Simulation of Clinical Trials with Mathematical Models ... - CBI

Simulation of Clinical Trials with
Mathematical Models:
Applications to
Comparative Effectiveness Research
November 18, 2010
Badri Rengarajan, M.D.
www.archimedesmodel.com
... a KAISER PERMANENTE Innovation

Goals
• Describe role of simulating clinical trials in
comparative effectiveness research (CER)
– Use Archimedes Model as example
• Show examples
– Case studies of diabetes treatments
– Describe a tool for comparing >20 interventions for cardiometabolic
risk
2

Topics
• Rationale, mechanisms, and challenges of simulating
clinical trials
• Archimedes Model
– Overview
– Validation
– Capabilities and applications
• Case Studies
– DPP expansion and extension
– A-L-L
– Results of 20 simulated trials of interventions for
cardiometabolic risk
3

Ideally use clinical trials to compare
different interventions
• Not feasible
– Large populations
– Long follow-up times
– High cost, budget constraints
– Ethical issues
– Provider and patient participation
– Large number of interventions and options to be studied
• Best available alternative is to simulate the desired
trials
– Design trials
– Replace trials
4

There are several ways to try to predict results
of trials, each has limitations
5
Mechanism
Comparison of trials for different
treatments
(also tests, devices, procedures)
Registry data
Questionnaires for physicians,
patients
Potential Limitations
• “Apples vs. Oranges” comparisons
̶ Different patient populations
̶ Different interventions
̶ Different endpoints/outcomes
• Confounding factors
• Missing/incomplete data
• Calibration of responses
• Inability to verify
PK/PD modeling
Pilot trial
• May not be predictive of clinical
outcomes – too far upstream
• Insufficient powering
• Requires time for protocol development,
enrollment, and analysis

Simulation methods have their own challenges,
and differ in their capabilities
Potential Challenges
• Limited to current knowledge of drug action, efficacy/effectiveness,
safety, dosing
• May not capture off-target toxicities
• Not a substitute for a real trial
Differences Among Methods
• Level of accuracy in capturing physiological and pathophysiological
relationships
• Level of integration across different organs and disease processes
• Ability to capture clinical care processes
• Ability to capture healthcare system costs
• Ability to capture downstream/ripple effects on outcomes and costs
6

Topics
• Rationale, mechanisms, and challenges of simulating
clinical trials
• Archimedes Model
– Overview
– Validation
– Capabilities and applications
• Case Studies
– DPP expansion and extension
– A-L-L
– Results of 20 simulated trials of interventions for
cardiometabolic risk
7

Archimedes uses mathematics to
enhance decision making
• The Archimedes Model is a mathematical
model of human physiology, diseases,
interventions, and healthcare systems
– Physiology-based
– Realistic
– Comprehensive
– Clinically and administratively detailed
– Rigorously validated
• Enables decision makers to understand likely
outcomes of various interventions
8

