1

appraisal
(interval validity)
(external validity)
appraise

4
appraise RCT

3
1. abstract abstract
2. Method

3.
manuscript EMPA-REG OUTCOME



appraise



Internal validity
RCT Bias

allocation bias randomised ? concealment? stratified?

Method randomised baseline characteristic Result

ascertainment bias blind (single-blinded, double-blinded) open-label
intention-to-treat per-protocol analysis
intention-to-treat preserve
randomization minimize allocation bias underestimate true effect
drop out compliance intention-to-treat analysis
non-inferiority trial reject null non-inferiority intention-totreat compliance drop out rate perprotocol explanatory
pragmatic ( external
validity) initial randomization
outcome primary secondary outcome pre-specified
outcome reflective
outcome placebo (active control)

 non-inferiority trial
minimal mean/median follow up time outcome

event rate primary outcome
sample size underpower
non-inferiority trial non-inferiority margin set
non-inferiority trial
https://www.facebook.com/371793389694429/photos/a.371873473019754.1073741828.371793389694429/52364685117
5748/?type=3
crude event rate outcome primary, secondary, tertiary safety
EMPA-REG rate/1000 patient-years N
magnitude outcome absolute risk reduction NNT
relative risk "clinical significance"
95% 95% CI
p-value "statistical significance"
survivial curve cumulative Hazard
ratio p-value ( survival curve
)
subgroup analysis interaction p value sig sig

External validity
inclusion exclusion criteria
intervention protocol

 outcome outcome clinical significance outcome

Internal validity magnitude
protocol external validity
Discussion Appraisal
Discussion
Final Thought

Appraisal of Clinical Trial (LEADER)

 ... . 1412
. 1412 .Rapp LEADER NEJM 24
.1412 Chaisiri Jarvis

1

https://www.facebook.com/371793389694429/photos/pb.371793389694429.2207520000.1465229381./525677197639380/?type=3
10
1412
full paper
Full Text
http://www.nejm.org/doi/pdf/10.1056/NEJMoa1603827
LEADER iPad
abstract ..... ( ) ....
non-inferiority trial GLP1 recepter agonist liraglutide A1c 7
CV disease CV risk liraglutide standard regimen NI margin 1.3
cardiovascular composite outcome survival analysis 1 outcome time to first CV death
nonfatal MI nonfatal stroke N 9340 3.8 1 outcome
NI margin superiority 2outcome all-cause CV death
hypoglycemia GI symptom
... 2 LEADER 8 ?
critical appraisal
Method manuscript

Method ..... ( 3 ... ) .... experimental research
, blind steering committee sponsor 4 11

 2 A1c
> 7% 50 60 60 CV risk placebo
run-in 2 adherence 1:1 eGFR liraglutide 1.8 mg
placebo 1,3,6,6,6 ... 42 60
N expected annual event rate 1.8% margin 1.3 RR AR margin
annual rate primary outcome composite endpont drive N
8754 power 90% p value NI 0.025 survival analysis Cox
proportional-hazards Hazard ratio interval intention-totreat analysis primary analysis sensitivity analysis 7 per-protocol analysis

... method 3
appraise 5
Table manucript ( )
1. Internal validity
blind concealed 1:1 eGFR intention-to-treat analysis
allocation bias ascertainment bias

intention-to-treat drop out adherence
result adherence rate 83 - 84%
per-protocol
analysis magniude intention-to-treat
NI trial
intention-to-treat
robustness generalizibility
primary analysis "sensitivity analysis"
7
per-protocol event 120
event

 on treatment event per-protocol event

on treatment + 30 days event 30
excluding events after visit 15 event follow-up
adjusted for additional covariates confounder
adjusted for random country effect
stratified for severe renal impairment primary analysis eGFR
subgroup analysis
eGFR 60 ... HR 0.83 - 1.07
eGFR 30 -59 ... HR 0.54 - 0.83 (sig)
eGFR<30 ... HR 0.51 - 1.54
eGFR 30 - 59 subgroup analysis
7 sensitivity analysis primary analysis primary outcome
noninferiority (on treatment, censoring data
confounder)
Hazard ratio outcome
1, 2, 3 safety outcome pre-specified ? pre-specified
reflective outcome? reflective outcome outcome proven benefit control
outcome clinical benefit standard regimen
benefit outcome LEADER outcome

minimal mean/median follow up time outcome

? minimum 42 60 3.8 45
primary secondary outcome event
LEADER

 event rate primary outcome

sample size ? event rate no. of event per 100 patientyear annual rate 1.8% N
( ) primary outcome event 3.9 per 100 patient-year
1.8% underpower NI margin
RR margin
... Hazard ratio

