BCH / CCHMC Site Report Ingrid A. Holm, MD, MPH John Harley, MD, - - PowerPoint PPT Presentation

BCH / CCHMC Site Report

Ingrid A. Holm, MD, MPH John Harley, MD, PhD June 29, 2015 eMERGE Steering Committee Meeting

Aim 1. Demonstrate real-time execution of phenotypic selection across two distinct pediatric institutions with disparate EMR (electronic medical record) systems as a model for ensuring phenotypic standardization and for national scalability.

The need: To pool the data assets of multiple institutions into a larger virtual or physical infrastructure to facilitate discoveries that might not otherwise be detected

Results

• CCHMC and BCH have created de-identified extracts of their EHRs using

i2b2 which can be queried at each institution using i2b2.

• The eMERGE investigators from BCH and CCHMC have used the i2b2

workbench for the development and validation of phenotype algorithms.

• To enable cross-institutional query capabilities, we implemented the

Shared Health Research Information Network (SHRINE) framework on a subset of the i2b2 databases (those patients with a sample in dbGaP).

• SHRINE allows local queries to be broadcast simultaneously to the i2b2

instances at BCH and CCHMC, returning the aggregate results.

• Used SHRINE for the eMERGE phenotypes: ADHD, Appendicitis,

Asthma, Atopic Dermatitis, Autism, C. difficile, and Severe Early Childhood Obesity.

• Manuscript under development – Keith Marsolo (CCHMC) and Jonathan

Bickel (BCH)

Aims 3 and 4

CCHMC Cynthia A. Prows, MSN, CNS, FAAN Melanie Myers, PhD, MS, LGC BCH Ingrid A. Holm, MD, MPH Cassandra Perry, MS, CGC

Aim 3. Use return of CYP2D6 research results to explore parents responses to and use of their

children’s research results and better

understand the factors that influence their incidental findings decisions

We hypothesized that parents of children with prior narcotic analgesia history would be more likely to share research results than parents of children without exposure histories.

Methods – CCHMC and BCH

• Results of CYP2D6 genotyping returned to parents

– Categories of metabolizers:

• Poor metabolizer
• Intermediate (CCHMC only)
• Extensive (Normal) metabolizer
• Ultra-rapid metabolizer

– Actionable vs. non-actionable

• Result delivered by telephone and parents participated in a survey

– Telephone scripts for result disclosure

• Subset of parents invited to participate in interview

– To more fully understand responses to results

• Cases vs. Controls

– Cases – previously exposed to narcotics – Controls – no previous exposure to narcotics

Methods Differences between CCHMC and BCH

• CCHMC

– DNA samples were assayed for 20 different forms of CYP2D6 and full gene duplication, gene deletion – Follow-up telephone surveys at 3 and 12 months – Compare intent to share with reported sharing at 3 and 12 months

• BCH

– Hypothetical results rather than actual results

• BCH Pharmacogenomic Oversight Committee deemed CYP2D6 testing

not yet ready for BCH population

– Equal random assignment of hypothetical results to the 3 metabolizer groups – Follow up surveys not being done

Enrollment

CCHMC Target CCHMC Actual BCH Target BCH Actual

Enrolled 200 203 200 99 Genotyped 200 180 N/A N/A Cases 100 92 100 (assigned) 58 Controls 100 104* 100 (assigned) 41 Actionable 133 (assigned) 61 Not actionable 67(assigned) 38 Survey at RoR 200 148 200 99 Interview at RoR Up to 60 61 Up to 60 Reached saturation at 22 Cases Up to 30 30 Up to 30 12 Controls Up to 30 31 Up to 30 10 Survey at 3 months 100 80 N/A N/A Survey at 12 months 100 49 N/A N/A

*So far, 12 parents who were enrolled as controls, became cases by time results returned and survey conducted (case = parent of child exposed to opioid)

Parent/Doctor Can Use Results to Care for Child

100 99 100 100 1 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Parent Boston Parent Cincinnati Doctor Boston Doctor Cincinnati Strongly or Somewhat Disagree Strongly or Somewhat Agree

Likelihood of Sharing Results

100 98 52 68 49 73 2 48 32 51 27 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

Doctor Boston Doctor Cincinnati Pharmacist Boston Pharmacist Cincinnati Child Boston Child Cincinnati

Extremely or Somewhat Likely Not Very or Not At All Likely

p < 0.01

Qualitative Interview Preliminary Results

• Despite CYP2D6 results, parents perceived

their children as normal.

• Parents’ emotional reactions were influenced

by their expectations about the results.