9
Physiological Pathways are Represented as
Equations in Virtual People
Family
history
Sex
Race/
ethnicity
Age
BMI
Height
Type 1
Diabetes
feature
Type 2
Diabetes
feature
Age, sex,
race/
ethnicity
Weight
Diabetes
blood
pressure
factor
Peripheral
resistance
Cardiac
output
Gliburide
Arterial
compliance
Insulin Insulin
treatment
Systolic
Cardiac
output
blood
pressure
Pulse \
pressure
Insulin
Insulin
Insulin
level level level level level
Unexplained
variance in
OGT
FPG
OGT
Ketoacidosis
Random
plasma
glucose
Insulin
Amount
Insulin
Joint Insulin
Glucose
Random
Amount
Diabetes Amount Insulin
uptake by
error and
feature Amount
muscle uptake by
variation
Amount uptake by
muscle uptake by
Efficiency of
muscle FPG
insulin use
Glucose muscle uptake by FPG
HbA1c
production muscle
FPG
by by liver Glucose
FPG
production Glucose
Untreated
by production liver
FPG
Glucose insulin level
Care
liver
insulin level
processes
Metformin
Normal liver by production
liver Unexplained
Family
Insulin
OGT
OGTT test
Gliburide
Insulin level
variance
GlucoseUntreated
in
history
treatment
glucose
OGT
production productionby production
liver insulin Untreated level
Random
Random
Diabetes
plasma
plasma
Sex
diagnosis
Normal liver
Fractional
Type 1
glucose
glucose test
Insulin
Diabetes
production
insulin level
feature Joint (Pancreas) Glucose
glucose Normal liver by liver Untreated change in
Race/
Random
FPG test
Diabetes
uptake by
ethnicity
error and
Type 2 feature
muscle
Insulin
variation
Insulin
Diabetes
efficiency production FPG
glucose
insulin level
feature
(Muscle)
Glucose
HbA1c test
Age
Normal production liver
Untreated
Insulin
by liver
efficiency
Untreated
Care
Urine
Diet and
Age, sex,
(Liver) glucose insulin level
processes insulin ketone test level
BMI
race/
Normal liver
To
exercise
FPG
ethnicity
glucose
treatment
production
Normal liver Fractional
Ketoacidosis
models
Metformin
change in
UKPDS
Insulin
data
glucose
Height Weight
Diet and
Hypoglycemia
FPG
exercise
LDL
production Diabetes Propensity
HDL
Triglyceride
Patient
Diabetes
takes action
LDL
cholesterol cholesterol s
Smoking Diabetes Propensity
blood
HDL
Triglyceride
cardiac risk
Blurred
to blurred
Smoking cardiac risk to blurred
pressure cholesterol cholesterol s
vision
factor
vision
factor
factor vision
Mean
arterial
pressure
Peripheral
resistance
Mean
arterial
pressure
Arterial
compliance
Systolic
blood
pressure
Pulse \
pressure
To the
Retinopathy
model
Coronary
artery
stenosis
Coronary
To the To the
Nephropath artery Neuropathy
y model model
stenosis
FPG
To the
Coronary
artery
disease
model
Propensity
to polyuria
Propensity
to fatigue
Propensity
to thirst
Polyuria
Fatigue
Thirst
Memory
Perception
Propensity
to polyuria
Propensity
to fatigue
OGTT test
Random
plasma
glucose test
FPG test
HbA1c test
Urine
ketone test
Hypoglycemia
Blurred
vision
Polyuria
Fatigue
Diabetes
diagnosis
To
treatment
models
Patient
takes action
Memory
Perception
To the
Retinopathy
model
To the
Nephropath
y model
To the
Neuropathy
model
To the CAD
model
Propensity
to thirst
Thirst

Type 1
10
The Model is comprehensive
Family
history
Sex
Race/
ethnicity
Age
BMI
Height
Diabetes
feature
Type 2
Diabetes
feature
Age, sex,
race/
ethnicity
Weight
Diabetes
blood
pressure
factor
Insulin
Gliburide
Insulin level
treatment
Insulin
production
Joint (Pancreas) Glucose
Diabetes
uptake by
feature
muscle
Insulin
efficiency
(Muscle)
Glucose
production
Insulin
by liver
efficiency
(Liver)
Normal liver
glucose
production
Metformin
UKPDS
data
Diet and
exercise
HDL
LDL Triglyceride
cholesterol cholesterol s
Unexplained
variance in
OGT
FPG
Untreated
insulin level
Diabetes
Smoking cardiac risk
factor
Mean Systolic Coronary
Peripheral
arterial blood artery
resistance
pressure pressure stenosis
FPG
OGT
Random
plasma
glucose
Random
error and
variation
Care
processes
Fractional
change in
Insulin
FPG
Propensity
to blurred
vision
Propensity
to polyuria
OGTT test
Random
Diabetes
plasma
diagnosis
glucose test
FPG test
HbA1c test
Urine
ketone test
To
treatment
Ketoacidosis
models
Hypoglycemia
Patient
takes action
Blurred
vision
Memory
Polyuria
Perception
Behaviors
Symptoms
Signs
Diseases
Outcomes
Cardiac
output
Arterial
compliance
Pulse \
pressure
Propensity
to fatigue
Fatigue
To the
To the To the To the Coronary
Retinopathy Nephropath Neuropathy artery
model y model model disease
model
Propensity
to thirst
Thirst