Hazard Ratio crude event rate 1 outcome 13 14.9%
1.9 % follow up time 3.8 composite outcome AR NNT
66 event 3 2 outcome NNT 98
1 clinical significance
crude
event rate
statistical significance outcome
survival curve HR Cox proportional Hazards model
p < 0.001 for NI p = 0.01 for superiority sensitivity analysis 7 2
outcome all-cause CV death microvascular event (renal + retinal)
nephropathy outcome hospitalization for
heart failure event rate liraglutide heart failure

manuscript survival curve cumulative incidence
Kaplan-Meier method plot cumulative Hazard ratio HR
interval survival curve non-parametric Cox Proportion Hazard
outcome diverge
postulate outcome LEADER diverge EMPA-REG
CV outcome survival curve
minimal follow up time survival curve

 subgroup analysis subgroup analysis LEADER primary

outcome subgroup overall result
interaction p value sig subgroup
overall result sig LEADER sig eGFR 60
CV diseae at baseline
safety profile safety outcome sever hypoglycemia
liraglutide hypoglycemic agents insulin
liraglutide GI symptom pancreatic
disease
2. External validity
LEADER
LEADER high risk diabetic high
risk external validity adherence LEADER
drop out 20% GI symptom liraglutide pancreatitis
pancreatic cancer long-term complication
Final thought LEADER
1 1
..... ()
empagliflozin class proven CV benefit

glycemic control A1c 0.4% 3.8 outcome EMPA-REG

cumulative Hazard ratio diverge empagligozin EMPA-REG ?

CV event ?

, Multiplicity rule Bonferroni

p-value
PLATO 2009
PLATO ticagrelor reversible direct-acting P2Y12 inhibitor clopidogrel
acute coronary syndrome persistent ST-elevation PCI
composite cardiovascular outcome (time to first event) survival analysis 12


1. primary outcome primary outcome
primary outcome appraise event rate
control sample size statistical
power (underpower)
negative failure to reject a false null hypothesis (type II error)
PLATO 11% 12 11.7%
2. primary outcome crude event rate study control event rate
12 12 survival analysis PLATO
plot interval 12 A
event plot interval 3 A B event
plot interval 2 plot survivial curve event
12 event event rate
absolute risk reduction NNT


3. intention-to-treat time to first event survival cumulative hazard
ratio hazard ratio survival curve (proportional hazards model)
95% CI p-value
event rate hazard ratio relative risk reduction caveat

4. primary outcome 3
survival curve y cumulative Hazard ratio median follow-up time (
Result) x curve
curve caveat diverge survival
curve diverge ralative risk
ratio 12 PLATO

5. primary outcome composite endpoint clinical trial

primary outcome composite
composite outcome ? composite outcome outcome
1 event event N power
event-driven multiplicity ( )
PLATO primary outcome composite vascular death, MI stroke
ticagrelor vascular death, MI stroke clopidogrel
secondary outcome composite
primary outcome benefit driven
multiplicity? multiplicity 1
statistical power p-value
secondary tertiary outcome 1
positive result by chance ?
outcome ( independent outcome) stroke
coronary revascularization d significance level positive result by
chance = 1(1)d = 0.05 10 positive by chance = 1[1
005]10 = 0.40 40%
40% multiplicity S. Nissen ACC.16
panel cataract surgery HOPE-3
outcome primary outcome CV
death MI multiplicity outcome MI cataract
PLATO secondary outcome
composite hierarchical test sequence
Bonferroni p-value
secondary outcome p-value primary outcome
0.05 5 , p-value outcome = 0.05/5 = 0.01
false positive conclusion type I error,
5 p-value p-value 0.05 p
= 0.04 0.04 x 5 = 0.20

composite outcome multiplicity outcome composite

outcome
manuscript full report
outcome outcome
20 outcomes 7 positive 20 outcome
prespecified outcome original paper post-hoc type I error
now, we see PLATO in a different angle

