Advanced Topics In SMART Design and Data Analysis-- Part 1
Transcript of Advanced Topics In SMART Design and Data Analysis-- Part 1
Advanced Topics In SMART Design and Data Analysis-- Part 1
Module 6
How to use repeated outcome measures from a SMART to compare embedded AIs
β’ A taste of how data from a SMART can be analyzed to address various scientific questionso How to frame scientific questions
o Experimental cells to be compared
o Resources you can use for data analysis
General Objectives
Outline
β’ Brief review of using end-of-study outcome to compare embedded AIs
β’ Learn how to use repeated outcome measures from a SMART to compare embedded AIs
β’ Review three types of scientific questions you can answer with repeated outcome measures o Difference in end-of study outcomeo Difference in Area Under the Curve (AUC)o Delayed effects
β’ Sample size considerations for planning SMARTs to compared embedded AIs with repeated outcome measures
Outline
β’ Brief review of using end-of-study outcome to compare embedded AIs
β’ Learn how to use repeated outcome measures from a SMART to compare embedded AIs
β’ Review three types of scientific questions you can answer with repeated outcome measures o Difference in end-of study outcomeo Difference in Area Under the Curve (AUC)o Delayed effects
β’ Sample size considerations for planning SMARTs to compared embedded AIs with repeated outcome measures
PI: PelhamADHD SMARTFor simplicity assume
response status was assessed at one time point
(week 8)
PI: PelhamADHD SMARTFor simplicity assume
response status was assessed at one time point
(week 8)
A1= 1
R=1
A1=-1
R=0
A2=-1
A2= 1
A2= .
A2= .
A2=-1A2= 1
R=1
R=0
4 embedded adaptive interventions
AI #2:Start with BMOD; if non-responder AUGMENT, else CONTINUE
AI #1:Start with MED; if non-responder AUGMENT, else CONTINUE
AI #3:Start with MED; if non-responder INTENSIFY, else CONTINUE
AI #4:Start with BMOD; if non-responder INTENSIFY, else CONTINUE
PI: PelhamADHD SMART
Recall Typical Primary Aim 3
AI #2:Start with BMOD; if non-responder AUGMENT, else CONTINUE
AI #1:Start with MED; if non-responder AUGMENT, else CONTINUE
vs.
Compare 2 embedded AIs
Comparison of Subgroups A+B vs. D+E
End of Study Primary Outcome AnalysisReview
Step 1: Assign weights and replicate the data
Weightingβ’ Accounts for over/underrepresentation of responders
or non-respondersβ’ Because of the randomization scheme
Replicatingβ’ Allows us to use standard software to do simultaneous
estimation and comparisonβ’ Because participants are consistent with more than
one AI
Step 2: Select and fit a model, such as
πΈ[π|π΄1, π΄2] = π½0 + π½1π΄1 + π½2π΄2 + π½3π΄1π΄2
End of Study Primary Outcome AnalysisReview
Step 3: Estimate model parameters and linear combinations to compare AI #1 and AI #2
πΈ[π|π΄1, π΄2] = π½0 + π½1π΄1 + π½2π΄2 + π½3π΄1π΄2
A1 A2
1 BMOD 1 INTENSIFY-1 MED -1 AUGMENT
End of Study Primary Outcome AnalysisReview
Step 3: Estimate linear combinations of parameters to compare AI #1 and AI #2
πΈ[π|π΄1, π΄2] = π½0 + π½1π΄1 + π½2π΄2 + π½3π΄1π΄2
Step 3: Estimate linear combinations of parameters to compare AI #1 and AI #2
πΈ[π|π΄1, π΄2] = π½0 + π½1π΄1 + π½2π΄2 + π½3π΄1π΄2
Mean Y under (MED, AUG) = π·π + π·π βπ + π·π βπ + π·π(βπ)(βπ)
A1 A2
1 BMOD 1 INTENSIFY-1 MED -1 AUGMENT
End of Study Primary Outcome AnalysisReview
Step 3: Estimate linear combinations of parameters to compare AI #1 and AI #2
πΈ[π|π΄1, π΄2] = π½0 + π½1π΄1 + π½2π΄2 + π½3π΄1π΄2
Mean Y under (MED, AUG) = π·π + π·π βπ + π·π βπ + π·π(βπ)(βπ)
Mean Y under (BMOD, AUG) = π·π + π·π π + π·π βπ + π·π( π)(βπ)
A1 A2
1 BMOD 1 INTENSIFY-1 MED -1 AUGMENT
End of Study Primary Outcome AnalysisReview