Aim 3 Preliminary Conclusions

• Parents feel they and their child’s physician can use the

child’s CYP2D6 research result to provide care for child

– Child’s previous exposure to opioids and actionability of research result may be associated with perceived utility of result

• Parents report that they are likely to share their child’s

CYP2D6 research result with their child’s physician

– Actionability of CYP2D6 research result was associated with parents likelihood of sharing the result with the child’s physician

• r pharmacist but not with the child

Work in Progress

• Finishing surveys

– BCH – initial surveys – CCHMC – follow-up surveys at 3 and 12 months

• Interviews completed and coded
• Manuscripts

– Qualitative interviews – Joint manuscript to be submitted this summer – Manuscripts under development

• The type of information parents request when being told a

pharmacogenetic research result

• Factors that contribute to differences regarding intent to share CYP2D6

results

• Similarities and differences between parents’ responses when told actual

versus hypothetical CYP2D6 results

Aim 4. To explore clinician perceptions of pharmacogenetic research results after EHR integration

• To date, CCHMC has distributed surveys to 67

unique providers and 32 have responded.

– Survey distribution requires participant permission at 3 month follow up survey.

• BCH distributing hypothetical results and a

modified survey to 32 pediatricians, 7 have responded.

CCHMC Physician Surveys

• 57 out of 63 parents gave permission at 3 month follow up to

send results to child’s provider

• 42 surveys with results (39 EM, 1 IM, 1 PM, I UM) mailed &

22 providers responded

– 41% felt result could potentially improve child’s health outcomes, 55% unsure – 82% plan to enter result in child’s EMR – 91% agree research results that could potentially improve health

• utcomes should be returned to research participant
• Will assess for differences between physicians who receive

actual CYP2D6 research results and physicians who receive hypothetical results

Aim 4 Preliminary Conclusions

• Providers consider it important to return research

results that could potentially improve their patients’ health outcomes

Acknowledgements Aims 3 and 4

Parents / Participants eMERGE Team CCHMC Cynthia A. Prows, MSN, CNS, FAAN Melanie Myers, PhD, MS, LGC Matt Veerkamp, BA Sarah Adelsperger, BA Beth Cobb, MBA Laura Gaines, MS Diane Kissell, BA John Lynch, PhD Brooke McLaughlin, MS Andrea Murad, BA Ashton Roach, BA Carrie Weaver, MS Kejian Zhang, MD, MBA Nicole Dalessandro, BA Rachel Supinger, BA eMERGE Team BCH Ingrid Holm, MD, MPH Cassandra Perry, MS Ariel Chandler, MS Wendy Wolf, PhD Sonja Ziniel, PhD Pharmacogenetics Survey Susanne Haga, PhD

Aim 2: Contribute to phenotype selection and perform the GWAS & PheWAS selected by the eMERGE II Steering Committee

Aim 2: Contribute to phenotype selection and perform the GWAS & PheWAS selected by the eMERGE II Steering Committee

• Early Childhood Obesity – Completed (2 papers in development)
• Autism – Completed (revisions requested JAMIA)
• Appendicitis – Completed
• CCHMC/BCH-PheWAS study – Completed (paper published)
• QTL- GWAS analyses for Obesity-BMI% – Completed

(paper published)

• ADHD – Completed (secondary validation)
• Asthma – Completed (secondary validation)
• Atopic Dermatitis – Completed (secondary validation)
• C-Diff – Completed (secondary validation)
• GERD – Completed (secondary validation)
• Bilirubin and LFTs- Completed- (draft circulating)

Early Childhood Obesity (ECO)

• Algorithm final, tested (PPV=90%), validated (Vandy, CHOP)

– Geisinger – site specific implementation & future projects

• EMR-linked childhood BMI. Namjou et al 2013

– COL6A5, a novel association, p<10E-8, MAF ~5%. – FTO, MCR4, CCK, IL15, LRP1B, TMEM18, all previously known, also identified 2 for the first time in pediatric cases.

• Two manuscript concept sheets for early obesity

– Algorithm construction (Lingren) – Clinician variation in EMR documentation (Thaker, Brady)

• Organizing for Sequencing and GWAS for binary trait and QTL.

– Controls, <85% BMI; cases, >99% BMI (Vanderbilt [cases=109, controls=369]; CCHMC [cases=130, controls = 478] [total=1,086]).

Namjou et al Front Genetics 10.3389, 2013

2860 subjects, 28% obese 9.8 yrs, average age OR=1.61 for FTO Best FTO variant, p<10-8

BMI in Children

Extracted BMI from the EHR & performed QTL

>5 yr slope=0.25/A allele <5 yr slope= 0.075/A allele

Namjou et al Front Genetics 10.3389, 2013

First Pediatric PheWAS

(Namjou, Front Genet 2014)

• 4268 GWAS records
• 2476 PheWAS SNPs
• 93,724 ICD-9 codes, 1402 groups, & 14 concepts
• Findings FDRq<0.05 & : 24 Replications

Pediatric thyroiditis with PTPN22(OR=3.5 q<0.001) Uveitis with HLA rs477515 (OR=6.5, q<0.0001) T1DM with HLA-DRB1 (OR=2.7, q<0.00002) JIA with HLA-DRB1 (OR=0.36, q<10E-13) Food allergy with CLEC16A (OR=0.67, q<0.00001)

Pediatric PheWAS (cont.)