 Survival Curve

New England Journal of Medicine

.. Journal Club
placebo


ADA 2016
New Orleans LEADER 1 LEADER appraise
LEADER EMPA-REG renal outcome
CV outcome empaglifozin
standard Heart Failure Guideline

Full paper
http://www.nejm.org/doi/pdf/10.1056/NEJMoa1515920

35%

RAAS



1. glucagon-like peptide 1 (GLP-1) agonists FDA
liraglutide exenatide lixisenatide 2007
meta-analysis Steven Nissen CV outcome trial
cardiologist endocrinologist GLP-1 agonists CV trial
LEADER (liraglutide) nejm
EXSCEL (exenatide) 2018 - 2019
ELIXA (lixisenatide) ESC 2015
2. dipeptidyl peptidase 4 (DPP-4) inhibitors GLP-1 agonist GLP-1 breakdown
insulin FDA saxagliptin, sitagliptin, alogliptin linagliptin
EXAMINE (alogliptin) nejm 2013
SAVOR-TIMI 53 (saxagliptin) landmark trial 2013
TECOS (sitagliptin) nejm metaanalysis . Heart Failure
meta-analysis TECOS 6 trials meta-analysis

 Jarvis
3. sodiumglucose cotransporter 2 (SGLT2) inhibitors FDA canagliflozin, dapagliflozin
empagliflozin
EMPA-REG Outcome (empagliflozin) nejm CV outcome part renal
outcome
CANVAS (canagliflozin) 2017
DECLARE-TIMI 58 (dapagliflozin) 2018
SGLT2 inhibitor glucose proximal tubule sodium-glucose
cotransporter (SGLT) SGLT 2 glucose

renal protection?
phase FDA
approval renal protection human
Diabetologia Circulation glomerular hemodynamic response 1
Glucose Sodium SGLT Na depletion tubule
Sodium glucose RAAS macula densa RAAS
hydrostatic pressure glomerulus intraglomerulra pressure efferent
arteriole hyperinflation SGLT2 inhibitor
hard clinical endpoint renal protection renal outcome
prespecified outcome EMPA-REG renal outcome
EMPA-REG high risk (established CV disease) Type 2 Diabetes
full paper

 survival curve

outcome event rate
"time to first event" event outcome outcome
event
"censoring" event censored curve No. at Risk
event censored
EMPA-REG empaglifozin placebo 2
10 25 mg randomized 1:1:1
curve 8 full HD
plot interval plot event plot
y cumulative probability of survival conditional probability
survivor the voice
plot Edward Kaplan
Paul Meier 1958 plot Kaplan-Meier method curve Kaplan-Meier
curve KM curve ()
survival curve
survival probability p value log-rank
test survivial curve log-rank
confounder assumption KM curve log-rank
KM curve log-rank test
assumption KM ? (
) appraise survival analysis
survival curve log-rank test adjust
covariables confounder adjusted Hazard ratio plot
x y cumulative survival cumulative Hazard ratio
KM curve
Kaplan-Meier method survival curve
Cox proportional-hazards model p-value



HOPE-3 cumulative
probability 0 - 100% cumulative prob 10%
y (expanded y axis) "inset"
small graphic presentation incorporated into a large one
cumulative HR p-value 0.001
Empaglifozin renal composite outcome time to event placebo

Post Hoc Post Hoc

hyphen Post Hoc "after the event"
Post Hoc analysis
prespecified analysis

EMPA-REG prespecified renal outcome composite 4 incident
worsening nephropathy
progression to macroalbuminuria (urine albumin-to-creatinine ratio > 300)
doubling of serum Cr + eGFR 45
initiation of renal replacement therapy
death from renal disease
modified intention-to-treat empaglifozin macroalbuminuria at baseline prespecified
outcome survival curve

