Avoiding Fake Diseases and
Biomarkup In Pediatrics
Isaac S. Kohane, MD, PhD
- HARVARD OEPARTMENTOFv MEDICAL SCHOOL Biomedical Informatics
Google Maps: GIS mayers Organized by Geographic-al Positioning
Information Commons Organized Around Individual Patients
ExpoJ:oma
Signs and Symptoms
Genome
Toward Precision Medicine: Building a Knowledge Network
for Biomedical Research and a New Taxonomy of Disease
Report from National academy of science, USA, 2011
CErA.RTMEHTOr
lliomedical Inform atics
Some of the points I will make
Most population genetics and clinical medicine addicted to categorical diagnosis
Fake/Wrongly taxonomized diseases makes social & commercial construction of disease likely.
Genomics-first/Genomics only unnecessarily limiting
Clinical predictive accuracy does not imply shared physiology
Disease overlaps demonstrates a fundamental problem
Clinical data can be used to lesson the confusion.
But multi-modal approaches are the most robust.
- HARVARD OEPARTMENTOFv MEDICAL SCHOOL Biomedical Informatics
Blood Gene Expression Detection
A
'.3
"l Q) 0
ro 0::: Q) "'0> ~ oo 0 0.. '
Invited to HLS Meeting
"I think when Ari [Ne'eman] talks
about autism and I talk about
autism, we're talking about people
with different clusters of autism. I
know he doesn't like the word
'cure.' If my daughter could
function the way Ari could, I would
consider her cured," says Singer. "I
have to believe my daughter
doesn't want to be spending time
peeling skin off her arm."
DEPARTMENT OF Iii HARVARD Biomedical Informatics
MEDICAL SCHOOL
Extractand pool
pttu1tary g1ands Homogenize Inject
30-year CJD and APe incubabon deposits.
CEPARTMENT OfIii HARVARD Biomedical Informaticsv MEDICAL SCHOOL
GIANT study
lOO's of genes
Not like diseases at tails
So what do you call short stature and is that a diagnosis?
v DEPARTMENT orIll HARVARD Biomedical InformaticsMEDICAL SCHOOL
Criteria for Treatment
''Growth hormone deficiency (GHD)''
''Idiopathic short stature (ISS), defined by
height standard deviation score ~-2.25''
associated with growth rates unlikely to result in normal adult height, in whom other causes ofshort stature have been excluded
and a little story from 25 years ago
v DEPARTMENT orIll HARVARD Biomedical InformaticsMEDICAL SCHOOL
"-.._, ........ lt ..1 1 ......... "11 1 111 1"'-"11--&,.;t..1'1 1 '1 1'1.1 ..... l 1 17 1 ...1'-lm J #U ...., VII \..11-t.:I 1 1.0-Y-'"" _i, ...,.,'IO.I' l !l.1.'t -1 I 11'-' UO~ .J i,""' J
NEWS Election U.S. World Entertainment Health Tech
But about once a mornth, an ambitious parent with cash to burn asks 44 SHARES Desrosiers. a pediatric endrocrinologist in Florida, if he would be willing
to g ive growth hormones to a short but otherwise healthy ch ild.
0 0 Desrosiers turns these patients away, but he says the requests still
come.
0 null Just last week. the father of a
young baseball player ~ a 14-yea old who was already 5 feet 6
inches tall - expected Desrosiers to prescribe recombinant growth hormone rGH to add hei ht to his biuddin athlete. rlHe wanted to make his kid big, and he th inks he1 s going to walk out
with the shots,1 r said Desrosiers, director of pediatrric endocrinollogy at
Arnold Palmer Children's Hospital in Orlando. "He was willing to pay
more than $45,000 a year, and didn1t even bat an eyelash."
Desrosiers said he 1even gets requests for growth hormone from 11Jolly
Green Gianr families, where children are likely to be tan.
In 2003, the U.S. Food and Drug Administration approved the use of
rGH for children with idiopathic - or unexplained -- short stature,
without a diagnosed metabolic hormone deficiency.
Pathological
Interaction Between
Clinical Annotation and
Genetics.
