Kuan-Po Peng MD Kuan-Po Peng MD Adjunct Lecturer
National Yang-Ming University, Taiwan Taipei Veterans General Hospital, Taiwan
Disclosure
• Nothing to declareg
Experience from the Taiwan National H lth I R h D t bHealth Insurance Research Database
Today’s Overview
• Secondary data analysis on migraine-claims database-study design and statistical model-limitation and pitfalls
• Validation of diagnosis• Drug persistence, economic cost, or
adverse eventsadverse events
Taiwan National Health Insurance Research Database (NHIRD)Research Database (NHIRD)
Comprehensive reimbursement
Secondary data analysis
Strength WeaknessStrength WeaknessLarge number Population-
basedDiagnosis code ≠ Diagnosis
based
Extensive coverage gprescription, hospitalization, comorbidities.
ReimbursementPurpose
Coding error(inaccuracy)
i.e. NHIRD migraine cohort (n=759,027)
p ( y)
Clinical observation
3 migraine patients Hypothesis Migraine patients
developed yp
linked to Bell’s palsy?Bell’s palsy Bell s palsy?
Why use claims database?
LargeLarge l
Disease with low incidencesample
size
Statistical model I: cohort studyA i ti St d
Migraine cohort
Association Study
Follow-up
Comparison cohort(i.e. non-headache)
Outcome incidence comparison
Follow-up
p
Migraine and Bell’s palsy
Case Selection Comparison Selection
Case ExclusionComparison ExclusionExclusion
Case - Comparison Matching
Peng et al. Neurology 2015; 84: 116-24
Key issue:Validation of diagnosisValidation of diagnosis
Validation study Sensitivity analysisValidation studyNeurologist diagnosed migraine 91.6% accurate
Sensitivity analysisBell’s palsy aOR P-value
all 1 91 <0 001gusing ICHD-2
Ch t i
-all 1.91 <0.001-without pregnancy 1.91 <0.001-dx by neurologist 1.78 <0.001
Wang et al. J Formos MedAssoc. 2008;107:485–94.
Chart reviewy g
-steroid prescribed 1.53 <0.001-all of the above 1.59 <0.001
Yang et al. Ophthalmology. 2016 Jan;123(1):191–7.
Peng et al. Neurology 2015; 84: 116-24
Validation: alternative strategy
Improve coding Combine coding with p gaccuracy• Consecutive coding:
prescription / procedure• Pulmonary embolism +
e.g. ≥3 migraine diagnosis codeC d d b
anticoagulant prescription
• Codes made by specialist
• Myocardial infarction + cardiac catheterization
Key issues in cohort study
Comparison Statistical PossibleComparison cohort
Statistical model
Possible pitfalls
Matched on• Sex• Age
e.g. Cox Regression
• Unaccounted covariates
• Age • Propensity
score
ganalysis, with correction
• Medical surveillance
• …... bias
Uncorrected comorbidities
https://commons.wikimedia.org/wiki/File:Migraine_Comorbidities.PNG
Unaccounted covariates
Surveillance bias
SES
DIA
GN
OS
mbe
r of D
Num
Number of CLINICAL VISITS
Overcome surveillance biasP it S M t hiPropensity Score Matching
Age, Sex Social demographic, behavior
Migraine Control Comorbidities Adapted from http://www.summitllc.us/propensity-score-matching
What have we found?What have we found?
Migraine is associated CN disordersO hth l l i T i i l l i SSNHL
HR: 2.67 HR: 4.23 HR: 3.37 HR: 6.72
Ophthalmoplegia Trigeminal neuralgia SSNHL
HR: 1.8
BPPV
Y t l O hth l l 2016 123 191 7
HR: 2.03
Yang et al. Ophthalmology. 2016;123:191–7.Torpy et al. JAMA. 2013; 309:1058Lin et al. Cephalalgia. 2015 Dec 20. (advanced online) Chu et al. J Headache Pain. 2015; 16: 62.Chu et al. Cephalalgia. 2012;33(2):80–6.
Migraine and ischemic stroke
Peng et al. Cephalalgia 2016, Epub ahead of print
Migraine and venous thromboembolism
Accepted in Headache
Statistical model 2: Nested case control studyNested case-control study
ExposedExposed
Non‐exposed
Migraine patients without stroke
Non exposed
Exposure Comparison
Migraine patients
Exposed
Migraine patients with strokeNon‐exposed
Triptan use and ischemic stroke
Unpublished data
Nested-case control approach
Strength and statistical model selectionmodel selection
Sh t t E S i i lShort-term exposure on outcome
Exposure over different time interval
Statistical correction of covariatesoutcome
(e.g. drug)interval covariates
Implication from our studies
Associated with all migraine
Migraine with aura-specific associationsall migraine spec c assoc at o s
• Cranial neuropathy • Ischemic strokerelated: CN3,4,6, CN5, CN7, CN8D i
• Venous thromboembolism
MA specifically • Depression (not listed)
• Renal calculi (not listed)
p ylinked to vascular comorbidities?
Why vascular disease and MA ?
Central Hypertension
Vascular Comorbidity
Vascular ResistanceHypertension
(Schillaci, 2010)Comorbidity(Scher, 2005)
Resistance(Schillaci, 2010 )
Poorer Patent ProthromboticPoorer cholesterol profile
Patent foremen ovale(Carod-Artal, 2006)
Prothromboticstate (Hering-Hanit, 2001)profile
(Scher, 2005)
Drug related studies
Drug adherence(Triptan)
PPI and headache(Triptan)
• Better when prescribed by • Esomeprazole and neurologist
• Worse at local medical li i
lansoprazole increased the risk of headache
clinic
Chen et al. J Headache Pain. 2014;15:48. Liang et al. Cephalalgia. 2015;35:203-10
SummarySummary
TED speaker / data analyst Kenneth CukierKenneth Cukier
Big data is better data
BUT
JUNK IN JUNK OUTJUNK IN JUNK OUT
Methodology mattersMethodology matters
Acknowledgement:• My mentor: Prof. Shuu-Jiun Wang• My fellow neurologist / photographer:• My fellow neurologist / photographer:
Vincen Chuang, for all the Taiwan photos