11
The Model is clinically and administratively detailed,
enabling detailed costing and economic analysis
Patient:
• Has chest pain and presents to the ER
• Receives EKG, chest x-ray, and blood tests
• Is diagnosed with MI, admitted to the hospital, and given an angioplasty with drug-eluting stent
• Remains in the hospital for 2.1 days and is discharged
Who is reimbursed Service performed Codes
Cardiologist
Reimbursement
MS-DRG 247: Percutaneous
Cardiovascular Procedure with
Consultation
Read EKG
Perform cardiac catheterization
Drug-Eluting Stent
Perform angioplasty with drug-eluting stent
•2008 Medicare base payment:
$10,365
CPT
CPT (professional component only)
CPT
CPT
Radiologist Read x-ray CPT (professional component only)
Anesthesiologist Provide anesthesia during angioplasty RVG code and time patient is anesthetized
(can be converted to a CPT)
•Medicare average length of stay:
2.1 days
Other physicians Inpatient consultations CPT
Hospital All services performed during stay DRG

The Model spans several diseases
• Coronary artery disease
• Stroke
• Diabetes and complications
• Obesity
• Metabolic syndrome
• Hypertension
• Dyslipidemia
• Congestive heart failure
• Atrial fibrillation
• Asthma
• Lung cancer
• Breast cancer
• Colon cancer
Includes a model of family pedigrees,
enabling studies of risk-prediction,
genetic and genomic testing,
and cost-effectiveness of screening
strategies.
12

Over 50 trials have been used to validate
the Model
4S DCCT Lifetime Risk of Diabetes SEATTLE
ADVANCE DPP Look-AHEAD SHEP
AGE DREAM MERIT SOLVD
ALLHAT
ANBP2
European Orlistat
Obesity Study
FHS
MicroHOPE
Minnesota FOBT
Screening Trial
ARIC Flechtner-Mors National Polyp Study TNT
STENO-2
SYSTEUR
ASCOTT LLA GLAI PPP TRACE
ATBC HOPE PROactive UKPDS 33
CAPRICORN HPS PROSPER UKPDS 34
CARDS IDEAL RENAAL UKPDS 45
CARE IDNT RIO-Europe UKPDS 74
CPS-II Breast
Lieberman Colonoscopy
Screening
RIO-Lipids UKPDS 80
CPS-II Colon LIFE all RIO-North America VALUE
CPS-II Lung LIFE diabetes SAVE WESDR
13

The Model can reveal health and economic effects of
multiple interventions under different scenarios
• Can include active comparators of interest, e.g.
– Multiple drugs (diagnostics, devices)
– Different regimens, schedules, frequencies
– Different care protocols and guidelines
• Can create and tailor specific relevant populations,
settings and costs
– Populations
• Demographics, behaviors, test results, symptoms, past medical histories,
current conditions, current treatments
• All at individual level
– Protocols and practice patterns
– Costs
• Tests, treatments, visits, hospitalizations, procedures
14

The Model can enhance trial design,
CER, and product strategy
• Clinical Trial Design
– Appropriate powering (with knowledge of baseline/control arm event
rates)
– Observations on primary and secondary endpoint responses
– Patient population segmenting
– Examination of different permutations of I/E criteria, trial size, timelines,
treatment effects
• Comparative Effectiveness
– Comparison on health outcomes
– Comparison on costs
– Examination of different permutations of patient subsets, practice
guidelines, treatment compliance, clinical guideline adherence, costs
• Product Strategy (Indication statement, Positioning, Pricing, etc.)
15

Topics
• Rationale, mechanisms, and challenges of simulating
clinical trials
• Archimedes Model
– Overview
– Validation
– Capabilities and applications
• Case Studies
– DPP expansion and extension
– A-L-L
– Results of 20 simulated trials of interventions for
cardiometabolic risk
16

Only simulation modeling could have enabled
expansion and extension of the DPP trial
Situation
• American Diabetes Association (ADA) sought to
understand cost-effectiveness of screening and
management guidelines to prevent/delay
development of T2DM in high-risk individuals
• Three-year DPP (Diabetes Prevention Program) trial
comparing current care, metformin, and lifestyle
modification was nearly complete
• However, ADA and Department of Health and Human
Services (HHS) were interested in long-term health
and economic outcomes of different strategies, as
well as several questions outside scope of DPP trial
• Existing trial had already cost over $300M
17

The approach involved matching populations &
protocols, adding an arm, and extending duration
Approach
• Created a simulated population matching DPP
inclusion/exclusion criteria and patient baseline
characteristics
• Conducted prospective simulation of DPP trial (same
duration, interventions) to validate Model’s ability to
reproduce population, interventions, and outcomes
• Added intervention arm (lifestyle intervention
initiated after diagnosis – FPG >125)
• Extended duration of simulated trial to 30 years
18