- HARVARD DEPARTMENTOF MEDICAL SCHOOL Biomedical Informatics
---- E1Cli8cellu4ar mallix
Dystrophln Associated Protein Complex
-lAMA4
Ul83 MYOZ2 NEBL NEXN TCAP VCl
CRYAB
Cytoplasm
- NUCiear membrane TMPO
Hypertrophic Cardiomyopathy
(HCM}
- Heart failure - Arrhythmias - Obstructed blood flow - Infective endocarditis - Sudden cardiac deat h
4 6503 individuals
NHLBIESP
ttttittttiittii itttittiittttii tiiitttitttttti iittititttiiitt tiiit
expected i100 individuals
6503 individuals
NHLBIESP
iiitttitiittttt
tittititttiittt
iittiitttttttti
ttiiitiiiiittit
ittii
observed
HCM Prevalence= 1 :500 HCM Inheritance = Autosomal Dominant
- HARVARD DEPARTMENTOF ,., MEDICAL SCHOOL Bio medical Informatics
Genetics-induced Health Disparities
30.00%
25.00%
Cl> u c 20.00% ~ IO >Cl> .... 0.
~ >. 15.00% .... 0 c: Cl> Ol Cl>
"'ijj
IO 10.00%
.... 0
-cE
5.00%
0.00%
European Americans
African Americans
European Americans
African Americans
!NE
Potential
JG UST 18, 2016
_J _I ... TNNT2 OBSCN MYBPC3 Remaining 89
TNNl3 (P82S) JPH2 (G505S) (K247R) (R4344Q) (G278E) mutat ions
2.88% 0.33% 0.03% 0.02% 0.80% 6.67%
27.14% 15.27% 4.07% 3.15% 2.92% 7.1 8%
DEPARTMENT OF 13Iii HARVARD Biomedical InfoTmatics
MEDICAL SCHOOL
46 Unavailable 2005 p B Cli nical Diagnosis of HCM
75 Unavailable 2005 p B Family Hi story and Clinical
Symptoms of HCM
32 Black or African American 2005 p B Clin ical Diagnosis of HCM
34 Black or African American 2005 u B Clinical Diagnosis and Family History of HCM
12 Black or African American 2006 u B Family History of HCM
40 Black or African American 2007 u B Clinical Diagnosis of HCM
45 Black or African American 2007 u B Clinical Features of HCM
16 Asian 2008 u B Clinical Diagnosis and Family History of HCM
59 Black or African American 2006 p B Clin ical Features of HCM
15 Black or African American 2007 p B Clinical Diagnosis of HCM
16 Black or African American 2007 p B Clinical Diagnosis of HCM
22 Black or African American 2007 p B Clinical Diagnosis and Family Hist ory of HCM
48 Black or African American 2008 u B Clinical Diagnosis of HCM
Pro82Ser
Gly278Glu
P = Pathogenic and Presumed Pathogenic U = Pathogenicity Debated and Unknown Significance
DEPARTMENT OF " HARVARD Biomedical InformaticsMEDICAL SCHOOL
Example: ' Heart attack" AND 'Los Angeles
Clinical Trials.gov Search for studies: J Seorch A service of the U.S. National Institutes of Health Advanced Search Help Studies by Topic Glossary
Find Studies About Clinical Studies Submit Studies Resources About This Site
Home > Find Studies > Study Record Detail Text Size "'
Valsartan for Attenuating Disease Evolution In Early Sarcomeric HCM (VANISH)
This study ls currently recruiting participants. (see Contacts and Locations) ClinicalTrials.gov Identifier:
NCT01912534Verified July 2013 byNew England Research Institutes First received: June 5, 2013
Sponsor: Last updated: October 5, 2015
New England Research Institutes Last verified: July 2013 History of Changes Collaborator:
National Heart, Lung, and Blood Institute (NHLBI)
Information provided by (Responsible Party): New England Research Institutes
Full Text View Tabular View No Study Results Posted Disclaimer El How to Read a Study Record
0(11,t,fl:TM(l';T (/" iill HARVARD Biomedic~J InformJticsv MED ICAL SCllOOL
http:ClinicalTrials.gov
Group 1 (Overt HCM Cohort)
1. LV wall thickness 2:12 mm and :S25 mm or z score 2:3 and :S18 as determined by rapid assessment by the echocardiographic core laboratory
2. NYHA functional class I or II; no perceived or only slight limitations in physical activities
3. No resting or provokable LV obstruction (peak gradient :s 30 mmHg) on clinically-obtained Exercise Tolerance Test (ETT)-echo within the past 24 months or transthoracic echo with Valsalva maneuver within the past 12 months
4. Age 8-45 years
5. Able to attend follow-up appointments, complete all study assessments, and provide written informed consent
Group 2 (Preclinical HCM Cohort (G+/LVH-))
1. LV Wall Thickness
Is Prediction the Acid
Test of Diagnosis?