outcome progression to macroalbuminuria composite 3

macroalbuminuria

 post-hoc analysis clinical question

clinical trial TIMI SAVOR-TIMI 53
DECLARE-TIMI 58 registerted trial TIMI study group HQ Brigham
Eugene Braunwald
TIMI study TIMI TIMI 15b clinical trial
(
IMPROVE-IT TIMI 40)
OPUS-TIMI 16
INTIME II-TIMI 17
TACTICS-TIMI 18
ER-TIMI 19
INTEGRITI-TIMI 20
A2Z-TIMI 21
PROVE IT-TIMI 22
ENTIRE-TIMI 23
FASTER-TIMI 24
EXTRACT-TIMI 25
JUMBO-TIMI 26
PROXIMATE-TIMI 27
CLARITY-TIMI 28
ADVANCE MI-TIMI 29
PROTECT-TIMI 30
TIMI 31
ANTHEM-TIMI 32
DISPERSE2-TIMI 33
TITAN-TIMI 34
PROMPT-TIMI 35
MERLIN-TIMI 36
TIMI 37A

TRITON-TIMI 38
EARLY ACS-TIMI 39
SEPIA ACS-TIMI 42
AVANT GARDE-TIMI 43
PRINCIPLE-TIMI 44
ATLAS ACS-TIMI 46
IC TITAN-TIMI 47
ICET-TIMI 49
IMPROVE-IT (TIMI 40)
ENGAGE AF-TIMI 48
TRA 2P-TIMI 50
ATLAS ACS2-TIMI 51
SOLID-TIMI 52
SAVOR-TIMI 53
ELEVATE-TIMI 56
PEGASUS-TIMI 54
LAPLACE-TIMI 57
ongoing
HPS3/TIMI 55 REVEAL
DECLARE-TIMI 58
FOURIER (TIMI 59)
LATITUDE-TIMI 60
CAMELLIA-TIMI 61

Appraisal of Meta-Analysis

meta-analysis
1
appraise RCT
oral

RCT observational studies
systematic review
meta-analysis
meta-analysis

" appraise meta-analysis RCT"
validity meta-analysis RCT

A B AF 20

angry birds

meta-analysis
? 3

appraise meta-analysis RCT
(1) Heterogeneity statistical between studies heterogeneity

 by chance
heterogeneity
angry birds
by
chance heterogeneity Forest
Plot Test of Heterogeneity
Cochran's Q I2

clinical difference

methodological difference bias

Heterogeneity Hazard Ratio pooled HR
"random-effect model" true treatment effect
Heterogeneity "fixed-effect model"
Forest Plot
random-effect model Heterogeneity pooled HR
random-effect model Heterogeneity
random-effect model Funnel Plot

(2) Small study effect sampling error

Forest Plot
upper lower boundary
meta-analysis
sensitivity analysis
(3) Reporting bias Publication bias Reporting bias
Publication bias Outcome reporting Analysis reporting bias
Publication bias negative study
citation metaanalysis
Outcome reporting bias
Analysis reporting bias p value outcome
revascularization HOPE-3 pre-specified outcome

 meta-analysis HOPE-3
reporting bias
Funnel Plot

Non-inferiority Trial

Non-Inferiority trial ?
Liverpool is non-inferior to
Manchester United
?
2
" 2 " non-inferiority margin

 4
1 3 1-1, 0-1, 2-1, 3-2

4 "sampling error"
margin " 5 " 4
margin 5 4 5
4 falsely conclude
margin
margin 2


N non-inferiority trial
Sample Size N non-inferiority trial margin margin
N margin N expected event rate
(: non-inferiority trial placebo placebo
"active control") expected event rate
placebo event rate N event rate N
low risk population N composite outcome outcome
, event-driven margin N
appraise NI study
placbeo null hypothesis ""
dose-response
relationship 1990 "superiority trial"
generic ""
null hypothesis " (equivalence
margin)" generic equivalence margin
"equivalence trial"
NOAC Warfarin placebo
placebo anticoagulant
margin
"non-inferiority trial"
non-inferiority margin
margin control meta-analysis 50% lower
boundary 95% CI placbeo placebo margin

 placebo non-inferiority trial placebo

margin RE-LY
margin absolute risk difference (ARD) relative risk (RR)
event rate active control N
non-inferior , margin "liberal margin" margin
set margin RR event rate RR
inflation margin RR falsely
conclude non-inferiority margin "conservative margin"
US FDA EMA margin approval process 1.3
margin ( 1.8)
post-marketing trial
non-inferiority ()
1. mean upper bound margin "inferior" reject null hypothesis
2. mean lower bound margin "non-inferior" reject null hypothesis
3. 2 lower bound superiority
non-inferior meet non-inferiority superiority margin
superiority superiority trial
4. mean margin upper lower margin
"inconclusive"
5. mean margin
upper lower margin "noninferior inferior" sample size set margin
.
.
journal superiority,
equivalence non-inferiority trial Lancelot
journal