Step 3: Estimate linear combinations of parameters to compare AI #1 and AI #2
πΈ[π|π΄1, π΄2] = π½0 + π½1π΄1 + π½2π΄2 + π½3π΄1π΄2
The difference between (MED, AUG) and (BMOD, AUG):π·π β π·π β π·π + π·π β π·π + π·π β π·π β π·π = βππ·π + ππ·π
A1 A2
1 BMOD 1 INTENSIFY-1 MED -1 AUGMENT
End of Study Primary Outcome AnalysisReview
Outline
β’ Brief review of using end-of-study outcome to compare embedded AIs
β’ Learn how to use repeated outcome measures from a SMART to compare embedded AIs
β’ Review three types of scientific questions you can answer with repeated outcome measures o Difference in end-of study outcomeo Difference in Area Under the Curve (AUC)o Delayed effects
β’ Sample size considerations for planning SMARTs to compared embedded AIs with repeated outcome measures
Repeated Outcome Measures From SMART
Repeated Outcome Measures in a SMART = >before and after each randomization
Step #1: Assign weights and replicate the person-period data
Weightingβ’ Accounts for over/underrepresentation of responders
or non-respondersβ’ Because of the randomization scheme
Replicatingβ’ Allows us to use standard software to do simultaneous
estimation and comparisonβ’ Because participants are consistent with more than
one AI
Longitudinal Outcome Analysis
MonthODD at
baselineResponse
StatusSchool
PerfID t O11 R Y A1 A2 W13 0 1 0 1 -1 1 413 2 1 0 5 -1 1 413 8 1 0 3 -1 1 414 0 0 1 2 1 1 214 0 0 1 2 1 -1 214 2 0 1 5 1 1 214 2 0 1 5 1 -1 214 8 0 1 4 1 1 214 8 0 1 5 1 -1 2
Longitudinal Outcome Analysis
Step #1: Weight and replicate the person-period data Fake weighted and replicated data would look like this:
MonthODD at
baselineResponse
StatusSchool
PerfID t O11 R Y A1 A2 W13 0 1 0 1 -1 1 413 2 1 0 5 -1 1 413 8 1 0 3 -1 1 414 0 0 1 2 1 1 214 0 0 1 2 1 -1 214 2 0 1 5 1 1 214 2 0 1 5 1 -1 214 8 0 1 4 1 1 214 8 0 1 5 1 -1 2
Longitudinal Outcome Analysis
Step #1: Weight and replicate the person-period data Fake weighted and replicated data would look like this:
Step #2: Select and fit a model
πΈ[π|π΄1, π΄2] = π½0 + π½1π΄1 + π½2π΄2 + π½3π΄1π΄2
β’ This model is for a single end-of-study outcome β’ But we can extend for additional repeated outcome
measures
Longitudinal Outcome Analysis
Step #2: Select and fit a model for the repeated outcome measurements
β’ The model we select should allow the outcome at each stage to be impacted only by intervention options that were offered prior to that stage.
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Longitudinal Outcome Analysis
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Step #2: Select and fit a model for the repeated outcome measurements
β’ The model we select should allow the outcome at each stage to be impacted only by intervention options that were offered prior to that stage.
Longitudinal Outcome Analysis
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Step #2: Select and fit a model for the repeated outcome measurements
β’ The model we select should allow the outcome at each stage to be impacted only by intervention options that were offered prior to that stage.
Longitudinal Outcome Analysis
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Step #2: Select and fit a model for the repeated outcome measurements
β’ The model we select should allow the outcome at each stage to be impacted only by intervention options that were offered prior to that stage.
Longitudinal Outcome Analysis
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Step #2: Select and fit a model for the repeated outcome measurements
β’ The model we select should allow the outcome at each stage to be impacted only by intervention options that were offered prior to that stage.
