Prediction Model Template from OHTSEGPS Pooled Analyses

1

• Prediction Model Template from OHTS-EGPS Pooled
  Analyses
• Todays version is November 14

2
A Prediction Model for Managing Ocular
Hypertensive Patients

• Presenter Name
• The Ocular Hypertension Treatment Study Group
  (OHTS)National Eye Institute, National Center
  for Minority Healtlh and Health Disparities, NIH
  grants EY 09307, EY09341, EY015498, Unrestricted
  Grant from Research to Prevent Blindness, Merck
  Research Laboratories and Pfizer, Inc.
• The European Glaucoma Prevention Study (EGPS)
• European Commission BMH4-CT-96-1598 and Merck
  Research Laboratories

3
Ocular hypertension

• Ocular hypertension occurs in 4-8 of people in
  the United States over age 40 (3-6 million
  people)
• The number of affected people will increase with
  the aging of the population
• Associated with large costs for patient
  examinations, tests and treatment

4
Ocular hypertension

• Elevated IOP is a leading risk factor for
  development of POAG
• Only modifiable risk factor for POAG
• Patients can lose a substantial proportion of
  their nerve fiber layer before POAG is detected
  by standard clinical tests
• Quigley HA, et al. Arch Ophthal 198199635

5
Why do we need a prediction model?

• 2002 OHTS publication showed that early treatment
  reduces the incidence of POAG by more than 50
• However, only 1 of ocular hypertensive
  individuals develop POAG per year
• Clear that treating all ocular hypertensive
  patients is neither medically nor economically
  justified

6
Why do we need a prediction model?

• Common in the past to base management decisions
  on a single predictive factor usually IOP
• What level of IOP do you treat?
• IOP 24 mmHg?
• IOP 26 mmHg?
• IOP 28 mmHg?
• IOP 30 mmHg?
• This approach ingores other important predictive
  factors

7
Why do we need a prediction model?

• A prediction model stratifies ocular hypertensive
  individuals by level of risk
• To guide the frequency of visits and tests
• To ascertain the benefit of early treatment

8

• In 2002, the Ocular Hypertension Treatment Study
  (OHTS) published a prediction model for POAG
  based on...
• Data from 1,636 ocular hypertensive participants
  randomized to either observation or topical
  hypotensive medication
• Median follow-up 6.6 years
• Gordon et al, Arch Ophthalmol. 2002 120
  714-720.

9
Factors predictive for the development of POAG
in 2002 OHTS model

• 5 baseline factors increased the risk of
  developing POAG
• Older age
• Higher Intraocular pressure
• Thinner central cornea
• Larger vertical cup/disc ratio by contour
• Higher pattern standard deviation
• Diabetes decreased the risk of POAG
• .

10
2002 OHTS model needed to be confirmed in a
large, independent sample

• 2002 prediction model based on data from treated
  and untreated ocular hypertensive individuals
• A prediction model should be based solely on
  untreated individuals
• OHTS sample included 25 African American
  participants
• Is the prediction model valid in other groups?
• OHTS was 1st study to report central cornea
  thickness as a powerful predictor of POAG
• Can this finding be confirmed?

11

• A large indepent sample available through the
  European Glaucoma Prevention Study (EGPS)
• EGPS is a randomized clinical trial of 1,077
  ocular hypertensive individuals randomized to
  either placebo or dorzolamide
• Median follow-up 4.8 years

12
Purpose of collaboration with EGPS

• To test the 2002 OHTS prediction model for the
  development of glaucoma in a large, independent
  sample
• Before undertaking a collaboration with EGPS, the
  two study protocols were compared

13
Comparison of OHTS and EGPS Study design
Similarities between OHTS and EGPS
14

• Collaborative analysis uses data only from
  participants not receiving medication
• OHTS Observation Group n819
• EGPS Placebo Group n500

15
OHTS vs EGPS Eligibility criteria
Similarities between OHTS and EGPS
16
OHTS vs EGPS Eligibility criteria
Similarities between OHTS and EGPS
17
OHTS vs EGPS Exclusion criteria Similarities
between OHTS and EGPS
18
OHTS vs EGPS Corneal thickness measurement
Similarities between OHTS and EGPS

19
OHTS vs EGPS POAG endpoint criteria
Similarities between OHTS and EGPS

20
Collaborative analysis is feasible

• OHTS and EGPS protocols are similar enough to
  test the validity of the prediction model after
  resolution of study differences
• Different enough in measures, geographic
  distribution and patient characteristics to test
  the generalizability of the OHTS prediction model

21
ResultsOHTS vs EGPS control groups Baseline
characteristics(Univariate analyses)
22
ResultsOHTS vs EGPS control groups Definition
of baseline IOP (mmHg)
23
OHTS vs EGPS control groups Baseline
characteristics
24
OHTS vs EGPS control groups 1st eye to develop
POAG endpoint
25
Why was the incidence of POAG higher in EGPS than
in OHTS?