• Findings at FDRq<0.05 & β>0.8: 14 novel results

Allergic Rhinitis & GCKR (OR=1.4, pP<0.0001) Allergic Rhinitis & JAZF1 (OR=0.75, pP<0.0001) Eosinophilic Esophagitis & IL5-IL13 (OR=1.7, pP<0.00001) Mental Retardation & NDFIP1 (OR=1.7, pP<0.000002) Developmental Disorders & PLCL1 (OR=0.65, pP<0.000002) Depression & near RGS1 (OR=1.8, pP<0.000002) Supporative Otitis & ~RGS1 (OR=1.8, pP<0.000002)

Early Childhood Obesity Case Recruitment

• BCH

– IRB Approval Process (12 Months) – Standard recruitment procedure (letters, calls) – Targeted Clinics – 194 approached, 35 enrolled (18%) – $400/patient recruitment cost

• CCHMC

– EHR based recruitment – Targeted Clinics – 23 approached, 17 enrolled (73.9%) – $53/patient recruitment cost

Comparative Analysis of Electronic Health Record (EHR)-driven and Conventional Cohort– driven Genomic Research, Thaker, Lingren, et al ASHG 2014.

Autism – Rule Based

BCH 147 patients PPV: 0.89 Notes: 10+ years

CCHMC 152 patients PPV: 0.84 Notes: 3+ years

26% 55%

Yes (Case) No Maybe Unknown

60 % 18 %

Larger set of longitudinal data provides a clearer set of ‘yes’ cases. Heuristic algorithm is sensitive to minimal signal (>1 feature of autism)

Autism – Machine Learning

Best Model PPV BCH

0.86

CCHMC

0.83

Combined

0.84

Two-Stage SVM Classifier

ICD9 Baseline PPV BCH

0.27

CCHMC

0.65

Combined

0.46

Autism – Comorbidity Clusters

• 25,469 Patients (3 sites)
• 20,658 ASD cases from

Rule Based Algorithm

• 100-200 PheWas Codes
• Key Clusters cases

Psychiatric Developmental Seizure

Appendicitis

• NLP to parse Pathology reports
• Rule Based Algorithm – Prediction of Appendicitis

PPV -- CCHMC: 93%, BCH: 100%

• Network-wide collection of case/control

– Case: 890 , Control: >20,000

• Working on Genetic analysis:

– >500 cases from eMERGE samples have been evaluated – Genotyping effort on up to 500 cases with collaboration of PSU – Secondary confirmation using External dataset (23&Me)

Manhattan plot of Total Serum Bilirubin in 3294 European samples

Strong effect in UGT1A1 (P<10-118)

*The effect was solid in both adult and pediatrics.

• finalizing manuscript

PGRN-Seq

• CCHMC submitted 811 samples from 6 different study

cohorts from which, 776 PGRN-seq results were received and uploaded into Sphinx and dbGap. – 402 contribute acute pain and opioid response phenotype – 139 contribute carefully phenotyped methylphenidate response – 34 have malignant hyperthermia susceptibility

• 201 Pre-emptive sequencing 6 – 21 year olds at risk for

requiring opioid therapy for surgical pain management

Enrolle d ROR CLIA Genotyp e PGRNs eq Data Phenotype

BCH Pre-emptive 109 17 109 Epilepsy CCHMC Pre-emptive Opioid CYP2D6 testing + PGRNseq) 251 232 201 Surgical Pain (Bone) CCHMC Stored Research Samples 590 N/A

• Tonsillectomy

374 Surgical pain / MH controls*

• Malignant

Hyperthermia - possible 34 MH susceptibility – trigger agents avoided*

• Oxycodone PK

28 Surgical pain

BCH, CCHMC PGx Projects

*IRB approval to be sought for known pathogenic, clinically relevant RYR1, CACNAS1 variants

Screening algorithm (43 records): MH 995.86 ICD-9 (but no dx), family hx of MH, Central Core Disease, King Denborough Syn, Multiminicore disease by NLP Chart review: 34 accepted RYR1 (ryanodine receptor) and CACNAS1, two genes in PGRN related to MH

• Table. Preliminary results in 34 potential MH cases-CCHMC-PGRN.

