parameters
- blood pressure
- blood sugar
Examples of Interval/ Ratio variables
Living siblings and personal age
Height in cm
Speed in m/s
LDL in mg/ dL
Mean
Average value
Median
Middle value
Mode
Most common value
This is the most useful measurement for descriptive statistics
What is descriptive statistics?
Tells us about our population
Describes our population
Range
Maximum - minimum
Interquartile Range
Top 25% = Q3
Bottom 25% = Q1
Middle 50% = Q3 - Q1
- represented the 25% above and below the mean
Variance
The average of the squared differences in each individual measurement value and the groups mean
Describes the spread of data
Variance from the mean
Standard Deviation
Square root of variance
Restores units of mean
Describes spread of data
Normal Distribution
Symmetrical
Mean and median are (almost) equal
Equal dispersion of curve (tails) to both sides of mean
Statistical tests useful for normal- distributed data are known as _____
Parametric Tests
Required assumptions of interval/ ratio data for proper selection of parametric tests
- Normal distribution
- Equal variances
- use Levene’s Test
- Randomly derived and independent
Levene’s Test
Test used to calculate if data is normally distributed and has equal variance
Used to assess if the variances are different between groups
Null Hypothesis: groups are equal
Tries to show that there is a difference between groups
How to handle interval data that is not normally distributed
- Use a statistical test that does not require the data to be normally distributed
- non-parametric tests
- step down and run ordinal test
- Transform data to a standardized value
- hope that the transformation allows data to be normally distributed
- z score or log transformation
Positively Skewed
Asymmetric distribution with one tail longer than the other
Mean > median
- mean is higher than median
Tail points to the right
Negatively Skewed
Asymmetric distribution with one tail longer than the other
Mean < median
- mean is lower than medium
Tail points to the left
What effect do outliers have on skewness?
Outliers pull the tails out farther
Contributes to skewness
Skewness
A measure of the asymmetry of a distribution
Perfectly normal distribution is symmetric
- skewness = 0
Negative skewness = negatively skewed data
Positive skewness = positively skewed data
Kurtosis
A measure of the extent to which observations cluster around the mean
- how peaked a value is
Kurtosis of normal distribution curve = 0
Positive kurtosis = more cluster (around a number)
- closer to a positive value
Negative kurtosis = less cluster (around a number)
- closer to a negative value
68%
1 standard deviation away from the mean
95%
2 standard deviations away from the mean
99.7%
3 standard deviations away from the mean
Null Hypothesis
Research perspective which states there will be no true difference between the groups being compared
Most conservative and most commonly utilized
At end of study, either need to accept or reject the null
Can take on the superiority, noninferiority, and equivalency perspectives
Alternative Hypothesis
Research perspective which states there will be a true difference between the groups being compared
Type 1 error
Alpha
Not accepting the null hypothesis when it is actually true and you should have accepted it
Rejecting the null hypothesis when you shouldn’t have
Ex: Telling a man he is pregnant
Type 2 error
Beta
Accepting the null hypothesis when it is actually false and you should not have accepted it
Not rejecting the null hypothesis when you should have
Ex: Telling a pregnant woman she is not pregnant
P value
Probability value (alpha)
Based on the probability, due to chance alone, a test statistic value as extreme or more extreme than actually observed if groups were similar (not different)
Represents your chances of being wrong
If p < 0.05, risk of experiencing a type 1 error is acceptably low
T/F: If p value is lower than the pre-selected alpha (5% or 0.05), it is statistically significant
True
Do you accept or reject the null hypothesis if p value < alpha?
REJECT
What is the interpretation of a p value?
The probability of making a type 1 error if the null hypothesis is rejected
If the data is statistically significant and there are 3+ groups, the p value tells you what?
Tells you that there is at least 1 difference present
Guaranteed difference between control and most extreme value
Lowest and highest value represent the difference
At baseline, do we want groups to be equal?
Yes
At start of study/ baseline, p value should be 1.0
Want p values to start above 0.05
Power
The statistical ability of a study to detect a true difference if only one truly exists between group comparisons and therefore the level of accuracy in correctly accepting/ not accepting the null hypothesis
If there is truly a difference between groups, study has high power
Studies are set up to have 80% power
When lose people, lose power
What is the mathematical representation of power?