• Differences in entry criteria
• Differences in POAG endpoint criteria
• Differences in risk characteristics of
  participants

26
Steps in testing the validity of the OHTS
prediction model

• Perform separate analyses of OHTS Observation
  Group and EGPS Placebo Group
• (Multivariate Cox proportional hazards models)
• Compare results of the two analyses

27
Results of independent multivariate analyses
OHTS vs EGPS

• Separate predictive models in OHTS and in EGPS
  identified the same 5 predictors for POAG
• Age
• IOP
• CCT
• PSD
• Vertical cup/disc ratio by contour
• The predictive factors in the OHTS model and the
  EGPS model have similar hazard ratios
• All comparisons of hazard ratios by t-test, p
  values gt 0.05
• DAgostino et al., JAMA2001 180-187

28
Multivariate Hazard Ratios for OHTS Observation
group and EGPS Placebo group
HR 95 CI
Age Decade EGPS OHTS
1.37 (1.00, 1.88) 1.16 (0.94, 1.43)
IOP (mm Hg) EGPS
OHTS
1.11 (0.98,1.27) 1.21 (1.11, 1.31)
CCT (40 µm decrease) EGPS OHTS
2.07 (1.49, 2.87) 2.00 (1.59, 2.50)
1.27 (1.04,1.54) 1.26 (1.12, 1.41)
Vertical CD ratio EGPS
by contour
OHTS
1.05 (0.95, 1.16) 1.16 (0.95,1.41)
PSD (per 0.2 dB increase) EGPS

OHTS


29
OHTS prediction model for POAG is confirmed in
EGPS

• Prediction model is validated...
• In an independent European study population
• In ocular hypertensive individualsnot on
  treatment
• Thinner central corneal measurement is confirmed
  as a predictive factor for POAG

30
Next step was to pool OHTS and EGPS data in the
same prediction model

• To increase the sample size to 1,319 participants
  (165 POAG endpoints)
• To tighten 95 confidence intervals for estimates
  of hazard ratios for POAG

31
Multivariate Hazard Ratios OHTS Observation
Group, the EGPS Placebo Group Pooled OHTS and
EGPS dataset
Age Decade EGPS OHTS Pooled
IOP (mm Hg) EGPS

OHTS Pooled
CCT (40 µm decrease) EGPS
OHTS Pooled
Vertical CD Ratio (per 0.1 increase) EGPS

OHTS

Pooled
PSD (per 0.2 dB increase) EGPS

OHTS
Pooled
32
Factors not in the prediction model Heart disease

• In univariate analyses, history of heart disease
  was a significant predictive factor in OHTS but
  not in EGPS
• In multivariate analyses, heart disease was not a
  significant predictive factor in OHTS, EGPS or
  the pooled sample

33
Factors not in the prediction model Diabetes

• History of diabetes reduced the risk of
  developing POAG in the 2002 OHTS prediction model
• The effect of diabetes was difficult to estimate
  in current OHTS models data based solely on
  self-report
• Diabetes was not significant in univariate or
  multivariate EGPS prediction models
• Because of poor statistical estimation, diabetes
  was not included in the final prediction models

34
Which model performs best?

• A model averaging data from both eyes?
• A model using data from the worst eye?
• A model using data from both eyes including
  asymmetry between the eyes?

These models all perform similarly and
correlation coefficients ranging from 0.94
0.98.
35
The OHTS and EGPS pooled data were reanalyzed
using tree analyses to look for predictive
factors that might be missed in Cox model

• Results from tree analyses
• Identified the same 5 predictive factors for POAG
  (Age, IOP, CCT, Vertical C/D, PSD)
• Confirmed that heart disease, diabetes,
  hypertension, myopia and self-identified race had
  no detectable effect on risk of developing POAG

36
How accurate is the OHTS-EGPS prediction model
for POAG?

• The accuracy of prediction models in
  discriminating between patients who do and do not
  develop a disease is measured using the C
  statistic
• C statistic ranges from 0.50 (random agreement)
  to 1.00 (perfect agreement)

37
Accuracy of prediction models for POAG compared
to Framingham Heart Study
DAgostino et al. JAMA, 2001.
38
Comparision of observed vs. predicted 5 year
incidence of POAG for the OHTS-EGPS pooled sample
Decile of Predicted Risk (112 participants per
decile)
39
Using the prediction model

• Available on web free of charge
•  https//ohts.wustl.edu/risk

40
Home Page
41
(No Transcript)
42
(No Transcript)
43
Benefits of risk stratification to clinicians
and patients

• Decide on frequency of visits and tests
• Ascertain the benefit of early treatment
• Potentially reduce medical costs

44
Cost Utility Analysis

• Kymes et. al., reported that it was cost
  effective to treat ocular hypertensive
  individuals with gt 2 per year risk of developing
  POAG
• Kymes et al., AJO, 2006141 997-1008.

