applying meta-analysis to trauma registry ammarin thakkinstian, ph.d. clinical epidemiology unit...
TRANSCRIPT
Applying meta-analysis to trauma registry
Ammarin Thakkinstian, Ph.D. Clinical Epidemiology Unit
Faculty of Medicine, Ramathibodi Hospital
Tel: 2011269,2011762 Fax: 02-2011284
e-mail: [email protected]
Meta-analysis
• A tool for pooling results/data of the same topics from different sources/centres in order to – estimates treatment/intervention effects– leading to reduces probability of false
negative results– potentially to a more timely introduction of
effective treatments/intervention/program– Objective evidence & quantitative conclusion
Type of meta-analysis
• Summary data– Unit of analysis is study– Mean (SD) – Count/frequency data by intervention &
outcome– Person-time data
Summary-data
• Continuous data
Studyi N MeanSD
Rx/Exp+ N1 Mean1 SD1
Cont/Exp- N2 Mean2 SD2
Summary-data
• Categorical data
Studyi Case Control
Rx/Exp+Cont/Exp-
ac
bd
Type of meta-data
• Individual patient data (IPD)– Raw databases – Unit of analysis is patient– Analogous to multi-centre trials – More retrospective than prospective – Data registry
IPD
– Carry out data checking (data validation)– Better standardization of information
•Categorization of eligible participants•Definition of Outcomes•Variables’ Classification
– ICD-10– Type of trauma – AIS
IPD– Flexible to apply statistic modeling – Better adjust for confounders & adjust for the
same confounders simultaneously– More flexible to assess interaction effects – More flexible and capable in assessing cause
of heterogeneity– Allow to assess which subgroup of patients
(centre) that intervention/program may/may not work
– Establishment of international networks of collaborating investigators
IPD• Disadvantage
– Data quality •Missing data•Data validation
– More cost & time consuming– Substantial effort and infrastructure require to
•Develop & administer a standardized protocol
•Collect, manage, & data management •Communicate with collaborators
Data collection & management– Data Registry
Databases
Data coding
Data entry
Cleaning Checking
Validate data
QC
Hospitals
Data manager
Validated Data
Retrieve databases
Combine data
Re-check data
Analyse data
Statistician
Report results
Writing report (manuscript)
Publish (annual, twice/year)
Data analysis
• Heterogeneity test – Different source data are homogeneous?
• Homogeneity
n/programinterentiojth theofeffect thet
centreith theofeffect the
j
i
ijkijijk
c
ctY
Analysis • Heterogeneity
hospital j in individual ifor error individuale
effect) ion(intervent slope regressionβ
hospital jfor intercept
j hospital in i individual
ththij
1
0j
β
ij
exββY ijijojij 1
Outcomes
• Death/alive • Disability/Non-disability • Complications
– Infection– Fracture
• Hospitalization • Hospital days• QoL• Cost
Count (discrete) outcome • Poisson regression
– Number of death – Number of infection – Number of disability – Number of fracture
Intervention Period Pop No. of death
Death /1000
RR
Audited 2001-2005 87870 3926 45 0.65 (0.62, 0.67)
Non-audited 1999-2003 228243 13120 57 1
RRe
ex
jj
j
j
ij
ijijjij
11
1
101
0
11
1
hospital-audited if 1hospital audited-non if 0i
1
lnln
:hospital audited-non versus-Audited
ln
:hospital-Audited-Non
ln
:hospital-Audited
j hospitalin ion interventi of rate]log[death ln
effecton interventi
oninterventix
ln
Hospital standardised mortality ratio
HSMR
• Definition – The ratio of actual number of deaths to
expected number of deaths in the hospital
Expected number of deaths
k
iij xββ
P
P10ln
• Original HSMR• X
– Age in year – Sex– Admission category
• Emergency versus elective
– Length of stay – Diagnosis group
• Account for 80% of death
– Co-morbidity• Chalson’s index • Might be able to use AIS scores
– Transfer• Patient was transferred from acute care
Step of analysis
– Fit logistic regression with death as the outcome – Estimate probability of death from the logit
model
– E = sum(p)
k
jjj
k
jjj
Xββ
Xββ
k
jjj
e
e
Xββ
Yp
1
0
10
1
exp1
1)1(
10
Modified HSMR• age in year • sex• Length of stay • Admission category
– Emergency vs elective • Transfers
– Acute care
• Diagnosis group – Account for 80% of death
• Co-morbidity– Chalson’s index
• age in year • sex• Length of stay • Patient transferring
– Ambulance – Non-ambulance
AIS scores Add• Risk behavior
• Alcohol • Transquilizer/sedation
• Type of trauma
Un-audited-hospital
Audited-hospital
HSMR=1
01
002
003
004
005
006
00H
SM
R
1999 2000 2001 2002 2003 2004 2005
YearHSMR= Actual death/Expected death x100
Hospital standardised mortality ratio by Audited-unaudited hospitals and years
Problem
• Missing – Diagnosis – Co-morbid– Length o stay
• Data validation??