Biostatistics Core A team of faculty and staff biostatisticians with diverse and extensive experience conducting a broad range of research projects 15 staff biostatisticians 53 faculty in the Biostatistics and Bioinformatics Department ID: 698136
Download Presentation The PPT/PDF document "Clinical and Translational Science Award" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Clinical and Translational Science AwardBiostatistics Core
A team of faculty and staff biostatisticians with diverse and extensive experience conducting a broad range of research projects15 staff biostatisticians 53 faculty in the Biostatistics and Bioinformatics DepartmentOver 40 collaborative teams throughout Duke
1Slide2
Clinical and Translational Science AwardBiostatistics Core
Mission:To support an interdisciplinary network of clinical investigators conducting research at Duke by providing expertise in study design, implementation of statistical methodology, and interpretation of results 2Slide3
Core Collaborative Research Process
Investigator submits a request for statistical collaborationhttp://biostat.duke.edu/biostatistics-coreClinical Research Unit (CRU) statisticians:
Maragatha
Kuchibhatla
,
Ph.D
Ben A. Goldstein, Ph.D Gina-Maria Pomann, Ph.D Hui-Jie Lee, Ph.D Alfredo E. Farjat, Ph.D Tracy Truong, MSBeiyu Liu, MS
3Slide4
Schedule a MeetingSlide5
SCORES Office HoursMonday:
Hui-Jie LeeWednesday: Tracy TruongThursday: Alfredo FarjatFriday: Beiyu LiuSlide6
Biostatistics Collaboration through a Case Study
CTSA Biostatistics CoreSpeaker: Hui-Jie Lee, PhDOctober 27, 2016Slide7
Article Example:
Lymphovascular Invasion is Associated with Survival for Papillary Thyroid Cancer Lauren N Pontius, Linda M Youngwirth, Samantha M Thomas, Randall P Scheri, Sanziana A Roman and Julie A SosaEndocrine-related Cancer, 2016 Slide8
Scientific ProcessObservation/Pattern
Scientific QuestionHypothesis Study Design/Data CollectionStatistical AnalysisResults/ConclusionSlide9
Scientific ProcessObservation/Pattern
Impact of lymphovascular invasion (LVI) on survival is important for several cancers.Limited data regarding the impact of LVI on patient outcomes in papillary thyroid cancer (PTC).Slide10
Scientific ProcessScientific Question
Do PTC patients undergoing thyroid resection with LVI have compromised survival compared to those without LVI?What are the current practice patterns associated with LVI in PTC patients?Slide11
Primary Outcome
Greatest importanceClinically relevantMeasurableArticle example:Slide12
Primary Outcome
Greatest importanceClinically relevantMeasurableArticle example: Overall survivalTime from diagnosis of papillary thyroid cancer to death or the last follow-up visitSlide13
Scientific ProcessScientific Question
Do PTC patients undergoing thyroid resection with LVI have compromised survival compared to those without LVI?What are the current practice patterns associated with LVI in PTC patients?Slide14
Scientific ProcessScientific Question
Do PTC patients undergoing thyroid resection with LVI have compromised survival compared to those without LVI?Slide15
Scientific ProcessScientific Question
Do PTC patients undergoing thyroid resection with LVI have compromised survival compared to those without LVI?Slide16
Scientific ProcessScientific Question
Do PTC patients undergoing thyroid resection with LVI have compromised survival compared to those without LVI?Slide17
Target Population
Population
Adult patients diagnosed of
papillary thyroid cancer who
underwent thyroid resection.Slide18
Representative Sample
Population
Adult patients diagnosed of
papillary thyroid cancer who
underwent thyroid resection.
Sample for the studySlide19
Inference
Population
Adult patients diagnosed of
papillary thyroid cancer who
underwent thyroid resection.
Sample
InferenceSlide20
Scientific ProcessObservation/Pattern
Scientific QuestionHypothesis Study Design/Data CollectionStatistical AnalysisResults/ConclusionSlide21
Hypothesis of Interest
Article example: Presence or absence of LVI on patient survival.
Null hypothesis
:
LVI is NOT a risk factor for patient survival.
Alternative hypothesis
:
?Slide22
Hypothesis of Interest
Article example: Presence or absence of LVI on patient survival.
Null hypothesis
:
LVI is NOT a risk factor for patient survival.