1 - beta
= 1 - type 2 error rate
We allow a type 2 error rate of 20%
- accept 20% of risk of finding an error
1 represents sample size
The _____ people the study has, the _____ the power
More
Higher
What are the common elements utilized in determining sample size of a study?
Minimum difference between groups deemed significant
- the smaller the difference between groups necessary to be considered significant, the greater the sample size needed
Expected variation of measurement
Alpha and beta error rates and confidence interval
How does sample size affect power?
The larger the sample size, the greater the likelihood of detecting a difference if one truly exists
Increases power
What is the number one way you can ensure your study has power?
To show the difference between groups if a difference is really present
When is it okay for p values to be non-statistically significant?
Start of study
Levene’s Test
What is a caveat of p values?
They do not tell us about spread/ dispersion
Confidence Intervals
Precision measurement
CI around a group’s differences help reader understand where true value may lie
Calculated at an a priori percentage of confidence that statistically includes the real (yet unknown) difference or relationship being compared
What advantage do confidence intervals have over p values?
Tells us about statistical significance and spread
What are confidence intervals based on?
Variation in sample
Sample size
If you don’t use the same directional word when interpreting CI, is the data statistically significant?
NO
If CI values cross 1.0, data is not statistically significant
When the numbers that make up the CI are on the same side of 0, are the numbers statistically significant?
YES
Why is it okay to have a high p value for a Levene’s Test?
We are okay with accepting the null because it means we are saying the variances are equal so we can thus use an interval test with our data.
The bigger the CI, the ____ confidence we have in the precision of our estimate. Why do we see this?
Less
With 99% CI, it is harder to show significance
- less likely to cause type 1 error - more confident that true value lies in interval
On Forest plots, the horizontal lines represent _____ and the boxes represent _____.
Confidence interval
Values (our data point)
What is an important question to consider once you have determined your data is statistically significant?
Does statistical significance actually confer meaningful “clinical” significance?
What questions do we ask when selecting the correct statistical test?
- What data level is being recording?
- Does the data have order or magnitude
- Does the data have an equal, consistent distance along the entire scale?
- What type of comparison/ assessment is desired?
- Frequencies/ counts/ proportions
- How many groups are being compared?
- 2 or 3 or more groups
- Is the data independent or related (paired)?
- data from the same groups = paired
- data from different groups = independent
Correlation
Provides a quantitative measure of the strength and direction of a relationship between variables
Tells you if there is a relationship or not
- direction ( + or -)
- magnitude (strong or weak)
Partial Correlation
A correlation that controls for confounding variables
What does a correlation of - 1.0 tell us and look like graphically?
Perfectly negative
As 1 variable goes up, the other goes down
- 45 degree slope
What does a correlation of + 1.0 tell us and look like graphically?
Perfectly positive
As 1 variable goes up, the other goes up
+ 45 degree slope
Nominal Correlation Test
Contingency Coefficient
Ordinal Correlation Test
Spearman Correlation
Interval Correlation Test
Pearson Correlation
p > 0.05 for a Pearson Correlation means there is no linear correlation but there could be non- linear correlations present
Survival Tests
Deal with event occurrence / time-to-event
Compares the proportion of events over time or time-to-events between groups
Predict changes in event over time
Asks ‘does frequency of events between groups differ over time?’
- amounts can be different
- rates can be different
Used when we want to see changes over time
Survival refers to an event that has or has not occurred yet?
Has NOT occurred yet
Examples: Death, hospitalization
What kind of curve represents survival tests and what are the features of it?
Kaplan- Meier curves
If everyone starts at 100, means every patient is free of event
If everyone starts at 0, means no one has had the event
Nominal Survival Test
Log- rank test
Ordinal Survival Test
Cox- proportional hazards test
Interval Survival Test
Kaplan- Meier Test
Assesses time
- time as a variable
Regression Test
Used to predict the likelihood of an outcome/ association
Allows us to used multiple variables at one time to determine an outcome
- predicting an outcome given a bunch of variables
Provide a measure of the relationship between variables by allowing the prediction about the dependent (outcome) variable knowing the value/ category of independent variables
Can be used to calculate the odds ratio for a measure of association
Nominal Regression Test
Logistic Regression
Ordinal Regression Test
Multinomial Logistic Regression
Interval Regression Test
Linear Regression
Pearson’s Chi- square Test
Compares 2 groups of independent nominal data
Compares group proportions and if they are different from those expected by chance
Not powerful for small numbers
- no expected cell count of less than 5 observations
Asks “is the proportion of yes or no responses between groups any different?”