Longitudinal Outcome Analysis
Step #2: Select and fit a model for the repeated outcome measurements
β’ The model we select should allow the outcome at each stage to be impacted only by intervention options that were offered prior to that stage.
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Longitudinal Outcome Analysis
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Step #2: Select and fit a model for the repeated outcome measurements
β’ The model we select should allow the outcome at each stage to be impacted only by intervention options that were offered prior to that stage.
Longitudinal Outcome Analysis
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Step #2: Select and fit a model for the repeated outcome measurements
β’ The model we select should allow the outcome at each stage to be impacted only by intervention options that were offered prior to that stage.
β’ Failing to properly account for the ordering of the measurement occasions relative to the intervention options can lead to bias (see Lu et al., 2016).
Longitudinal Outcome Analysis
Step #2: Select and fit a model for the repeated outcome measurements
β’ Piecewise (segmented) linear regression models can be useful in this setting.
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Longitudinal Outcome Analysis
Step #2: Select and fit a model for the repeated outcome measurements
β’ Piecewise (segmented) linear regression models can be useful in this setting.
β’ What is a piecewise linear regression?
Longitudinal Outcome Analysis
Step #2: Select and fit a model for the repeated outcome measurements
β’ Piecewise (segmented) linear regression models can be useful in this setting.
β’ What is a piecewise linear regression? β’ Itβs just a regression
β’ Where you can fit a separate line for different intervals
β’ The boundary for the time intervals can form a transition point.
Longitudinal Outcome Analysis
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Step #2: Select and fit a model for the repeated outcome measurements
β’ In the current example we have 2 intervals of primary interest:
Longitudinal Outcome Analysis
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Interval 1: First-stage intervention
Step #2: Select and fit a model for the repeated outcome measurements
β’ In the current example we have 2 intervals of primary interest:
Longitudinal Outcome Analysis
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Interval 2: Second-stage intervention
Step #2: Select and fit a model for the repeated outcome measurements
β’ In the current example we have 2 intervals of primary interest:
Longitudinal Outcome Analysis
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Interval 2: Second-stage intervention
Interval 1: First-stage intervention
Transition point
Step #2: Select and fit a model for the repeated outcome measurements
β’ In the current example we have 2 intervals of primary interest:
Longitudinal Outcome Analysis
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Interval 2: Second-stage intervention
Interval 1: First-stage intervention
Transition point
Step #2: Select and fit a model for the repeated outcome measurements
β’ In the current example we have 2 intervals of primary interest.
β’ The linear trend in the outcome during the first stage can vary from second-stage and be impacted by different variables
Longitudinal Outcome Analysis
Step #2: Select and fit a model β’ To fit a separate line for each interval: meet π1 and π2
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Interval 2: Second-stage intervention