Sensitivity Analysis

Sensitivity Analysis ?
meta-analysis

oral
sensitivity analysis meta-analysis
meta-analysis
generalizibility (robust)

1.

2. small
study effect reporting publication bias meta-analysis

3. pooled HR fixed effect model
Heterogeneity Test of Heterogeneity power reject null
random effects model
meta-analysis Digoxin European Heart Journal
sensitivity analysis
systematic review 19 Digoxin outcome all-cause
mortality AF CHF AF CHF
CHF sinus AF AF CHF
digoxin plasma level
digoxin

 1 Forest Plot meta-analysis 19

CHF AF appraise meta-analysis Heterogeneity
Funnel plot Funnel plot

Test of Heterogeneity Cochran's Q I2 Q p < 0.001 2 distribution

reject null hypothesis "" Heterogeneity reject ""
Heterogeneity I2 = 85.7% quantitative test Heterogeneity Heterogeneity
pooled HR random effect model
treatment effect overall
HR = 1.21 (1.07 - 1.38), p < 0.01 pooled HR CHF
AF
2 sensitivity analysis 3 19
CHF AF
pooled HR sig AF sig HF
3 sensitivity analysis 6 19
digoxin sig HR = 1.26 (0.91 -1.74), p =
0.16
sensitivity analysis meta-analysis overall HR digoxin
CHF
AF overall analysis sensitivity analysis
digoxin digoxin

Subgroup Analysis

subgroup analysis
endocrine subgroup analysis, p - interaction, p for trend
? ( )

dig up
journal club subgroup analysis

subgroup analysis
subgroup analysis ?
subgroup analysis treatment effect ( HR, RR, AR ) specific end point
primary, secondary tertiary efficacy safety end point
treatment effect ?
outcome outcome

subgroup analysis
?

 ticagrelor CV event study placebo

CV event
Heterogeneity Heterogeneity treatment
effect Heterogeneity
RCT Heterogeneity subgroup result
were consistent across all specified subgroup interaction p value sig
appraise ()
HOPE-3 primary outcome study
control

Heterogeneity statistical test for interaction null
hypothesis (Ho) "" significant p < 0.05
reject null hypothesis ( ) treatment effects
p value p value for interaction p value for trend
overall treatment effect HR 0.92 (0.79 - 1.06) significant
P Trend = 0.005 reject null hypothesis treatment
effect > 143.5 mmHg
appraise subgroup analysis
1 subgroup analysis type I error power
sample size
primary outcome
subgroup analysis subgroup analysis primary outcome
secondary tertiary outcome
2 subgroup analysis primary secondary outcome
Multiplicity Steven Nissen
NEJM Circulation
subgroup analysis Editor
3 prespecified post hoc analysis? prespecified
post hoc analysis bias
interval validity prespecified analysis
controlled analysis post hoc analysis

Ingelfinger Rule

Ingelfinger Rule ?
()
1969 Franz Ingelfinger New England Journal of Medicine
masthead

"Articles are accepted for consideration with the understanding that they are contributed for publication solely in this
journal"
Ingelfinger Rule manuscript
peer-review process

1969 Ingelfinger Rule

violation NEJM
1995 presubmission publication manuscript
peer-review process

redundant publication inflation publication record
1980
pre-publication voluntary peer review
arXiv preprint servers figshare, PeerJ, Github ResearchGate

reviewer 3-4 peer-review process


Ingelfinger rule 1980 Ingelfinger Rule revisit
NEJM editors
" 11 2015 NIH Bethesda SPRINT
20 2015 Data Safety and Monitoring
Board