Compares real numbers with what they expect to find based on a skewed Chi-square curve
Chi- square Test of Independence
Used for 3+ groups of independent nominal data
Compares group proportions and if they are different from those expected by chance
Not powerful for small numbers
- no expected cell count of less than 5 observations
Compares real numbers with what they expect to find based on a skewed Chi-square curve
Fisher’s Exact Test
Used for 2+ groups with expected cell counts of less than 5
Nominal data test
Powerful for small numbers
Still Chi- square like test but handles the problem of small numbers
Bonferroni Test of Inequality
Used for 3+ groups of independent nominal data
Used to find which groups are different in nominal data set
Adjusts the p-value for the number of comparisons being made
= p -value / number of comparisons you are making
Very conservative
Helps us reduce risk of type 1 error
If have statistically significant results in 3+ comparisons, you need to perform _______ ?
Subsequent analysis to determine which groups are different
- subsequent analysis = post- hoc testing
Cannot use multiple chi- square/ multiple statistical tests
Why can’t you use multiple chi-square/ multiple statistical tests to determine which groups are different?
Risk of type 1 error increases with each additional test
McNemar Test
Used for 2 groups of paired/ related nominal data
Just tells us that there is at least 1 difference between groups
- doesn’t tell us where but can use a Bonferroni test of inequality/ correction to find where differences are
Same principles/ assumptions as Chi- square
Cochran Q Test
Used for 3+ groups of paired/ related nominal data
Just tells us that there is at least 1 difference between groups
- doesn’t tell us where but can use a Bonferroni test of inequality/ correction to find where differences are
Same principles/ assumptions as Chi- square
What are examples of key words for paired data?
Pre vs post
Before vs after
Baseline vs end
Start vs end
Mann- Whitney
Used to compare the median values between 2 groups of independent ordinal data
Also used for interval data not meeting parametric requirements
Kruskal- Wallis Test
Used to compare the median values between 3+ groups of independent ordinal data
Also used for interval data not meeting parametric requirements
If 3+ group comparison is significant, must perform a post- hoc test to determine where differences are
Wilcoxon Signed Rank
Used to compare median values between 2 groups of paired/ related data
Also used for non-normally distributed interval data or data that don’t meet all parametric requirements
Neg sign = rank moved down
Pos sign = rank moved up
Friedman Test
Used to compare the median values between 3+ groups of paired/ related data
Also used for non-normally distributed interval data or data that don’t meet all parametric requirements
If 3+ group comparison significant, must perform a post- hoc test to determine where differences are
Student- Newman- Keul Test
Post- hoc test that is used for 3+ group comparisons of ordinal and interval data when they are statistically significant
Compares all pair-wise comparisons possible
All groups must be equal in size
Dunnett Test
Post- hoc test that is used for 3+ group comparisons of ordinal and interval data when they are statistically significant
Compares all pair-wise comparisons possible against a single control
All groups must be equal in size
Dunn Test
Post- hoc test that is used for 3+ group comparisons of ordinal and interval data when they are statistically significant
Compares all pairwise comparisons possible
Useful when all groups are not of equal size
- Account for lost to follow ups
Tukey Test
Post- hoc test that is used for 3+ group comparisons of interval data when they are statistically significant
Compares all pairwise comparisons possible
All groups must be equal in size
More conservative than Stident- Newman- Keul test
Scheffe Test
Post- hoc test that is used for 3+ group comparisons of interval data when they are statistically significant
Compares all pairwise comparisons possible
All groups must be equal in size
Less affected by violations in normality and homogeneity of variances
Most conservative
What are the Post- hoc tests that can be used when 3+ group comparisons of ordinal data are statistically significant?
Student- Newman- Keul Test
Dunnett Test
Dunn Test
What are the Post- hoc tests that can be used when 3+ group comparisons of interval data are statistically significant?