Interval 1: First-stage intervention
Longitudinal Outcome Analysis
Step #2: Select and fit a model β’ To fit a separate line for each interval: meet π1 and π2β’ π1: indicator for the number of months spent so far in the
first stage by time π‘,
β’ π2: indicator for the number of months spent so far in the second stage by time π‘
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Interval 2: Second-stage intervention
Interval 1: First-stage intervention
Longitudinal Outcome Analysis
Step #2: Select and fit a model
β’ π1: How many months spent so far in stage 1 by time π‘,
β’ π2: How many months spent so far in stage 2 by time π‘
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Interval 2: Second-stage intervention
Interval 1: First-stage intervention
π‘ 0 2 8
π1 0 2 2
π2 0 0 6
Longitudinal Outcome Analysis
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Interval 2: Second-stage intervention
Interval 1: First-stage intervention
Step #2: Select and fit a model
β’ π1: How many months spent so far in stage 1 by time π‘,
β’ π2: How many months spent so far in stage 2 by time π‘
π‘ 0 2 8
π1 0 2 2
π2 0 0 6
Longitudinal Outcome Analysis
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Interval 2: Second-stage intervention
Interval 1: First-stage intervention
Step #2: Select and fit a model
β’ π1: How many months spent so far in stage 1 by time π‘,
β’ π2: How many months spent so far in stage 2 by time π‘
π‘ 0 2 8
π1 0 2 2
π2 0 0 6
Longitudinal Outcome Analysis
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Interval 2: Second-stage intervention
Interval 1: First-stage intervention
Step #2: Select and fit a model
β’ π1: How many months spent so far in stage 1 by time π‘,
β’ π2: How many months spent so far in stage 2 by time π‘
π‘ 0 2 8
π1 0 2 2
π2 0 0 6
Longitudinal Outcome Analysis
ODD at baseline
Response Status Month
Time on stage 1
Time on stage 2
School Perf A1 A2 W
ID O11 R t S1 S2 Y13 1 0 0 0 0 1 -1 1 413 1 0 2 2 0 5 -1 1 413 1 0 8 2 6 3 -1 1 414 0 1 0 0 0 2 1 1 214 0 1 0 0 0 2 1 -1 214 0 1 2 2 0 5 1 1 214 0 1 2 2 0 5 1 -1 214 0 1 8 2 6 4 1 1 214 0 1 8 2 6 5 1 -1 2
Step #2: Select and fit a model
The (fake) weighted and replicated data would look like this:
Longitudinal Outcome Analysis
Step #2: Select and fit a model β’ Letβs ignore treatment assignment for now
β’ i.e., imagine everyone is on the same adaptive intervention
πΈ ππ‘ = π½0 + π½1π1 + π½2π2
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Longitudinal Outcome Analysis
Step #2: Select and fit a model
πΈ ππ‘ = π½0 + π½1π1 + π½2π2
π½0: Expected SP at beginning of school year
π½1: Stage 1 slope: Expected monthly change in SP during stage 1
π½2: Stage 2 slope: Expected monthly change in SP during stage 2
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Longitudinal Outcome Analysis
Step #2: Select and fit a model β’ Now, letβs incorporate the treatment assignment in:
πΈ ππ‘ β£ π΄1, π΄2 = π½0 + π½1π1 + π½2π2
β’ Slope at each stage should depend only on intervention options that have been assigned prior to that stage
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Longitudinal Outcome Analysis
Step #2: Select and fit a model β’ Now, letβs consider incorporating the treatment assignment in:
πΈ ππ‘ β£ π΄1, π΄2 = π½0 + π½1π1 + π½2π2
β’ Recall π½0 is the expected SP at beginning of school year
β’ π½0 should not vary depending on π΄1 or π΄2
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Longitudinal Outcome Analysis
Step #2: Select and fit a model β’ Now, letβs consider incorporating the treatment assignment in:
πΈ ππ‘ β£ π΄1, π΄2 = π½0 + π½1π1 + π½2π2