- HARVARD DEPARTMENTOF MEDICAL SCHOOL Biomedical Informatics
Survival 3 Years After a WBC Test (White, Male, 50-69 Years; Using Last VVBC Between 7128105 and 7127106)
Repeat
Interval
Result
Time
WBCValue Patients
Low Normal High Any
< 1 Day 12a-8a 43.33% 84.68% 63.24% 76.39% 1830
8a-4p 54.55% 86.61% 79.40% 83.15% 1442
4p-12a 77.30% 67.53% 72.49% 229
< 1 Year 12a-8a 47.83% 79.58% 66.67% 74.39% 1644
8a-4p 76.96% 90.73% 80.80% 88.53% 8812
4p-12a 81 .65% 92.99% 86 .01% 91.69% 2769
> 1 Year 12a-8a 95.65% 96.97% 96.00% 175
8a-4p 97.30% 98.13% 91.98% 97.83% 4280
4p-12a 92 .68% 97.35% 96.67% 97.20% 1932
Any 73.17% 91.79% 78.11% 88.95% 23113
Patients 1286 18775 3052 23113
v ota-.~itTMe:r.ro'1111 HARVARD Biomedical Informatics MEDICAL SCHOOL
Healthcare System Dynamics
Clinical data reflect both patients' health AND their interactions with the healthcare system.
Patient P athophysiology Healthcare System Dynamics Data Quality
Patient Demographics
Diagnoses
Laboratory Test Results
Vital Signs
Genetic Markers
Number of Observations
Time of Day of Observations
Time Between Observations
Cost of a Test or Treatment
Clinical Setting I Clinician Type
Data Entry Errors
Dictation Mistakes
Data Compression Loss
Unstructured Data
Missing Data
Clinical Encounter
Patient + Clinician
Electronic Health Record Data
Healthcare System Dynamics
Normal
Abnormal
Patient Pathophysiology
Normal
Best
Outcomes
Moderate Outcomes
Abnormal
Moderate
Outcomes
Worst Outcomes
DEPARTMENT or
Biomedical Informatics
Predicting Survival from Ordering a Lab Test
--
100.00%
Q)
CtS CtS .::!:: a:: ~~ ::J = 10.00% en CtS -~ 0 0 Q) :!E (.) "C c: Q)CtS ...... 'i:: II) CtS :::::J>:o
"C < > I c: Q)
- C)c:
Predicting Survival Using Lab Value & HSD
"C =Cl> C1> .... "C C'lS 0
~:e O"C (.) ~ cu :c .~ E c:: 0 ::::sU
"' - C1>o~ C1> a:: g~ ta=
.::: C'lS
~~ "C :e
C1> "C .~ C1> .!!! ti c. ::::J >< Cl> "C c: < ::> Q,c: ()
- C'lSc: a::0 I - >< - C1>ucn::::J I "C C1>
C1> C'> a:: < ~o 0 ~
80%
70 %
60%
50%
40%
30%
20%
10%
0%
_______.I +. ..I______,
MHCT
URIC -
UIUCRE
ALB -. ROW """'e MPHS ALKP CDS
I POLYS i.. e TBILI GOT T3 e ~ EOS ~ IL/
T/U~RE H B IT~ B~~RE~' MONOSW ANI N:} SGPT ALYMPHS FTP "lTIBC ;-.e RBC NA~. e e
I PTH e HCT CAe -'e ._ CL [ MC K _; :A-~co GLU WBC FSP/DD FMONO FER PL T -{ MCHC fr L MC
F-NON JG __. FE POL~ ___!. I e-- MONO ALC-TSTES -. - e ESR_ e e e e e uAS-RBC UA-PH LYM ~ ~12- FOLATa A osN+ TRIG UIUNA UA-SPGR