45
Benefits of risk stratification

• Approximately 30-40 of the participants in the
  pooled sample have lt1 per year risk of
  developing POAG
• Many of these individuals could be seen and
  tested once a year
• Most of these individuals do not require
  treatment
• Potential cost savings

46

• LIMITATIONS AND CAUTIONS
• There is no guarantee that the predicted risk is
  accurate for a specific patient.
• The predictions are more likely to be accurate
  for patients who are similar to the patients
  studied in the OHTS and the EGPS, and if your
  testing protocols for your patients resemble
  those used in the studies.
• The model predicts the development of early POAG.
  It is not clear whether the model also predicts
  progression of established disease or the
  development of visual disability.
• The model is based on baseline parameters.
  Changes during follow-up will alter the risk of
  developing POAG.

47
Limitations and Cautions Application of
prediction models to individual patients must
include information outside the model

• THE PREDICTIONS ARE DESIGNED TO AID BUT NOT TO
  REPLACE CLINICAL JUDGMENT.
• Need to consider factors such as health status,
  life expectancy and patient preferences
• An 18 year old ocular hypertensive with a low
  5-year risk of developing POAG might be a
  candidate for treatment
• A seriously ill 63 year old ocular hypertensive
  with a high 5-year risk of developing POAG might
  not be a candidate for treatment

48
Summary

• 5 baseline factors accurately stratify ocular
  hypertensive individuals by their risk for
  developing POAG
• Age
• IOP
• Central corneal thickness
• PSD
• Vertical cup/disc ratio by contour

49
Summary

• OHTS prediction model for POAG has demonstrated
  high external validity
• OHTS model validated in EGPS sample and
  Diagnostic Innovations in Glaucoma Study sample
  (Medeiros FA, et al., Archives of Ophthalmology,
  2005.)
• Model accurately predicts development of POAG in
  ocular hypertensive individuals not on treatment.
• Predictive model is accurate in self-identified
  whites and African Americans

50
Next Steps

• Clarify the effects of diabetes, cardiovascular
  disease, ethnic origin, myopia and family history
  of glaucoma on the risk of developing POAG
• Test the generalizability of the predictive model
  in other populations
• Add new diagnostic technology
• Quantitative assessments of disc and nerve fiber
  layer parameters
• Psychophysical tests
• Identify new predictive factors
• Diet
• Environmental exposures
• Genetic factors
• Predictive models will evolve with new
  information

51
Collaborative Group

• Ocular Hypertension Treatment Study
• Mae Gordon
• Michael Kass
• Phil Miller
• Julie Beiser
• Feng Gao
• Ralph DAgostino
• Consulting Statistician, Boston University

• European Glaucoma Prevention Study
• Valter Torri
• Stefano Miglior
• Irene Floriani
• Davide Poli
• Ingrid Adamsons

52
OHTS Clinical Centers

• Bascom Palmer Eye Institute
• Eye Consultants of Atlanta
• Eye Physicians and Surgeons
• Cullen Eye Institute
• Devers Eye Institute
• Emory Eye Institute
• Henry Ford Hospitals
• Johns Hopkins University
• Krieger Eye Institute
• Howard University
• University of Maryland
• University of California, Los Angeles
• Charles Drew University
• Kellogg Eye Center
• Kresge Eye Institute
• Great Lakes Eye Institute
• University of Louisville

• Mayo Clinic
• New York Eye Ear Infirmary
• Ohio State University
• Ophthalmic Surgeons Consultants
• Pennsylvania College of Optometry
• MCP/Hahnemann University
• Scheie Eye Institute
• Keystone Eye Associates
• University of California, Davis
• University of California, San Diego
• University of California, San Francisco
• University Suburban Health Center
• University of Ophthalmic Consultants
• Washington Eye Physicians Surgeons
• Eye Associates of Washington, DC
• Washington University, St. Louis

53
EGPS Clinical Centers

• Belgium
• University of Antwerpen
• University of Buxelles
• University of Gent
• Germany
• University of Leuven
• University of Mainz
• University of Freiburg
• University of Heidelberg
• University of Wuerzburg
• Portugal
• Coimbra, AIBILI
• Viseu, S. Teotonio Hospital
• Lisbon, S. Jose Hospital

• Italy
• University of Milan, S. Paolo Hospital
• University of Milan, L. Sacco Hospital
• University of Verona
• University of Parma
• Oftalmico Hospital, Rome
• S. Giovanni Hospital, Rome
• Fatebenefratelli Hospital, Rome