β’ Recall π½1 is the stage 1 slope
β’ π½1 can vary depending on π΄1
Replace π½1 by (π½10 + π½11π΄1).
= π½0 + (π½10+π½11π΄1)π1 + π½2π2
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Longitudinal Outcome Analysis
Step #2: Select and fit a model β’ Now, letβs consider incorporating the treatment assignment in:
πΈ ππ‘ β£ π΄1, π΄2 = π½0 + π½1π1 + π½2π2
β’ Recall π½2 is the stage 2 slope
β’ π½2 can vary depending on π΄1and π΄2
Replace π½2 by (π½20 + π½21π΄1 + π½22π΄2 + π½23π΄1π΄2)
= π½0 + (π½10+π½11π΄1)π1 + (π½20 + π½21π΄1 + π½22π΄2 + π½23π΄1π΄2)π2
Beginning of school year Week 8 End of school year
Y1 A1 Y2 / R Status A2 Y3
Longitudinal Outcome Analysis
Step #3: Estimate linear combinations of parameters to compare AI #1 and AI #2
Final example model:πΈ ππ‘|π΄1, π΄2 = π½0 + (π½10+π½11π΄1)π1 + (π½20+π½21π΄1 + π½22π΄2 + π½23π΄1π΄2)π2
A1 A2
1 BMOD 1 INTENSIFY-1 MED -1 AUGMENT
Longitudinal Outcome Analysis
A1 A2
1 BMOD 1 INTENSIFY-1 MED -1 AUGMENT
Step #3: Estimate linear combinations of parameters to compare AI #1 and AI #2
πΈ ππ‘|π΄1, π΄2 = π½0 + (π½10+π½11π΄1)π1 + (π½20+π½21π΄1 + π½22π΄2 + π½23π΄1π΄2)π2
Mean π3 under AI #1 (MED, AUG)= π½0 + (π½10+π½11 β β1) β 2 + (π½20+π½21 β β1 + π½22 β β1 + π½23 β 1) β 6
= π·π + ππ·ππ β ππ·ππ + ππ·ππ β ππ·ππ β ππ·ππ + ππ·ππ
Longitudinal Outcome Analysis
Step #3: Estimate linear combinations of parameters to compare AI #1 and AI #2
πΈ ππ‘|π΄1, π΄2 = π½0 + (π½10+π½11π΄1)π1 + (π½20+π½21π΄1 + π½22π΄2 + π½23π΄1π΄2)π2
Mean π3 under AI #2 (BMOD, AUG)= π½0 + (π½10+π½11 β 1) β 2 + (π½20+π½21 β 1 + π½22 β β1 + π½23 β β1) β 6
= π·π + ππ·ππ + ππ·ππ + ππ·ππ + ππ·ππ β ππ·ππ β ππ·ππ
A1 A2
1 BMOD 1 INTENSIFY-1 MED -1 AUGMENT
Longitudinal Outcome Analysis
Step #3: Estimate linear combinations of parameters to compare AI #1 and AI #2
πΈ ππ‘|π΄1, π΄2 = π½0 + (π½10+π½11π΄1)π1 + (π½20+π½21π΄1 + π½22π΄2 + π½23π΄1π΄2)π2
Difference in mean π3 between AI#1 and AI#2
= (π½0 + 2π½10 β 2π½11 + 6π½20 β 6π½21 β 6π½22 + 6π½23)
β (π½0 +2π½10 + 2π½11 + 6π½20 + 6π½21 β 6π½22 β 6π½23)
= β4π½11 β 12π½21 + 12π½23
A1 A2
1 BMOD 1 INTENSIFY-1 MED -1 AUGMENT
Longitudinal Outcome Analysis
Outline
β’ Brief review of using end-of-study outcome to compare embedded AIs
β’ Learn how to use repeated outcome measures from a SMART to compare embedded AIs
β’ Review three types of scientific questions you can answer with repeated outcome measures o Difference in end-of study outcomeo Difference in Area Under the Curve (AUC)o Delayed effects
β’ Sample size considerations for planning SMARTs to compared embedded AIs with repeated outcome measures
Area Under the Curve (AUC)
β’ Scientific Question: Is AI#1 better than AI#2 in terms of school performance averaged over the course of the intervention?
Longitudinal Outcome Analysis
Area Under the Curve (AUC)
β’ AI#1 AUC
***This example is hypothetical
Expe
cted
Sch
ool P
erfo
rman
ce
Longitudinal Outcome Analysis
Area Under the Curve (AUC)
β’ AI#1 AUC
***This example is hypothetical
Expe
cted
Sch
ool P
erfo
rman
ce
Longitudinal Outcome Analysis
Area Under the Curve (AUC)
β’ AI#2 AUC
Expe
cted
Sch
ool P
erfo
rman
ce
***This example is hypothetical
Longitudinal Outcome Analysis
Area Under the Curve (AUC)
β’ Difference in AUC AI#1-AI#2:
Expe
cted
Sch
ool P
erfo
rman
ce
***This example is hypothetical
Longitudinal Outcome Analysis
Area Under the Curve (AUC)
β’ If you compare the two AIs in terms of end-of-school-year outcome, what would you conclude?
Expe
cted
Sch
ool P
erfo
rman
ce
***This example is hypothetical
Longitudinal Outcome Analysis
Area Under the Curve (AUC)
β’ If you compare in terms of AUC...