"
Eric Topol Harlan Krumholz associate editor Circulation Lancet
2015;386:2447-2449 embargo

 peer-review
?
Valentine Fuster Editor-In-Chief JACC JACC 2016;67(7):883-884

Krumholz Topol
Hippocratic Oath
(1)
SPRINT
Data Safety and Monitoring Board 9400
manuscript editorial board
NEJM peer-review 9 2015
82 ?
? SPRINT NEJM
?
(2)

peer review
(3) peer-reivew process
peer-review
Eugene Garfield Impact
Factor Chicago 2005
embargo

Ingelfinger Rule SRINT

in-depth analysis SPRINT
https://www.facebook.com/notes/1412cardiology/%E0%B8%A7%E0%B8%B4%E0%B9%80%E0%B8%84%E0%B8%A3%E0%B8%B2%E0%B8%B0%E0%
B8%AB%E0%B9%8C-sprint-in-depth-review%E0%B8%AA%E0%B8%B3%E0%B8%AB%E0%B8%A3%E0%B8%B1%E0%B8%9A%E0%B8%AA%E0%B8%AD
%E0%B8%9A%E0%B8%9A%E0%B8%AD%E0%B8%A3%E0%B9%8C%E0%B8%94/484234981783602

Forest Plot

F cardio
...
()

"... ?"




meta-analysis randomised controlled trials evidencebased medicine
meta-analysis observational studies
mortality outcome severe AS TAVI European Heart Journal
( )


" ... transcatheter aortic valve replacement

intial analysis PARTNER survival benefit

 ... systematic review observational studies

TAVI all-cause mortality TAVI
European Heart Journal 2016
... ?"
1.

2. " ?" appraise meta-analysis yes-no question
" meta-analysis "
3.
"Forest plot meta-analysis 12 observational studies ( odds ratio

) mortality outcome TAVI event
1 -2 Finkelstein Williams
Hayashida crude event rate Forest plot
box odds ratio box 95% CI
margin 1.0 box sample size 95% CI margin
Hayashida, Humphries Sherif estimated pooled odds ratio Mantel-Haenszel method
random-effects model ( M-H Random, M-H pooled OR 2 x 2
tables case-control observational studies strata fixed random-effects model) 0.70 95%
CI 0.59 - 0.82 margin 1.0 ( pooled OR )
" TAVI "
4. ? validity meta-analysis

(1) study selection endpoints criteria systematic review population
TAVI approach transfemoral, transapical, transaortic, transaxillary, transsubclavian device
Edwards
CoreValve targeting sex difference
12
endpoints mortality outcome main outcome analysis
subgroup analysis meta-analysis validity
(2) publication bias bias population

sample size systematic review
meta-analysis Forest plot
Eres publication bias

 plot logarithm estimated effect standard errors

precision inverted cone
bias plot p value
Egger's test < 0.05 publication bias
meta-analysis publication bias p value Egger's sig
(3) heterogeneity meta-analysis variation study outcome
heterogeneity ( homogeneity heterogeneity)
statistical power pooled OR HR meta-analysis
heterogeneity heterogeneity Cochrans Q (
p value < 0.10 heterogeneity) I2 index degree heterogeneity
I2 > 75% heterogeneity I2 < 25% heterogeneity
quantitative qualitative Q meta-analysis
Q power I2
pooled OR HR heterogeneity fixed-effect model heterogeneity
random-effects model Q I2
Random Forest plot random-effects model 12 metaanalysis heterogeneity

Funnel Plot

Angry Birds
Lancelot


non-inferiority, superiority equivalence trial sensitivity
analysis Forest Plot Bland-Altman plot appraise RCT meta-analysis


appraise meta-analysis
3 meta-analysis
Forest Plot
appraise meta-analysis "Funnel Plot"

