PreDx Diabetes Risk Score Overview CDC Diabetes Translation Conference Michael McKenna, PhD April 24

1
PreDx Diabetes Risk Score OverviewCDC Diabetes
Translation ConferenceMichael McKenna,
PhDApril 24, 2009
2
Agenda

• Background
• PreDx Diabetes Risk Score Research and
  Development
• Clinical Performance

3
Agenda

• Background
• PreDx Diabetes Risk Score Research and
  Development
• Clinical Performance

4
Our Focus
5
Millions at Risk for Preventable Diseases,but
Few Are Treated
Populations at Risk
12 Treated
8 Treated
lt5 Treated
Source IMS Health / PharMetrics, 2005, US Data
6
Why Diabetes?

• Diabetes is a worldwide epidemic that results in
  a chronic disease state and is eminently
  preventable
• The associated yearly cost of diabetes exceeds
  175 billion.
• People with diabetes are vulnerable to multiple
  and complex medical complications including
  cardiovascular disease (heart disease, stroke,
  and peripheral vascular disease) and
  microvascular disease (blindness).

7
Risk of CV Events is Significantly Increased
among Pre-diabetics
8
Pre-diabetics Incur Incremental Costs more than
5 Years Prior to Diagnosis of Diabetes
Incremental Medical Expenses
Nichols G, et al. Type 2 diabetes incremental
medical care costs during the 8 years preceding
diagnosis. Diabetes Care 2000 23 (11)
1654-1659.
9
Prevention Works Decrease in Diabetes Risk with
Medication and Lifestyle Intervention
10
The Unmet Need How to Identify the Patient at
Greatest Risk of Diabetes Conversion?
57 MM Pre-diabetics
How to identify those at greatest risk of
diabetes conversion?
11
Agenda

• Background
• PreDx Diabetes Risk Score Research and
  Development
• Clinical Performance

12
Biomarkers Available for Diagnostic Tests
13
Risk Factor Counting
A Population of People at Risk
Risk Factor 1 Age
Risk Factor 3 BMI
Risk Factor 2 Hypertension
Multiple Risk Factors increase the risk of
diabetes or fracture
14
Biomarker Panels are Equivalent to Risk Factors
Protein 1
Protein 2
Number of Risk Factors
Protein 7
Level 1 Risk (low) Level 2 Risk Level 3
Risk Level 4 Risk Level 5 Risk Level 6
Risk Level 7 Risk (high)
Protein 3
Protein 6
Protein 4
Protein 5
By monitoring multiple pathways, a molecular
physiology fingerprint stratifies patient risk
15
Protein Biomarkers and PreDx

• Powerful disease predictors
• Ubiquitous in commercial diagnostics
• Stable in banked samples
• Major constraint Sample volumes

16
Molecular Counting Technology

• Micro ELISA format
• 1ul sample per well
• Single molecule detection
• Standard ELISA configuration with fluorescence
  reporting
• Very high sensitivity (to sub pg)
• CV average of 0.14
• Automated, high-throughput infrastructure
• gt10,000 384 well plates/year
• Screening capacity gt one million patient samples
  per year
• Fully tracked LIMS system
• Automated data processing

With ½ ml of serum, over 100 protein biomarkers
can be screened in triplicate.
sensitive and uses very small serum volumes
Tethys has exclusive rights to Molecular
Counting Technology from Singulex Inc.
17
Protein Biomarker Identification and Validation
Process

• Biomarker Candidates
• Search literature/patents
• Standardize nomenclature, eliminate redundancy
• Annotate with KEGG, molecular function, cellular
  localization
• Prioritization
• Assign evidence levels based on publications
• Vet list with KOLs, get priority input, preferred
  sources
• Sourcing
• Identify sources, qualify antibodies and protein
  standards
• Consolidate, order, receive
• Assay Development
• Do concentration matrix, identify best S/N
• 8 point standard curve serum dilution factor
• MCT Validation
• Clinical sample processing
• Algorithm development
• Assay components identified for production (5-20
  EIA assays)

Biomarker Candidates 200-1000
Prioritization 100-500
Sourcing 100-300
Assay Development 50-200
MCT Validation 50-200 gt 5-20
18
Summary of study acquisitions
19
Protein Biomarker Identification and Validation

• Literature searches based on multiple pathways
  involved in diabetes were used to develop
  candidate biomarker list of over 260 markers
• 89 markers were tested using a highly sensitive
  immunoassay technology which requires about 1µL
  per biomarker
• 58 assays passed QC, providing quantified protein
  levels
• Lower limit of detection was typically 0.01 ng/mL
• Intra-plate replicate CV was 6-10
• Assay dynamic range was 100-1000 fold (varies by
  assay)
• An algorithm based on 7 biomarkers was validated
  across studies and in a CLIA laboratory

20
Approach to Model Development

• Marker Selection
• Markers are selected using a variety of
  statistical methods
• Predictive models are built from selected markers
  