In RYR1, 5 samples were carriers (HTZ) for a potential deleterious mutations

Gene Name Substitution dbSNP ID Prediction SIFT Score* HMZ HTZ Compound HTZ Global MAF - Hapmap

RYR1 F4126S novel DAMAGING 0.02 1 NA

RYR1 D4500H rs150396398 DAMAGING 0.01 3** 0.003**

RYR1 R5025H novel DAMAGING 1 NA CACNA1S L1800S rs12139527 DAMAGING 0.02 1 8 1 0.22 CACNA1S R1539C rs3850625 DAMAGING 0.01 10 0.08 CACNA1S S606N novel DAMAGING 2 2 NA

*SIFT Score, Ranges from 0 to 1. The amino acid substitution is predicted damaging is the score is <= 0.05, and tolerated if the score is > 0.05.

** D4500H has OR=25.5, p=0.0001 (binomial test)

PGRN-Malignant Hyperthermia

Acknowledgements for Aim 2 & PGx

• Bahram Namjou
• Beth Cobb
• Robert Carroll
• Nancy Crimmins
• Josh Denny
• Tanya Froehlich
• Ingrid Holm
• Sara Lazaro
• Rongling Li
• Todd Lingren
• Ken Kaufman
• Stephanie Kennebeck
• Zak Kohane
• Leah Kottyan
• Lisa Martin
• Keith Marsolo
• Yizhao Ni
• Cassandra Perry
• Marylyn Ritchie
• Imre Solti
• Michael Wagner
• Wendy Wolf

40 CCHMC/BCH site pubs 2012-2105

• Algorithm for medication discrepancies –

Li et al 2015

• Strategy to penetrate the GWAS firewall –

Kottyan et al & Lu et al 2015

• Automated clinical trial screening – Li et al 2015
• Pharmacogenomics of opioids at FAAH –

Sadhasivam et al 2015

• The 8q22 inversion – Namjou et al 2014
• Macrophage Activation Syndrome is related to

Hemophagocytic Lymphohistiocytosis through rare genetic variants – Kaufman et al 2014

40 CCHMC/BCH Pubs (continued)

• Machine learning prediction of intensive care

transfer for newly hospitalized children – Zhai et al 2014

• Safety: Detecting drug adverse events &

medical errors using the EMR in the neonatal intensive care unit – Li et al 2014

• GWAS of Eosinophilic Esophagitis –

Kottyan et al 2014

40 CCHMC/BCH Pubs (continued)

• Complications of subject de-identification –

Deleger et al 2014

• Codeine adverse drug reactions in children –

Prows et al 2013

• Automated algorithm to identify appendicitis in

the pediatric emergency room – Deleger et al 2013

• Crowdsourcing for gold standard natural

language processing – Li et al 2013

Inversion at 8p32.1

BLK associated with lupus, scleroderma, Sjögren’s, rheumatoid arthritis, anti-phospholipid, Kawasaki’s, & dermatomyositis/polymyositis

• BLK = B lymphoid tyrosine kinase
• BLK is in the largest human common inversion,

4.5 mb & 12-60% MAF

• Non-inversion status (OR=1.18, p<10-6) and risk

haplotype at BLK are additive for lupus risk from 7,416 Europeans & 267 in HapMap

• Namjou et al, PLOS One, 10.1371, 2014

…as of Aug 2014 11,115 GWAS hits at p<10-5 in 866 phenotypes & traits

GENETIC CAUSE KNOWN &

MECHANISM EXPLAINED

for disease risk …

…as of Aug 2014 11,115 GWAS hits at p<10-5 in 866 phenotypes & traits

~1% causal

• f variants

identified, cell type for differential action known, & Mechanism Understood ! GENETIC CAUSE KNOWN &

MECHANISM EXPLAINED

for disease risk …

Identifying independent genetic effects from the trans- ancestral meta-analysis of the IRF5-lupus association

Study Design

(from LLAS2)

8,395 SLE Cases 7,367 Controls 113 SNPs (CGGGG)3 vs 4 DNA Sequencing Imputation (6,016; 1,434)

7,560 markers

Cap aptur ture e al all l Var ariation! ation!

Variant 1 Variant 2

No adjustment

No adjustment Identifying independent genetic effects from the trans- ancestral meta-analysis of the IRF5-lupus association

IRF5 Promoter

IRF5/TNPO3

Variant 1 Variant 2

IRF5 (kottyan et al 2015)

• Principles for model optimization

– Complete genotyping in multiple ancestries – Parsimony with Bayesian and frequentist methods

• IRF5: SLE, RA, IBD, MS, Sjogrens, Scleroderma

– 7,530 in and around IRF5 (~200kb) variants to:

• 4 variants in promoter in all ancestries one variant is

differentially bound by ZBTB3

• 24 variants in an 85.5 kb Neanderthal haplotype IRF5-

TNPO3 found only in Europeans

• Four previously attributed functional variants excluded.

• THE END… !