Evidence Based Practice Final

What tests would you use to answer the question “is variable D associated with variable E”?

Pearson correlation: each participant has a value for two continuous variables

Spearman correlation: the non-parametric version where each participant has a value for two variable, one or mroe of which may be ordinal

what is a correlation?

a measure that expresses the extend to which two variables are related or “change together”

can range from -1 to +1

what tests would you use to answer the question “can the values of variable D (and E, etc) predict the value of Variable F?

Linear regression: each participant has a continuous value for variables D, E, and F. Beta values then quantify the strength of the relationship. R² quantifies the proportion of the variability in variable F that is predictec by variables D and E

Logistic regression: Each participant has a continuous value for variables D and E. Each participant has a categoraical value for F (yes or no) . Beta values quanitify the strength of the relationship.

what is a regression?

a method to estimate the relationship between a dependent (response) variable and one or more independent variables (predictors)

what tests could you use to answer the question “is population G different from population H in terms of some characterstic or outcome measure?”

unpaired t-test: if comparing two populations

one way ANOVA (not repeated measures): if comparing more than two populations

fisher exact test: if omparing only two populations with only two possible outcomes (yes or no)

Chi square test: if comparing more than two populations or outcomes with only two possible outcomes (yes or no)

what are the parametric vs non-parametric tests?

what are categorical statistical tests?

organize the available data by categories— basically puts the data in piles

ex of categories: control group vs experimental group (group) ; before the intervention vs after the intervention (time)

ex of tests: t-test, ANOVA, chi square

what are ordinal statistical tests?

you measure at least two variables that can be represented numerically

asking of the variables scale together: are larger values of a variable consistently accompanied by larger or smaller values of the other variable?

ex of continuous measures: age, height, weight, baseline levels

ex of tests: correlation, regression

in general, when your data is brouped in terms of 1 categorical variable and has 2 levels (your comparing 2 groups OR pre vs post) then use what tests?

t test (unpaired or paired), fisher exact test

in general, when your data is grouped in terms of 1 categorical variable and has more than 2 levels (your comparing 3 groups, or pre vs post vs follow up) then use what tests?

1 way ANOVA, chi squared

in general, when your data is grouped in terms of more than one categorical variable (your comparing 2 groups AND pre vs post) then use what tests?

2 way ANOVA, 3 way ANOVA, etc

in general, when you are comparing values of mutliple variables to each other then use what tests?

correlation, regression

what is an ANCOVA tests?

it takes a t-test/ANOVA and correlation and does them at the same time to get results of both.

what is an effect size?

a quantitative measure of the magnitude of the experimental effect. helps us to compare across different studies or measures

effect size accounts for the ______ of the effect and the ________ of the effect.

mean, variability

a big effect size is achieved with a _______, ______ change in the outcome measure of interest.

large, consistent

so we would want a small standard deviation because it means there is less variability

what is cohens d?

a common effect size reported for categorical statistical tests

larger values indicate a larger change caused by an intervention

what is pearson’s correlation coefficient?

an effect size reported for correlations (between -1 adn +1)

larger calues indicate a stronger relationships between two variablesho

how do we interpret pearson’s correlation coefficient?

large effect > 0.5

medium effect >0.3

small effect >0.1

if you have a _____ effect size, you can identifiy a statistically significant effect with a small sample size.

large

if you have a ______ effect size, you would need a larger sample size to identifiy a statistically significant effect

small

if a 95% CI for each group does not overlap with the mean value for the other group, what does this mean?

there is a most likely a statistically significant difference between groups

if a 95% CI does not overlap with zero, what does that mean?

the effect of that intervention alone is probably significant

what is an odds ratio?

the relative probability that an individual with a given characteristic will experience the event of interest

what is logistic regression?

predicts what category an individual belongs ins

what is a linear regression?

predicts the value of a continueous measure

what happens when you reach test threshold?

performing tests to rule in/out conditions

what happens when you reach treatment threshold?

no further testing is performed because you have the information you need and know how to move forward with treatment

how do you read a 2×2 table?

what is sensitivity?

“true positive” test results— correctly identifying patients with a conditions

what is specificity?

“true negative” test results— correctly identifying patients who do not have a condition

how do we use specificity and sensitivity to make diagnostic decisions?

SnOut: when a test with high sensitivity is negative, the condition is ruled out

SpIn: when a test with high specificity is positive, the condition is ruled in

what is the goal when quantifying diagnostic accuracy?

to minimize false negatives and false positives

• sensitive tests rarely have false negatives

• specific tests rarely have false positives

what is the ideal threshold for sensitivity and specificity?

close to 1 or 100%

>80 is acceptable for many PT orthapaedic tests

what is a positive likelihood ratio?

probability that a positive test result will be obtained in an individual with the condition compared to an individual without the competition

how do you interpret a positive likelihood ratio?

range from 0 to infinity, the larger the better

LR+ = 1, test produces result that is no better than chance

what is a negative likelihood ratio?

probability that a negative test result will be obtained in an individual with the condition compared to an individuals without the condition

how do you interpret a negative likelihood ratio?

range from 0 to 1, smaller is better

LR- = 1, test produces results that are no better than chance

what is pre-test probability?

probability that an individual has a condition based on their clinical presentation. a diagnostic test has not yet been administered

what is a post-test probability?

probability that an individual has a condition based on the findings from a diagnostic test

what is a nomogram?

how do we know an indivduals pre test probability?

1. known prevelance in population

2. prevalence in your clinic/setting

3. gut instict about probability individual has a condition

what is a positive predictive value?

proportion of patients who have the condition and test “positive” out of all the patients who test positive

what are negative predictive values?

proportion of patients who do not have the condition and test “negative” out of all patients who test “negative”

what are the limitation of PPV and NPV?

1. they are based on teh overall prevelance of a condition which is defined by the number of people in the study

2. prevelance of conditions changes overtime

3. generalizable only when prevelance in your scenario is same as the prevelance in the study

what are reciever operating curves?

they plot the true positive rate vs the false positive rate

the area under the curve quantifies diagnostic accuracy

AUC=1 is perfect

AUC- .05 is like flipping a coin

what is the limitation of a reciever operator curve?

often the values are calculated using a “cut point” on the diagnostic test