U/UOSM LDH 25VITD PHOS EOSP JI ~ /CPK e VPC02 CHOL e e ~ ,[ T H 9 .,, e UA-KET 1 e NEUT e ...- e e UA-BLD
OSM HDL \ CORT e , e e DIG MONOS e LDL e e CREACJ ~ VLDL e -[!_-UNID CBASOUUN T
e e e B~HGHBA~C e ; CWBC UK ~ OI L LACT F-WBC e GLU e Vi2 _,,,. U~~~~I~~ ~ ~ I .t\fC 2 PTT ._A YPS
UCL ' e e e e e AMY e PT \.. ~2 BAND e
/ ' _A '9 e MG F-A YP VPH
MALBCR~ F-L Y e e BP0 2 e e ~ae:: IC PT-LNRN_e FBLAST.....--. H3 HC03 UNA ~ F-MONO a e ~J 02 Sat '.. ,r ~ BPH e AP02 PSA ._ ABANDS META e F-EOS
APH TRVA C CP:..MB ~ TROP-T MYELO UA-UROBI
UA-NIT
~ TROP-I __T CNRBC
UA-WBC
0.1 0.2 0.3 0.5 2 3 1005 20 30 50 DEPARTMENT OF Improvement from Only HSD I lmproveme .' ~ .t"J~VM.D Biomedical Informatics
MEDICAL SCHOOL
Clinical Data To
Clarify Diagnostic
Boundaries?
- HARVARD DEPARTMENTOF MEDICAL SCHOOL Biomedical Informatics
Mothers told me about bowel
problems but pediatricians told me...
code counts code counts code counts 0-6 months 6-12 months 12-18 months ___A A A I T' V' --,
-"' -
-,, ' I I I I
patient clustering
patients :
DEPARTMENT OFlliv H.PtRVARD Biomedical InformaticsMEDICAL SCHOOL
Autism or
Autisms?
0 co
0 N
0
t:::.
I \
I /:;.
\
/:;. !:;
o POD \ o CP D. t:::. Epilepsy
t:::.t:::. I
I t:.
;\ / t:.
I /:.,
t:.
t:. /:;.
0
D 0 , \
a . D ~ : o/ o \ ,o'o' a D D o,,o II I
l:. I 0t:::. t:::. B \ I 0
0
I t:::. I 0t:::. ~ 0I ' t:::. I0 0
t::. I I \ 0 /:;. 0 0 0~ I 0 I0 0
0 0 a I aI I 0 00
Ill 0 0 0
0 5 10 15
lli HAaRVARD DEPARTMENT OF MEDICAL SCHOOL
Biomedical lnfmmat1cs
t:. I \ t:./:i 6
+
0 co
0 CD
0 ~
0 N
0
a POD o GP t:. Epilepsy I Oti!ism. , SpecificDD viratChhrn
D.I \ \ D.t:. , I 6
;\/ !:.
.6.
0 5 10
DEPARTMENT OF lli HA~VARD Biomedical lnfmmat1cs
MEDICAL SCHOOL
15
"" ' +
0 a>
0
What about
increasing overlaps
across diagnoses?
- HARVARD DEPARTMENTOF MEDICAL SCHOOL Biomedical Informatics
What genes are shared across co-
Nazeen et al. Genome Biology 17 (1): 228, 2016.
ITOLL-LIKE RECEPTOR SIGNALING PATHWAy I
- ASD [=:J .-U1hm [=:J a lil!!!!!!I IBD [=:J lafctiou i=J URI
Cheniolao:t>: effects (NutropNI. lnunahlre DCLPS(O-) ----- - .\SD,II(NK cell) - ASD, 1ar..doa
~ Alduna,D(Li~=~) Htitl--liiiiilf'.--~~~-r - A>thma,IBDC:::J :Ulbu,Wttdoa - CXD, Dr:::J DC.a
( Flogn.t ass.nilly ) Cosbmulalory !OOletW.. (=:J D,IBD
FlogeDin ----i I CD-II I [=:J D, W..Uoa i=J IBD, ~IDI CDSO 1- [=:J mo, 1.rttdoa.