19
This enabled generation of longitudinal health
data on a larger population
Results
Actual Trial
Simulated Trial
Size N=3,234 N=10,000
Mean trial duration 3.2 years 30 years
Rates of diabetes • Current care: 28.9%
• Metformin: 21.7%
• Lifestyle: 14.4%
30-year data
27.4%
21.9%
13.2%
Risk of DM 72%
Benefits of Lifestyle Program (over entire period)
Risk reduction
Risk of serious complication
(high-risk group)
Risk of death from DM
complication
Cost per QALY (vs. no
intervention)
Down 15% (relative)
38% to 30%
13% to 11%
$143K

Cumulative Incidence of Diabetes
20
The simulation was prospectively validated
against the original trial
0.5
DPP: Diabetes Progression
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
Current
28.9% 27.4%
0 0.5 1 1.5 2 2.5 3 care 3.5 4
Time (years)
3-Yr
Timepoint
Actual
control
metformin
lifestyle
Simulated
Metformin 21.7% 21.9%
Lifestyle 14.4% 13.2%

In simulation, the DPP trial was simultaneously
expanded (fourth arm) and extended (30 yrs)
Fraction developing diabetes
0.8
0.7
Effect of Four Programs on Progression to Diabetes
Postpone one decade
Prevent 11%
0.6
0.5
0.4
21
0.3
0.2
0.1
0
Baseline
LifeStyle
Metformin
Lifestyle when FPG>125
0 5 10 15 20 25 30
Time (years)
21

Estimating longitudinal health outcomes
generated CE insight (Baseline vs. Lifestyle)
Without Lifestyle (baseline)
Difference with Lifestyle
Years of follow-up 10 20 30 10 20 30
Diabetes 56.9% 68.6% 72.2% -14.3% -11.6% -10.8%
CAD//CHF
Have an MI 4.0% 8.5% 12.0% -0.4% -1.1% -1.7%
Develop CHF (systolic or diastolic) 0.2% 0.7% 1.2% -0.1% -0.1% -0.1%
Stroke (ischemic or hemorrhagic) 2.9% 7.0% 11.6% -0.5% -1.0% -1.4%
Some serious complication 11.2% 26.1% 38.2% -3% -6.6% -8.4%
Deaths
22% 17% decrease
CHD 2.2% 6.6% 11.9% -0.6% -1.1% -2.0%
Stroke 0.4% 0.9% 1.5% -0.1% -0.3% -0.3%
Renal disease 0.00% 0.02% 0.1% 0.00% -0.01% -0.04%
Death from any complication 2.6% 7.6% 13.5% -0.7% -1.3% -2.3%
Life Years 24.032 0.288
22
22

Estimating longitudinal cost outcomes also
generated CE insight (Baseline vs. Lifestyle)
Expected Costs in a Health Plan with 100,000 Members
$Millions
23
Without Lifestyle
(Baseline)
Difference with Lifestyle
Years of
follow-up 5 10 20 30 5 10 20 30
Admissions $10 $23 $58 $96 $0.8 $0.7 $1.5 $2.2
Visits $8.3 $16 $33 $48 -$0.4 -$11 -$1.7 -$2
Procedures $7.4 $16 $38 $57 -$0.8 -$2 -$4 -$5.5
Interventions $3.4 $6.6 $16 $26 $14 $26 $48 $64
Total $29 $62 $144 $227 $14 $24 $44 $59
PMPM for
high-risk $57 $50 $46 $41
PMPM for all
members $2.29 $2.00 $1.83 $1.64
23

30-year QALY/person
The cost effectiveness of each arm
over 30 years was revealed
11.50
QALY vs Cost for Four Programs
11.48
11.46
$201,800
11.44
11.42
11.40
11.38
11.36
11.34
11.32
$24,523 $35,523
$62,600
No Program
Lifestyle when FPG>125
DPP Lifestyle
Metformin
24
11.30
$37,000 $39,000 $41,000 $43,000 $45,000 $47,000 $49,000
30-year Cost/person
24

Simulation enabled trial expansion and
extension at a fraction of the cost of a real trial
25
Results
Actual Trial
Simulated Trial
Size N=3,234 N=10,000
Trial arms 3 3 + 1
Mean trial duration 3.2 years 30 years
Rates of diabetes • Current care: 28.9%
• Metformin: 21.7%
• Lifestyle: 14.4%
27.4%
21.9%
13.2%
Trial cost >$300M