 Funnel Plot


meta-analysis
combinable

" "
heterogeneity, small study effect reporting bias
Forest Plot


x treatment effect Odd ratio Hazard Ratio
1 neurtal 1 (
1 ) protective effect 1
event Odd ratio
logarithm logarithmic scale 1 0
1 ?
( log)
y precision study
( ... ) sample size
sample size y precision
sample size Standard Error

reverse scale
0 1/SE y
meta-analysis (
) plot
plot




meta-analysis ?
sample size

 meta-analysis pooled estimated Odd Ratio ( X 1

) X
plot X

heterogeneity bias by chance alone
(horizontal scattering)
inverted funnel shape
plot plot pooled OR ( 1 )
symmetrical funnel plot combinable meta-analysis
( )
Heterogeneity (I2 50% p 2 < 0.10) plot
asymmetrical funnel plot random-effect
model pooled OR fixed-effect model
correct Heterogeneity
random-effect model Heterogeneity
Reporting Bias Small Study Effect Meta-Analysis
combinable
Sensitivity Analysis
Angry Birds Funnel Plot
Funnel Plot

Box and Whisker Plot

box-and-whisker plot
...
5
( ) mean
sd box-and-whisker plot ?


outlier
(1) 'box' box arbitrary
box lower quartile (Q1)
box upper quartile (Q3) box median (Q2)
box sample data !! box interquartile range
Q3 - Q1 box
(2) box sensitive outlier mean box
box preserve center spread sample data
(3) box
box box
box

(4) box 'whisker'

 whisker maximum minimum upper lower whisker

spear style 1.5 interquartile range (IQR) median
Tukey style
(5) 1.5 IQR? normal distribution 1.5 x
IQR 2.7 s.d. whisker 99.3% Tukey style
normal distribution spear
type whisker
(6) whisker outlier tail outlier
mean box and whisker
outlier
(7) spear type outlier 1.5 x IQR upper lower fence
far upper far lower fence 3 x IQR far outlier
extreme
(8) outlier normal distribution outlier
outlier normal distribution
outlier (skew to the right)
outlier (skew to the left)
outlier skew box and whisker

(9) box plot notch

median 95% CI median

m +/- 1.58 x IQR/square root n

Bland and Altman Plot

Bland and Altman Plot ?

"" 1 ( )
A B


A 25 B 28
A 34 B 33
A 18 B 28
.
.
.
"limit of Agreement"
"Correlation"
Correlation Pearson Correlation linear regression
r ratio of covariance -1.0 +1.0

 p value r -1.0 +1.0 strong correlation strong

correlation good agreement r A B 0.9, p
= 0..02 strong correlation acceptable agreement
agreement Bland-Altman method
Bland-Altman method agreement quatitative data plot Y
X A B gold
standard plot
"Bland-Altman Plot" agreement visual estimation
plot A B
15 A B 2
plot X 15 Y 2 100
A B agreement
X ideally 0 practically
acceptable agreement 1.96 SD
mean difference bias negative bias
positive bias -27.2
mean difference bar
1.96 SD 1.96 x SD
assumption normal distribution normal
distribution oral
normal distribution p value Shapiro-Wilk Kolmogorov-Smirnov test for
normal distribution p sig reject null hypothesis normal
distribution

Meta-analysis
10 appraise meta-analysis 30
( )


appraise meta-analysis appraise clinical trial
Seminar appraise clinical trial

clinical trial
meta-analysis ( meta-analysis clinical trial
) meta-analysis

appraise meta-analysis "
"
step-by-step
1. meta-analysis
primary outcome meta-analysis treatment effect OR, RR
HR
2. inclusion criteria meta-analysis meta-analysis
all-cause mortality rate chronic symptomatic HFrEF II-IV digoxin inclusion criteria
observation cohort studies 1970 2015
digoxin outcome all-cause death
recruit database

meta-analysis appraise meta-analysis narrative
review
3. pooled treatment effect meta-analysis
treatment effect mean HR 95% CI p-value
crude
event rate, treatment effect 95% CI % weight
Forest Plot %weight
standard error 1/SE N
N SE weight N SE weight

 Forest plot N SE standard error N

fixed effect model
4. appraise meta-analysis Heterogeneity
pooled treatement effect Heterogeneity
true treatment effect treatment effect
by chance only % weight 1/SE weight
pooled treatment effect "fixed effect model" Heterogeneity
clinical methodological difference metaanalysis
fixed effect model true treatment effect
Heterogeneity
meta-analysis narrative review
trim and fill
sensitivity analysis meta-analysis
meta-analysis Heterogeneity subgroup Heterogeneity