1 CD86 I (=:J ASD, II, loft
Classification accuracy reveals shared
biology
1 -.------------------0. '9 -+------------------
0.8 -------
~ 0.7 ------- 'a 0.16 u
What is the disease?
Shared Genes?
Shared Symptoms?
v DEPARTMENT orIll HARVARD Biomedical InformaticsMEDICAL SCHOOL
Autism(s)
Implications for study and treatment
Currently not obvious from a geneticsfirst/only approach
Why clinicians might miss it
Unsupervised vs self-referential & circular supervised
Multi-modal-first exploration.
v DEPARTMENT orIll HARVARD Biomedical InformaticsMEDICAL SCHOOL
What about when
there is NO
diagnosis?
- HARVARD DEPARTMENTOF MEDICAL SCHOOL Biomedical Informatics
n UDN D1ata1 P'rocess Overview Undiagnosed Diseases Network
Data Mgt. Step #1_;_ PublicII UDN Patient AppIi catio nr summaries Data& inina l response
Warehouse
M'etncs outcomes
II UDN Record Archive and Electronic Tri al Mgt.
.__rn.__, f .lclh~~ -'~~~~ ~~~WR O.ata IMgt. St!ep #2 :
Data Mgt. Step #4-: pos.-t v isit reporting and reviewref erra I and data tran sfer to a UDN ..
------ ~ --------~ - -clin ical cente r Digita( files, workflowJ admJn reports, phenotype, genotvpe
~T~"'~-~~-~ ~=- I[!][j]~__ . - ~~,;, I~~- D DEPARTMENT OF IData1 Mgt. Step #3: irlllr~RM'AlRD Biomedical Informatics
MEDICAL SCHOOL
-
Seven clinical sites 1 BaylorCollegeofMedicine and
TexasChildren's Hospital
2 Duke Medicinewilh COiumbia univt'l'sity Medical center
3 Harvard Teaching Hospitals (BCH, BWH, MGH)
4 NaUooallnstitutes ofHeallh
5 Stmford Medicine
6 UCLA SChool ofMedicine
7 l/anI CAL SCHOOL.
When there is no diagnosis
If there is a high probability causal variant:
- Diagnostic label links clinical findings in that
patient to that variant
Do all individuals with that variant have disease?
- How does genetic background/environment contribute in the general population
- p(D IV) --t- p(V ID) {cf. HFE & Hemochromatosis}
- HARVARD OEPARTMENTOFv MEDICAL SCHOOL Biomedical Informatics
Why is medicine so
dependent on
categorical diagnoses?
- HARVARD DEPARTMENTOF MEDICAL SCHOOL Biomedical Informatics
RESEARCH LETTER
Medicine's Uncomfortable Relationship With Math: Calculating Positive Predictive Value
Survey Responses n-61
Most common answer - 950/o n=27
. () 0.4 0.6 0.8 1.0
HARVARD DEPARTMENT OFResponse MEDICAL SCHOOL Biomedical Informatics9
Summary
Addiction to categorical diagnoses is cognitively useful
- But our patients have beaten us to the Google-reflex
Categorical diagnosis can result in spurious biological & clinical inference - Often manipulated for$$$, 2ndry agenda - Single measurement modality (incl. genomics) easier to manipulate
Prediction of diagnostic class not necessarily evidence of biological etiology
Diagnoses are more robust and useful when Data-driven Formally model health systems dynamics Statistically-informed Multi-modal Unsupervised or lightly supervised.
v DEPARTMENT orIllHARVARD Biomedical InformaticsMEDICAL SCHOOL
Acknowledgments
Nathan Palmer
Susanne Churchill
Arjun "Raj" Manrai
Andrew Beam
Denis Agniel
MattMight
Griffin Weber
Rachel Ramoni
Sumaiya Nazeen
Bonnie Berger
Alexa McCray
R HARVARD Biomedtc.al lnf