Topics
• Rationale, mechanisms, and challenges of simulating
clinical trials
• Archimedes Model
– Overview
– Validation
– Capabilities and applications
• Case Studies
– DPP expansion and extension
– A-L-L
– Results of 20 simulated trials of interventions for
cardiometabolic risk
26

A-L-L: Clinical trial simulation enabled testing
of a novel hypothesis
Situation
• A group of physicians at Kaiser Permanente
suggested that a combination of aspirin, lovastatin,
and lisinopril would reduce complications of
diabetes
• However, there was no direct evidence and no way
to determine effects on health outcomes and costs
• Yet, data would be needed to support adoption
27

The approach involved matching populations, care
protocols, and costs
Approach
• Created a simulated population representing all
diabetics over 55 years in Kaiser Permanente (CA)
• Constructed trial with three groups:
– Current care (based on national guidelines and realistic
performance and compliance)
– A-L-L regimen on top of Current Care
– Reference group: 1% HbA1c reduction
• Used KP’s costs for drugs, lab tests, office visits, and
admissions for MIs and strokes
• Set duration of simulated trial at 30 years
********
• KP’s Care Management Institute conducted an
independent real trial (2 years) to evaluate A-L-L
28

29
A-L-L has a bigger effect than 1%
HbA1c reduction
0.045
0.04
0.035
0.03
Average annual risk of various events
Nothing
HbA1c control
ALL
0.025
0.02
0.015
0.01
0.005
0
MI Stroke ESRD Blind Dying

30
The effect begins immediately
0.12
Annual risk of four complications or death
0.1
0.08
Nothing
ALL
HbA1c
0.06
0.04
0.02
0
0 5 10 15 20 25
Time since start of program

31
CMI Independent Evaluation (2007)
• In 2004-2005 28% of KP’s eligible study population in northern +
southern CA (n=170,024) had received A-L-L at low exposure, 13%
at high exposure (59% no exposure)
• Modeled results consistent with actual findings by CMI:
– By 2006, heart attacks and strokes decreased by 15 per 1000
members for the low exposure group (p

Topics
• Rationale, mechanisms, and challenges of simulating
clinical trials
• Archimedes Model
– Overview
– Validation
– Capabilities and applications
• Case Studies
– DPP expansion and extension
– A-L-L
– Results of 20 simulated trials of interventions for
cardiometabolic risk
32

A population dataset with pre-run interventions serves as a reference tool
for development and implementation of novel approaches/medications
Alternate Simulation Tool for CE: Cardio-Metabolic Risk Dataset
Population
Cardiometabolic
interventions
Treatment
scenarios
Settings
• 100,000 individuals
• Drawn from NHANES data 1999-2006
• Subpopulations by age, gender, risk factors (e.g., CAD, DM)
• Weight management
• BP management
• Cholesterol management
• Medications (e.g., aspirin)
• Glycemia management
• Smoking cessation
• Guidelines followed perfectly
• Medications initiated at study start with no future adjustments
• “Magic pills” - Medications that maintain outcome goal (e.g., LDL

The results of a variety of interventions can be
captured
Interventions
Target Populations
Weight Management
Lower BMI to 30 BMI > 30
Lower BMI to 25 BMI > 25
Lower BMI by 5% not to go below 25 BMI > 25
Take weight loss intervention
As per guidelines
Blood pressure management
Lower BP to 130/80 SBP > 130 or DBP > 80
Take BPC meds (ACE-I/ARB, diuretic, beta blocker, CCB) As per guidelines
Cholesterol management
Lower LDL to 100 LDL > 100
Lower LDL to 130 LDL > 130
Lower LDL to 160 LDL > 160
Take statin
As per guidelines
Take medication
Take polypill
Is diagnosed with diabetes or CAD
Take ASA
Has had MI or ischemic stroke
Glycemia management
Lower FPG to 125 mg/dL
FPG > 125 mg/dL
Lower HbA1c to 9% Is diagnosed with diabetes and HbA1c > 9%
Lower HbA1c to 8% Is diagnosed with diabetes and HbA1c > 8%
Lower HbA1c to 7% Is diagnosed with diabetes and HbA1c > 7%
Take diabetes medication (metformin, glitazone,
As per guidelines
sulfonylurea, insulin)
Lower FPG by 10% not to go below 100%
FPG between 100 and 125 mg/dL
Lower FPG with Metformin
FPG between 100 and 125 mg/dL
Smoking Cessation
Stop smoking
Smokers
34