Alternative hypothesis
:
LVI is a risk factor for patient survival.Slide23
Inference
Hypothetical Population
LVI/no LVI have the same survival probability distributionSlide24
Inference
Sample
Hypothetical Population
LVI/no LVI have the same survival probability distributionSlide25
Inference
Sample
7.9%
LVI presence
LVI absence
5-year
survival
86.6%
94.5%
Hypothetical Population
LVI/no LVI have the same survival probability distribution
?Slide26
Inference
Sample
7.9%
Sample
0.7%
Hypothetical Population
LVI/no LVI have the same survival probability distribution
?Slide27
Inference
Sample
7.9%
Sample
0.7%
Sample
1.5%
Hypothetical Population
LVI/no LVI have the same survival probability distribution
?Slide28
Inference
Sample
7.9%
Sample
0.7%
Sample
1.5%
Sample
-0.3%
Hypothetical Population
LVI/no LVI have the same survival probability distribution
?Slide29
P-value
0Slide30
P-value
0
Observed difference in our sample 7.9%Slide31
P-value
0
Observed difference in our sample 7.9%
-0.3%
1.5%Slide32
P-value
0
Observed difference in our sample 7.9%
-0.3%
1.5%
p-valueSlide33
Hypothesis Testing: Types of ErrorsSlide34
P-value
“ The ASA statement is intended to steer research into a ‘post p < 0.05 era. ”
-Ron WassersteinSlide35
P-value
“We teach it because it's what we do; we do it because it's what we teach.”Slide36
Scientific ProcessObservation/Pattern
Scientific QuestionHypothesis Study Design/Data CollectionStatistical AnalysisResults/ConclusionSlide37
Article Example
Study DesignPopulation-based cohort studyData CollectionNational Cancer Data Base (NCDB)Data from 2010 to 2011Inclusion/Exclusion criteriaVariables to collectSlide38
Data Collection: REDCap
The data should be reproducible and must be associated with a data dictionary that explains the meaning of each variable.DOCR can help you.
Patient
ID
Outcome
Age
Gender
Treatment11.32
65
F
1
2
5.66
33
M
2
3
NA
48
M
1Slide39
Scientific ProcessObservation/Pattern
Scientific QuestionHypothesis Study Design/Data CollectionStatistical AnalysisResults/ConclusionSlide40
Scientific Process
Statistical AnalysisOverview/quality control of the dataSummarize main information about the dataIdentify outliers, validate assumptionsHypothesis testing, statistical modelingSlide41
Introduction to Survival Analysis
Basic principlesKaplan-Meier estimatesLog-rank testCox proportional hazards modelSlide42
Survival Analysis: Basic Principles
Outcome: Time until the occurrence of an event of interestAccount for event times that cannot be accurately determined (censoring) Article example:Slide43
Survival Analysis: Basic Principles
Outcome: Time until the occurrence of an event of interestAccount for event times that cannot be accurately determined (censoring) Article example: Time from diagnosis to deathSlide44
Censoring
Right censoring: Event is not observed because of loss to follow-up, drop-out, or study terminationEvent has not yet occurred at the last time point the patient was observedEnd of study period
Time 0
EventSlide45
Statistical Analysis: Kaplan-Meier PlotSlide46
Statistical Analysis: Kaplan-Meier Plot
Probability of surviving past time tSlide47
Statistical Analysis: Kaplan-Meier PlotSlide48
Statistical Analysis: Kaplan-Meier PlotSlide49
Statistical Analysis: Kaplan-Meier Plot
=
= 80%
Slide50
Statistical Analysis: Kaplan-Meier Plot
=
= 80%
Slide51
Statistical Analysis: Kaplan-Meier Plot
Censored observationSlide52
Statistical Analysis: Kaplan-Meier PlotSlide53
Statistical Analysis: Kaplan-Meier Plot
Group 1
Group 2Slide54
Log-rank Test: Compares Survival
When are two survival curves statistically equivalent?Null hypothesis: No difference between populations in the probability of a death event at any time point.
P-value < 0.0001
Group 1
Group 2Slide55
Log-rank Test: Compares Survival
Log-rank tests can fail when survival curves cross.
P-value = 0.09
Group 1
Group 2Slide56
Article: Fig 2. Unadjusted Survival for patients with PTC undergoing thyroid resection based on presence of LVISlide57
Hazard Rate
Hazard rate λ(t) : The probability that if a patient survives to time t, an event will occur in the next instant of time.It is a rate per unit of time similar in meaning to reading a car speedometer at a particular instant and seeing 45 mph.Slide58
Hazard Ratio
The ratio of hazard rate of death in patients with LVI compared to the hazard rate of death in patients without LVI at a given time.Slide59
Hazard Ratio
The ratio of hazard rate of death in patients with LVI compared to the hazard rate of death in patients without LVI at a given time.Slide60
Statistical Analysis: Cox Proportional Hazards Regression
Model hazard rate of an event.