Expe
cted
Sch
ool P
erfo
rman
ce
***This example is hypothetical
Longitudinal Outcome Analysis
Area Under the Curve (AUC)
β’ Consider when outcome values towards the middle (in the course of the school year) are considered more informative than are values that are at the end
Longitudinal Outcome Analysis
Delayed Effects
β’ Scientific Question: Does the initial intervention options in AI #1 vs. AI #2 impact school performance differently before vs. after these AIs unfold?o before vs. after these AIs unfold before vs. after second-stage
options are introduced
Longitudinal Outcome Analysis
Delayed Effects
β’ AI#1 & AI#2 lead to the same outcome at the end of school year
β’ But the process is different
Expe
cted
Sch
ool P
erfo
rman
ce
***This example is hypothetical
Longitudinal Outcome Analysis
Delayed Effectsβ’ Definition:
o Difference between long-term effect and short-term effecto Difference between two differences
Longitudinal Outcome Analysis
***This example is hypothetical
Delayed Effectsβ’ Difference between two differences
Expe
cted
Sch
ool P
erfo
rman
ce
Long-term: Difference in mean Y3
between AI#1 and AI#2
Longitudinal Outcome Analysis
***This example is hypothetical
Delayed Effectsβ’ Difference between two differences
Expe
cted
Sch
ool P
erfo
rman
ce
Short-term: Difference in mean Y2
between AI#1 and AI#2
***This example is hypothetical
Longitudinal Outcome Analysis
Delayed Effectsβ’ Difference between two differences
Expe
cted
Sch
ool P
erfo
rman
ce Long-term: Difference in mean Y3
between AI#1 and AI#2
-Short-term: Difference in mean Y2
between AI#1 and AI#2
***This example is hypothetical
Longitudinal Outcome Analysis
Delayed Effectsβ’ Consider when there is scientific/practical rationale for positive
or negative synergies between first and second stage options
Longitudinal Outcome Analysis
Outline
β’ Brief review of using end-of-study outcome to compare embedded AIs
β’ Learn how to use repeated outcome measures from a SMART to compare embedded AIs
β’ Review three types of scientific questions you can answer with repeated outcome measures o Difference in end-of study outcomeo Difference in Area Under the Curve (AUC)o Delayed effects
β’ Sample size considerations for planning SMARTs to compared embedded AIs with repeated outcome measures
πΏ is the standardized effect size for the comparisonπ is the (compound-symmetric) within-person correlationπ is the probability of response to first-stage treatment
π =
4 π§1β
πΌ2+ π§1βπ½
2
πΏ2Γ 1 β π2 Γ (2 β π)
Longitudinal Outcome Sample Size for End-of-Study Comparisons
40% response rateπΌ=0.05 (two-sided)80% target power
Within-Person CorrelationStd. Effect Size
π = 0 π = 0.3 π = 0.6
πΏ = 0.3 559 508 358πΏ = 0.5 201 183 129
Longitudinal Outcome Sample Size for End-of-Study Comparisons
Longitudinal Outcome Sample Size for End-of-Study Comparisons
π =
4 π§1β
πΌ2+ π§1βπ½
2
πΏ2Γ 1 β π2 Γ (2 β π)
πΏ is the standardized effect size for the comparisonπ is the (compound-symmetric) within-person correlationπ is the probability of response to first-stage treatment
π =
4 π§1β
πΌ2+ π§1βπ½
2
πΏ2Γ (2 β π)
πΏ is the standardized effect size for the comparisonπ is the probability of response to first-stage treatment
Sample Size for End-of-Study Comparisons
Citations
β’ Lu, X., NahumβShani, I., Kasari, C., Lynch, K. G., Oslin, D. W., Pelham, W. E., ... & Almirall, D. (2016). Comparing dynamic treatment regimes using repeatedβmeasures outcomes: modeling considerations in SMART studies. Statistics in Medicine, 35(10), 1595-1615.
β’ Dziak, J.J., Yap, J.R., Almirall, D., McKay, J.R., Lynch. K.G., & Nahum-Shani, I. (2019) A Data Analysis Method for Using Longitudinal Binary Outcome Data from a SMART to Compare Adaptive Interventions. Multivariate Behavioral Research, 54(5), 613-636
β’ Nahum-Shani, I., Almirall, Yap, J.R., D. McKay, J., Lynch, K., Freiheit, E., & Dziak, J.J. (in press). Longitudinal Analysis: A Tutorial for Using Repeated Outcome Measures from SMART Studies to Compare Adaptive Interventions. Psychological Methods.
β’ Seewald, N. J., Kidwell, K.M., Nahum-Shani, I., Wu, T., McKay, J.R., Almirall, D. Sample Size Considerations for the Analysis of Continuous Repeated-Measures Outcomes in Sequential Multiple-Assignment Randomized Trials. Statistical Methods in Medical Research