Transcript Slide 1

PTP 560
• Research Methods
Week 12
Thomas Ruediger, PT
April 5th is last 5 chapters-comprehensive review
April 12th Final
Bulk of final 80%
Chap. 4,5,6 Underpinning for Scale, reliability,
validity, chap 8 Sampling, 10 Experimental Designs,
17-21, 23, 24, 26, 27
Last 20%: 16, 22,25,28,29,32,34
Independent T-test
• If the top row of Levene’s Sig is >.05, then do NOT
assume equal variances and use the bottom row
of chart.
• If we research sig. then the t-stat has to be bigger
than the critical value. If t-stat is bigger than
critical then REJECT the NULL (because there is a
difference)
• The bigger the t-stat then will have a better
chance of being bigger than the critical value.
Validity
Truth
+
1-Sn = - LR
Sp
+
a
b
-
c
d
Test
Sn = a/a+c
Sn
+ LR = 1-Sp
Sp = d/b+d
Validity
Ruling in/Ruling Out
• SpPin
– With high Specificity,
– a Positive tests rules in the diagnosis
• SnNout
– With high Sensitivity,
– a Negative tests rules out the diagnosis
Validity
Pretest Posttest Probability
• Pretest
–
–
–
–
What we think might be the problem
Conceptually a “best guess”
However, it is enhanced by pertinent literature
Influenced by your clinical experience
• Posttest
– Revised probability based on test outcome
– Likelihood ratios widely used in PT literature
• +LR
– How many more times a positive test will be seen in those with the
disorder than without the disorder
• -LR
– How many more times a negative test will be seen in those with the
disorder than without the disorder
Receiver Operating Characteristic
(ROC) Curves
 Strikes a balance between
 Sensitivity
 Specificity
 So that we can trade-off over and under diagnosing.
 Construction
 Set several cutoff points
 Plot Sensitivity and 1-Specificity
Interpret



Visually - which is best diagnostic tool?
Mathematically the Area under the curve is best diagnostic trade-off
 Decide on Cutoff
 Based on the impact of incorrect decision
Receiver Operating Characteristic
(ROC) Curves
50:50
Clinical Prediction Rules
• Incorporates Sensitivity, Specificity
• Quantifies the contributions of different variables
• Used to increase diagnostic utility
– Is the patient at risk for a certain outcome?
– Does the patient have this pathology
• Ottawa ankle rules a good example
Measuring Change
 MDD=can we find a difference one test to another
 MCID=can you find a difference being made for patients
 Distribution based methods (normalized data)




Effect Size Index
Standardized Response Mean
Guyatt’s Responsiveness Index
Standard Error of the Measurement
 Anchor Based Methods (like a pain scale)
 Global Rating of Change
 Ordinal scale based on subjective rating of change
 Global Rating Scale common Scale
Epidemiology
• Distribution and determinants of:
– Disease
– Injury
– Dysfunction
• Descriptive
• Analytic
Descriptive Epidemiology
 Incidence: the amount of new cases
 May be cumulative
 Number of new cases
(during a given period)
Total population at risk
 May be in person-time (used to be
 Number of new cases
(during a given period)
Total person time
 Prevalence: the amount of all cases (new & old)
 Number of existing cases
(during a given period)
Total population at risk
 Relationship between Incidence and Prevalence
Analytic Epidemiology
• Relative vs. Absolute Effects
– Ratio vs. Actual difference
• Relative Risk
– Likelihood that exposed person gets disease
• Odds Ratio
– Analogous to RR
– Applicable to Case-Control Situation
Analytic Epidemiology
• Event Rates and Risk Reduction
• Experimental Event Rate (EER): with exposure
• Control Event Rate (CER): without exposure
• EER/CER = Relative Risk (RR)
• CER-EER/CER = RRR (RR reduction)
• CER-EER = ARR (Absolute Risk Reduction)
Analytic Epidemiology
• CER-EER = ARR (Absolute Risk Reduction)
• 1/ARR = (Number needed to treat) NNT
– If represents the number of patients that would
be needed to be treated to make a change in their
disorder as big as that in the study.
Multivariate Analysis
A one time read thorough is warranted
 Examine several variables for interrelationships
 Applications to correlation
 Partial correlation coefficient
 Regression
 Multiple independent variables
 Beta weights are standardized values for relative weighting
 R2 (coefficient of determination) is amount of total variance explained by all
IVs
 Adjusted R2 corrects for chance
 Discriminate Analysis
 Analogue to multiple regression
 Used with categorical variables