Allows
to adjust for covariates of
interest
Provides a hazard ratio to compare the risk of the event between groups
Assume constant hazard ratio over timeSlide61
Hazard ratio
=
Statistical Analysis: Cox Proportional Hazards RegressionSlide62
Statistical Analysis: Cox Proportional Hazards Regression
Hazard ratio =
=
1
Slide63
Statistical Analysis: Cox Proportional Hazards Regression
Hazard ratio =
=
1
No difference in survival between two groups.Slide64
Statistical Analysis: Cox Proportional Hazards Regression
Hazard ratio =
<
1
Slide65
Statistical Analysis: Cox Proportional Hazards Regression
Hazard ratio =
<
1
Survival is worse in no LVI group.Slide66
Statistical Analysis: Cox Proportional Hazards Regression
Hazard ratio =
>
1
Slide67
Statistical Analysis: Cox Proportional Hazards Regression
Hazard ratio =
>
1
Survival is worse in LVI group.Slide68
Statistical Analysis: Cox Proportional Hazards Regression
≈
LVI +
age
+
gender +
+
RAI + error
.
Slide69
Article Example: Table 4
LVI is associated with an increased risk of death.
The rate of death for patients with LVI is 88% higher at any given time point studied than the rate for patients without LVI.Slide70
RecapKaplan-Meier
→ survival estimates that can be graphed, cannot control for covariatesLog-rank test → used to compare survival between 2+ groups, cannot control for covariatesCox proportional hazards models → model survival outcomes, can include multiple predictorsSlide71
Scientific ProcessObservation/Pattern
Scientific QuestionHypothesis Study Design/Data CollectionStatistical AnalysisResults/ConclusionSlide72
Results: Data Visualization
Figure 1 : Distribution of patients with LVI by age.Slide73
Results: Data VisualizationSlide74
Results: Data Visualization
Many different ways to display data.Talk with your collaborating statisticians!Slide75
The End
Questions?hui-jie.lee@duke.edusiyun.yang@duke.eduBiostatistics Core Website: http://biostat.duke.edu/biostatistics-core
Walk-in Office Hours: Monday 9-10AM at Hock Plaza 11016Slide76
DOCR and Data Support
Jason LonesIT Analyst, Sr.Duke Office of Clinical ResearchSlide77
Duke Office of Clinical Research
Research support services for the Duke communityProtocol developmentRegulatory and IRB submissionsData planning and database buildDEDUCE, data managementCRC services (screening, consent, recruitment, etc.)Contracts, budgets, CT.gov, MaestroCareTraining and educationSlide78
Data planning: The time to think about the end is at the beginning…
Consult your statistician early!Project goals drive data collectionVariable namingData coding - avoid free textDefining terms
78
Collecting identifiers
Collecting summary data – collect raw values
User access
Field typesSlide79
Tragic consequences of poor planningSlide80
REDCap - Research
Electronic Data CaptureHIPAA compliantLicensed from Vanderbilt for freeData stored behind institutional firewallWeb-based, intuitive interfaceAudit trailAuto-validation, branching logic, calculated fieldsData import/exportData export to statistical packages (SAS, SPSS)Data dictionary
Secure alternative to ExcelSlide81
REDCap @ Duke
Nearly 4000 projects!No fees for DIY buildersDOCR and ORI provide:Upgrade validation servicesServer support and maintenanceOversight of user rights, post-production changesSlide82
Resources
Office hours
Every Tuesday 11:00am – 12:00pm, 9
th
Floor Hock Plaza &
1st Thursday of the month from
10:00am-11:00am,DMP
2W96
Meet with programmers for questions and guidance
One-on-one REDCap demonstrations and free consultations with PIs and study teams
Full service contract builds
https://redcap.dtmi.duke.edu/redcap/
redcap-docr@duke.edu
Training videos available
at
http://
projectredcap.org
REDCap Workshops (LMS)
Jason Lones
IT Analyst, Sr.
Duke Office of Clinical Research
Office:
(919) 668-1787
jason.lones@duke.eduSlide83
Statistical Analysis: Cox Proportional Hazards Regression
Categorical predictor: LVI≈
LVI + error
Slide84
Statistical Analysis: Cox Proportional Hazards Regression
Categorical predictor: LVI ≈
LVI
+ error
≈
+ error
≈
error
Slide85
Statistical Analysis: Cox Proportional Hazards Regression
Categorical predictor: LVI
-
Slide86
Statistical Analysis: Cox Proportional Hazards Regression
Categorical predictor: LVI
-
Slide87
Statistical Analysis: Cox Proportional Hazards Regression
Categorical predictor: LVI
-
Hazard ratio:
=
exp
(
)
Slide88
Statistical Analysis: Cox Proportional Hazards Regression
Univariate Analysis∝
LVI + error
Slide89
Statistical Analysis: Cox Proportional Hazards Regression
Univariate Analysis∝
LVI + error
=
Slide90
Statistical Analysis: Cox Proportional Hazards Regression
Univariate Analysis∝
LVI + error
=
CONSTANT OVER TIME