Health and cost outcomes can be captured
Outcomes in Dataset
Biomarkers
Health
CE
• Proportion smokers, BMI, weight, SBP, DBP, LDL, HDL, total cholesterol,
triglycerides, FPG, HbA1c, serum creatinine
• MI, stroke, MACE, revascularization, diabetes, diabetic proliferative
retinopathy, bilateral blindness, ESRD, diabetic foot ulcer, diabetic foot
amputation.
• Total deaths, CHD deaths, stroke deaths
• Medication usage: aspirin, antihypertensives, ACE-inhibitor, beta blocker,
CCB, diuretics, dyslipidemia medications, oral DM agents, metformin,
sulfonylurea, TZD, insulin, weight loss interventions, polypills
• Life years
• Medical costs
• QALYs, Discounted QALYs
• Cost/QALY
• Cost per averted event
35

Clinical Trial Simulation modeling is an essential
component of comparative effectiveness analysis
• Overcomes cost, time, and
recruiting/operational challenges of real
trials
• Improves trial design
• Generates data quickly
• Provides preview of real trial outcomes
• Facilitates testing of unlimited scenarios
• With integration of physiology/disease
with healthcare processes and costs, can
provide CE insights and evidence
Meaningful
component of
CE Toolkit,
but not a
substitute for
real trials
36

Appendix
www.archimedesmodel.com
... a KAISER PERMANENTE Innovation
37

Archimedes is a health care modeling
company
• Healthcare Modeling Company
• HQ in San Francisco
• Core technology - Archimedes Model
– Mathematical model of human physiology,
diseases, interventions and healthcare
systems
• Highly detailed
• Carefully validated
– In development since 1993
• David Eddy MD, PhD
• Len Schlessinger PhD
– Owned by Kaiser Permanente
• Spun out as independent organization 2006
38

Publications
• Age at initiation and frequency of screening to detect type 2
diabetes: a cost-effectiveness analysis
[»The Lancet, 3/30/2010 ]
• Preventing Myocardial Infarction and Stroke with a Simplified
Bundle of Cardioprotective Medications
[ »Am. Journal of Managed Care, 10/1/2009]
• Effect of Smoking Cessation Advice on Cardiovascular Disease
[ »Am. Journal of Medical Quality, Vol. 24, No. 3, (2009)]
• The Relationship between Insulin Resistance and Related
Metabolic Variables to Coronary Artery Disease: A
Mathematical Analysis
[ »Diabetes Care Publish Ahead of Print, 11/18/2008 ]
• The potential effects of HEDIS performance measures on the
quality of care
[ »Health Affairs, 9/15/2008 ]
• The Impact of Prevention on Reducing the Burden of
Cardiovascular Disease
[ »Circulation, 7/29/2008 ]
• Validation of Prediction of Diabetes by Archimedes and
Comparison with Other Predicting Models.
[ »Diabetes Care, 5/28/2008 ]
• The Metabolic Syndrome and Cardiovascular Risk: Implications
for Clinical Practice.
[ »International Journal of Obesity, 5/1/2008 ]
• Diabetes Risk Calculator: A Simple Tool for Detecting
Undiagnosed Diabetes and Prediabetes.
[ »Diabetes Care, 5/1/2008 ]
39
• Cure, Care, and Commitment: What Can We Look Forward To?
[ »Diabetes Care, 4/15/2008 ]
• Reflections on science, judgment, and value in evidence-based
decision making: a conversation with David Eddy
[ »Health Affairs, 6/19/2007 ]
• Medical Decision-making: Why it must, and how it can, be
improved
[ »Expert Voices, 5/15/2007 ]
• Linking Electronic Medical Records To Large-Scale Simulation
Models: Can We Put Rapid Learning On Turbo?
[ »Health Affairs, 1/26/2007 ]
• Accuracy versus transparency in Pharmacoeconomic
modeling: finding the right balance.
[ »Pharmacoeconomics, 6/6/2006 ]
• Bringing health economic modeling to the 21st century.
[ »Value in Health, 5/30/2006 ]
• Clinical outcomes and cost-effectiveness of strategies for
managing people at high risk for diabetes.
[ »Annals of Internal Medicine, 8/16/2005 ]
• Earlier intervention in type 2 diabetes: The case for achieving
early and sustained glycaemic control.
[ »International Journal of Clinical Practice, 11/28/2005 ]
• Evidence-based medicine: a unified approach.
[ »Health Affairs, 02/15/2005 ]
• Validation of the Archimedes diabetes model.
[ »Diabetes Care, 11/15/2003 ]
• Archimedes: a trial-validated model of diabetes.
[ »Diabetes Care, 11/15/2003 ]