"random effect model" pooled treatment effect %weight
inter-trial variance 2 (tau-square) %weight = 1/(SE - 2)
Forest plot N inter-trial variance ( 2 3 ) pooled
treatment effect 95% CI fixed effect model
meta-regression
5. Heterogeneity 4 ?
Forest plot Funnel plot ( 1
)
Cochran Q 2 p-value 0.10 0.05 relatively low power
Heterogeneity reject null Heterogeneity
p value 0.10
I2 quantitative analysis Heterogeneity 25% Heterogeneity
50% 75% Heterogeneity 2
degree of freedom Heterogeneity I2 = ( 2 - df)/ 2
6. Plot Heterogeneity Funnel plot Funnel
plot Angry Birds Funnel plot asymmetry
asymmetry Heterogeneity horizontal scattering plot
treatment effect (small study effect) asymmetry Funnel plot
Reporting Bias ( 2 differential asymmetry Funnel plot Heterogeneity Reporting Bias)

 Bias 100%
Seminar
Egger Horbold-Egger test test
power fail to reject null hypothesis asymmetry
10 Egger Test
Test for Asymmetry p-value sig asymmetry
sig reject null asymmetry

Interim analysis
10 17
Haybittle-Peto 0.026 paragraph Method PEGASUS TIMI 54
10
seminar
Interim analysis
treatment effect safety issue

( boundary )


(accrual)

(study integrity) fraud misconduct
criteria manuscript
supplementary appendix early premature termination
criteria manuscript full report
SPRINTER Eric Topol Full Report
Ingelfinger rule
Interim analysis NIH
Data and
Safety Monitoring Board DSMB completely blind NIH
""

Interim analysis boundary significance level ()

boundary significance level reject null hypothesis
(type I error) "Frequentist"

1. Bayesian approach treatment effect

(probability distribution) prior distribution
likelihood function prior distribution posterior distribution Bayesian
approach treatment effect

2. Frequentist approach treatment effect
treatment effect boundary 95% confidence interval
treatment effect sample
1 type I error false conclusion
Frequentist Interim anlaysis
Group Sequential Approach
R interim analysis
p value

OBrien-Fleming method boundary p value interim analysis

p value final analysis
overall significance level p = 0.049
manuscript overall significance level 0.05

Haybittle-Peto method PEGASUS TIMI 54 fix p value final analysis

overall significance level boundary p value interim
p value PEGASUS 0.05 final analysis
0.05 p interim 0.026 reject null hypothesis
boundary, p-value interim

Pocock method boundary p value interim final analysis

interim analysis reject Ho terminate
Interim analysis interim

 Kaplan-Meier Curve
10 appraise KM curve
Meta-Analysis
KM curve

LAA occluder warfarin
time-to-event analysis rate
Kaplan-Meier risk estimate event rate 12
time-to-event analysis event
censored Endpoint number of event N
% KM risk estimate
step-by-step
1. unit of measurement y cumulative survival, cumulative
probability (event rate) cumulative Hazard rate ( Hazard ratio
) x y cumulative rate
y inset
2. No. at risk
? survival function attenuated
protocol cumulative survival
cumulative risk 12 protocol survival curve
minimal follow up time x
survival rate

3. survival curve


log-rank test reject null hypothesis reject (p <
0.05) pool
interval event union interval
probability survival interval
n estimated event rate
event rate sum 2 degree of freedom = 1 ( n 1)
Cox Proportional Hazard model log-rank test
null hypothesis Hazard ratio = 1.0 reject (p < 0.05) Hazard ratio 1
, Hazard ratio Hazard rate , Hazard rate ?
Hazard rate event next interval interval 0
instantaneous Hazard rate
Cox model Hazard ratio proportional hazards regression
assumption Hazard ratio HR
adjust time-dependent covariate

4. ARISTOTLE Kaplan-Meier method Hazard ratio Cox Proportional Hazard

model mean HR primary composite endpoint apixaban warfarin 0.79 (0.66 - 0.95), p <
0.001 apixaban event
0.79 warfarin magnitude
Hazard rate velocity 0.79 event
warfarin 100/79 = 1.26 outcome
Hazard ratio slope
magnitude x

 Hazard ratio
Cox model