Student- Newman- Keul Test Dunnett Test Dunn Test Tukey Test Scheffe Test
Student t- Test
Used to compare the means of 2 groups of independent interval data
ANOVA (Analysis of Variance)
Used to compare the means of 3+ groups of independent interval data
Also compares intra- and inter-group variations against a dependent variable
If 3+ group comparisons are significant, must perform a post- hoc test to determine where the differences are
ANCOVA (Analysis of Co- Variance)
Used to compare the means of 3+ groups of independent interval data with confounders
Compares intra- and inter- group variations of related data against a dependent variable while also controlling for the co-variance of confounders
Repeated Measures ANOVA
Used to compare the mean values of 3+ groups of paired/ related interval data
Also compares intra- and inter-group variations of related data against a dependent variable
If 3+ group comparison is significant, must perform a post-hoc test to determine where differences are
Repeated Measures of ANCOVA
Used to compare the mean of 3+ groups of paired/ related interval data with confounders
Kappa Statistic
A correlation test showing relationship or agreement between evaluators
Consistency of decisions/ determinations
K = +1.0
Observers perfectly classify everyone exactly the same way
Good agreement
K = 0.0
No relationship at all between the observers’ classifications above the agreement that would be expected by chance
K = -1.0
Observers classify everyone exactly the opposite of each other
Poor agreement
What is the National Clinical Trials number (NCT)?
A unique identifier number assigned by clinicaltrials.gov once research protocol is submitted prior to study initiation
A number all studies are given once registered that allows you to track progress of study
Allows us to know what’s going on and why its going on
What is the purpose of the NCT number?
Reduce publication bias
- not sharing results because they are bad/ don’t show anything
Clinicaltrials.gov is an ___ ___ ___ ___ ___ ___ which offers what kind of information?
International Committee of Medical Journal Editors
Acceptable public registry that offers up to date information for locating interventional studies
What do readers need in order to accurately assess a study?
Complete, clear, and transparent information on the study’s methodology and findings
What does a checklist help when reviewing medical literature?
Provides a stepwise, systematic review of the published medical literature
What type of studies is a CONSORT checklist used for?
Randomized interventional trials
Interventional Studies
What does CONSORT stand for?
Consolidated Standards of Reporting Trials
CONSORT can be extended and used for _____?
Non-inferiority Trials
Equivalence Trials
Cluster Trials
Pragmatic Trials
Define: Pragmatic Trials
Randomized, controlled trial whose purpose is to inform decisions about clinical practice
Philosophy as a continuum not a dichotomy
What type of studies is a PRISMA checklist used for?
Systematic reviews of multiple randomized trials
Interventional Studies
What does PRISMA stand for?
Preferred Reporting Items for Systematic Reviews and Meta-Analyses
What type of studies is STROBE used for?
Observational Studies
- cohort, case-control, cross-sectional
What does STROBE stand for?
Strengthening the Reporting of Observational Studies in Epidemiology
STROBE can be extended and used for _____?
Molecular Epidemiology Studies (STROBE-ME)
Genetic Association Studies (STREGA)
- Strengthing the reporting of genetic association studies
What type of studies is TREND used for?
Reporting evaluations with non-randomized designs of behavioral and public health interventions
Non-randomized studies
What does TREND stand for?
Transparent Reporting of Evaluations with Non-randomized Designs
What type of studies is REMARK used for?
Tumor marker prognostic studies
What does REMARK stand for?
Reporting Recommendations for Tumor Marker Prognostic studies
What type of studies is GRIPS used for?
Genetic risk prediction studies
What does GRIPS stand for?
Genetic Risk Prediction Studies
What type of studies is STARD used for?
Diagnostic Studies
Single diagnostic study
What does STARD stand for?
Standards for the Reporting of Diagnostic Accuracy Studies
What type of studies is QUADAS-2 used for?
Systematic reviews of multiple diagnostic studies
Diagnostic Studies
What does QUADAS-2 stand for?
Quality Assessment of Studies of Diagnostic Accuracy in Systematic Reviews, 2nd edition
DOI number
Digital Object Identifier number
Gives us location of study on internet
Tells us where we can find print version of article
Unique and specific to each article