• Model Evaluation
• Discrimination evaluated with ROC analysis
• Resampling methods (bootstrap, cross-validation,
  permutation tests) ensure robust results
• Validation (Research)
• Models validated by bootstrap sampling and
  independent populations

Clinical Study
Select Markers
Build Model
Bootstrap Model Performance
Fitted Model Performance
20
21
Multiple Biomarkers were Evaluated to Identify
Most Predictive Set
ACE AGER BCL2 C3 CCL2
CD40 CTSB EGF FAS GH1
GPT HGF ICAM1 IGF1 IGF1R
IGFBP1 IGFBP3 IL18 IL6 IL6R IL8
LEP PLAT POMC RETN SELE SELP SERPINE1 SHBG
TGFB1 TIMP2 VCAM1 VEGFA
Muscle Insulin resistance Metabolism
Pancreas Insulin Secretion Inflammation Metabolic
signaling
ACE C3 CCL2
CTSB DPP4 EGF FAS GH1 HGF
IGF1 IGF1R IGFBP3 IL8 INHBA LEP POMC
RETN SHBG TGFB1 VEGFA VWF
ADIPOQ
glucose
CRP
HBA1c
ADIPOQ
CRP
glucose
INS
INS
APOA1 APOB APOE BCL2
CCL2 CD40 DPP4 EGF FAS GH1
GPT HBA1C HGF ICAM1 IGFBP2 IL6 IL8
LEP POMC SELP SGK1 TGFB1
Small Intestine Incretin function Glucose
regulation
Liver Insulin resistance Glucose
regulation Glycogen synthesis
AGER AHSG APOA1 APOB
APOE BAX BCL2 C3 CCL2 CD14 CD40
CTSB DPP4 EGF FAS FGA
GH1 GPT HBA1 HGF HSPA1B ICAM1 IGF1
IGF1R IGFBP1 IGFBP2 IGFBP3 IL18
IL6 IL6ST IL8 INHBA LEP PLAT POMC
RETN SELE SELP SERPINE1 SHBG TGFB1 TNFRSF1B
VCAM1 VEGFA
ADIPOQ
glucose
ADIPOQ
INS
glucose
CRP
FTH1
ACE APOE C3 CCL2
EGF FAS GH1 GPT HP ICAM1
IGF1 IL6 IL6ST IL8 LEP PLAT POMC
RETN SERPINE1 TGFB1 TIMP2 TNFRSF1B VCAM1
VEGFA
Adipose Insulin resistance FFA metabolism Inflamma
tion
IL2Ra
INS
glucose
ADIPOQ
CRP
INS
22
Algorithm Incorporates Multiple Biomarkers to
Provide a Single Risk Score
Multiple Biological Processes Detected
ALGORITHM
Single Risk Score Produced
Weighted contribution of individual biological
pathways
23
Agenda

• Background
• PreDx Diabetes Risk Score Research and
  Development
• Clinical Performance

24
Developing a Predictive Diagnostic Test

• The Inter99 Study
• A large Danish natural history cohort
• Focused on cardiovascular disease and diabetes
  endpoints
• Collaboration with the Steno Diabetes Centre

25
Performance of PreDx DRS validated in CLIA
laboratory
Training Performance
Test Performance
ROC analysis based on Inter99 case-cohort results
26
PreDx performance comparison
27
DRS Stratifies Better than FPG
IFG (55.6)
NFG (44.4)
Low (53.6)
High (10.3)
Medium (36.1)
Fasting Glucose Status ( Population)
DRS Risk Stratum ( Population)
28
Risk of T2DM Based on DRS in Subjects with IFG
IFG (55.6)
NFG (44.4)
Low (28)
Medium (53)
High (19)
Fasting Glucose Status ( Population)
DRS Risk Stratum ( Population)
29
Test Validation in the Botnia Cohort

• The Botnia Study
• A large Finnish natural history cohort
• Focused on diabetes prediction
• Collaboration with Leif Groop, Lund University

30
PreDx DRS Tracks With Disease Progression
31
Performance of DRS Model on Inter99 and Botnia
Inter99
Botnia
DRS AUC 0.812 FPG AUC 0.738
DRS AUC0.850 FPG AUC 0.674 (Converters lt5 yrs)
32
DRS Discriminates Early and Late Converters
There is a significant difference in the DRS
scores of early and late converters in the
Botnia cohort
33
Kaplan-Meier shows risk stratification
5-year Conversion to Diabetes by Diabetes Risk
Score
34
Reclassification better than glucose tests
35
Conclusions

• Proven platform for discovering biomarkers and
  developing innovative diagnostics
• The PreDx DRS provides a personal assessment of
  conversion to diabetes within five years
• The model is validated across studies and in a
  CLIA laboratory
• This model was based on the quantitative analysis
  of multiple biomarkers covering multiple
  metabolic pathways.
• This Model
• Has better performance than any single baseline
  risk factor (e.g. fasting glucose or BMI)
• Has better performance than any non-invasive
  assessment of diabetic risk
• Provides additional stratification even in
  subjects with IFG