The Model is clinically and administratively
detailed in following care processes
40
Modeling health care processes and clinical practice guidelines (e.g. ATP III)

41
The validation approach is rigorous
People
Treatments
Randomized
Controlled
Trials
Real
Outcomes
Same
(Virtual)
People
Same
(Virtual)
Treatments
Same?
Virtual
Outcomes

Fraction of patients
42
The Model has undergone external,
independent validation
0.14
0.12
Major Coronary Events in Heart Protection Study
Solid lines are the trial results
0.1
0.08
Placebo
0.06
0.04
0.02
Treated with Simvastatin
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Years

Fraction of patients
43
The Model yielded a reasonable
approximation of the HPS Study
0.14
0.12
0.1
0.08
Major Coronary Events in Heart Protection Study
Solid lines are the trial results
Dotted lines are the Model’s results
Placebo
0.06
0.04
0.02
Treated with Simvastatin
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Years

Fraction
44
It has also prospectively predicted CARDS
Trial results
CARDS Trial: Major Coronary Events
0.14
0.12
0.1
CARDS Integrated Hazard: control
Archimedes Integrated Hazard: control
CARDS Integrated Hazard: treated
Archimedes Integrated Hazard: treated
0.08
0.06
0.04
0.02
0
0 1 2 3 4
Years

In the DPP simulation, populations and interventions
were matched, and identical outcomes were set up
45
• The people: “high risk of diabetes”
– BMI >24
– FPG 5.3 – 6.9 mmol/L = 95 – 125 mg/dL
– OGTT 7.8 – 11 mmol/L = 140 – 199 mg/dL
• The interventions
– Metformin (originally $742/year, now $260/year)
• Reduces FPG
• Reduces LDL and TG
• Retards weight gain
– DPP Lifestyle ($1356 in 1 st year, then $672/year thereafter)
• Reduces weight
• Reduces FPG, BP, LDL, Total Cholesterol
• Raises HDL
– Add Lifestyle after diagnosis of diabetes
• Outcomes
– Progression to diabetes
• FPG > 7 mmol/L (125 mg/dl) or OGTT > 11 mmol/L (199 mg/dl)
– Myocardial infarction and stroke
– End stage renal disease, retinopathy
– Quality of life
– Costs
45

Net 30-year Cost
(Discounted) to Health
Plan
DPP simulation revealed net
cost of intervention
Net Cost to Health Plan (Discounted) of two
Programs, as Function of Annual Cost of Lifestyle
Program
5000
4000
3000
2000
DPP Lifestyle vs No
Program
DPP Lifestyle
is cost neutral
at $100/year
Lifestyle after
diabetes diagnosis
is cost neutral at
Lifestyle after FPG+125
vs. No Program $240/year
1000
0
-1000
0 200 400 600
46
-2000
Annual Cost of Lifestyle Program
46

Cost/ QALY
DPP simulation also enables Cost/QALY
calculation over 30 years
Cost/QALY of Three Comparisons as Function of
Cost of Lifestyle Program
$250,000
DPP Lifestyle vs. No Program
$200,000
$150,000
Lifestyle after FPG=125 vs. No
Program
DPP Lifestyle vs. Lifestyle after
FPG+125
$100,000
$50,000
47
$0
-$50,000
0 200 400 600
Annual Cost of Lifestyle Program
47

48
A-L-L regimen saves more money
$5,000
Average annual cost per person
$4,500
$4,000
$3,500
Nothing
HbA1c control
ALL
$3,000
$2,500
$2,000
$1,500
$1,000
$500
$0
Cost

A-L-L: The savings begin immediately
$6,000
Annual cost per person
$5,000
$4,000
$3,000
$2,000
$1,000
$0
Nothing
ALL
HbA1c
0 5 10 15 20 25
Years after start of program
49

Summary of Dataset with Pre-Run
Interventions
• Community of 100,000k
• 8 subpopulations
• 20 intervention arms
• 15 control arms
• 20 time points for following outcomes
– 12 biomarker outcomes
– 51 health outcomes
– 1 lifeyears outcome
– 19 cost effectiveness outcomes
– 83 differences, 83 ratios
50